Potential protein biomarkers were initially screened using immunohistochemical stains to archived formalin-fixed paraffin-embedded tissue microarrays. Tissue microarrays included a variety of high grade primary pediatric brain tumors including initial diagnosis of medulloblastoma (n = 28), ependymoma (n = 17), high-grade glioma (n = 10), and CNS embryonal tumors, NOS (n = 6). In addition, 13 paired tumors obtained at the time of relapse included 2 medulloblastoma, 8 ependymoma, 1 high-grade glioma and 2 CNS embryonal tumors. Laboratory-developed immunohistochemical testing was performed for the following proteins: EGFR, HER2, CKIT, PDGFRA, pS6, pERK and scored using the methods detailed below. All patients included in these initial screening studies had appropriate consent and IRB approval was obtained for this analysis. Survival analysis was performed using data extracted from the electronic medical record system. The log-rank test was used to evaluate the significance of associations between survival, tumor type, age at diagnosis, extent of resection, and immunohistochemical staining pattern. Stepwise Cox regression was used to conduct multivariable analysis. Hazard of death associated with each biomarker was adjusted for tumor histology, age, and extent of resection.

This study was conducted as a prospective phase 1 clinical trial. FDA risk determination was requested regarding complex laboratory-developed testing, and the immunohistochemical panel was determined to pose a non-significant risk in the context of this study. The study was subsequently approved by the Seattle Children’s Institutional Review Board (IRB). Informed consent, and assent where applicable, was obtained from study participants or their legal guardians according to institutional standards, including IRB-approved process for phone consent for patients not residing in the region.

Eligibility for tumor testing included age less than 30 years, Karnofsky or Lansky performance score $\ge$50%, and documented progression or recurrence of brain tumor by MRI or CSF since completion of last tumor-directed medical therapy, for which there was no known curative therapy. Histologic confirmation of brain tumor was required from either initial diagnosis or relapse. Most recent surgical tissue was preferred and utilized for study treatment selection if discrepant results were found between initial diagnostic and relapse tissue. Additional eligibility criteria to begin study therapy included adequate organ function and recovery from prior therapy. Adequate organ function included bone marrow function defined as absolute neutrophil count $>$750/uL and platelet count $>$75/uL; renal function defined as normal creatinine for age; liver function defined as bilirubin and alanine aminotransferase $\le$1.5 times upper limit of normal; and neurologic function defined as well-controlled seizures on non-enzyme inducing antiepileptics, and no increase in steroid dose within past seven days. Adequate recovery from prior therapy was considered three weeks from surgery or myelosuppressive chemotherapy; seven days from hematopoietic growth factors or biologic therapy; 12 weeks from craniospinal radiation and 2 weeks from focal radiation therapy. Exclusion criteria consisted of pregnancy, breastfeeding, uncontrolled infection, within one year of allogeneic transplant, bleeding disorder or more than punctate intratumoral hemorrhage, other anti-neoplastic agents, immunosuppressive agents, or strong CYP3A4 inducers or inhibitors.

Treatment was recommended but not required for patients consented to study. Treatment consisted of 28-day cycles of oral temozolomide 150 mg/m${}^2$ daily on days 1–5, oral etoposide 50 mg/m${}^2$ daily on days 1–12 and continuous kinase inhibitor based on arm assignment to either Sorafenib (Arm A) 150 mg/m${}^2$/dose twice a day, Everolimus (Arm B) 3 mg/m${}^2$/dose daily, Erlotinib (Arm C) 85 mg/m${}^2$/dose daily, or Dasatinib (Arm D) 60 mg/m${}^2$/dose twice a day. Eligible subjects were permitted to begin study treatment with temozolomide and etoposide alone prior to laboratory testing results if treatment was deemed to be urgent based on disease status.

All cases were reviewed by the study pathologist to confirm the diagnosis. The schema for assignment of treatment arm (see Fig. 1) was based on specificity of targeted agent with more specific agent selected if target was present. If more than one tumor specimen was available from different surgical procedures, the results of the most recent relapse specimen were prioritized. Additionally, if a sample was positive for more than one marker the drug priority was first Erlotinib, then Dasatinib, Everolimus, and last Sorafenib. Priority was based on drug specificity from most specific to less specific.

Fig. 1.

CONSORT diagram with study arm assignment.

Four-$\mu$m-thick paraffin sections of each case were stained using a Ventana Benchmark Stainer (Tucson, AZ). Sections were incubated with primary antibodies in the following concentrations: EGFR, 1:100 (Invitrogen, Carslbad, CA, USA); CD117, 1:2000 (Dako, Carpinteria, CA, USA); pERK, 1:50 (Cell Signaling Technology, Danvers, MA, USA); PDGFRA, 1:100 (Santa Cruz Biotechnology Inc., Dallas, TX, USA); and pS6, 1:150 (Cell Signaling Technology, Danvers, MA, USA) diluted in phosphate-buffered solution (PBS). Slides were incubated with biotinylated secondary antibodies, followed by incubation with the streptavidin and biotinylated peroxidase complex. Sections were counterstained with hematoxylin and mounted. For HER2 testing only, unstained slides were sent to Phenopath Laboratories, Seattle, WA and stained with HER2. All testing was performed in a CAP accredited CLIA approved setting with appropriate control tissues.

Stains were interpreted as either positive or negative according to the following algorithm. Each stain was first scored according to both staining intensity (weak, moderate, to strong) and based on the percentage of tumor cells staining positive (0 for no staining, 1+ for 10%, 2+ for 10–50, 3+ for 50–90% and 4+ for $>$90%). Tumors with no staining were scored as 0. A score of 1+ or greater was considered positive for PDGFRA, EGFR, CD117, and HER2. A score of 2+ or greater was considered positive for pERK and pS6.

The study was subsequently amended to offer optional correlative targeted tumor Next Generation Sequencing (NGS). Sequencing was performed using UW Oncoplex, a clinically validated method as previously reported [21]. This panel was designed to detect most classes of mutations, including single nucleotide variants, small insertions and deletions, gene amplifications, and selected gene-fusions. Sequencing libraries were prepared from DNA samples and hybridized to a custom set of complementary RNA (cRNA) biotinylated oligonucleotides targeting the exons of 262 cancer related genes and select intronic regions. NGS was performed using a HiSeq 2500 instrument system (Illumina, San Diego, CA, USA).

Response was evaluated by MRI imaging every two cycles. For subjects with measurable disease, protocol-defined tumor measurements included the product of the longest diameter and the next longest perpendicular measurement. Complete response was defined as the disappearance of all abnormal signal, partial response was defined as $\ge$50% decrease, minor response was defined as $\ge$25% decrease and progressive disease was defined as $>$25% increase in two-dimensional tumor measurement. Disease was considered stable if no more than 25% increase or decrease in tumor measurement.

The statistical design of the study defined two feasibility endpoints, namely (1) the feasibility of rapid testing, and (2) the feasibility of subjects receiving treatment on assigned study arm. Biologic selection by immunohistochemical testing was defined as feasible if at least 80% of study subjects received results of testing within two weeks of study enrollment. To meet this endpoint, tumor blocks were identified locally or acquired from another institution, tissue quality had to be adequate in amount and quality for staining and interpretation by study pathologist. Treatment on study was considered to be feasible if at least 50% of subjects started therapy with an assigned treatment arm. The treatment feasibility endpoint therefore depended on both the presence of at least one of the pre-specified protein targets as well as patient and/or treating physician decision-making.

In the archival tissue microarray studies eighty-nine percent of patients (54 of 61) evaluated had at least one positive immunohistochemical marker. The percent of positive tumors was 16% for EGFR, 25% for HER2, 31% for KIT, 73% for PDGFRA, 43% for pERK and 54% for pS6. Each marker was positive in a subset of tumors of each histologic type, with the exception of HER2, which was negative in all high-grade gliomas; and EGFR, which was negative in all ependymomas. In addition to tumor histology, survival was associated with extent of resection and expression of 3 of the 6 antigens evaluated: EGFR, PDGFRA and pS6 (log rank p 0.05). Multivariable analysis was conducted to evaluate the independence of clinical and IHC variables. The magnitude of association between EGFR, PDGFRA and pS6 positivity and survival persisted when adjusted for tumor type, age, and extent of resection (Supplementary Table 1).

Twenty subjects were enrolled on this phase 1 clinical trial. One additional subject was screened and found to be ineligible due to no documented disease progression following most recent medical therapy. Subject demographics and prior therapy are detailed in Table 1. Median age at the time of enrollment was 8 years (range 1–28 years), and 11 subjects (55%) were male. All patients had received prior multi-modality therapy. The number of prior surgeries was one in eight subjects, two in eight subjects, and three in four subjects. All 20 subjects received prior radiation therapy, including seven (35%) with prior craniospinal radiation and six subjects (30%) who had received two prior radiation therapy courses. Fifteen subjects (75%) received at least one prior chemotherapy regimen. The subjects without prior chemotherapy included diagnoses of ependymoma (n = 4) and chordoma (n = 1) who had received treatment with surgery and radiation.

The rapid return of testing results was considered feasible, as results were returned within the 14-day study target goal in 90% of subjects (n = 18/20). Median time to return of results was 11.5 days (range 6–22 days). Only seven study subjects (35%) had tissue available at the primary study institution, and tissue was obtained from an outside institution for the remaining 13 subjects (65%). Testing was returned for the two subjects who did not meet the 14-day study target goal at 19 and 22 days. The delay to testing results was due to time to obtain tissue from outside institution in both cases. Targeted sequencing was performed for 18 subjects (90%), and oncogenic mutations or copy number changes were found in 15 of 18 tested (83%). Details of patient characteristics, prior treatment, and testing results are provided in Table 1. The highest levels of protein overexpression (3–4+) were often observed to be associated with gene amplifications or pathway mutations (Fig. 2). In six cases, 33% of cases sequenced, the immunohistochemical staining results identified cases with genetic alterations predicted to respond to targeted drug therapy.

Fig. 2.

Examples of high immune positivity associated with gene amplification, subject 3. (A) PDGFRA overexpression associated with gene amplification; subject 7 (B) EGFR overexpression associated with gene amplification; subject 14 (C) pS6 overexpression associated with TSC2 mutation.

No subjects were assigned to treatment arm A (Sorafenib). Four, six, and ten subjects were assigned to treatment arms B (Everolimus), C (Erlotinib) and D (Dasatinib), respectively. While all 20 subjects were assigned to a study treatment arm (see Fig. 1), only two subjects (10%) elected treatment with study-assigned therapy within the pre-specified four-week time-period (subjects 2 and 5). Of the two subjects who elected treatment on study, only one was able to receive kinase inhibitor; the other subject experienced a delay in start of targeted therapy due to insurance denial of targeted agent, and subsequently experienced disease progression while receiving oral chemotherapy alone. Two subjects received study-directed therapy off study, one who elected for other treatment with re-irradiation prior to initiation of study testing-directed therapy (subject 7); the other decided not to travel for study therapy but received treatment at home institution (subject 3). Both subjects remained on therapy for 9 months until disease progression. Subject 3 experienced initial response to combination study treatment which was sustained for $>$6 months (Fig. 3).

Fig. 3.

Response to therapy containing dasatinib in a recurrent tumor with PDGFRA immunopositivity and amplification. MRI images at study enrollment (A), 3 months later demonstrating response (B) sustained at six months (C). Patient remained on therapy for 9 months until disease progression.

Five other subjects received targeted therapy alone based on study testing results, without recommended concurrent chemotherapy, one of whom received two sequential targeted therapies (subject 14). While a total of 8 subjects (40%) received targeted therapy based on study results, only one (5%) met the second feasibility endpoint according to study design. Details of subject therapy and follow-up are provided in Table 2.

The remaining subjects who did not receive targeted treatment received either chemotherapy alone (n = 7), surgery alone (n = 2) or no further tumor-directed therapy (n = 3). One subject who did not plan initial post re-resection therapy was subsequently lost to follow-up after return of study results. At the time of study completion, 12 subjects had died of disease, one died of secondary malignancy (acute myelogenous leukemia), one died of infectious complications of surgery, and three were alive with disease. Only two patients were in follow-up with no evidence of disease, one of whom was later deemed to have probable pseudoprogression.