Immune checkpoint inhibitors in Wilms' tumor and Neuroblastoma: What now?

Abstract Background Therapeutic activation of tumor‐infiltrating lymphocytes using monoclonal antibodies targeting PD1 or PD‐L1 (immune checkpoint inhibitors—ICIs) has revolutionized treatment of specific solid tumors in adult cancer patients, and much hope has been placed on a similar effect in relapsed or refractory solid pediatric tumors. Recent clinical trials have disappointingly shown an almost nonexistent response rate, while case reports have demonstrated that some pediatric patients do achieve durable responses when treated with this type of drug. Aim To elucidate this paradox, we mapped the landscape of expressed neoantigens as well as the levels of immune cell infiltration in the two most common extracranial solid pediatric tumors: Wilms tumor and neuroblastoma using state‐of‐the‐art in silico analysis of a large cohort of patients with these tumors. Methods By integration of whole exome sequencing and RNA‐sequencing, we mapped the landscape of neoantigens in the TARGET cohorts for these diagnoses and correlated these findings with known genetic prognostic markers. Results Our analysis shows that these tumors typically have much lower levels of expressed neoantigens than commonly seen in adult cancers, but we also identify subgroups with significantly higher levels of neoantigens. For neuroblastomas, the cases with higher levels of neoantigens were confined to the group without MYCN‐amplification and for Wilms tumor restricted to the TP53‐mutated cases. Furthermore, we demonstrate that neuroblastomas have an unexpectedly high level of CD8+ tumor‐infiltrating lymphocytes, even when compared to adult tumor types where ICI is an approved treatment. Conclusion These results could be important to consider when designing future clinical trials of ICI treatment in pediatric patients with relapsed or refractory solid tumors.


| INTRODUCTION
Immune checkpoint inhibitors (ICIs), targeting either programmed death-1 (PD1) or programmed death-ligand-1 (PD-L1), have revolutionized treatment of some tumor types in adults, such as lung cancer and melanoma. This success has fueled an interest in using ICI in relapsed and/or refractory pediatric cancers. 1 At least three clinical studies evaluating the efficacy of ICI in this setting have recently been published [2][3][4] ; disappointingly showing that the objective response rate for PD1 inhibition (pembrolizumab and nivolumab) and PD-L1 inhibition (atezolizumab) for pediatric cancer patients is low. For pembrolizumab, the objective response rate (ORR) according to REC-IST v1.1 for solid tumors was 5.5%. 2 For nivolumab and atezolizumab, there was no objective response in any of the patients with solid tumors. 3,4 At the same time, case reports have shown that there exist pediatric patients that have a durable response on ICI, at least in combination with other antineoplastic agents. 5 This apparent paradox could be due to the fact that in general, tumors in pediatric patients are less immunogenic than in adults, but that there exist patients with more immunogenic tumors that do respond to ICI. Children with such tumors would be strong candidates for future clinical trials. and small insertions and deletions (indels) was performed using Mutect2 7 version 4.1.3.0, according to the GATK4 best practice guidelines, with a normal reference panel consisting of all normal samples from the specific cohort. Filtering of putative variant calls was performed using the Fil-terMutectCalls tool from GATK4, including the ReadOrientationBiasFilter, to remove potential Oxo-G artefacts, 8 that are known to be present in a subset of TARGET neuroblastoma cases. 9 Purity and ploidy estimation as well as detection of allele-specific copy number alterations were performed using Sequenza, 10 with standard settings. All proposed Sequenza solutions (purity and ploidy) were manually inspected to ensure the solution did not imply biologically unreasonable scenarios such as large-scale homozygous deletions, and if so, refitted using one of the alternative solutions proposed by Sequenza.

| Processing of RNASeq data
Bam-files containing all sequencing reads from these experiments (ie, even reads that failed to map to the reference) were downloaded from the GDC. Raw paired end fastq-files were extracted from these bam-files using bazam. 11 The raw reads where then pseudoaligned and transcript-level abundances were quantified using kallisto. 12 Transcript-level abundances were merged into gene-level estimates using the Bioconductor package tximport. For mutant allele expression detection, all variants underlying putative neoantigens were genotyped in their corresponding RNASeq bam-file using freebayes, 13 and variants with more than three reads supporting the ALT allele were considered expressed.

| Combined cohort
The dataset named TCGA_TARGET_pancan was downloaded from UCSC Xena (UCSC toil data hub, dataset version 21 January 2017).
We removed samples that were annotated to be from metastatic locations and also removed all primary brain tumor-, leukemic-, or lymphoma-samples (the exact samples used are listed in Table S5).
We then utilized QuantiSeq for estimation of infiltrating CD8+ TIL.

| Clonality analysis
As we were only focused on primary tumors and typically only had access to data from a single sample per primary tumor, we opted to perform a simplified dichotomization of variants into clonal and subclonal instead of trying to perform a full clonal deconvolution. Briefly, we calculated the product of cancer cell fraction (CCF) and mutation multiplicity (m) following 14 as: This was followed by parametric bootstrapping assuming a binomial distribution of read counts to calculate a 95% confidence interval.
Mutations whose 95% CI for cancer cell fraction contained 1.0 were classified as clonal, all others as subclonal.

| Estimation of immune cell infiltration
The gene-level abundance estimates from tximport were used as input to QuantiSeq, 15  binding affinity <500 nM and a rank percentage score <2%.

| RESULTS
To assess the existence of such subgroups from a biological point of view, we set out to map the prevalence in solid pediatric tumors of biomarkers known from adult cancer to predict response to ICI. The most commonly used biomarkers for ICI response in adult cancers are tumor mutational burden (TMB), expressed neoantigens, PD-L1 and PD1 protein expression, various gene signatures, microsatellite instability, and specific somatic genetic alterations. 21 Large-scale genomic analysis of pediatric solid tumors has consistently demonstrated very low TMBs compared to adult cancers. 22 We thus chose to focus on evaluating PD1/PD-L1 expression, the numbers of expressed neoantigens, as well as CD8+ TILs (tumor-infiltrating lymphocytes) in the two most common extracranial solid tumors in children, Wilms' tumor (WT), and neuroblastoma (NBL). We started by analyzing the PD1/PD-L1 mRNA expression across a combined cohort of patients from the WT and NBL TARGET (excluding Stage 4S NBL patients, as Stage 4S represents a distinct clinical entity with partly differing biology) cohorts and adult cancers from the TCGA pan cancer cohort, processed through a unified computational pipeline to ensure comparability. 23 After keeping only primary solid extracranial tumors, this cohort consisted of 6083 patients. We grouped the TCGA tumors into cases from cancer types where there is an FDA approval for ICI and those where no such approval exists. This analysis revealed that both PD1 and PD-L1 mRNA expression levels were significantly lower in the pediatric tumors than in adult cancers ( Figure 1A,B). This was the case for adult cancers both with and without FDA approval of ICI (Mann-Whitney U test; largest P-value = 8,2 × 10 −4 , Table 1). Furthermore, cases with WT had significantly lower levels of PD-L1 and PD1 than NBL. In concordance with this, immune cell deconvolution applied to the RNASeq data from the same set of samples revealed that the median level of CD8 TIL in WT was significantly lower than in NBL (Table 2). Surprisingly, the NBL cohort in fact had significantly higher levels of CD8+ TILs than both groups of adult tumors ( Figure 2); the median absolute fraction of CD8+ TILs was more than three times higher than in the cohort of adult cancers with ICI approved by FDA, and more than four times higher than the median absolute fraction in the non-ICI approved group.
Next, we mapped the landscape of expressed neoantigens in NBL and WT. We annotated all expressed neoantigens as either clonal or subclonal based on allele-specific copy number data, as clonal neoantigens are known to typically elicit a stronger immune response in lung cancer. 24 For this analysis, we selected cases that had  (Table 1). B, The same comparisons as in A, for PD1 expression; see Table 1 for P-values T A B L E 1 Comparison of mRNA levels for PD-L1 and PD1 in TCGA to WT and NBL cases from TARGET, for a graphical representation of these data see Figure 1 Comparison PD-L1 P-value PD1 P-value Note: All P-values are from Mann-Whitney U tests between the two groups designated in the comparison column. the host immune system. The level of PD-L1-expression in WT and NBL was also significantly lower than in adult tumors where ICI is commonly used. We found, rather unexpectedly, that CD8+ TIL levels in NBL are on the same level as in some adult tumors, but that PD1 and PD-L1 expression are still significantly lower. This could potentially be due to low levels of MHC-1 expression seen in NBL, 33 prohibiting the chronic stimulation of CD8+ TILs that causes them to up-regulate PD1. 34 However, the correlation between clonal neoantigens and PD1 expression levels in non-MNA NBL cases argues against this model and could signal that there in fact exists an immune response of activated CD8+ TILs targeting these neoantigens.
Whether this response is of sufficient magnitude to be modulated with significant clinical effects remains to be seen. One interesting ongoing study is the INFORM2-NivEnt, 35 Table 2 clonal

ACKNOWLEDGEMENTS
The results published here are in whole or part based upon data generated by the TCGA Research Network: https://www.cancer.gov/ tcga. The results published here are in whole or part based upon data generated by the Therapeutically Applicable Research to Generate Effective Treatments (https://ocg.cancer.gov/programs/target) initiative, phs000218. The data used for this analysis are available at https://portal.gdc.cancer.gov/projects. This study was supported by grants from the Swedish Research Foundation, the Swedish Cancer Society, the Swedish Childhood Cancer Foundation, the Royal Physiographic Society and by grants from the Swedish state under the agreement between the Swedish government and the county councils, the ALF-agreement.

CONFLICT OF INTEREST
The authors report no conflict of interest.

ETHICAL STATEMENT
The study was approved by the regional ethics review board under permit numbers L289-11 (genomic analyses; updated as L796-2017).

DATA AVAILABILITY STATEMENT
Raw data can be obtained through the controlled access mechanism at dbGAP (accession phs000218). Processed data (neoantigen counts, clinical info, CD8+ TIL-levels) are available in the supplementary tables to this article.