Mutational spectrum by phenotype: panel‐based NGS testing of patients with clinical suspicion of RASopathy and children with multiple café‐au‐lait macules

Children with neurofibromatosis type 1 (NF1) may exhibit an incomplete clinical presentation, making difficult to reach a clinical diagnosis. A phenotypic overlap may exist in children with other RASopathies or with other genetic conditions if only multiple café‐au‐lait macules (CALMs) are present. The syndromes that can converge in these inconclusive phenotypes have different clinical courses. In this context, an early genetic testing has been proposed to be clinically useful to manage these patients. We present the validation and implementation into diagnostics of a custom NGS panel (I2HCP, ICO‐IMPPC Hereditary Cancer Panel) for testing patients with a clinical suspicion of a RASopathy (n = 48) and children presenting multiple CALMs (n = 102). We describe the mutational spectrum and the detection rates identified in these two groups of individuals. We identified pathogenic variants in 21 out of 48 patients with clinical suspicion of RASopathy, with mutations in NF1 accounting for 10% of cases. Furthermore, we identified pathogenic mutations mainly in the NF1 gene, but also in SPRED1, in more than 50% of children with multiple CALMs, exhibiting an NF1 mutational spectrum different from a group of clinically diagnosed NF1 patients (n = 80). An NGS panel strategy for the genetic testing of these two phenotype‐defined groups outperforms previous strategies.

An NGS panel strategy for the genetic testing of these two phenotype-defined groups outperforms previous strategies. The presence of few skin CALMs at an early age is common among children 4,5 but in multiple form could also be an indication of the presence of an inherited disease. The sole presence of six or more CALMs at early childhood constitutes a high risk for having NF1 and it is one of the main clinical criteria used. [6][7][8] However, multiple CALMs could also be indicative of other monogenic diseases, such as some RASopathies or other syndromes like PTEN hamartoma tumor or Cowden syndrome, Carney Syndrome or constitutive mismatch repair deficiency (CMMRD). 4, 5,9,10 The RASopathies constitute a clinically defined group of genetic conditions caused by germline mutations in genes that encode for components of the RAS/mitogen-activated protein kinase (MAPK) pathway. In addition to NF1, RASopathies include: Noonan syndrome (NS), NS with multiple lentigines (NSML), Costello syndrome (CS), Legius syndrome (LS), cardio-facio-cutaneous (CFC) syndrome and capillary malformationarteriovenous malformation (CM-AVM). 11 Each RASopathy exhibits a particular group of clinical manifestations, but due to the common underlying RAS/MAPK pathway deregulation, many of these conditions exhibit numerous overlapping phenotypic features, especially during early childhood (reviewed in Reference 12). Overlapping clinical manifestations include cutaneous, musculoskeletal, and ocular abnormalities; craniofacial dysmorphology; cardiovascular abnormalities; neurocognitive impairment; hypotonia; and increased risk of tumor development. NF1, LS and CM-AVM are caused by loss-of-function mutations in NF1, 13 SPRED1 14 or RASA1 genes, 15 respectively. Others such as NS, NSML and CFC syndromes exhibit genetic heterogeneity and together with CS are caused by activating mutations in PTPN11, SOS1, RAF1, BRAF, KRAS, NRAS, SHOC2, BRAF, MAP2K1, MAP2K2, CBL, HRAS and RIT1. 11,[16][17][18][19] There are evidences that support other genes to be potentially associated with RASopathies (reviewed in Reference 17) such as RASA2, 20 A2ML1, 21 PPPC1B, 22 SOS2, 18 MRAS, 23 RRAS 24 and LZTR1, 18,25 although not all of them have the same gene-disease association supporting evidences. 26 At an early age, uncertain clinical diagnostics can emerge due to the presence of clinical manifestations that are common to NF1 as well as to other RASopathies or CALM-associated diseases. One example is the solely and combined presence of CALMs and skin-fold freckling in LS-and NF1-affected children. [27][28][29] Thus, children sharing a similar initial clinical presentation can bear mutations in different genes that predispose to very different clinical courses, that need to be managed in distinct ways. In this context, genetic testing can help in confirming a clinical suspicion, facilitating an early adequate surveillance. For instance, genetic testing has been recommended to confirm NF1 in children fulfilling only pigmentary features of the diagnostic criteria. 30 Current comprehensive genetic testing strategies for these inherited diseases consist of either multigene panels or whole-exome sequencing to solve problems such as genetic heterogeneity or inconclusive clinical presentations. 31 In addition, due to the power and cost-effectiveness of NGSbased strategies and the demonstrated clinical utility, 32 inconclusive but suggestive clinical presentations are increasingly being accepted for genetic testing, accounting for a significant raise in genetic tests, especially at pediatric ages. 33 On the other side of the NGS coin, there is the exponentially growing number of identified genetic variants and the problem related to their pathogenicity analysis and interpretation. To standardize this process, the The American College of Medical Genetics and Genomics/ Association for Molecular Pathology (ACMG/AMP) guidelines 34 are now widely adopted into clinical practice. However, these guidelines need to be adjusted according to the specificities of the genetic conditions being tested. In the case of RASopathies, an additional complication is the existence of diseases caused by both loss-of-function and gain-of-function mutations. Recently, the ClinGen RASopathy Expert Panel has published new guidelines to improve gene-disease association and variant interpretation for NS, NSML, CS, CFC and Noonan-like syndromes. 26,35 In this study, a new version of the I2HCP (ICO-IMPPC Hereditary Cancer Panel), a custom NGS-based diagnostic strategy, 36 was validated for its use in the molecular testing of all RASopathy-related genes. Once validated, the performance and implementation of this panel into routine diagnostics was evaluated for its use in the genetic testing of two groups of individuals sent to the diagnostics laboratory with an inconclusive clinical diagnostic: patients with a clinical suspicion of a RASopathy or children presenting multiple CALMs. We describe here a mutational spectrum for the two groups of individuals and demonstrate the appropriateness of our NGS panel in this clinical scenario.

| Subjects
All procedures performed were in accordance with the ethical standards of the IGTP Institutional Review Board, who approved this study, and with the 1964 Helsinki declaration and its later amendments.
Informed consent was obtained from all genetically diagnosed individuals who were included in the study. Genomic DNA from 259 unrelated individuals were obtained from blood lymphocytes using standard pro-

| Sample preparation and sequencing
Library preparation and enrichment were performed as previously described. 36 Briefly, sample preparation was performed following the SureSelect XT protocol for MiSeq (Illumina) and enriched with custom I2HCP v2.2 library baits, which contained 135 genes including 22 genes involved in the RAS/MAPK pathway (Table S1). A total of 16 equimolar indexed samples were pooled after capture and sequenced in a MiSeq (Illumina) with Reagent Kit v3, 2×300. For each gene, we defined the regions of interest (ROIs). 36 NGS data were processed and filtered according to the clinical indication using a custom data analysis pipeline as previously described. 36 The validation analysis of I2HCP was performed blindly using the RASopathy control group.

| Validation by Sanger sequencing
Any ROI for all genes tested with at least one base below 30× was Sanger sequenced using standard protocols (primer sequences available upon request). All pathogenic variants and variant of unknown significance (VUS) were also validated by Sanger sequencing. Human Genome Variation Society (www.hgvs.org) nomenclature guidelines were used to name the mutation at the DNA level and the predicted resulting protein.

| NF1 and SPRED1 additional mutational analysis
In NF1 control patients and in individuals with multiple CALM phenotype (or NF1-associated tumors) who tested negative with the custom I2HCP panel, NF1 and SPRED1 genes were additionally tested by multiplex ligation-dependent probe amplification (MLPA) (SALSA MLPA, MRC-Holland, P122, P081, P082 and P295 following the manufacturer's instructions). In addition, negative MLPA cases and those with inconclusive NF1 variants were further analyzed at RNA level.
The entire coding region of NF1 was amplified from cDNA in five overlapping polymerase chain reaction (PCR) fragments. PCR products were analyzed by electrophoresis and Sanger sequencing. A schematic representation of the whole testing workflow is represented in   (Table S1). These genes included those recommended by the ClinGen RASopathy Expert Panel, 26 with the addition of NF1, SPRED1 and CBL, 11 and RASA1 for its association with CM-AVM. 11 The whole validation process was performed blindly.

| Implementation of I2HCP into routine genetic testing for RASopathies and children with multiple CALMs
We used the I2HCP-RASopathy subpanel to genetically test two  (Table S1). Similarly, only genes that have been consistently related with a CALM phenotype were analyzed for children with multiple CALMs in a stepwise manner ( Figure 1,  (Table S3). The results obtained using the three-and five-algorithm combinations were exactly the same and performed slightly better than previous findings using ClinVar data. 39 The third combination included three highperformance algorithm predictors: REVEL, VEST3 and MetaSVM. In this case, the rate of concordance between all three algorithms was of 36% (16 out of 45) for NF1 data set and 59% (309 out of 521) for RASopathy data set (Tables S2 and S3), performing worse than previously described. 39 Therefore, we chose the five-algorithm combination for the interpretation of missense variants in NF1 and other RASopathy genes.
Using this algorithm, we still had a fair percentage of variants classified as VUS. To try improving their classification, we explored the possibility of using FoldX, a structure-energy-based predictor tested previously in the context of RASopathies, 37 and InterVar, a clinical semiautomatic interpreter of genetic variants based on ACMG/AMP guidelines for variant classification. 38 The information provided by FoldX was very interesting due to its complementarity, but the absence of three-dimensional (3D) protein structures for a significant number of RASopathy gene products or domains limited its use. Furthermore, a fair proportion of pathogenic or likely pathogenic variants were classified as VUS by InterVar. In our study, none of them added utility regarding variant interpretation in a routine diagnostics setting.

| Testing RASopathy-like patients
A total of 48 patients composed the RASopathy-like set. Overall, the I2HCP diagnostics strategy allowed to detect a disease-causing variant in about 44% of the cases (n = 21/48). In addition, in 12% of the individuals (n = 6), a VUS in one of the interrogated genes was identified. About 44% of the cases were negative after using the NGS RASopathy panel (n = 21) (Figure 2A, Table S4). Out of 48 cases of the RASopathy set, 16 had previously been tested for a few genes causing NS and CS, and represented patients who tested negative with previous approaches. Of those, we were able to detect a pathogenic variant in five cases (31%, 5/16) and one VUS (6%) in another individual. The remaining 32 were genetically analyzed for the first time. In this group, 16 disease-causing mutations were identified (50%) and five VUS (16%). Eleven patients (34%) tested negative for all RASopathy genes using the NGS-I2HCP panel strategy ( Figure 2B).
However, among the individuals who tested negative, we identified a deleterious PTEN mutation in two patients. In both cases, patients exhibited some clinical manifestations related to RASopathies but none of them fulfilled clinical criteria for any particular condition. Altogether, these results showed that I2HCP strategy exhibit a good diagnostic yield outperforming previous single gene-based strategies.  (Table S4). We analyzed the mutational spectrum for each specific suspected condition and for the RASopathy-like group (Figure 3, Table S4). Among patients with a suspicion of a specific condition (n = 18), five tested negative (28%), The mutational spectrum in the RASopathy-like group (Figure 3, inconclusive phenotype) was similar to the one considering the whole RASopathy set (Figure 2A) being NF1, PTPN11 and RAF1, the genes concentrating most of the disease-causing mutations identified.

Mutational spectrum by phenotype
Among the patients who tested negative in this group, there were the two individuals with a pathogenic mutation in PTEN.
If disease-causing mutations and VUS were considered together, NF1 was the gene with more genetic variants identified (n = 7) in the RASopathy-like group, even higher than PTPN11 (n = 5) ( Figure 2B) (Table S4). We were not able to perform segregation analysis for any of the VUS identified in these three patients.

| Testing children with multiple CALMs
The I2HCP NGS panel allowed the genetic testing of multiple CALM phenotypes in a stepwise manner as mentioned above (Figure 1). We analyzed NF1 and SPRED1 genes in 102 children who arrived from   Figure 4B, Table S5). Out of these 77 variants, 70 were unique in our set. The percentage of each mutation type was slightly different but in accordance to other published results. 40 As we did not test the effect of all mutations at RNA level, the slight discrepancy regarding splicing mutations could come from the fact that different types of pathogenic mutations at DNA level can cause splicing defects. 41,42 Any bias in the NF1 population analyzed would also affect the frequency of the different mutation types.
Remarkably, the different frequency of NF1 mutation types detected in the cohort of children with ≥6 CALMs (n = 102) compared to the control NF1 group was statistically significant, consisting of a lower proportion of nonsense and frameshift mutations and a higher proportion of missense variants, both, clearly pathogenic and VUS ( Figure 4B, Table S5).

| DISCUSSION
In this study, we first validated the use of the I2HCP V2.  48 In these situations, the use of a panel such as the I2HCP could be convenient due to the possibility to analyze all recommended genes at once.
The NF1 mutational spectrum identified in the group of children with ≥6 CALMs not fulfilling the NF1 NIH criteria was different from the one present in the control NF1 group ( Figure 4B). In the former group, the proportion of truncating mutations was lower than the control NF1 group. At the same time, the percentage of missense variants (both pathogenic and VUS) was higher, opening the possibility that these mutations could represent hypomorphic alleles generating incomplete or mild NF1 phenotypes. A prospective follow-up of this group of patients could provide insight into this possibility and contribute to a better genotype-phenotype understanding.
Using the I2HCP strategy in hereditary cancer patients, we previously identified patients containing a complex variation landscape in hereditary cancer genes coexisting with the disease-causing mutation. 36 This was not the case for RASopathy-like individuals in RAS/MAPK pathway genes, except in three patients in whom a missense variant was detected in the same disease-causing gene in addition to a pathogenic variant (Tables S4 and S5). All three patients presented a complex clinical presentation, although the functional impact of these two coexisting variants in the same gene is unknown and further studies would be required to reach any conclusion.
We followed the ACMG/AMP guidelines including modifications suggested by the ClinGen's RASopathy Expert Panel 35 for variant interpretation. ACMG/AMP guidelines recommend the agreement of all in silico programs tested for using them as supportive evidence. In our study, SIFT, PolyPhen, CADD, PROVEAN and MutationTaster had the highest ratio of true concordance when they were combined, seeming the most suitable for the assessment of RASopathy-related variants. In any case, it was also necessary to perform a manual curation, in addition to in silico predicting algorithms, along the process of variant interpretation.
We also explored the possibility of using FoldX and InterVar for variant classification but, in our study, none of them added utility regarding variant interpretation in a genetic testing context. For the whole set of 71 missense variants detected in both sets of patients, only 25 of them (35%) had an accurate 3D protein structure containing the variant we wanted to evaluate (Table S6). In addition, 41% of pathogenic or likely pathogenic variants classified following adopted guidelines were classified as VUS by InterVar, exhibiting a poor correlation for its use.
We have validated the use of the custom I2HCP NGS panel for the routine genetic diagnostics of RASopathy-related genes. We have shown the utility of this strategy for testing children with an inconclusive clinical suspicion of a RASopathy and children with multiple CALMs. An NGS panel strategy outperforms previous testing strategies for these two phenotype-defined groups. We identified pathogenic variants in the NF1 gene in several inconclusive RASopathy cases. In addition, we identified pathogenic variants in NF1 and SPRED1 in more than 50% of children with ≥6 CALMs, exhibiting an