Target actionability review to evaluate CDK4/6 as a therapeutic target in paediatric solid and brain tumours

Background: Childhood cancer is still a leading cause of death around the world. To improve outcomes, there is an urgent need for tailored treatment. The systematic evaluation of existing preclinical data can provide an overview of what is known and identify gaps in


Introduction
Cancer remains the leading cause of disease-related death in children and adolescents in Western Europe [1].Despite significant improvements in the overall outcomes of some paediatric cancers over the last decades, the discovery of novel, curative and less toxic therapies is hampered by the rarity and heterogeneity of these diseases (<1% of all cancers) [2].Small patient numbers and limited economic incentives complicate the development of cancer-specific drugs for children.However, global initiatives and recent changes in the regulation, such as the Research to Accelerate Cures and Equity for Children Act in the US and the obligatory paediatric investigation plan in Europe, now oblige companies to no longer ignore childhood cancers [3].There certainly have been advances in the targeted treatment of paediatric tumours in recent years [4], though not as big as in adult cancer treatment, compelling paediatric oncologists to turn to off-label use of drugs approved for adults.This off-label use may not only raise key ethical and legal concerns [5], but it also precludes the systematic evaluation of drug efficacy.This argues a strong case for the need to systematically review proof-of-concept (PoC) preclinical data to match paediatric tumour entities to the most promising therapeutic options.To address this, the target actionability review (TAR) methodology [6] was previously established as part of the innovative therapies for children with cancer paediatric preclinical PoC platform (ITCC-P4), an innovative medicines initiative 2-funded publiceprivate partnership between academic research institutions and pharmaceutical companies [7].In a pilot TAR evaluating the MDM2-TP53 pathway in primary tumour data and preclinical models of paediatric cancers, we demonstrated that the TAR methodology provided the most comprehensive overview of available preclinical data on targeting of MDM2 in paediatric cancer to date [6].To extend the TAR series within the ITCC-P4 project, we applied the TAR methodology to systematically review the published literature on CDK4/ 6 and its inhibitors across a broad panel of 16 paediatric solid and brain tumour types.
CDK4 and its homologue CDK6 are positive regulators of cell cycle progression.Upon binding cyclin D, the complex phosphorylates Rb protein, resulting in the release of E2F transcription factors and the transcription of genes involved in the G1/S transition.Currently, three CDK4/6 inhibitors are approved by the FDA for ER-positive, HER2-negative breast cancer: palbociclib, ribociclib and abemaciclib.In addition, CDK4/6 inhibition seems promising in other solid, as well as haematological, adult cancers [8,9] and gains attention in paediatric oncology.However, the systematic evaluation of preclinical PoC data are currently still lacking for CDK4/6 as a therapeutic target in paediatric tumours.
This TAR provides a comprehensive overview of the available preclinical data on CDK4/6 in paediatric cancers.By summarising and visualising the scores for each tumour type as a heatmap, our review highlights the strengths and gaps in the current preclinical knowledge on CDK4/6 as a paediatric cancer target.

Methods
The TAR method was applied as described previously, with four general steps: (1) extensive literature search for papers on the therapeutic target þ paediatric tumours of interest, (2) critical evaluation and scoring of the papers, (3) reviewer adjudication and (4) visualisation of PoC as a heatmap (Fig. 1a) [6].Briefly, the first and second reviewers searched PubMed for papers on CDK4/6 and their inhibitors in paediatric solid and brain tumour histologies.After reading the titles and abstracts of the identified papers, the two reviewers agreed on a final list of papers, which included all studies addressing at least one critical appraisal question (CAQ) (Supplementary Table 1).Both reviewers individually performed the full assessment of these papers, i.e. determining the scores for experimental quality and outcome (Tables 1 and 2) and reporting the evidence in the online platform R2.Subsequently, the two reviewers discussed scoring discrepancies and agreed on the final scores.Blinded to these scores, the third reviewer revised the same studies with discordant scores, after which the adjudicated scores of reviewer 1 þ 2 and those of reviewer 3 were compared.The remaining discrepancies were resolved by the three reviewers and the final heatmap was generated in R2 [r2platform.com/TAR/CDK4_6].
For this TAR, we made a few adjustments to the standard methodology as defined in [6].These changes are underlined in the scoring tables for experimental quality (Table 1) and experimental outcomes (Table 2).

Results
In this study, we applied the TAR methodology to evaluate the potential actionability of CDK4/6 in paediatric solid and brain tumours.To obtain a list of papers that was as complete as possible with studies addressing CDK4/6 or their respective inhibitors in paediatric malignancies, we used only minimal keywords as our search terms for PubMed (Table 3).
Using these search terms (search date: 24 November 2021), 394 unique papers were identified (Fig. 1b).Of these, 18 (4.6%)were review papers and 30 (7.6%) were case reports and thus excluded immediately.We further filtered out 38 papers (9.6%) published before 2000, based on our experience with previous TARs that older publications typically used experimental techniques that would score poorly on quality, thus having minimal impact on the final heatmap.After reading the titles and  The TAR revealed that the most studied cancers were osteosarcoma (OS), neuroblastoma (NBL), medulloblastoma (MB) and rhabdomyosarcoma (RMS), whereas no relevant studies on CDK4/6 (inhibitors) were found for hepatoblastoma (HB), inflammatory myofibroblastic tumour (IMT), extracranial germ cell tumour (GCT) and retinoblastoma (RB) (Fig. 1c).Only six studies (8.5%) addressed more than one tumour entity (Supplementary Figure 1a) and 13 studies (18.3%) included one or more tumour subtypes (e.g., different subtypes of MB) (Supplementary Figure 1b).The papers, it is assumed that the authors used similar definitions, CR: complete regression, the disappearance of tumour; PR: partial regression, 30% decrease of tumour volume; SD: stable disease, neither PR nor PD criteria met; PD: progressive, disease, 20% increase of tumour volume; TGI: tumour growth inhibition; criteria based on RECIST criteria [10]; underlined text indicates deviations from the original methodology as, described in ref. 6. RP2D: recommended phase 2 dose; DLT: dose-limiting toxicity; ORR: overall response rate; DoR: duration of response; PFS: progression-free survival; SOC: standard-of-care, NB: if publications did not address the experimental outcomes according to these criteria, the outcomes were estimated and scored based on this table.
a Clinically relevant concentration: the dose that corresponds to the maximum plasma concentrations reached in patients without signs of toxicity.
b Toxicity profile is acceptable if adverse events are not life-threatening (no higher than Grade 3 based on the Common Terminology Criteria for Adverse Events) [11].sensitivity of cell lines to CDK4/6 inhibition (PoC 4) was the most studied module, with a total of 36 out of 151 entries (23.8% -31 papers), closely followed by CDK4/6 amplification/gain/overexpression (PoC 1a) with 34 entries (22.5% -26 papers) (Fig. 2a).The final heatmap is shown in Fig. 2b.
For PoC 1, we grouped target amplification, gain, and overexpression into one module (PoC 1a) and distinguished it from target protein expression (PoC 1b) because studies of DNA/RNA typically show or imply concomitant protein overexpression.OS was the entity most frequently addressed in PoC 1a.However, outcomes were contradictory, which may partly be caused by mixed patient cohorts with both paediatric (18 years) and adult cases.In such cases, we lowered the quality scores of PoC 1 by one point to adjust for the fact that adult cases may inflate the actual occurrence of an aberration and consequently the scored outcome [12,13].CDK4 copy number variation frequencies of w10% were reported by three next-generation sequencing studies [14e16].In RMS, CDK4 amplification might be more frequent, especially in the alveolar subtype (26.1%) as opposed to the embryonal subtype (7.5%) [17e19].For NBL, CDK4 amplification was studied in larger cohorts (ranging from 82 to 628 paediatric patients per study) but seems to be rare (<1.3%) [20e22].Nonetheless, elevated CDK4 levels were shown to correlate with poor survival in NBL, which is why we increased result scores to þ1 [20].Evidence from this TAR suggests that amplification of CDK6 is more frequent than CDK4 in brain tumours, contrary to solid tumours [23].Overall, overexpression of CDK4/6 seems to be more frequent than gains, which are more frequent than amplification.Moreover, CDK4 status was studied almost 2.5 times more than CDK6 status.In summary, there was strongest evidence (average score of 3) supporting higher levels of CDK4/6 in RMS, malignant peripheral nerve sheath tumour (MPNST), Wilms tumour (WT), high-grade glioma (HGG) and low-grade glioma (LGG).
12 entries (7.9%) were included for tumour target dependence in vitro (PoC 2), compared to only three (2.0%) for tumour target dependence in vivo (PoC 3); two studies examined both.Eight papers addressed CDK4 knockdown/knockout, as opposed to CDK6 knockdown in seven papers.Overall, quality scores for PoC 2 were moderate due to the use of few cell lines, single knockdown methods or the absence of rescue experiments.While either knockdown resulted in decreased cell viability and proliferation, as well as cell cycle arrest and reduced levels of (phosphorylated) pRb, one NBL study found that the effect was lower for CDK6 knockdown [20].The biggest effect, i.e. >75% cell death upon CDK4 knockdown, was seen for ES and OS [24,25].
Tumour target dependence in vivo (PoC 3) was only studied in RMS (CDK4) and MB (CDK6).Mice intramuscularly injected with RMS cells with inducible CDK4 knockdown showed reduced tumour growth compared to control mice [19].Constitutive overexpression of CDK6 in orthotopic MB xenografts (MB subgroups were one SHH, one former PNET) led to tumour development and shorter survival times [23], whereas another study with transgenic models reported reduced tumour size and prolonged survival after Cremediated homozygous CDK6 knockout [26].The positive evidence for in vitro target dependence suggests that OS and ES should be further evaluated in an in vivo context.Notably, future studies should also aim to evaluate tumour target dependence in other tumour entities.
A total of 31 papers reported testing CDK4/6 inhibitors in vitro (PoC 4); 45.2% of these also included in vivo studies.Palbociclib was the most studied compound with 23 reports, whereas five studies tested ribociclib and six abemaciclib.Of these, three studies tested more than one CDK4/6 inhibitor.CDK4-specific inhibitors CAS 546102-60-7 and fascaplysin were each used in one study [27,28].
Palbociclib efficacy varied between studies addressing the same entity.There were only three studies that used more than five cell lines, reporting IC 50 values lower than 500 nM in >10% of NBL and HGG cell lines [29e31].Atypical rhabdoid tumour/malignant rhabdoid tumour (ATRT/MRT) cell lines were sensitive in three studies that scored lower for quality due to the number of cell lines used [32e34], whereas HGG cell lines seemed rather insensitive [35,36].For other tumour types, results were mostly conflicting or only based on one study.Ribociclib efficacy in vitro was studied in four high-quality studies, showing good responses (IC 50 < 500 nM) in NBL [31,37] and ES [24] cell lines but only moderate efficacy in RMS cells [19].Abemaciclib treatment was mainly effective in ES [38], NBL [31] and OS [39] cell lines.Two studies (in NBL and EPN) showed superior efficacy of abemaciclib compared with palbociclib or both other CDK4/6 inhibitors [31,40].Overall, the only entity that scored negatively for PoC 4 is GCT, all other studied entities have average scores between 0.8 (HGG) and 5.7 (ES).Most robust results were seen for NBL, ES and HGG.Studies that scored lower for quality may suggest that CDK4/6 inhibitors are less effective in these tumour entities, but this could be explained by the low number of cell lines included in these studies since we noticed that studies with more cell lines typically also had higher result scores.CDK4/6 inhibition may also be of value in ATRT/MRT, SS, OS, MB and RMS.
Overall, MB, MPNST, ES, RMS, ATRT/MRT and HGG received average scores 0 for in vivo (mouse) studies (all SD with the exception of CR in MB SHH and Group 3).Of these, only MB scored high (>3).Other tumour types received either an average score of 0 due to conflicting results (NBL, RMS, ATRT/MRT and HGG) or a negative score (OS and, based on a single study, EPN).
Our search identified three clinical studies, up to phase II.The phase I trial for ribociclib included 15 NBL and 15 MRT patients (we excluded the only RMS patient, as this would resemble a case study) and reported a maximum tolerated dose of 470 mg/m 2 and a recommended phase II dose (RP2D) of 350 mg/m 2 / d [53].Stable disease was reached in 7/15 NBL and 2/15 MRT patients.The same dose was used in phase I/II trial with 10 newly diagnosed DIPG (HGG K27M mutant) patients following radiotherapy [54].Nine patients progressed and one patient discontinued treatment after course 14.Both studies reported manageable adverse events, with neutropenia being the most frequent (up to 90%).The third trial was a phase II study examining palbociclib treatment in 34 patients with grade 3 oligodendroglioma (HGG) >18 years of age [55].Given the age of the patients, this study received a low quality score.Moreover, the study was discontinued early owing to a lack of efficacy.
Overall, the results of this TAR reveal that extensive preclinical work is still necessary to determine the relevance of targeting CDK4/6 in paediatric cancers.Information on CDK4/6 aberration frequencies is unknown for ATRT/MRT, HB, IMT, GCT, RB and EPN or based on a single publication in SS, MPNST, ES, WT and LGG.The dependency of tumours on these oncogenes is also barely investigated in paediatric cancers.Compound sensitivity should be (further) addressed in all tumour types, especially in vivo, and particularly in SS, WT, HB, IMT, GCT, RB, LGG and EPN.Future studies should also focus more on the identification of biomarkers and combinatorial approaches.

Discussion
The goal of the ITCC-P4 consortium is to accelerate evidence-driven paediatric cancer drug development by prioritising drugs currently undergoing preclinical investigation (or drugs repurposed from adult malignancies) for clinical development in children suffering from cancer.In this study, we applied the previously established TAR methodology to evaluate the potential actionability of CDK4/6 in paediatric solid and brain tumours.Based on our experience of having a high percentage of discordant scores between the two reviewers, we suggest to adapt the TAR methodology by performing a 'pilot adjudication' after the first ten papers that are fully scored (Supplementary Figure 2).This initial comparison will help in identifying pitfalls and different approaches at an early stage, aligning the scoring, and will ultimately result in fewer discordant scores.
Evidence from this TAR suggests that CDK4/6 aberrations occur in RMS, OS, HGG and MB, and at lower frequencies also in NBL.For most other indications, our search strategy did not capture any or more than one publication(s) reporting aberration frequencies.Overexpression seems to occur more frequently than gain or amplification of CDK4/6, suggesting that other mechanisms may contribute to higher levels.It is important to realise that lower incidence rates of certain tumour types could possibly result in smaller sample sizes, thus adding a bias to the quality and the overall score of modules.
There is still a lot of uncertainty regarding the correlation between higher CDK4/6 levels and drug sensitivity.Four studies examined this correlation, but the biomarker status of CDK4 could only be confirmed in OS [16].Cell lines included in PoC 4 and 5 had all sorts of genetic backgrounds and the relatively low incidence of CDK4/6 aberrations makes it difficult to draw conclusions.The ambiguous effect of CDK4/6, as well as p16/CDKN2A, was also reported in adult malignancies [56e59].As reviewed recently, CDK4/6 overexpression or amplification even correlated with resistance in some adult cancer models [57].Our included studies suggest only MYCN as putative biomarkers for CDK4/6 inhibitor sensitivity, indicating that further research on biomarkers is needed to select the best patient cohort for this intervention.Moreover, not only the dependency on CDK4/6 overexpression but also its exact contribution to the development or proliferation of tumours should be further investigated.
Frequently, tumour entities which scored positively in PoC 4 (in vitro sensitivity) scored negatively or at least much lower in PoC 5 (in vivo sensitivity), as was the case for NBL, RMS, ES, OS, ATRT/MRT, HGG and EPN.These findings suggest that in vitro studies alone are not always predictive of drug efficacy in vivo, highlighting the necessity of in vivo studies.Based on the included studies, CDK4/6 inhibition may be most promising in MPNST, ES and MB (especially SHH and Group 3).Clinical data showed that CDK4/6 inhibitors are tolerated at relatively high doses, with a maximum tolerated dose of 470 mg/m 2 for ribociclib and an RP2D of 350 mg/m 2 , which are comparable to those in adults [53].Moreover, ribociclib shows good central nervous system penetration [60].Therefore, entities that scored lower for these modules (mainly NBL, RMS, ATRT/MRT and HGG) may also still benefit from CDK4/6 inhibition, although secondary target inhibition should be examined and prevented.While two studies reported superior in vitro efficacy of abemaciclib, assumptions on the most efficient CDK4/6 inhibitor in vivo are not possible based on the results of this TAR.However, the MAST study (https://braid.stjude.org/masttour/),which was not found using our search terms, shows that ribociclib has superior efficacy over palbociclib in paediatric solid cancers [61].
That hardly any in vivo drug sensitivity studies or the clinical trials were able to achieve a response better than stable disease shows that combination therapies may be necessary to achieve objective responses.Based on current preclinical evidence, CDK4/6 inhibitors in combination with ALK inhibitors in ALK-driven tumours, radiation therapy or chemotherapy (mainly doxorubicin) should be prioritised for clinical development.The combination of CDK4/6 inhibitors with chemotherapy indeed shows clinical promise [62].A key finding reveals the mechanism by which CDK4/6 inhibitors impair recovery from DNA damage induced by chemotherapies that require cycling cells for their activity, suggesting that CDK4/6 inhibitors should be applied after and not before cytotoxic chemotherapy [63].Additionally, the search for other tumour-specific genetic dependencies that are synergistic with CDK4/6 inhibition should continue.
All three CDK4/6 inhibitors are currently tested in several paediatric clinical trials and included in different precision medicine programs for children with CDK4/6 amplification or a homozygous loss of CDKN2A.Several of these studies or programs use abemaciclib, even though our results show that published preclinical evidence for this drug is still sparse.This disproportion may indicate that results with palbociclib/ribociclib are sometimes extrapolated.Future (pre)clinical studies will have to show whether extrapolation is appropriate, especially given the broader target spectrum of abemaciclib [64].
For clinical trials to succeed, optimal target group selection, taking the molecular status into account, is pivotal.Unfortunately, the data from this TAR shows that preclinical evidence for a positive biomarker status of CDK4/6 aberrations and CDKN2A loss is still scarce and contradictory.Their exact influence on CDK4/6 inhibitor sensitivity will need to be further addressed in future studies.Given the complexity of cell cycle regulation, future studies may also want to look at predictive gene signatures instead of single gene biomarkers.
In conclusion, the heatmap generated from the CDK4/6 TAR reveals that preclinical data are still lacking for many paediatric tumour entities.Indicated by 2 publications, intensive work across all PoC data modules is necessary for WT, HB, IMT, GCT, RB, LGG and EPN, while for most other tumour types, research should mainly focus on unravelling the dependence of a tumour on CDK4/6 and the identification of biomarkers, resistance mechanisms and combination therapies.Researchers should also be encouraged to differentiate between tumour subtypes where this is applicable.Our data suggest that CDK4/6 inhibition might be most relevant for MPNST, ES and MB (SHH and Group 3) patients, but patients with NBL, RMS, ATRT/MRT and HGG may benefit from this targeted treatment as well.Whether this is indeed the case, will have to be addressed in future clinical studies.The full TAR data is summarised in one publicly accessible online application [r2platform.com/TAR/CDK4_6], where all data can be interactively explored and evaluated.

Fig. 1 .
Fig. 1.Overview of the methodology and the studies included in the CDK4/6 TAR.(a) Overview of the TAR methodology.Adapted with permission from Schubert et al. [6].(b) Study selection process.(c) Number of papers and entries per tumour entity.TAR, target actionability review.

Fig. 2 .
Fig. 2. Overview of the entries included in the CDK4/6 TAR.(a) Number of entries included per PoC module and tumour entity.(b) Heatmap showing the average scores of all entries made for this CDK4/6 TAR.Numbers indicate the number of included publications.Interactive versions of both figures are accessible through R2 [r2platform.com/TAR/CDK4_6].PoC, proof-of-concept, TAR, target actionability review.

Table 1
Rubric for scoring experimental quality.
more targets (e.g., pan-CDK inhibitor); patients 18 years with a paediatric tumour CI: combination index; strikethrough and underlined text indicate deviations from the original methodology as described in ref. 6. N.A. Schubert et al. / European Journal of Cancer 170 (2022) 196e208

Table 2
Rubric for scoring experimental outcomes.
50 > 1500 nM -3 No activity (IC 50 > 10 mM) PoC 5: in vivo activity of compound/drugIn vivo tumour response 3 Response comparable to PR/CR 1 Response comparable to SD -1 Very minor response (between SD and PD, data entries) that were scored.Of all data entries, 64 (42.1%) were discordant after the assessment by the first two reviewers.Following the third reviewer's assessment, 40 (26.3%)data entries still had discrepant scores.Subsequently, discrepancies were discussed between the three reviewers and a consensus was reached for all data entry scores.One additional paper was excluded because it did not clearly fit one of the PoC modules, resulting in a final heatmap with 151 data entries from 71 papers.

Table 2
>8 copies, based on next-generation sequencing (NGS) techniques, array CGH, FISH or Southern blotting; gain: 2,5e8 copies, based on NGS techniques, array CGH, FISH or Southern blotting; overexpression: z-score >2 in the related cohort.If definitions are not clearly mentioned in

Table 3
Search terms.
a "AND (pediatric OR child)" was added to the search terms in an attempt to exclude papers on adult gliomas.N.A. Schubert et al. / European Journal of Cancer 170 (2022) 196e208