Enhancing CDK4/6 inhibitor therapy for medulloblastoma using nanoparticle delivery and scRNA-seq-guided combination with sapanisertib.

The therapeutic potential of CDK4/6 inhibitors for brain tumors has been limited by recurrence. To address recurrence, we tested a nanoparticle formulation of CDK4/6 inhibitor palbociclib (POx-Palbo) in mice genetically-engineered to develop SHH-driven medulloblastoma, alone or in combination with specific agents suggested by our analysis. Nanoparticle encapsulation reduced palbociclib toxicity, enabled parenteral administration, improved CNS pharmacokinetics, and extended mouse survival, but recurrence persisted. scRNA-seq identified up-regulation of glutamate transporter Slc1a2 and down-regulation of diverse ribosomal genes in proliferating medulloblastoma cells in POx-Palbo-treated mice, suggesting mTORC1 signaling suppression, subsequently confirmed by decreased 4EBP1 phosphorylation. Combining POx-Palbo with the mTORC1 inhibitor sapanisertib produced mutually enhancing effects and prolonged mouse survival compared to either agent alone, contrasting markedly with other tested drug combinations. Our data show the potential of nanoparticle formulation and scRNA-seq analysis of resistance to improve brain tumor treatment and identify POx-Palbo + Sapanisertib as effective combinatorial therapy for SHH medulloblastoma.

INTRODUCTION

Patients with medulloblastoma, the most common malignant brain tumor in children, need new therapies. The current standard treatment for medulloblastoma with surgery, craniospinal radiation, and chemotherapy cures ~80% of patients but causes disabling untoward effects, including neurocognitive impairment, hearing loss, endocrine dysfunction, and secondary malignancies. Survivors remain at risk of recurrence after treatment, and recurrent medulloblastoma is presently incurable. Therapies that specifically target the biology of the tumor and spare normal tissues may improve the efficacy of medulloblastoma therapy and decrease long-term toxicities.

Cyclin-dependent kinase 4/6 (CDK4/6) inhibitors may be ideal targeted agents for medulloblastoma (, ). While medulloblastomas are a heterogeneous group of tumors with four subgroups, all four groups of medulloblastomas typically disable the Retinoblastoma (RB) tumor suppressor through CDK4/6–mediated RB phosphorylation (pRB) and RB mutations are rare in every subgroup (–). The CDK4/6 inhibitor palbociclib (Ibrance, Pfizer Inc.) effectively blocks pRB and arrests cells with intact RB in the G1 phase of the cell cycle () and is U.S. Food and Drug Administration–approved for specific breast cancers, in combination with hormonal therapy (, ).

Implementing palbociclib for medulloblastoma therapy may require both improving central nervous system (CNS) pharmacokinetics (PK) and identifying mutually enhancing drug combinations. The brain penetration of palbociclib is limited (), and dose-limiting toxicities restrict the potential for increasing systemic doses to improve CNS drug delivery. In addition, alternative mechanisms can compensate for CDK4/6 activity as cells proliferate effectively in Cdk4/6–deleted mouse embryos (). These mechanisms may support palbociclib resistance in tumors, and combinations of drugs may be needed to block resistance mechanisms (). Consistent with these obstacles, palbociclib efficacy has been limited by recurrence in mouse models of diffuse, intrinsic pontine glioma () and medulloblastoma (, ). Similarly, palbociclib was ineffective as a monotherapy in phase 2 clinical trials for recurrent glioblastoma patients with detectable RB expression (NCT01227434) (). Dose-limiting toxicities may have contributed to poor clinical efficacy, as the tested doses were low relative to doses in preclinical studies and many patients required dose reductions because of neutropenia. At the doses that were tolerable for patients, no pharmacodynamic (PD) effect was noted, and all patients recurred during therapy (). Addressing the problems of CNS drug delivery and mechanisms of resistance may realize the potential of palbociclib for medulloblastoma and, more broadly, for brain tumor therapy.

We have previously shown that poly(2-oxazoline) (POx)–based amphiphilic block copolymer micelles can act as nanoparticle carriers of diverse small molecules for CNS delivery (, ). POx delivery of the Sonic Hedgehog (SHH) inhibitor vismodegib increases brain and tumor drug exposure, reducing systemic toxicity and increasing efficacy (). On the basis of these prior studies, we tested whether POx delivery improved palbociclib. We designed a novel POx block copolymer to effectively load palbociclib and compared the resulting formulation (POx-Palbo) to conventional palbociclib in the treatment of SHH-driven medulloblastomas that developed spontaneously in Smo-mutant mice, analyzing PK, PD, toxicity, and efficacy. To address medulloblastoma recurrence during therapy, we investigated the cell cycle progression and single-cell gene expression profiles of proliferating tumor cells in POx-Palbo–treated medulloblastomas. We then tested combinations of POx-Palbo with specific agents suggested by our analyses. Our results show that POx delivery improved the CNS PK and efficacy of palbociclib, identify palbociclib-induced changes in gene expression in the tumor cells that proliferated during therapy, and show that combining POx-Palbo with the mammalian target of rapamycin (mTOR) inhibitor sapanisertib markedly increased efficacy compared to either agent alone.

RESULTS

Limited efficacy of conventional palbociclib in medulloblastoma suggests the need for nanoparticle drug delivery to the CNS

We found that conventional palbociclib (Palbo-HCl) was ineffective in a genetic model of aggressive, refractory SHH medulloblastoma, indicating the need for optimization. We generated mice with SHH medulloblastoma by breeding hGFAP-Cre mice that express Cre recombinase in CNS stem cells during development with SmoM2 mice that express a Cre-conditional oncogenic allele of Smo. The resulting G-Smo mice developed medulloblastoma with 100% penetrance by postnatal day 10 (P10) and, untreated, died of progressive tumors by P20, as in prior studies (, ). We determined the maximum tolerated doses (MTDs) for healthy wild-type (WT) mouse pups treated daily with oral Palbo-HCl starting at P10 or parenteral Palbo-HCl administered intraperitoneally daily from P10 to P14 and then every other day (Fig. 1A). We then treated G-Smo mice starting at P10 with similar regimens of Palbo-HCl. Compared to saline-injected controls, Palbo-HCl produced no statistically significant increase in the event-free survival of G-Smo mice when administered orally at either the MTD of 100 mg/kg per day or 50% of the MTD or when administered parenterally at the MTD of 10 mg/kg per day (Fig. 1B). Considering the poor efficacy of Palbo-HCl in our model, we determined whether nanoparticle drug delivery might improve drug performance.

Fig. 1.

Limited tolerability and efficacy of conventional palbociclib.

(A) MTD of Palbo-HCl following oral or systemic administration. Palbo-HCl was administered daily at the indicated doses starting at P10, and mice were weighed each day. The gray range indicates the mean weights ± SEM of uninjected littermate controls. The MTD was defined as the highest dose that did not reduce weight gain by >15%. (B) Kaplan-Meir curve of G-Smo mice treated with the indicated regimens. Oral Palbo-HCl was dosed daily starting on P10. Palbo-HCl IP was administered on postnatal days 10-14 (P10-14) and then every other day.

To develop a nanoparticle formulation, we attempted to load palbociclib into the POx copolymer, which we previously used to formulate vismodegib (POx-A) (). This amphiphilic, noncharged A-B-A type copolymer, [P[MeOx34-b-BuOx20-b-MeOx35] (Mn = 8.6 kg/mol)] showed very low loading capacity (Fig. 2A). Noting that palbociclib is a pyridopyrimidine that is hydrophobic and forms hydrogen bonds due to its secondary piperazine nitrogen [pKa (where Ka is the acid dissociation constant), 7.4] and pyridine nitrogen (pKa, 3.9) groups, which act as hydrogen donor/acceptors, we designed new POx block copolymers predicted to have stronger polymer-drug interactions. We synthesized three different copolymers with carboxylic groups on the side chains and varying hydrophobicity: POx-B, A-C-A type anionic copolymer (P[MeOx38-b-CEtOx27-b-MeOx38], Mn = 10.6 kg/mol) that does not contain hydrophobic block; POx-C, an A-B-C copolymer (P[MeOx33-b-BuOx21-b-CEtOx29], Mn = 9.9 kg/mol) with hydrophilic, hydrophobic, and anionic blocks; and POx-D, A-B-C copolymer (P[P[EtOx34-b-BuOx21-b-CEtOx31], Mn = 10.7 kg/mol), a more hydrophobic analog of POx-C due to methyl for ethyl side-chain substitution in the hydrophilic block (see fig. S1). We then assessed the ability of each polymer to encapsulate palbociclib using the thin-film method.

Fig. 2.

Specific block copolymers improve the loading of palbociclib into polyoxazoline micelle nanoparticles.

(A) Structures of POx triblock copolymers. The structures were confirmed by 1H NMR spectrum (in MeOD) by identifying NMR peaks. Detailed information on polymer structure is in fig. S1. (B) Particle size, size distribution, and loading parameters (LC%, LE%) of palbociclib-loaded micelles prepared from indicated polymers. Loading capacity (%) = Mdrug/(Mdrug + Mexcipient) × 100%, Loading efficiency (%) = Mdrug/(Mdrug added) × 100% (n = 3 ± SD). (C) Particle size distribution (z-average, Dz, by DLS), zeta potential, and morphology [by TEM; scale bars, 500 nm (left), 10 nm (right)]. (D) Palbociclib release profile from POx-Palbo incubated in 10% fetal bovine serum (FBS) solution at 37°C over time. Palbo-HCl was used as penetration control.

Introducing carboxylic groups alone, as in POx-B, was not sufficient to improve drug loading. Adding hydrophobic blocks, as in POx-C and POx-D, effectively improved drug loading to ~28% loading capacity. The further increase in hydrophobicity in POx-D compared to POx-C reduced the particle size and size distribution (Fig. 2B). The POx-D-palbociclib micelles were small, spherical particles with mean size of 55 nm and narrow size distribution [polydispersity index (PDI) <0.1], as determined by dynamic light scattering (DLS) and transmission electron microscopy (TEM) (Fig. 2C). We selected the POx-D formulation as optimal on the basis of loading capacity and particle size and designated it as POx-Palbo for further studies. In vitro release studies of POx-Palbo showed a sustained release profile without burst release, with approximately 50% of palbociclib released in 3 hours and 100% released within 24 hours (Fig. 2D).

POx micelle formulation reduced systemic toxicity after parenteral administration

We compared the tolerability of POx-Palbo versus Palbo-HCl in MTD studies. We injected healthy, non–tumor-bearing pups intraperitoneally with escalating doses of POx-Palbo daily on P10 to P14 and then every other day until P26, as in the parenteral Palbo-HCl regimen and analyzed toxicity, defined as 15% less weight gain compared to saline-treated littermate controls. The MTD for POx-Palbo was 50 mg/kg (Fig. 3A), markedly higher than the MTD of 10 mg/kg for Palbo-HCl. POx-Palbo therefore showed superior tolerability in parenteral administration.

Fig. 3.

POx-Palbo reduces toxicity.

(A) The weights of mice treated with the indicated formulations over time. The gray range indicates the mean weights ± SEM of littermate controls (*P < 0.05). ns, not significant. (B) Complete blood count and clinical chemistry parameters and (C) hematoxylin and eosin (H&E) staining of C57BL/6 mice treated with saline, Palbo-HCl, or POx-Palbo (equivalent palbociclib, 25 mg/kg). The square brackets indicate the fibrosis, and arrows indicate neutrophils in both the low-power and high-power micrographs. Mice were treated intraperitoneally daily on P10 to P14 and then every other day until P24, and blood samples and tissues were collected 24 hours after the last injection (P25) (n = 3 ± SD). RBC, red blood cells (1012/liter); WBC, white blood cells (109/liter); PLT, platelets (109/liter); Neutro, neutrophils (% in WBC); Mon, monocytes (% in WBC); Lym, lymphocytes (% in WBC); ALT, alanine aminotransferase (U/liter); AST, aspartate aminotransferase (U/liter); T.P., total protein (g/dl); BUN, blood urea nitrogen (mg/dl); Cr, creatinine (mg/dl).

To identify toxicities, we treated mice with Palbo-HCl at 25 mg/kg, the highest dose that did not produce fatal toxicity, or with an equivalent dose of POx-Palbo, using the regimen of daily administration on P10 to P14 and then every other day until P24. We found no notable changes in complete blood counts or serum levels of liver enzymes, total protein, albumin, blood urea nitrogen, and creatinine compared to saline-treated controls (Fig. 3B). Histologic examination of the heart, thymus, lung, liver, spleen, and kidney demonstrated fibrosis of the peritoneal organs in all Palbo-HCl–treated mice, with neutrophilic infiltrates, which was not seen in POx-Palbo–treated mice (Fig. 3C). The neutrophils and fibrosis indicated acute and subacute peritoneal inflammation that likely explains the poor tolerability of intraperitoneal Palbo-HCl. While intraperitoneal tolerability is not relevant to clinical implementation, the absence of peritoneal inflammation with POx-Palbo treatment provides important evidence that the drug remained in the nanoparticle carrier until after systemic uptake.

POx micelles improved palbociclib delivery to the brain and brain tumors

We compared the PK of Palbo-HCl and POx-Palbo in the blood, forebrain, and medulloblastomas of G-Smo mice. For these comparisons, we administered each formulation at a constant palbociclib dose of 25 mg/kg, selected as the highest intraperitoneal dose of Palbo-HCl that did not produce fatal toxicity. We incorporated tritiated palbociclib into POx-Palbo, and Palbo-HCl formulations administered these tritium-labeled agents to groups of replicate G-Smo mice on P10. We then harvested the mice at successive time points after administration and collected plasma and tissue samples. We measured palbociclib concentrations by scintillation counting and analyzed the results using Phoenix modeling software.

Palbo-HCl showed shorter times to peak drug concentrations (Tmax) in plasma and all tissues sampled and higher volume of distribution [8.45 ml compared to 3.07 ml for POx-Palbo (P < 0.05)]. While peak concentration (Cmax) in plasma was comparable between the formulations, in tumors, the Cmax of POx-Palbo was 75% higher compared to Palbo-HCl (5.45 μg/g compared to 3.28 μg/g), and the POx-Palbo area under the curve (AUC) was nearly twofold greater (P < 0.05) (Fig. 4 A and B). Palbo-HCl produced higher concentrations in the kidneys, consistent with renal clearance, while POx-Palbo reached higher peak concentrations in the liver (table S1), consistent with typical nanoparticle clearance in the hepatobiliary system (). Compartment modeling highlighted the differences between the free drug and micellar formulation. Palbo-HCl fit a two-compartment model, while the POx-Palbo micelles better fit a one-compartment model, suggesting reduced exposure to unintended tissues.

Fig. 4.

PK profile of POx-Palbo and Palbo-HCl in medulloblastoma-bearing G-Smo mice.

(A) Plots of palbociclib concentration in serum and major organs (tumor, forebrain, liver, kidney, and spleen over time). ID, injected dose. (B) PK parameters of given formulations in serum, tumor, and forebrain (n = 3 ± SD). Statistical analysis was performed with t tests,; *P < 0.05, **P < 0.01, and ***P < 0.001.

POx-Palbo shows a significant antitumor effect, consistently limited by recurrence

To determine whether the improved PK of POx-Palbo correlated with improved efficacy in vivo, we compared the event-free survival times of G-Smo mice treated with regimens of 50 mg/kg intraperitoneally (ip) daily (the MTD), 25 mg/kg ip daily (50% of MTD), or 25 mg/kg ip daily for 7 days, followed by 12.5 mg/kg thereafter (table S2). The lower-dose POx-Palbo regimens were used to determine whether less intense, less toxic doses would differently influence animal survival. All POx-Palbo regimens showed similar efficacy, with increased animal survival compared to G-Smo treated with saline as sham controls or treated with Palbo-HCl at the MTD of 10 mg/kg ip daily (Fig. 5A). We compared sections of replicate POx-Palbo–treated and control tumors at P11, 24 hours after a single dose, or P15, 5 days into treatment. At both P11 and P15, POx-Palbo–treated tumors were consistently smaller than age-matched controls; however, POx-Palbo–treated tumors were consistently larger at P15 than at P11 (Fig. 5B), indicating continued growth during POx-Palbo treatment. This continued growth in POx-Palbo–treated tumors was consistent with their eventual recurrence in survival studies of Fig. 5A. To gain insight into the mechanisms of growth during palbociclib therapy, we analyzed POx-Palbo PD and determined whether the PD changed over the course of therapy.

Fig. 5.

PD analysis of POx-Palbo in medulloblastoma-bearing G-Smo mice.

(A) Kaplan-Meier curve of POx-Palbo treatments (#1 to #3) and Palbo-HCl in G-Smo mice. (B) H&E of POx-Palbo–treated brains at 24 hours (P11) and five daily treatments (P15) compared to saline controls. (C) Two-dimensional flow cytometry analysis of DNA content (x axis) and pRB content (y axis) in POx-Palbo–treated (2 to 24 hours) G-Smo mice, quantifying pRB++ cells at G1. (D) Quantification of pRB++ cells in G1 (red square) in POx-Palbo–treated mice. P values were determined using one-way analysis of variance (ANOVA). (E) Uniform manifold approximation and projection (UMAP) qualitative map of tumor cells dissociated from G-Smo mice. (F) Cell cycle analysis of tumor cells in G-Smo mice at indicated intervals after treatment with POx-Palbo (25 mg/kg). P values were determined by comparisons of each cell cycle phase at each time point versus saline control using one-way ANOVA. (G) Quantification of pRB++ cells at G1 in POx-Palbo (25 mg/kg)–treated G-Smo mice after 5 days. Statistical analysis was performed with one-way ANOVA. (H) Longitudinal analysis of BLI G-Smo mice with indicated treatment, quantified in the right panel. *Mouse died before 72-hour imaging. In (A, D, F, and G), *P < 0.05, **P < 0.01, ***P < 0.001, and ****P < 0.0001. A.U., arbitrary units.

POx-Palbo PD remain stable as efficacy decreases

We determined PD after in vivo POx-Palbo treatment by quantifying pRB and cell cycle progression. We administered POx-Palbo intraperitoneally to G-Smo mice and then harvested tumors at defined intervals after drug administration, injecting 5-Ethynyl-2′-deoxyuridine (EdU) intraperitoneally 30 min before harvest to label cells at S phase. We then dissociated the tumors; fixed and stained the cells for pRB, EdU, and DNA content; and used flow cytometry to quantify cells at each phase of the cell cycle. We identified G0 as pRB− cells with 2 N DNA content, G1 as pRB+/EdU− cells with 2 N DNA content, S phase as EdU+/pRB+ cells, and G2-M as EdU− cells with 4 N DNA content (fig. S2). The fraction of pRB+ cells with 2 N DNA content, which included cells in G1 and early S phases, was markedly reduced in POx-Palbo–treated tumors at both 2 and 6 hours after administration (Fig. 5, C and D), demonstrating effective inhibition CDK4/6, as expected. By 24 hours, pRB+ cells with 2 N DNA were significantly lower compared to controls but significantly higher compared to 6 hours after administration, indicating waning CDK4/6 inhibition (Fig. 5D). Increasing the dose of POx-Palbo from 25 to 50 mg/kg did not increase pRB suppression (Fig. 5D), indicating that the dose of 25 mg/kg saturated the capacity of the system to respond.

In contrast to pRB, cell cycle progression decreased in the hours after POx-Palbo administration but more than fully resumed by 24 hours (Fig. 5, E and F). POx-Palbo–treated tumors showed significantly reduced cells at G1 at 2 hours after injection and reduced cells at S phase at 6 hours after injection, indicating that fewer cells proceeded from G0 to S phase (Fig. 5F). However, by 24 hours, POx-Palbo–treated tumors showed significantly increased G1, S, and G2-M cells, indicating a compensatory increase in cell cycling (Fig. 5F). While cell cycle progression resumed more rapidly than pRB, the suppression on both processes decreased more rapidly than tumor drug concentration, and the waning PD therefore could not be explained by PK alone.

To determine whether medulloblastoma sensitivity to palbociclib was affected by repeated exposure, we treated mice with daily POx-Palbo (25 mg/kg ip) for 5 days and then harvested replicate tumors at successive intervals after the fifth dose and analyzed PD using our flow cytometry methods. We found that medulloblastomas treated for 5 days with daily POx-Palbo showed similar temporal patterns of pRB suppression after the fifth dose, indicating that the magnitude and duration of CDK4/6 inhibition were not changed by repeated exposure over this period (Fig. 5G).

While repeated doses of POx-Palbo continued to inhibit CDK46, longitudinal studies of tumor growth during therapy showed diminishing effects over the same time period. We measured tumor growth noninvasively in vivo by breeding Gli-luc transgenic mice that report SHH activity through a GLI-sensitive luciferase reporter () into the G-Smo model as in our prior studies (). The resulting G-Smo mice developed medulloblastomas detectable by bioluminescence imaging (BLI), with BLI signal proportional to both the degree of SHH activation and the size of the tumor. BLI signal in untreated G-Smo mice increased progressively from P10 to P14 (Fig. 5H, control), indicating growth of SHH-activated tumors. Daily treatment of replicate G-Smo mice with POx-Palbo (25 mg/kg ip) produced a decrease in mean BLI, indicating initial tumor suppression. This decrease was detectable at 48 hours, later than the decreased tumor size noted in hematoxylin and eosin (H&E) sections (Fig. 5B), and this delay may result from integrating changes in both SHH activation per cell and the number of tumor cells. By day 4, however, the tumor-suppressive effect of POx-Palbo clearly diminished as BLI signal began increasing while on daily therapy (Fig. 5H). Tumor growth therefore resumed during the period of stable PD, indicating that the recurrence did not require loss of CDK4/6 inhibition or resistance to pRB suppression but rather a dissociation between pRB suppression and growth suppression. To define the processes that allow medulloblastoma cells to proliferate during ongoing palbociclib treatment, we compared the transcriptomic profiles of proliferative medulloblasted cells in untreated G-Smo tumors versus tumors in G-Smo mice on POx-Palbo therapy using single-cell RNA sequencing (scRNA-seq).

scRNA-seq studies show that proliferative tumor cells in palbociclib-treated tumors down-regulate ribosomal genes

We used Drop-seq () methods as in our prior studies (, ) to compare tumors isolated from three P15 G-Smo mice treated with POx-Palbo daily (25 mg/kg ip) for 5 days to tumors from five age-matched, untreated G-Smo mice. Briefly, tumors were harvested and dissociated, and individual cells were paired with bar-coded, oligo dT–coated beads using microfluidic methods, and bar-coded cDNA was synthesized on the beads. We then prepared and sequenced amplified libraries from the pooled cDNA and used cell-specific barcodes to identify the transcriptomes of individual cells in the resulting sequence data. We filtered cells identified by bead-specific barcodes to address unintentional cell-cell multiplexing and premature cell lysis (, ). A total of 1530 of 4959 cells from POx-Palbo–treated tumors and 8699 of 16,489 cells from control tumors met the criteria and were included in the analysis. To compare the two groups at similar sequencing depths, we randomly down-sampled the control transcript counts to 20% of the original depth ().

We performed an integrated analysis of scRNA-seq data from POx-Palbo–treated and control tumors using principal components analysis (PCA) as in our prior studies () and generated a two-dimensional uniform manifold approximation and projection (UMAP), in which individual cells were placed according to their similarities, forming clusters of mutually similar cells (Fig. 6A). As in our prior medulloblastoma scRNA-seq studies, the UMAP showed both discrete clusters and a set of clusters with shared borders. We identified each cluster using markers of proliferation and cerebellar granule neuron progenitor (CGNP) lineage to identify tumor cells and using markers of different types of stromal cells expected in the brain, as validated in our prior medulloblastoma scRNA-seq studies (, ). These markers showed that each discrete cluster comprised a different type of stromal cell, including astrocytes, oligodendrocytes, myeloid cells, endothelial cells, and fibroblasts (Fig. 6, A and B, and Table 1). Marker analysis identified the six clusters with shared borders as medulloblastoma cells in a range of differentiation states with opposite proliferative and differentiated poles. Proliferation markers Mki67 and Pcna and the SHH-driven transcription factor Gli1 identified the cells of clusters 0, 1, 3, and 5 as proliferative; similarly, intermediate differentiation markers NeuroD1 and Cntn2 and late differentiation markers Grin2b and Gria2 placed the cells of clusters 2 and 4 in successive states of differentiation (Fig. 6, C and D).

Fig. 6.

scRNA-seq shows key differences in the composition of POx-Palbo–treated tumors, including a population of proliferative medulloblastoma cells not found in control tumors, marked by up-regulation of Slc1a2 and down-regulation of mTOR complex 1 activity.

(A) UMAP of cells from POx-Palbo–treated and control tumors, grouped by transcriptomic similarities into color-coded clusters. (B) Expression of indicated markers color-coded onto the UMAP from (A), with stromal cluster identities indicated. (C) Expression of proliferation and differentiation markers is color-coded onto the UMAP from (A). The arrow indicates the overall direction of progression from the proliferative to the differentiated pole. (D) Dot plot shows the magnitude and frequency of the expression of indicated proliferation and differentiation markers in the indicated tumor cell clusters. (E and F) UMAP from (A), disaggregated by condition to show (E) cells from control tumors and (F) cells from POx-Palbo–treated tumors. (G) Representative sections of control and POx-Palbo–treated G-Smo tumors, immunostained for SLC1A2, with quantification. (H) Representative sections of control and POx-Palbo–treated G-Smo tumors, immunostained for phosphorylated 4EBP1 (p4EBP1), with quantification. The P values in (G and H) were determined by Student’s t test. Scale bars, 100 μm. DAPI, 4′,6-diamidino-2-phenylindole.

Table 1.

Cluster identities, markers, and population differences by condition.

Cluster no.	Best marker	Designation	Population differences
0	Rpl37, Barhl1	Proliferative tumor cells, high ribosomal genes	Depleted in POx-Palbo–treatedP = 0.024
1	Mik67, Barhl1	Proliferative tumor cells
2	Tubb3, Barhl1	Early differentiating tumor cells
3	Ccnb1, Barhl1	Mitotic tumor cells
4	Neurod1, Meg3	Late differentiating tumor cells and neurons	Enriched in POx-Palbo–treatedP = 0.023
5	Slc1a2, Barhl1	Proliferating tumor cells	Enriched in POx-Palbo–treatedP = 0.024
6	Olig1	Early oligodendrocytes
7	C1qa	Myeloid
8	Aqp4	Astrocytes
9	Cldn5	Endothelial
10	Dcn	Fibroblasts
11	Mag, Mog	Late oligodendrocytes

Disaggregating the data by treatment group showed that medulloblastoma cells were not evenly distributed in the tumor cell clusters (Fig. 6, E and F). Palbociclib-treated tumors contained significantly larger fractions of cells in cluster 4 (P = 0.023), the most differentiated cluster. Proliferative medulloblastoma cells were also differentially distributed across clusters. Cluster 5 specifically comprised palbociclib-treated cells (P = 0.024, Wilcoxon rank sum test), and cluster 0 comprised predominantly control cells (P = 0.024), while clusters 1 and 3 did not show statistical significance between treatment groups. POx-Palbo–treated tumors thus contained both more differentiated tumor cells, consistent with drug-induced growth suppression, and a distinct population of proliferating cells with a transcriptomic profile that was different from proliferating cells in control tumors.

We considered that the gene expression patterns that distinguished cluster 5 cells from the other proliferative cells may include both potentially growth-suppressive mechanisms that failed to block proliferation and resistance mechanisms that allowed proliferation to continue on therapy. To identify these patterns, we determined which genes were differentially expressed in cluster 5 cells, compared to the cells of clusters 0, 1, and 3, and then filtered to include only the set of genes that was expressed by a twofold or greater proportion of cluster 5 cells or the set expressed by a twofold or greater proportion of cells of clusters 0, 1, and 3 (data file S1). We subjected both sets to Gene Ontology (GO) enrichment analysis to identify biological processes discernably affected.

GO analysis of the set of genes enriched in clusters 0, 1, and 3 identified translation as the most strongly increased process (P = 1.2 × 10−39), with diverse biosynthetic processes also increased. GO analysis of the set of genes enriched in cluster 5 cells identified less specific terms, including “positive regulation of biologic process” and “positive regulation of metabolic process” as the most significantly increased (P = 6.5 × 10−9 and P = 6.8 × 10−9, respectively), with chromatin organization also significantly increased. Consistent with increased translation in clusters 0, 1, and 3, we noted increased expression of diverse ribosomal genes and Eef1b2 (data file S1). Consistent with metabolic and chromatin alterations induced by palbociclib, we noted increased expression of the glutamate transporter Slc1a2 and the chromatin modifier Smarca4 in cluster 5 cells (data file S1). We confirmed increased protein expression of solute carrier family 1 member 2 (SLC1A2) [aka glutamate transporter 1 (GLT1)] in replicate POx-Palbo–treated and control tumors using immunohistochemistry (Fig. 6G).

Signaling through mTOR complex 1 (mTORC1) regulates ribosomal gene transcription as part of a general regulation of translation (), and we therefore analyzed whether mTORC1 activity was reduced in cells that remained proliferative during chronic palbociclib treatment. Inhibition of mTORC1 has been shown to increase SLC1A2 expression in other cell types (), which also suggested reduced mTORC1 activity in POx-Palbo–treated medulloblastomas. To investigate mTORC1 activation, we analyzed phosphorylated EUKARYOTIC INITIATION FACTOR 4E BINDING PROTEIN (4EBP1) (p4EBP1) as a marker of mTORC1 activity. We compared p4EBP1 in replicate G-Smo mice treated for 5 days with POx-Palbo versus untreated G-Smo controls using immunohistochemistry. We found significantly fewer p4EBP1+ cells in POx-Palbo–treated tumors, indicating reduced mTORC1 activation, consistent with decrease in ribosomal genes in the scRNA-seq data (Fig. 6H).

These data show reduced mTORC1 activity and increased SLC1A2 protein expression in palbociclib-treated medulloblastomas, validating our scRNA-seq data. These correlations do not show whether reduced mTORC1 activity or increased SLC1A2 are direct effects of the drug or secondary effects of pRB suppression or whether the effects are mechanisms of resistance that allow continued proliferation. We hypothesized, however, that whether reduced mTORC1 activity was growth suppressive or growth enabling, the mTORC1 activity in POx-Palbo–treated tumors might already be at the lower limit of tolerability for proliferating cells, resulting in increased sensitivity to small-molecule mTORC1 inhibitors. We therefore tested whether combining POx-Palbo with the mTORC1 inhibitor sapanisertib, which, similar to palbociclib, shows medulloblastoma efficacy limited by recurrence (), would produce an efficacy greater than either agent alone.

Testing combination therapy with POx-Palbo plus multiple agents validated POx-(Palbo+Sapanisertib) and failed to show benefits of other combinations

In parallel with the palbociclib-sapanisertib combination, we tested the potential of agents targeting SHH signaling or DNA replication to limit resistance when combined with palbociclib. We theorized that vismodegib, which inhibits the SHH receptor component smoothened (SMO), would combine favorably with palbociclib by targeting SHH-driven proliferation at two distinct points, SMO and CDK4/6. We further theorized that replication-targeted agents gemcitabine or etoposide would enhance palbociclib efficacy by disrupting tumor cells that progress to S phase despite CDK4/6 inhibition and, reciprocally, that the compensatory increase in cycling cells 24 hours after palbociclib would increase gemcitabine and etoposide sensitivity. We tested these hypotheses by developing regimens for each agent and then treating cohorts of G-Smo mice with each agent, either individually or in combination with POx-Palbo.

To facilitate administration of vismodegib, etoposide, and sapanisertib, which were not water soluble, we developed POx-based micellar formulations. The use of POx formulations was supported by our prior findings that vismodegib formulated in polyoxazoline nanoparticles (POx-Vismo) was more effective than conventional vismodegib when administered to G-Smo mice () and that etoposide loaded into POx micelles (POx-Etop) was less toxic and more effective than free drug in mouse lung cancer models (, ). We characterized these formulations by size, size distribution, and drug loading (fig. S3A). For gemcitabine studies, we used a conventional rather than POx formulation, as gemcitabine is water soluble and did not load well into POx micelles. We confirmed that gemcitabine reached the brain in bioactive concentrations after intraperitoneal administration by injecting 50 mg/kg ip in P7 mice and then detecting DNA damage in brain progenitors using immunofluorescence for pH2AX (fig. S4). The dose of each single drug was determined on the basis of MTD evaluations in vivo in healthy mice without tumors. These studies identified gemcitabine (50 mg/kg) every 3 days as the MTD (fig. S3B). The MTD for POx-Etop was 5 mg/kg every 5 days (fig. S3C); however, at this dose, mice showed marginally acceptable weight gain, and we therefore used 2.5 mg/kg (50% of the MTD) for tumor treatment. The MTD for formulation of sapanisertib loaded into POx micelles (POx-Sapanisertib) was 0.2 mg/kg daily (fig. S3D); as with etoposide, we used 0.1 mg/kg dosing because of marginal weight gain at the MTD. For combinational studies, we coencapsulated the pairs of drugs in single-nanoparticle formulations. The drug ratios were varied to evaluate the effects of drug ratios on treatment outcomes (fig. S5). The drug release profile from each formulation was investigated (fig. S6). All single drug loaded POx micelles showed similar sustained drug release profile, with 50% drug released at 3 hours (fig. S6A). Coloading with palbociclib did not affect the release profiles of palbociclib, etoposide, and sapanisertib (fig. S6, B and C); however, the release of vismodegib from POx-(Palbo+Vismodegib) was significantly slower (40% for 5 hours), likely because of additional interactions between vismodegib, palbociclib, and the polymer (fig. S6D) (, ).

To combine vismodegib and palbociclib, we administered POx-(Palbo+Vismo) starting at P10 using three different treatment schedules, including reducing POx-Palbo to 12.5 mg/kg after five doses and increasing vismodegib dose to 100 mg/kg (Table 2 and table S3). We tested these different regimens to identify the most effective, tolerable combination. None of the tested regimens, however, were superior to either single-agent POx-Palbo (Fig. 7A) or single-agent POx-Vismo (). Next, we tested three different schedules of combined gemcitabine and POx-Palbo (Table 2 and table S4); all failed to improve survival compared to POx-Palbo alone (Fig. 7B).

Table 2.

Regimens used in in vivo testing.

QOD, every other day; Q3D, every third day; Q5D, every fifth day.

	Drugs combination
	Palbociclib+Vismodegib		Median survival
#1	25 mg/kg dailystarting on P10	50 mg/kg P10–12daily, then QOD	19.5
#2	25 mg/kg P10–14daily, then QOD	50 mg/kg P10–14daily, then QOD	19
#3	25 mg/kg P10–13daily, 12.5 mg/kgP14–16 daily,then QOD	100 mg/kg P14–16daily, then QOD	21
	Palbociclib+Gemcitabine
#1	25 mg/kg dailystarting on P10	50 mg/kg fromP12 Q3D	16.5
#2	25 mg/kg dailystarting on P10	25 mg/kg fromP14 Q5D	20
#3	25 mg/kg P10–13daily, then12.5 mg/kg daily	25 mg/kg fromP14 Q5D	22
	Palbociclib+Etoposide
#1	25 mg/kg dailystarting on P10	5 mg/kg fromP14 Q5D	20
#2	25 mg/kg P10–14daily, then12.5 mg/kg daily	2.5 mg/kg fromP10 Q5D	19
#3	25 mg/kg dailystarting on P10	2.5 mg/kg fromP10 Q5D	24
#4	25 mg/kg dailystarting on P10	2.5 mg/kg fromP14 Q5D	21
	Palbociclib+Sapanisertib
#1	25 mg/kg dailystarting on P10	0.1 mg/kg daily	19.5
#2	25 mg/kg P10–13daily, then12.5 mg/kg daily	0.1 mg/kg fromP14 daily	24
#3	25 mg/kg P10–13daily, 12.5 mg/kgP14–16 daily,then QOD	0.2 mg/kg P14–16daily, then QOD	>35 days

Fig. 7.

Therapeutic outcomes in medulloblastoma-bearing G-Smo mice.

Kaplan-Meier survival curves for G-Smo mice treated with (A) POx-(Palbo+Vismo), (B) POx-Palbo+Gemcitabine, (C) POx-(Palbo+Etop), and (D) POx-(Palbo+Sapanisertib) in the indicated regimens. The P values are determined by comparing each combination to POx-Palbo single-agent therapy using the log-rank (Mantel-Cox) test. *P < 0.05 and **P < 0.01. Regimens used are described in Table 2.

We tested multiple regimens with varied doses of etoposide and palbociclib (Table 2 and table S5). Treatment of G-Smo mice with single-agent POx-Etop (2.5 mg/kg ip) every 5 days improved survival compared to sham treatment (P = 0.03; Fig. 7C). The regimen of POx-(Palbo+Etop) [palbociclib at 25 mg/kg/etoposide at 2.5 mg/kg) every 5 days starting on P10 with POx-Palbo at 25 mg/kg daily between combined doses improved mouse survival relative to both POx-Palbo alone (P = 0.047) and POx-Etop alone (P = 0.001). Recurrence during therapy, however, remained a consistent limitation (Fig. 7C).

We tested multiple regimens of sapanisertib and palbociclib (Table 2 and table S6). Single-agent POx-Sapanisertib dosed at 0.1 mg/kg ip to G-Smo mice significantly improved survival compared to sham treatment (Fig. 7D). Mice on combined regimen 1 were ill appearing, with reduced spontaneous movement, and we therefore reduced the dose of palbociclib to 12.5 mg/kg after day 4 of treatment in regimens 2 and 3 and further reduced the frequency of administration in regimen 3 to every other day, increasing the sapanisertib component to 0.2 mg/kg to maintain the same sapanisertib dose over time. Regimens 2 and 3 both showed improved survival compared to POx-Palbo alone. Regimen 3 was most effective, superior to both POx-Palbo alone (P = 0.011) and POx-Sapanisertib alone (P = 0.006), and reduced recurrence during therapy from 100 to 40% (Fig. 7D). Combining palbociclib with mTORC1 inhibitor sapanisertib was therefore markedly more effective in reducing recurrence during treatment than all other combinations tested.

Palbociclib and sapanisertib show enhanced PD

PD studies showed that combining palbociclib and sapanisertib enhanced the mechanistic effects of each agent. We compared the temporal patterns of pRB suppression and p4EBP1 suppression after administration of POx-Palbo, POx-Sapanisertib, or POx-(Palbo+Sapanisertib) using flow cytometry and immunohistochemistry (Fig. 8). Six hours after administration, POx-Palbo, POx-Sapanisertib, and POx-(Palbo+Sapanisertib) significantly and similarly decreased pRB; however, pRB suppression by POx-(Palbo+Sapanisertib) was markedly more durable, remaining significantly reduced compared to controls 24 hours after administration and resembling controls by 36 hours (Fig. 8, A to C).

Fig. 8.

PD analysis of POx-(Palbo+Sapanisertib) in medulloblastoma-bearing G-Smo mice.

(A) Quantification of flow cytometry analysis of pRB+ G1 cells in POx-Palbo (red), POx-sapanisertib (blue), and POx-(Palbo+Sapanisertib) (purple) in G-Smo mice at 6 and 24 hours after administration. The P values were determined by one-way ANOVA. (B) Representative pRB immunofluorescence in G-Smo medulloblastomas 24 hours after the indicated treatment. (C) Quantification of pRB immunofluorescence in replicate mice as in (B). (D) Representative p4EBP1 immunofluorescence in G-Smo medulloblastomas 6 or 24 hours after the indicated treatment. (E) Quantification of p4EBP1 immunofluorescence in replicate mice as in (D). The P values were determined by Student’s t test. *P < 0.05, **P < 0.01, ***P < 0.001, and ****P < 0.0001.

Both POx-Sapanisertib and POx-(Palbo+Sapanisertib) significantly reduced p4EBP1 6 hours after administration, but p4EBP1 suppression by POx-(Palbo+Sapanisertib) was more durable and persisted over 36 hours (Fig. 8, D and E). In contrast, 24 hours after administration of single-agent POx-Sapanisertib, mTORC1 suppression completely waned and tumors showed a homeostatic increase in p4EBP1 (Fig. 8E). The enhanced PD effects of palbociclib and sapanisertib, delivered as POx-(Palbo+Sapanisertib), support the increased efficacy of the combination that was predicted by the scRNA-seq data and demonstrated by the survival studies.

DISCUSSION

CDK4/6 inhibitor therapy for brain tumors has been complicated by suboptimal PK and the development of recurrence during therapy. We developed a POx polymeric micelle–based nanoparticle formulation of the CDK4/6 inhibitor palbociclib, POx-Palbo, which allowed parenteral administration and improved CNS PK. Parenteral POx-Palbo improved mouse survival in the G-Smo model of aggressive, refractory SHH–driven medulloblastoma. In contrast, Palbo-HCl failed to improve survival of G-Smo mice with either oral or parenteral Palbo-HCl administration. Recurrent disease, however, consistently limited POx-Palbo efficacy. While PD analysis of G-Smo mice treated for 5 days showed that palbociclib continued to inhibit CDK4/6 and suppress pRB, cell cycle progression resumed by 24 hours after each dose despite reduced pRB, and tumor growth resumed within 5 days of the start of treatment. Analysis of medulloblastomas in POx-Palbo–treated mice using scRNA-seq showed specific transcriptional changes in the tumor cells that remained proliferative during palbociclib treatment, including up-regulation of the glutamate transporter Slc1a2 and suppression of ribosomal genes indicating mTORC1 inhibition. Further decreasing mTORC1 activity by adding the mTORC1 inhibitor sapanisertib to the palbociclib regimen markedly increased antitumor efficacy. Similarly, enhanced efficacy was not observed on combining palbociclib with vismodegib or gemcitabine; etoposide enhanced the efficacy of palbociclib but less effectively than sapanisertib despite several dose optimization attempts. These data show the potential for nanoparticle technology to optimize CNS drug delivery and for transcriptomic analysis with single-cell resolution to identify processes that, when targeted therapeutically, can reduce recurrence.

The reduced ribosomal gene expression pattern that we observed in G-Smo tumors treated with POx-Palbo in vivo matches the patterns of differentially suppressed transcripts previously reported in medulloblastoma cell lines with CDK6 deletion and in Cdk6-deleted medulloblastomas that form in Math1-Cre/SmoM2/Cdk6 mice (). Our data show that this transcriptomic change correlates with decreased mTORC1 activation, demonstrated by reduced p4EBP1. The reduced ribosomal function in Cdk6-deleted SHH medulloblastomas has been shown to sustain proliferation by inducing production of SMO-stimulating lipids (). This mechanism may explain the failure of palbociclib to combine well with the SMO inhibitor vismodegib, as SMO-activating lipids induced by POx-Palbo may antagonize the inhibition of SMO by vismodegib. In contrast, by reducing mTORC1, which allows medulloblastoma cells to proliferate in the context of reduced CDK4/6 activity, palbociclib increased susceptibility to mTORC1 inhibitors. These data show that when mechanisms of resistance involve down-regulation of fundamental biologic processes, such as reduced mTORC1 activity, resistance may be blocked by amplifying rather than inhibiting the resistance mechanism, as in the addition of sapanisertib to palbociclib.

The combination of palbociclib and sapanisertib, suggested by our scRNA-seq and p4EBP1 studies, has been similarly reported to produce enhanced antitumor activity in a mouse model of intrahepatic cholangiocarcinoma (ICC) induced by somatic gene transfer of mutant Akt and Yap (). In these ICC tumors, palbociclib alone produced a transient antitumor effect, followed by resistance mediated by increased Cyclin D1 (CCND1), and addition of sapanisertib reduced CCND1 expression and potentiated cell cycle inhibition. In medulloblastomas, we found that palbociclib resistance did not involve increased Ccnd1 expression. Sapanisertib, however, potentiated pRB suppression, as in ICC. In both G-Smo medulloblastomas and ICC, single-agent therapy with palbociclib reduced p4EBP1, indicating mTORC1 inhibition, and also potentiated p4EBP1 suppression when combined with sapanisertib (). A potential mechanism for the potentiating effect of palbociclib on sapanisertib-mediated mTORC1 inhibition is suggested by the suppression of CDK4-mediated activation of Insulin Receptor Substrate 2 (IRS2) and resulting inactivation of tuberous sclerosis complex 2 (TSC2) (, ). The mutually enhancing effects of palbociclib and sapanisertib in both SHH medulloblastoma and ICC show that this combination may be effective in diverse cancers driven by different oncogenic pathways in different tissues of origin.

For patients with SHH medulloblastoma, radiation plus chemotherapy is a highly effective, with 80 to 90% of patients surviving long term (), but overly toxic, leaving many survivors with lifelong disability (). Future preclinical studies are needed to determine whether POx-Palbo+Sapanisertib may combine well with standard radiation, potentially enabling lower, less toxic doses of radiation without increasing recurrence risk. For patients with recurrent SHH medulloblastoma, for whom there is no standard or effective treatment, POx-Palbo+Sapanisertib should be explored as a potential new approach.

MATERIALS AND METHODS

Materials

All materials for the synthesis of POx block copolymers, methanol, ethanol, and sodium lactate were purchased from Sigma-Aldrich (St. Louis, MO). Water and acetonitrile [high-performance liquid chromatography (HPLC) grade] were purchased from Fisher Scientific (Fairlawn, NJ). All materials for the synthesis of POx block copolymers, including methyl trifluoromethanesulfonate (MeOTf), 2-methyl-2-oxazoline (MeOx), 2-ethyl-2-oxazoline (EtOx), 2-n-butyl-2-oxazoline (BuOx), and 2-methoxycarboxyethyl-2-oxazoline (MestOx) were dried by refluxing over calcium hydride (CaH2) under inert nitrogen gas and subsequently distilled before use. Palbociclib-free base, palbociclib-HCl salt form, vismodegib, etoposide, and gemcitabine were purchased from LC Laboratories (Woburn, MA). Sapanisertib was purchased from MedKoo Biosciences (Morrisville, NC). [3H] Palbociclib was purchased from American Radiolabeled Chemicals (St. Louis, MO). Soluene-350 and Ultima Gold LLC scintillation cocktail were purchased from PerkinElmer Life and Analytical Sciences (Waltham, MA).

Synthesis of POx block copolymers

Triblock copolymers, POx(A) [P[MeOx34-b-BuOx20-b-MeOx35] (Mn = 8.6 kg/mol)], POx(B) [P[MeOx38-b-CEtOx27-b-MeOx38] (Mn = 10.6 kg/mol)], POx(C) [P[MeOx33-b-BuOx21-b-CEtOx29] (Mn = 9.9 kg/mol)], and POx(D) [P[EtOx34-b-BuOx21-b-CEtOx31] (Mn = 10.7 kg/mol)] were synthesized by step-by-step (A-B-C) living cationic ring-opening polymerization (scheme S1). Under dry and inert conditions, 1 equivalent of MeOTf and precalculated equivalent of corresponding block A monomer were dissolved in dry acetonitrile. The mixture was reacted in the microwave for 15 min at 150 W and 130°C. After cooling to room temperature, the monomer for block B was added and reacted again for another 15 min. The procedure was repeated with the monomer for block C, and the polymerization was terminated by addition of 3 equivalent of piperidine and incubating for 1 hour at 50 W and 40°C. To remove the methyl ester group from P(MestOx), each polymer was dissolved in methanol and mixed with 0.1 N NaOH (1 equivalent to methyl ester group) at 90°C for 3 hours. The final solution was dialyzed [Molecular weight cut off (MWCO), 3500] against distilled water for 2 days and freeze-dried.

Proton nuclear magnetic resonance (1H NMR) spectrum was obtained using INOVA 400 and analyzed using MestReNova (11.0) software. The spectra were calibrated using the Deuterated methanol (MeOD) solvent signals (4.78 parts per million). The number-average molecular weight was determined by 1H NMR by calculating the ratio of the initiator and each repeating unit using samples taken upon polymerization of each block and final block copolymer.

Preparation of drug loaded POx micelles

Drug-loaded polymeric micelles were prepared by thin-film hydration method as previously described. Each polymer and drug stock solutions in methanol were mixed together at the predetermined ratios, followed by complete evaporation of methanol under a stream of nitrogen gas. The well-dried thin films were subsequently rehydrated with normal saline and then incubated at room temperature to self-assembly into drug-loaded polymeric micelles. The resulting micelle solutions were centrifuged at 10,000g for 3 min (Sorvall Legend Micro 21R Centrifuge, Thermo Scientific) to remove nonloaded drug. The concentration of drugs in micelles was analyzed by reversed-phase HPLC (Agilent Technologies 1200 series) with a Nucleosil C18, 5-μm column (L × I.D. 250 mm × 4.6 mm). Samples were diluted 20 times in mobile phase (mixture of acetonitrile/water with 0.01% trifluoroacetic acid), and 10 μl of the diluted sample was injected into the HPLC, while the flow rate was 1.0 ml/min and column temperature was 40°C. The retention time of drugs, detection wavelength, and detailed mobile phase were presented in table S7. The drug loading efficiency (LE) and loading capacity (LC) were calculated using following equations

Characterization of drug loaded POx micelles

The particle size, PDI, and zeta potential of drug-loaded POx micelles were measured by photon correlation spectroscopy using a Zetasizer Nano-ZS (Malvern Instruments, Worcestershire, UK). Before measurement, each micelle was diluted to yield 1 mg/ml final polymer concentration, and the average values were calculated from three independent sample measurements. To observe the morphologies of POx micelles, one drop of micelle solution (diluted 100 times using distilled water) was placed on a copper grid/carbon film and stained with negative staining (1% uranyl acetate) before the TEM imaging.

The release profile of drug was determined as described earlier. Briefly, the POx-Palbo or Palbo-HCl were dispersed in saline (palbociclib, 0.1 mg/ml), transferred into floatable Slide-A-Lyzer Mini dialysis device with 3.5 kDa (Thermo Fisher Scientific), and dialyzed against 20 ml of phosphate-buffered saline containing 10% fetal bovine serum (FBS) in compliance with the perfect sink conditions requirements. Four devices were used for each time point. At predetermined time points, the samples were collected and the remaining amount of palbociclib were analyzed by HPLC.

Animals

SmoM2 mice were purchased from the Jackson Laboratory (Bar Harbor, ME, USA). hGFAP-cre mice were provided by E. Anton (University of North Carolina, Chapel Hill, NC, USA). Gli-luc mice, which express luciferase under the GLI promoter, were shared by O. Becher and E. Holland. Mouse genotyping was performed using Cre or SmoM2 primers. All mice were of the species Mus musculus and maintained on a C57BL/6 background over at least five generations. Mice were handled under a protocol approved by the University of North Carolina Institutional Animal Care and Use Committee (protocol 19-098).

Toxicology studies

For MTD studies, we evaluated expected growth of healthy WT C57BL/6 mice treated with escalating doses of the drug under study. Drugs were administered by intraperitoneal injection or oral gavage as indicated to groups of three replicate mice, daily from P10 to P14 and then every other day until P28. We recorded weights daily and included age-matched littermate controls to determine the expected weight gain over time. MTD was defined as the highest dose resulting in less than 15% weight gain compared to the control group.

For additional toxicity studies, we treated healthy mice with Palbo-HCl (25 mg/kg) or POx-Palbo (25 mg/kg) for 15 days (P10 to P14, daily; P15 to P24, every other day) and compared to saline-injected controls. On P25, mice were sacrificed, blood was collected, and comprehensive complete blood counts and blood chemistry panel were performed. Major organs including heart, thymus, lung, liver, spleen, and kidney were harvested; fixed in formalin; and subjected to pathological analysis by H&E staining.

PK analysis

Groups of three to four replicate G-Smo mice were administered via intraperitoneal injection on P10 with a single dose of POx-Palbo or palbociclib-HCl (25 mg/kg). All samples contained [3H] palbociclib (100 μCi/kg). At various time points at 0.083, 2, 4, 8, and 24 hours after injections, three to four G-Smo mice from each group were euthanized, and blood and organs (forebrain, tumor, liver, spleen, and kidney) were collected. Each organ was weighted and homogenized in 1 ml of Soluene-350. Serum (50 μl) from each sample was collected, and 1 ml of a mixture of soluene-350 and isopropyl alcohol at 1:1 ratio was added. A volume of 15-ml Ultima Gold LLC scintillation cocktail (PerkinElmer, USA) was added to each vial, vortexed, and counted in a liquid scintillation counter, TriCarb 4000 for [3H] palbociclib. Noncompartmental analysis (NCA) was performed using the Phoenix WinNonlin software. Parameters derived from the NCA were used as initial estimates in the model building.

Efficacy studies

G-Smo mice were randomized into the indicated treatment groups with 4 to 12 replicate mice per group and administered via oral or intraperitoneal on P10 with each regimen. The indicated formulations were administered until P35, unless mice first developed symptoms of tumor progression, such as hunched posture, ataxia, tremor, and seizures.

PD analysis

P10 to P12 tumor mice were administered drugs [POx-Palbo, Palbo-HCl, POx-Sapanisertib, and POx-(Palbo+Sapanisertib)] via intraperitoneal injections. EdU (40 mg/kg) was administered to mice 30 min before harvest. Tumor tissue was harvested and subjected to cellular dissociation via Worthington Papain Dissociation System kit. Dissociated tumor cells were fixed for 15 min on ice and washed with fluorescence-activated cell sorting wash buffer. Fixed cells were stained with fluorophore markers for DNA (FX Cycle stain; catalog no. F10347, Thermo Fisher Scientific), EdU (Click-it Edu Kit; catalog no. C10337, Thermo Fisher Scientific), and pRB content [phospho-RB (Ser807/811); catalog no. 8974S, Cell Signaling Technology]. Stained cells were prepared in appropriate strained for flow analysis.

Flow cytometry

Stained cells were resuspended in sheath fluid and ran on an LSR II flow cytometer provided by the UNC Flow Cytometry Core. Proper compensation controls were used.

Immunofluorescence imaging

Brains including tumors from G-Smo pups were harvested, fixed in 4% paraformaldehyde for 48 hours, and embedded in paraffin at the UNC Center for Gastrointestinal Biology and Disease Histology core. Sections were deparaffinized, and antigen retrieval was performed using a low-pH citric acid–based buffer. Staining was performed, and stained slides were digitally scanned using the Leica Biosystems Aperio ImageScope software (12.3.3) with assistance from the UNC Translational Pathology Laboratory. The primary antibodies used were anti-pRB (catalog no. 8516, Cell Signaling Technology), anti-p4EBP1 (catalog no. 2855, Cell Signaling Technology), and anti-GLT1 (catalog no. 701988, Thermo Fisher Scientific).

Drop-seq library preparation, sequencing, and analysis

We used Drop-seq () methods for scRNA-seq study as in our prior studies (, ). Briefly, Drop-seq libraries were prepared according to the Drop-seq protocol V 3.1 () with full details available online (http://mccarrolllab.com/dropseq/). Cell and bead concentrations were both set between 95 and 110/μl. Tumor cells were coencapsulated using a Dolomite-brand glass device. All cells were processed within 1 hour of tissue dissociation. Flow rates on the glass device were set to 2400 and 12,000 μl/hour for cells/beads and oil, respectively, with a 1 to 2.5% bead occupancy rate. From the obtained library, the raw sequence data were processed in a Linux environment using Drop-seq Tools V1.13 (https://github.com/broadinstitute/Drop-seq/releases) to generate a digital expression (DGE) matrix. DGE matrices were used to generate Seurat objects in R (https://satijalab.org/seurat/). Input data are raw sequences in Fastq format, demultiplexed by sample identity. We first convert Fastq to BAM/SAM format and merge samples that were sequenced across multiple lanes.

The data were normalized using the SCTransform method as implemented in Seurat. PCA was performed on the 3000 most highly variable using the RunPCA function. We used the FindNeighbors and FindClusters functions to identify cell clusters based on the Louvain algorithm. To identify differential genes between clusters of cells, we used the Wilcoxon rank sum test to compare gene expression of cells within the cluster of interest to all cells outside that cluster, as implemented by the FindMarkers function. UMAP was used to reduce the PCs to two dimensions for data visualization using the RunUMAP function. Biological processes implicated by differential gene expression profiles were identified using GO Enrichment Analysis (–).