Germline Pathogenic/ Likely Pathogenic Mutations and Subsequent Neoplasms among Childhood Cancer Survivors – A report from the Children’s Oncology Group ALTE03N1 Study

Abstract

Purpose:

There is emerging evidence that germline pathogenic/likely pathogenic (P/LP) mutations in cancer predisposition genes (CPGs) increase the risk of subsequent neoplasms (SNs) in childhood cancer survivors. However, clinical application of this observation is hampered by the lack of knowledge regarding sub-populations at risk for SNs who could potentially benefit from genetic screening.

Patients and Methods:

Whole exome sequencing was performed using germline DNA from 499 survivors with SNs (cases) and 625 survivors without (matched controls) in a Children’s Oncology Group study. Using conditional logistic regression, we estimated demographic/clinical/therapeutic characteristics and P/LP mutations associated with SN risk. We then randomly partitioned the case-control dataset using a 60/40 split to create training and test data respectively, and developed a clinical risk classifier. Model performance and its improvement with addition of P/LP mutation status was evaluated on the test data using the area under the receiver operating curve (AUC).

Results:

We found a 4.26-fold higher odds of SNs among P/LP mutation carriers (95%CI=2.36–7.69). The clinical risk classifier (including sex, primary cancer type and year of diagnosis, exposure to radiation and platinum compounds, and length of follow-up) showed a significant performance improvement with the addition of P/LP mutation status, and P/LP*platinum and P/LP*radiation interactions (AUC increased from 0.79 to 0.82, p=0.014). Using the clinical risk classifier, we classified survivors at low-risk (22%) and moderate-to-high-risk (78%) of developing SNs. Overall, 86.4% of the survivors with any P/LP mutations, and all TP53 and RB1 mutation carriers were partitioned into the moderate-to-high risk group.

Conclusion:

These findings provide a risk-based approach for identifying childhood cancer survivors who could be referred for genetic testing, informing surveillance strategies based on refined risk-classification.

INTRODUCTION

Subsequent neoplasms (SNs) are the leading cause of premature death in childhood cancer survivors.^1–3 Radiation-related solid tumors constitute the vast majority of SNs⁴, although there is increasing evidence that alkylating agents, platinum compounds, and anthracyclines may also contribute to SN risk⁵. The significant inter-individual variability in SN risk suggests a role for genetic susceptibility that likely modifies the treatment-SN association.⁶ There is emerging evidence that pathogenic/likely pathogenic (P/LP) mutations in cancer predisposition genes (CPGs)⁷ are associated with an increased risk of SNs in childhood cancer survivors⁸. However, unlike the general population, where CPGs are being increasingly utilized in risk-based cancer prevention^7,9, the clinical utility of CPGs in childhood cancer survivors has not been examined as there is no guidance as to which sub-populations of childhood cancer survivors are at risk for SNs. Nor is there guidance as to which sub-populations of childhood cancer survivors should undergo genetic screening to further refine their risk profile, such that risk-based surveillance/risk mitigation strategies can be instituted. We address these gaps in the current report.

The overarching goal of this study was four-fold: (1) determine the role of demographic/clinical/therapeutic factors and P/LP mutation status in the development of SNs in childhood cancer survivors; (2) develop a clinical risk classifier to identify survivors at high risk of developing SNs using demographic/clinical/therapeutic characteristics; (3) examine whether addition of P/LP mutation status improves the performance of the risk classifier; (4) determine the utility of the clinical risk classifier by describing the prevalence of survivors with P/LP mutations among those classified as being at high or low risk of SNs based on their demographic/clinical/treatment characteristics in order to aid clinicians in decisions regarding genetic screening.

METHODS

Study Participants

Study participants were drawn from a Children’s Oncology Group study, (COG-ALTE03N1, NCT00082745), which uses a matched case-control design to understand the pathogenesis of SNs in childhood cancer survivors (details in Supplemental Methods). Cases and controls were identified from individuals diagnosed with a primary cancer at age ≤21 years (y) at one of the 100 participating sites (Supplemental Data) after obtaining approval from local institutional review boards. Written informed consent/assent was obtained from participants and/or parents/legal guardians. For the current report, cases consisted of childhood cancer survivors who developed a histologically-distinct solid SN any time after childhood cancer diagnosis. For each case, 1–4 survivors with no SNs were identified as controls from the same childhood cancer survivor cohort, matched on primary cancer diagnosis, year of diagnosis (±10y), race/ethnicity, and receipt of radiation. Controls with a longer duration of SN-free follow-up period compared with time from cancer diagnosis to SN for the corresponding case were prioritized. Therapeutic summaries and pathology reports (to validate SNs) were provided by participating sites. Participants provided blood or saliva for germline DNA at study enrollment. Genomic DNA was isolated from blood or saliva. Whole exome sequencing (WES; 100x) was performed using Illumina NovaSeq (100 base-paired-end sequencing). Details about sample preparation, sequencing, and quality control (QC) are in Supplemental Methods.

Cancer predisposition genes, variant verification, and pathogenicity classification

We selected 60 CPGs (Supplemental Data) with well-established monogenic cancer risk inherited in an autosomal dominant fashion at moderate-to-high penetrance.⁸ After QC (Supplemental Methods), a total of 918 unique candidate variants (SNV/INDELs) were identified in 56 CPGs. Pathogenicity classification used the American College of Medical Genetics and Genomics Guidelines and was performed using the automated tool VarSome¹⁰ to classify each variant into five categories: pathogenic, likely pathogenic, benign, likely benign, and variant of uncertain significance. Pathogenic and likely pathogenic (P/LP) variants were combined for downstream analysis.

Statistical Analyses

All analyses were performed using R-4.3.2.^11–14

Risk of SNs associated with P/LP mutations and demographic/clinical/treatment characteristics:

Conditional logistic regression was used to estimate the association between the first occurrence of an SN and P/LP mutation status, sex, age at primary cancer diagnosis, anthracyclines (yes/no), alkylating agents (yes/no), platinum compounds (yes/no). Additional variables considered included radiation (yes/no: any site), year of primary cancer diagnosis and length of follow-up, to accommodate lack of perfect matching on these variables for all case-control sets. Four principal components (PCs) of common genetic variants were included in the model. PCs were generated based on common variants with minor allele frequency (MAF) ≥0.05 and genotype missing rate ≤0.2; number of PCs was determined by scree plots. We tested for interactions between P/LP mutations and demographic/clinical/therapeutic variables in their association with SNs and conducted stratified analyses for the variables that demonstrated significant interaction. We also explored the association between P/LP mutations and specific SN types. Site-specific radiation was considered for breast SN (chest radiation), central nervous system (CNS) tumors (glioma, meningioma, other: cranial radiation), and thyroid cancer (neck radiation). Variables were included in multivariable models if they showed a significant (p<0.05) association with SNs in univariate analysis.

Risk Classification Model:

The case-control dataset was randomly partitioned using a 60/40 split to create training and test data respectively, with matched sets preserved during the split to maintain the study design structure. A Clinical Model was developed using multivariable logistic regression on the training data and included sex, primary cancer type, year of primary cancer diagnosis, length of follow-up, exposure to radiation and platinum compounds. The Combined Model included all the variables in the Clinical Model, and P/LP mutation status (Supplementary Methods). Model performance was evaluated on the test data using the area under the receiver operating curve (AUC), using R package “pROC”.¹⁵ Given the matched case-control study design, a conditional C-index¹⁶ was used to adjust the AUC for the matched sets. Confidence intervals for the AUC were obtained using 9999 stratified bootstrap replicates. Improvement in performance of the AUC with the addition of P/LP mutation status to the Clinical Model was evaluated by using the bootstrap method to compare AUCs.

Using a methodology described previously¹⁷, the odds ratios in the Clinical Model were converted to integer risk scores (odds ratio<1.3, 1.3–1.9, 2.0–2.9, 3.0–4.9, and ≥ 5.0 corresponded to risk scores 0, 1, 2, 3, and 4, respectively). Risk scores were summed for each individual. The interquartile range (IQR) of the risk score was used to determine the threshold value to classify the survivors’ individual risk of developing an SN. An online clinical risk calculator was built using R shiny package.¹⁸

Sensitivity Analyses:

Recognizing that cases and controls were not perfectly matched on all matching criteria, sensitivity analyses were performed retaining case-control sets perfectly matched on race/ethnicity, primary cancer diagnosis, and radiation status.

RESULTS

The study included 1,124 survivors (499 with SNs [cases], 625 without [controls]); samples with adequate DNA from 533 survivors with SNs were submitted for WES: of these 34 failed QC, yielding 499 survivors with SNs and WES data (Figure S1). The 499 survivors with SNs included 64 with ≥2 histologically distinct SNs, yielding 575 SNs. Table 1 summarizes the demographic, clinical, and therapeutic characteristics of the case-control dataset. The median time from primary cancer diagnosis to first SN was 14.11y (IQR=8.31–22.92), and the median age at first SN was 21.02y (IQR=15.51–32.86). Prevalent SN types included basal cell carcinoma (BCC: n=92), thyroid cancer (n=89), meningioma (n=61), breast cancer (n=60), osteosarcoma (n=56), soft tissue sarcoma (n=44), and glioma (n=35).

Table 1.

Characteristics of Study Participants

Variable	Cases (n=499)	Controls (n=625)	p-value*
Age of primary cancer diagnosis in years
Median (interquartile range)	6.27 (2.85 – 12.68)	5.75 (2.71 – 12.00)	0.12
Age at SN diagnosis (cases) or age at study enrollment (controls) in years
Median (interquartile range)	21.02 (15.51 – 32.86)	19.44 (15.28 – 25.54)	<0.001
Sex, n (%)
Female	286 (57.31%)	308 (49.28%)	0.012
Race/Ethnicity, n (%)
Non-Hispanic White	400 (80.16%)	484 (77.44%)	Matched
Black	31 (6.21%)	43 (6.88%)
Hispanic	41 (8.22%)	56 (8.96%)
Asian	15 (3.01%)	25 (4.00%)
Other/ multiracial	12 (2.40%)	17 (2.72%)
Year of Primary Cancer Diagnosis, n (%)
<1990	220 (44.09%)	106 (16.96%)	<0.001
1990 – 2000	166 (33.27%)	247 (39.52%)
>2000	113 (22.65%)	272 (43.52%)
Length of Follow-up in years £
Median (interquartile range)	14.11 (8.31–22.92)	12.26 (8.51 – 17.16)	<0.001
Primary Cancer Diagnosis, n (%)
Acute lymphoblastic leukemia	100 (20.04%)	152 (24.32%)	Matched
Acute myeloid leukemia	7 (1.40%)	17 (2.72%)
Hodgkin lymphoma	98 (19.64%)	74 (11.84%)
Non-Hodgkin lymphoma	24 (4.81%)	40 (6.40%)
Central nervous tumor: Glioma	25 (5.01%)	54 (8.64%)
Central nervous system tumors: non-glioma	55 (11.02%)	35 (5.60%)
Soft tissue sarcoma	49 (9.82%)	56 (8.96%)
Ewing sarcoma	19 (3.81%)	51 (8.16%)
Osteosarcoma	15 (3.01%)	40 (6.40%)
Retinoblastoma	39 (7.82%)	17 (2.72%)
Neuroblastoma	30 (6.01%)	40 (6.40%)
Wilms tumor	20 (4.01%)	27 (4.32%)
Germ cell tumor	9 (1.80%)	11 (1.76%)
Other**	9 (1.80%)	11 (1.76%)
Radiation, n (%) ∀
Yes	406 (81.36%)	298 (47.68%)	<0.001
Anthracyclines, n (%) ∀
Yes	268 (53.71%)	396 (63.36%)	0.65
Platinum Compounds, n (%) ∀
Yes	151 (30.26%)	158 (25.28%)	<0.001
Alkylating Agents, n (%) ∀
Yes	351 (70.34%)	436 (69.76%)	0.31
P/LP mutations, n (%)
Yes	72 (14.43%)	38 (6.08%)	<0.001
Subsequent neoplasms, n (%) §
Basal cell carcinoma	92 (16.00%)
Thyroid cancer	89 (15.48%)
CNS tumor: Meningioma	61 (10.61%)
Breast cancer	60 (10.43%)
Osteosarcoma	56 (9.74%)
Soft tissue sarcoma	44 (7.65%)
CNS tumor: Glioma	35 (6.09%)
Salivary gland tumors	21 (3.65%)
Melanoma	17 (2.96%)
Kidney cancer	17 (2.96%)
CNS tumors (except glioma or meningioma)	12 (2.09%)
Ewing sarcoma	11 (1.91%)
Squamous cell carcinoma	9 (1.56%)
Colorectal carcinoma	7 (1.22%)
Lung cancer	6 (1.04%)
Other subsequent neoplasms¶	38 (6.61%)

SN denotes subsequent neoplasm; P/LP denotes pathogenic/likely pathogenic; CNS denotes central nervous system

^*P-values were calculated using conditional logistic regression

^∀These comparisons were adjusted for the year of primary cancer diagnosis

^**Other primary diagnoses include liver (cases: n=2; controls: n=4), squamous cell carcinoma (cases: n=2; controls: n=0), thyroid cancer (cases: n=1; controls: n=1), other (cases: n=4; controls: n=6).

^£represents time from cancer diagnosis to first SN for cases and time from cancer diagnosis to enrollment for controls.

^§Multiple SNs per patient were considered when counting SNs; total SNs are 575 in 499 patients.

^¶Other SN includes: gynecological cancer (n=12), other (non-osteosarcoma, non-Ewing sarcoma) bone tumors (n=4), bladder cancer (n=3), acute lymphoblastic leukemia (n=2), gastrointestinal cancers (n=2), other SNs (n=15)

Seventy unique P/LP mutations in 21 CPGs were detected in 72 survivors with SNs. Forty-two unique P/LP mutations in 24 CPGs were detected in 38 survivors without SNs. The most frequently-mutated CPGs were RB1 (n=21; 4.21%) and TP53 (n=18; 3.61%) among cases, and NF1 (n=8; 1.28%) and BRCA1 (n=4; 0.64%) among controls (Figure S2). Despite the genetic heterogeneity and pleiotropic effects (Figures S3, S4, S5), a marked overrepresentation of SNs was observed among survivors of certain primary cancers with P/LP mutations vs. those without mutations; these primary cancers included retinoblastoma (95.45% vs. 52.95%, p <0.001), soft tissue sarcoma (70.59% vs. 42.05%, p=0.03), and osteosarcoma (63.64% vs. 18.18%, p=0.010) (Figure S6). The prevalence of mutations by SN type ranged from 4.35% (BCC) to 34.09% (soft tissue sarcoma) (Figure S7).

Risk of SNs associated with P/LP mutations and demographic/clinical/treatment characteristics

Factors associated with SN risk included female sex (adjusted odds ratio [aOR]=1.58, 95%CI=1.15–2.18), exposure to radiation (aOR=26.47, 95%CI=9.67–72.47) and platinum compounds (aOR=2.11, 95%CI=1.28–3.5), and presence of P/LP mutations (aOR=4.26; 95%CI=2.36–7.69) (Table 2, Figure 1). The association between P/LP mutation status and specific SN types is shown in Figure 1 and Tables S1 to S7. The odds of subsequent glioma (aOR=12.61; 95%CI=2.12–75.08), subsequent osteosarcoma (aOR=6.53; 95%CI=1.51–28.36), subsequent meningioma (aOR=4.94, 95%CI=1.002–24.36), and subsequent soft tissue sarcoma (aOR=4.59; 95%CI=1.34–15.77) were increased among P/LP mutation carriers. When compared with survivors carrying no P/LP mutations, those carrying TP53 mutations had an 18.54-fold greater odds of developing SNs (95%CI=3.86–89.08); survivors carrying non-TP53 P/LP mutations had a 2.57-fold greater odds of developing an SN (95%CI=1.45–4.55). Overall, 86% of the TP53 mutations carriers had SNs, 61% of the non-TP53 mutation carriers had SNs, while 42% of the non-P/LP mutation carriers had SNs (Figure S8). The most common SNs among survivors with TP53 mutations included subsequent osteosarcoma, glioma, breast cancer, and soft tissue sarcoma (Figure S4).

Table 2.

Association between P/LP mutations and SN

Variable	Any solid SN
	aOR (95% CI)**	p-value*
P/LP mutations
No (427 cases; 587 controls)	Ref
Yes (72 cases ; 38 controls)	4.26 (2.36 – 7.69)	<0.001
Age at diagnosis of primary cancer
Per year increase	1.01 (0.97 – 1.05)	0.66
Sex
Male	Ref
Female	1.58 (1.15 – 2.18)	0.005
Radiation
No	Ref
Yes	26.47 (9.67 – 72.47)	<0.001
Anthracyclines
No	Ref
Yes	1.3 (0.8 – 2.1)	0.29
Platinum compounds
No	Ref
Yes	2.11 (1.28 – 3.5)	0.004

Abbreviations: SN: subsequent neoplasm; P/LP: pathogenic/ likely pathogenic; aOR: adjusted odds ratio; 95%CI: 95% confidence interval

^*P-values were calculated using multivariable conditional logistic regression model adjusted for P/LP mutation status, age of primary cancer diagnosis, sex, year of primary cancer, radiation (y/n), anthracycline exposure (y/n), platinum compounds exposure (y/n) and principal components.

^**adjusted for year of primary cancer diagnosis

Figure 1. Odds of developing an SN by P/LP mutation status.

Multivariable conditional logistic regression was used to calculate the odds ratios (OR) and 95% confidence intervals (95%CIs) for the association. Both OR and 95% CIs are plotted along the x-axis in log₁₀ scale. The dotted line represents an OR of 1.

Abbreviations: SN: subsequent neoplasm; P/LP: pathogenic/ likely pathogenic

Multivariable conditional logistic regression model was created for each SN type. The clinical/demographic variables included in each model were selected based on level of significance in the univariate analysis.

We identified a significant P/LP*platinum interaction associated with increased odds of SN (p=0.047). Among those exposed to platinum compounds, P/LP mutation carriers had a 10.66-fold higher odds of developing an SN (95%CI=4.06–27.96) vs. those not carrying P/LP mutations. In contrast, among those not exposed to platinum compounds, P/LP mutation carriers had a 1.99-fold higher odds of an SN (95%CI=1.03–3.84) vs. those not carrying P/LP mutations (Figure 2). We also identified a significant P/LP*radiation interaction (p<0.001). Among those exposed to radiation, P/LP mutation carriers had a 2.75-fold higher odds of SN (95%CI=1.44–5.25) vs. those not carrying P/LP mutations. In contrast, among those not exposed to radiation, P/LP mutation carriers had a 7.21-fold higher odds of SN (95%CI=2.56–20.27) vs. those not carrying P/LP mutations (Figure 2).

Figure 2. Odds of developing an SN by P/LP mutation status and exposure to radiation or platinum compounds.

Multivariable conditional logistic regression was used to calculate the odds ratios (OR) and 95% confidence intervals (CIs) for the association. Both OR and 95% CI are plotted along the x-axis in log₁₀ scale. The dotted line represents an OR of 1.

Abbreviations: SN: subsequent neoplasm; P/LP: pathogenic/ likely pathogenic

*Multivariable conditional logistic regression was adjusted for year of primary cancer diagnosis, anthracycline/alkylating agent status and principal components.

**Multivariable conditional logistic regression was adjusted for year of primary cancer diagnosis, anthracycline/alkylating agent status and principal components.

^Multivariable conditional logistic regression was adjusted for age of primary cancer diagnosis, sex, year of primary cancer diagnosis, anthracycline, platinum and principal components.

^^ Multivariable conditional logistic regression was adjusted for age of primary cancer diagnosis, sex, year of primary cancer diagnosis, anthracycline, platinum and principal components.

In an analysis restricted to perfectly matched case-control sets (269 cases; 341 controls), the associations were essentially unchanged. P/LP mutation carriers had a 4.91-fold greater odds of developing SN (95%CI=2.46–9.83) (Figure S9). Additional factors associated with SN risk included female sex (aOR=1.5, 95%CI=1.05–2.14), anthracyclines (aOR=1.93, 95%CI=1.06–3.5), and platinum compounds (aOR=2.49, 95%CI=1.38–4.47).

Risk Classification Model

A Clinical Model that included sex, primary cancer type, year of primary cancer diagnosis, radiation, platinum compounds, and length of follow-up, yielded an AUC_clinical of 0.79 in the test set (95%CI=0.73–0.84). Addition of P/LP mutation status to the clinical model resulted in a significant improvement in model performance (AUC_combined=0.82, 95%CI=0.76–0.87, P=0.014), and ΔAUC=0.03, 95% CI=0.006–0.04 (Figure 3A and 3B). In the perfectly matched case-control set, addition of P/LP mutation status to the Clinical Model also resulted in a statistically significant improvement in model performance (AUC_combined=0.81, 95% CI=0.73–0.88 vs. AUC_clinical=0.78, 95% CI=0.70–0.86, p=0.0476).

Figure 3. Classification of SN risk in survivors of childhood cancer.

In Figure 3A, Y-axis shows the AUC estimates (95% CI) of classification models in the test data for the entire cohort. Clinical model included sex, primary cancer type, year of primary cancer diagnosis, length of follow-up, exposure to radiation and platinum compounds. In Figure 3B, Y-axis shows the delta AUC estimates (95% CI).

Abbreviations: SN: subsequent neoplasm; P/LP: pathogenic/ likely pathogenic

Risk scores (median=6; IQR=4–8) were assigned to each patient using the Clinical Model in the training set (Table S8) and applied to the test set. Study participants were classified into low-risk (score <4) and moderate-to-high-risk (score: ≥4) groups. Clinical/demographic/therapeutic characteristics of the low-risk and moderate-to-high risk groups are in Tables S9 and S10. Overall, 22% of the study participants were classified as being at low risk of developing SNs (n=247), while 78% were classified as being at a high risk (n=877) (Table S11 and Figure 4A). While 12.55% of the low-risk group developed SNs (n=31), 53.36% of the moderate-to-high risk group developed SNs (n=468). The prevalence of P/LP mutations in the low-risk group was 6.1%, with no difference in the prevalence between cases and controls (p=0.93). In contrast, the prevalence of P/LP mutations in the moderate-to-high risk group was 10.8%, and the prevalence was higher among cases (14.96%) vs. controls (6.11%) (p <0.001). Most importantly, 86.36% of all P/LP mutation carriers partitioned into the moderate-to-high risk group and included all TP53 and RB1 mutation carriers (Table S12 and Figure 4B). Among moderate-to-high risk individuals with P/LP mutations, 73.7% developed SNs. An online clinical risk calculator (https://subsequent-neoplasms.shinyapps.io/risk-calculator/) incorporating sex, primary cancer type, year of primary cancer diagnosis, and exposure to radiation, and platinum chemotherapy was created to identify childhood cancer survivors at risk for solid SNs.

Figure 4A. Distribution of childhood cancer survivors in the study by clinical risk group.

Abbreviation: SN: subsequent neoplasm ; P/LP: pathogenic/ likely pathogenic

Figure 4B. Proportion of patients with P/LP mutations by clinical risk group – entire cohort.

Abbreviation: SN: subsequent neoplasm ; P/LP: pathogenic/ likely pathogenic

DISCUSSION

We found that childhood cancer survivors with P/LP mutations in CPGs carried a 4.3-fold higher odds of developing an SN. Additional risk factors included female sex, and exposure to radiation and platinum compounds. The P/LP-SN association was greater among those exposed to platinum compounds vs. among platinum-naïve survivors. In contrast, the P/LP-SN association was greater among those not exposed to radiation than among those exposed. Addition of P/LP mutation status to a clinical risk classifier significantly improved the performance of the model. A clinical risk score and an online risk calculator were generated to identify those at low and moderate-to-high risk of SNs based on demographic, clinical, and therapeutic characteristics. Over 86% of all P/LP mutation carriers partitioned into the moderate-to-high risk group and included all TP53 and RB1 mutation carriers. This clinical risk classifier could aid clinicians in decisions regarding further genetic screening and subsequent risk-based intervention strategies.

A St. Jude Lifetime cohort study (SJLIFE) of 439 survivors with SNs⁸ reported a 1.8-fold higher risk of SN among those with P/LP mutations. In comparison, the current report of 499 survivors with SNs found a 4.3-fold greater odds of SNs among those with P/LP mutations. The observed difference likely stems from differences in patient population characteristics. The majority of SNs in the SJLIFE cohort consisted of BCC (51%) and meningioma (21%), with few subsequent gliomas (<0.1%), osteosarcoma (0.5%), or soft tissue sarcoma (0.9%).¹⁹ The longer interval between primary cancer diagnosis and study enrollment in the SJLIFE cohort (median 28y) vs. the current study (13y) is the likely reason why SNs captured in the SJLIFE cohort represent late-occurring and non-lethal events (meningiomas and BCC), thus possibly resulting in an underestimation of the association between P/LP mutations and SNs that have a shorter latency and poorer prognosis (glioma, sarcoma) and are more likely to carry P/LP mutations.²⁰

We found that survivors carrying TP53 mutations were at an 18.5-fold higher odds of developing an SN. Loss of function germline genetic variation in TP53 results in a hereditary cancer syndrome – the Li-Fraumeni syndrome;²¹ ~50% of individuals develop cancer by age 30y²², including osteosarcoma, adrenocortical carcinoma, central nervous system tumors, breast cancer and soft tissue sarcoma²³. Along these lines, most of the TP53 mutation carriers in the current study developed subsequent CNS tumors, osteosarcoma, soft tissue sarcoma or breast cancer.

We make a novel observation of a significant P/LP*platinum interaction, resulting in a greater magnitude of association between P/LP mutation status and SN risk among platinum-exposed survivors vs. those not exposed. Platinum-induced damage is repaired by nucleotide excision repair and the long-term persistence of platinum compounds in blood in combination with mutations in DNA repair genes such as TP53 could be the likely mechanism for the increased SN risk.^24–26

Similar to the SJLIFE study⁸, we found that the magnitude of association between P/LP mutations and SNs was greater among non-irradiated survivors than among irradiated survivors. Given that radiation is a major risk factor for solid SNs, the relative contribution of P/LP mutations may be lower among radiation-exposed survivors. In contrast, in the absence of radiation exposure, the association between P/LP mutation status and SN emerges as a strong risk factor.

Identification of P/LP mutations in CPGs has been used to risk-stratify management of primary cancer in the general population.^7,9 Identifying childhood cancer survivors at greater risk for SNs may inform a targeted approach to intensified surveillance and early detection of SNs, thus potentially reducing morbidity and mortality.²⁰ However, among childhood cancer survivors, a clear guidance as to who should be referred for genetic screening is lacking. The current COG guidelines²⁷ have identified clinical scenarios where assessment for cancer predisposition should be considered, including certain primary neoplasms (list in Supplemental Data), bilateral primary neoplasms, a family history of cancer in a first degree relative, cancer predisposition syndrome in a relative, or consanguinity; however, these scenarios are relatively rare and do not take therapeutic exposures into consideration. In the current report, we used demographic/clinical/therapeutic predictors to classify childhood cancer survivors at low risk and moderate-to-high risk of SNs and showed that over 85% of the P/LP mutation carriers (and all RB1 mutation and TP53 mutation carriers) were part of the moderate-to-high risk group. These findings suggest that childhood cancer survivors identified to be at moderate-to-high risk of SNs based on the clinical classifier could be referred for genetic counseling to discuss potential risks and benefits of genetic testing and then use the mutation status to generate a personalized screening roadmap. We have created an online clinical risk calculator to facilitate identification of childhood cancer survivors at moderate-to-high risk of solid SNs and consideration for referral for genetic counseling.

The current report needs to be placed in the context of its limitations. First, the prevalent case-control design of the study necessitated participation by only those alive at time of enrollment in order to provide a biological specimen for germline DNA. The downstream effect of this study design is that P/LP mutations associated with greater mortality from primary cancer and/or SN may be under-represented. Second, there was no mechanism to document participation rate or collect family history. Third, we did not validate mutations using deep-read sequencing, thus running the risk of over-calling mutations. However, the prevalence of mutations in the cancer controls was similar to a prior study.⁸

These limitations notwithstanding, we make several novel observations. We report a statistically-significant interaction between platinum exposure and P/LP mutations in increasing the risk of SNs in childhood cancer survivors. We demonstrate that adding P/LP mutation status leads to a statistically significant improvement in the performance of risk classification. Finally, we propose a phased approach to genetic testing of childhood cancer survivors using readily available clinical characteristics. These findings can be used in personalizing risk-based surveillance of childhood cancer survivors.

Key objective:

Develop a clinical risk classifier to identify childhood cancer survivors at high risk for subsequent neoplasms (SNs) based on their clinical, demographic, and treatment characteristics and determine the utility of the clinical risk classifier in aiding clinicians in decisions regarding screening for pathogenic/likely pathogenic (P/LP) mutations in cancer predisposition genes to further refine risk for SNs.

Knowledge generated:

Using the clinical risk classifier, we were able to classify survivors at low-risk (22%) and moderate-to-high-risk (78%) of developing SNs. Overall, 86.4% of the survivors with any P/LP mutations, and all TP53 and RB1 mutation carriers were partitioned into the moderate-to-high risk group.

Relevance (written by Jonathan Friedberg):

Using a classifier including female sex, exposure to radiation and platinum compounds, and presence of P/LP mutations enables enhanced risk prediction for SNs. These results should enable a precision approach to survivorship care.