Diagnostic yield and clinical impact of exome sequencing in early-onset scoliosis (EOS)

Abstract

Background:

Early-onset scoliosis (EOS), defined by an onset age of scoliosis less than 10 years, conveys significant health risk to affected children. Identification of the molecular etiology underlying EOS patients could provide valuable information for both clinical management and prenatal screening.

Methods:

In this study, we consecutively recruited a cohort of 447 Chinese patients with operative EOS. We performed exome sequencing (ES) screening on these individuals and their available family members (totaling 670 subjects). Another cohort of 13 idiopathic EOS patients from the United States (US) who underwent ES was also recruited.

Results:

After ES data processing and variant interpretation, we detected molecular diagnostic variants in 92 out of 447 (20.6%) Chinese EOS patients, including 8 patients with molecular confirmation of their clinical diagnosis and 84 patients with molecular diagnosis of previously unrecognized diseases underlying scoliosis. One out of thirteen idiopathic EOS patients from the US cohort was molecularly diagnosed. The age at presentation, the number of organ systems involved, and the Cobb angle were the three top features predictive of a molecular diagnosis.

Conclusion:

ES enabled the molecular diagnosis/classification of patients with EOS. Specific clinical features/feature pairs are able to indicate the likelihood of gaining a molecular diagnosis through ES.

Introduction

Early-onset scoliosis (EOS), defined as clinical presentation of scoliosis before 10 years of age, can be associated with progressive pulmonary compromise and severe dysmorphic gross skeletal appearance if left untreated. The deformed spine results in impaired pulmonary function and impairment or inability to perform activities of daily life, which is an extremely debilitating type of disease in the pediatric age group [1].

The etiology of EOS includes congenital scoliosis (CS) due to structural spine defects, neuromuscular scoliosis (NMS), syndromic condition, and idiopathic EOS (IS) [2 3]. However, it is difficult to provide a definite etiological classification or diagnosis solely based on clinical evaluation, especially for syndromic EOS with variable expressivity, such as Ehlers–Danlos syndrome [4] and neurofibromatosis [5], and other neuromuscular conditions like Charcot-Marie-Tooth disease that can have an age-dependent penetrance[6]. Identification of the molecular etiology underlying EOS patients could provide valuable information for both clinical management and prenatal screening.

In previous studies, we identified that TBX6 compound variants, a rare loss-of-function (LoF) allele and a common variant noncoding haplotype, could explain around 9.6% of CS cases [7–9]. In addition to TBX6, variants in DDR2 [10] (Spondylometaepiphyseal dysplasia, MIM: 271665), FLNB [11] (Spondylocarpotarsal synostosis syndrome, MIM: 272460), RUNX2 [12] (Cleidocranial dysplasia, MIM:119600) and numerous other genes are thus far known to cause early-onset scoliotic phenotypes. Moreover, for other diseases and loci, only certain alleles or genotypes may be more likely to cause the EOS phenotype (e.g. homozygous CMT1A duplication and heterozygous CMT1A triplication)[6].

In this study, we aim to investigate the diagnostic performance of ES in EOS patients, and to dissect the genetic architecture of the clinical entity of EOS. We report the molecular diagnostic rate by ES among two cohorts: 1) a Chinese cohort ascertained and studies in Beijing and consisting of 447 clinically diagnosed EOS patients (encompassing 670 subjects) who underwent orthopedic surgery and 2) a Texas (United States) cohort that consisted of 13 patients with idiopathic EOS. The potential association between clinical phenotypic features and molecular diagnostic status of the patients was also investigated.

Methods

Cohort collections

Chinese cohort-

In the Chinese cohort, we consecutively recruited 447 EOS patients of Chinese Han ethnicity who underwent spinal surgery at Peking Union Medical College Hospital (PUMCH) from 2009 to 2016, as a part of the Deciphering Disorders Involving Scoliosis and COmorbidities (DISCO) study (http://www.discostudy.org/). Samples on unrelated probands and closest relatives (kinships) for 105 patients were available, including six parents (male and female) also affected by scoliosis, presenting a potential dominant disease trait pattern (Figure S1). The remaining patients/research subjects, who denied a family history, were counted as sporadic cases. Physical examination, X-ray, computed tomography (CT), and magnetic resonance imaging (MRI) were performed on each patient to give a prior clinical diagnosis [2]: (1) CS (CS I, vertebral malformations, CS II, segmentation defects, CS III, mixed type); (2) NMS; (3) Syndromic EOS; and (4) IS (Supplementary Methods).

United States Texas cohort –

A US population-based cohort of 13 patients specifically diagnosed as idiopathic EOS (either operated or non-operated) from Texas Scottish Rite Hospital for Children in Dallas (TSRH) was also recruited [13]. Clinical details for each subject are provided in Table S1.

Exome sequencing, data processing and analysis

Chinese cohort-

For the Chinese cohort, ES was performed on DNA extracted from peripheral blood from 342 assumed sporadic singleton cases and 105 cases along with their family members, totaling 670 subjects (Table S2). The detailed sequencing process and in-house developed Peking Union Medical College Hospital Pipeline (PUMP) [14] are described in the Supplementary Methods.

US Texas cohort-

For the US cohort, ES of 42 samples from 13 EOS families were performed at the Human Genome Sequencing Center (HGSC) at Baylor College of Medicine through the Baylor-Hopkins Center for Mendelian Genomics initiative. Detailed sequencing methods are described in the Supplementary Methods.

Variant interpretation

For both cohorts, interpretation of single nucleotide variants (SNVs) and insertions/deletions (indel variant alleles) was adapted from the American College of Medical Genetics and Genomics (ACMG) guidelines [15] (Supplementary Methods) and demonstrated in Figure S2. Variant pathogenicity, expected mode of disease inheritance (e.g. autosomal dominant [AD] for monoallelic and autosomal recessive [AR] for bi-allelic variants), and patient phenotype were taken into consideration. Potential clinically actionable secondary findings from 59 ACMG-recommended genes were analyzed [16].

Copy number variant (CNV) allele calling from ES data

For both cohorts, CNVs were computed from ES data using two software packages: XHMM [17] and Copy Number Inference from Exome Reads (CoNIFER) [18]. Detailed calling and analysis methods are described in Supplementary Methods and Figure S3. Identified Pathogenic CNVs were validated using comparative genomic hybridization microarray (aCGH), as described in Supplementary Methods.

Statistics

Phenotypic analysis among different diagnostic groups was performed based on the random forest algorithm [19] (Supplementary Methods). Mean comparison of relevant features was conducted using the Student’s t-test or Wilcoxon Signed Rank Test.

Results

Diagnostic findings in the Chinese cohort of operative EOS

In the Chinese cohort, 447 unrelated EOS patients of Chinese Han ethnicity were recruited, including 342 singletons and 105 cases with first-degree relative samples (Table 1). Six families have more than one generation affected by scoliosis, presenting a potential dominant AD or XL) pattern (Figure S1). Among the recruited patients, most have congenital (i.e. structural) scoliosis; NMS, syndromic conditions, and presumed IS are less common. The average age at presentation is 3.1y, and 217/447 (48.5%) are male (Table 1).

Table 1.

Constitution and overall molecular diagnostic yield of two cohorts

	Chinese cohort					US cohort
Subgroup	CS	NMS	Syndromic	IS	Overall	IS
No. patients	424	5	8	10	447	13
Age of onset, mean	3.1y	3.2y	2.8ỳ	3.1y	3.1y	1.5y
Male (%)	202/424 (47.6%)	2/5	6/8	6/10	217/447 (48.5%)	7/13
TACS rate (%)	41/424 (9.7%)	0	0	0	41/447 (9.2%)	0
SNV/indel positive rate (%)	35/424 (8.2%)	3/5	7/8	2/10	47/447 (10.5%)	1
CNV positive rate (%)	4/424 (0.9%)	0	1/7	0	5/447 (1.1%)	0
Overall diagnostic rate (%)	79/424 (18.6%)	3/5	8/8	2/10	92/447 (20.6%)	1/13

Abbreviations: CS, congenital scoliosis; NMS, neuromuscular scoliosis; NFS, neurofibromatosis; ACH, achondroplasia; IS, idiopathic scoliosis; TACS, TBX6-associated scoliosis; SNV, single nucleotide variant; CNV, copy number variation.

After ES data processing and variant interpretation, we detected molecular diagnostic variants in 92/447 (20.6%) Chinese EOS patients (Table 1), encompassing 33 disease-causing genes and 5 genomic regions (Table 2).

Table 2.

Genetic architecture and disease trait of EOS revealed by exome sequencing.

Disease trait	CS			NMS	Syndromic EOS	IS	Overall
	CS I	CS II	CS III
Autosomal dominant	RUNX2 (2)*, JAG1 (1), FLNB (1), COMP (1)	RYR1 (2)*, MYH3 (1), POGZ (1), FBN1 (1), SOX9 (1), MYH7 (1)	TCF12 (1), PTPN11 (1), COL11A1 (1), SUFU (1), MYH3 (1), CELSR1 (1), TRPV4 (2), MMP13 (1), COL5A2 (1), BMP2 (1), CHD7 (2)*	LMNA (1), RYR1 (1)	NF1 (3)*, FGFR3 (3)*	COL5A2 (1), COMP (1)	34
Autosomal recessive	PLOD1 (1), DDR2 (1), DCHS1 (1)	FLNB (1), B3GALT6 (1)	FLNB (1), HERC1 (1), COL27A1 (1)	TTN (1)	PLOD1 (1)	0	10
X-linked	FLNA (1), EBP (1)	0	ZC4H2 (1)	0	0	0	3

Abbreviations: CS I, congenital scoliosis type I, vertebral malformations; CS II, congenital scoliosis type II, segmentation defects); CS III, congenital scoliosis type II, mixed type); NMS, neuromuscular scoliosis; NFS, neurofibromatosis; ACH, achondroplasia; IS, idiopathic scoliosis. Bold: de novo changes.

^*One of two RUNX2 variants arose de novo; one of two CHD7 variants arose de novo, one of two RYR1 variants arose de novo, two of three NF1 variants arose de novo, two of three FGFR3 variants arose de novo.

In the CS group, 79/424 (18.6%) patients had identified a molecular diagnosis and potential molecular etiology, including 41 (9.7%) patients with TBX6-associated congenital scoliosis (TACS) and 38 (9.0%) patients with other pathogenic variant alleles/CNVs (Table 1). The molecular diagnostic rate was higher in the CS I (45/130, 34.6%) patient population than that of either CS II (7/41, 17%) or CS III (27/253 ,10.7%), due to the fact that TACS is predominantly observed in CS I cases (35/130, 26.9%) (Table S3). For NMS, pathogenic variants in TTN, LMNA, and RYR1 were found in three of five patients. As anticipated, eight patients with clinically diagnosed syndromic EOS, including four patients diagnosed with neurofibromatosis type 1 (NF1) three patients diagnosed with achondroplasia (ACH) and one patient with Ehlers-Danlos syndrome (EDS), were all solved by the identification of a specific variant allele in the given disease genes (NF1, FGFR3 and PLOD1) which could explain their complete phenotype (Table 2). Interestingly, one of the NF1 patients was found to have a 573 kb deletion CNV spanning the NF1 gene (Table 1), which would not be detected using a single-gene test and might be missed by clinical exome sequencing (cES) that can sometimes focus exclusively on SNV alleles. In the IS subgroup, causative variants in 2 out of 10 patients were detected (Table 1).

TBX6-associated scoliosis (TACS)

TACS, defined by a combination of a heterozygous 16p11.2 deletion/TBX6 null variant and a hypomorph allele in trans [7 8], accounts for 41/422 (9.7%) CS patients in this study, which is consistent with the result of our previous study [7–9] and recent studies of multiple cohorts [20 21].

Of the 41 patients with TACS, 31 harbored a 16p11.2 deletion CNV identified by computational CNV analysis and validated by additional experiments (Table S4), while the other 10 were carrying TBX6 truncating alleles (nonsense, frameshift, or canonical splicing site variants) (Table S4). All 41 patients carried the T-C-A (rs2289292, rs3809624, rs3809627) haplotype (frequency = 44.4%, Chinese Han population) in trans with the 16p11.2 deletion or null allele (Table S4), consistent with our previous findings regarding the compound inheritance of TACS [7–9].

From the clinical phenotypic perspective, all (41/41) of the patients presented with hemivertebra or butterfly vertebra at the lower half of the spine, which are in accordance with the reported clinical phenotypic characteristics of TACS and the animal model with Tbx6 compound inheritance [7 8 22].

Pathogenic SNVs/indels identified by ES

In addition to the contribution of TACS to diagnostics, 36 pathogenic and 21 likely pathogenic SNVs/indels in 33 genes were identified in 47 patients in the Chinese cohort (Table S5).

Variant types include missense (32/57), nonsense (6/57), frameshift (12/57), in-frame-indel (1/57) and splice site (6/57) alleles. Disease traits in the 47 patients included autosomal dominant (34/47), autosomal recessive (10/47), and X-linked (3/47) (Table 2, Figure S5). Of the 34 monoallelic autosomal dominant conditions, 12 variants arose as de novo variants as confirmed by Sanger sequencing of family trios (Figure S6); 8 variants were inherited, including 5 inherited from affected parents and 3 from unaffected asymptomatic parents without apparent parental mosaicism, implicating incomplete penetrance. This conclusion could not be made for sure due to the lack of radiographs from parents; the remaining 16 were identified in proband-only cases with unknown inheritance. In addition, dual diagnoses resulting in blended phenotypes [23] were found in 2/93 (2%) patients with pathogenic variants and molecular diagnostic findings (Table S8).

Of the 34 disease-causing genes, only 11 genes (TBX6, NF1, RUNX2, RYR1, COL5A2, PLOD1, MYH3, TRPV4, FGFR3, FLNB, CHD7) were implicated in more than one unrelated patient (Table 2), suggesting that the genetic heterogeneity underlying EOS is rather substantial. Critical biological processes such as skeletal system development, extracellular matrix organization, and ossification were identified by Gene Ontology enrichment from analysis of the 34 causative genes (Figure S7).

ES-based CNV analysis

ES-based CNV prediction was performed by computational tools and confirmed by array comparative genomic hybridization (aCGH, Supplementary Methods). Besides 16p11.2 deletions, pathogenic CNVs were identified in five patients, including two 16p13.1 duplications, one 22q11.2 deletion, one 5q35.3 deletion, and one 17q11.2 deletion (Table S6, Figure S8). NF1 likely acts as a key driver gene within the 17q11.2 deletion CNV, given the type 1 neurofibromatosis phenotype in this patient (XH821). XH821 had cafe-au-lait spots at birth, left inguinal hernia, and mild intellectual disability, a more severe phenotype compared with the other three patients with type 1 neurofibromatosis solved by NF1 single nucleotide variant alleles (Table S6). The patient with the 5q35.3 deletion, the leading cause of Sotos syndrome-1 (MIM: 117550) in reported Japanese patients [24], had congenital heart disease, macrocephaly and rapid growth, consistent with the driver gene being NSD1 (Table S6). We are unable to localize the causative gene(s) for 16p13.1 duplication and 22q11.2 deletion, but their pathogenic roles and causal relationship with scoliosis are supported by other cases from the Database of Chromosomal Imbalance and Phenotype in Humans using Ensembl Resources (DECIPHER) [25].

Clinically relevant secondary findings

Secondary findings involving the 59 ACMG-recommended genes were identified in 9 patients [16]. Relevant variants were identified in 7 genes, including BRCA1, BRCA2, APC, RET, TSC2, COL3A1, and LDLR (Table S7). Notably, RYR1 is one of the 59 ACMG-recommended genes for secondary findings, but in the present cohort was identified as a primary finding (i.e. related to the scoliosis phenotype) in three cases (Table S5). In addition, one patient (XH984) with hyperglycemia was found to carry a pathogenic GCK variant that could lead to Glucokinase-maturity-onset diabetes of the young (GCK-MODY, MIM: 125851) (Table S7).

Diagnostic yield in the US cohort of IS

For the US cohort, one (non-Hispanic white) of the thirteen patients received a molecular diagnosis of Sotos syndrome (MIM: 117550), caused by a de novo heterozygous frameshift variant: c.2051_2055TAAAGdel in NSD1 (Figure S4). This patient presented premature birth, mild developmental delay, atrial septal defect (ASD) and ventricular septal defect (VSD). Interestingly, one patient in the Chinese cohort who was firstly diagnosed with CS was also found to carry a 5q35 deletion encompassing NSD1, thus molecularly diagnosed with Sotos syndrome.

Molecular classification of EOS patients

Through identification of the pathogenic variants/CNVs and their corresponding Mendelian diseases in patients with a primary clinical diagnosis of CS/NMS/IS, 84/447 (18.8%) patients from the Chinese cohort (Figure 1) and 1/13 patients from the US IS cohort were molecularly classified as either TACS or a syndromic form of EOS. The secondary identification of TACS/syndromic EOS by molecular diagnosis indicates the efficacy of genetic testing when patients only exhibit mild phenotypes rather than the whole phenotypic spectrum of specific Mendelian syndromes, and when regular methods to ascertain a diagnosis have been exhausted. Our results also exemplify the cohort-level changes in EOS classification enabled by ES.

Figure 1. Molecular classification of early-onset scoliosis after exome sequencing.

Molecular classification for the 92 patients with molecular diagnosis in the Chinese cohort after exome sequencing. Abbreviations: ES, exome sequencing; IS idiopathic scoliosis; TACS: TBX6-associated congenital scoliosis; NMS, neuromuscular scoliosis; CS, congenital scoliosis.

Clinical features of cases with different diagnostic status/categories

In an effort to explore the potential medical indications for genetic testing in EOS patients, we selected 12 clinical features of significance (Supplementary Methods). We hypothesized that the observed pattern of phenotypic features might be used to identify individuals more likely to benefit from genomic testing. All patients were divided into three groups according to their diagnostic status and categories: the group without a current molecular diagnosis, the TACS group, and the syndromic EOS group. The ability of the 12 clinical features mentioned above to distinguish these three patient groups was analyzed using the random forest algorithm (Supplementary Methods). A mean reciprocal rank (MRR, 0–1) was determined for each clinical feature, demonstrating the degree of relevance between that feature and each of the patient groups (the more relevant, the higher the rank score) (Table S9).

As a result, the age of presentation (MRR = 0.53), the number of organ systems involved (MRR = 0.50), and the Cobb angle (MRR = 0.49) were ranked as the most distinct features among the three groups. By independently analyzing each of the three features, we found that TACS patients are associated with younger age at presentation (1.56±2.17 y, P = 0.001, Wilcoxon Signed Rank Test), involvement of fewer organ systems (0.07±0.26, P = 0.002, Wilcoxon Signed Rank Test), and smaller Cobb angle (51.41±19.85°, P = 0.032, Student’s t-test) than patients without a molecular diagnosis (Table S10, Figure 2A/2B/2C). Whereas, patients with syndromic EOS are associated with a younger age at presentation (2.50±2.52, P = 0.038, Wilcoxon Signed Rank Test) and involvement of more organ systems (0.54±0.68, P = 0.026, Wilcoxon Signed Rank Test) than the undiagnosed group (Table S10, Figure 2A/2B).

Figure 2. Phenotypic analysis of patients diagnosed by TBX6 or other genes.

(A-C) The distribution of three clinical features: number of organ systems involved, age of presentation and Cobb angle in three diagnostic groups. Y-axis indicates the probability density. Gray curve/column indicates undiagnosed patient group (n = 355); Orange curve/column indicates TBX6-associated scoliosis (TACS) patient group (n = 41); Blue curve/column indicates the syndromic EOS group (n = 51).

Discussion

Hereby, we provide evidence of the important utility of genetic testing in molecularly diagnosing and classifying a large cohort of EOS patients. By performing ES on Chinese operative EOS patients (N=447), followed by rare variant family-based genomics, variant interpretation and computational CNV analysis, a molecular diagnostic rate of 20.6% was achieved. As the major subgroup of EOS, CS patients received a diagnostic rate of 18.6%, which is higher compared to a recent genetic study on CS using both ES (2/28, 8%) and panel-based sequencing of 5 genes (10/73, 13.7%) [26]. The increased diagnostic rate observed in CS patients by ES is possibly due to the more expansive gene coverage in the exome analysis and the utility of computational CNV analysis. The positive rate for neuromuscular EOS (3/5) is consistent with a previous study on a cohort of neuromuscular disease (18/38, 47.4%) [27]. As anticipated, ES confirmed the diagnosis of syndromic EOS in 8/8 patients. In our study, we also identified the molecular basis of idiopathic EOS in 3/23 patients from both the Chinese and the US cohorts, which demonstrated the power of ES in cases that have eluded a clinical diagnosis.

One of the main obstacles for seeking a molecular diagnosis in EOS patients is the heterogeneous genetic predisposition within this broad clinical description. Indeed, 93 molecularly diagnosed patients were explained by variants in 34 genes and five distinct CNV regions (Table 2/Table S6). Because our study is driven by a core clinical endophenotype of EOS, patients in our cohort may arguably tend toward the milder end of the clinical spectrum of each molecularly diagnosed syndrome. Strong genetic heterogeneity and limited clinical phenotype expressivity brought challenges to locating the disease-causing genes and characterizing gene-specific phenotypes in the patients.

In addition to the molecular diagnosis made for individuals, ES on a large sample size in our study also provided insights into the genetic and genomic architecture of EOS. Several highly relevant biological processes were identified (Figure S7), which are instructive for both clinical ES analysis and future discovery of novel disease-associated genes.

By phenotype analysis, we identified that the Cobb angle, the number of multi-systemic defects, and the age of onset as the three top associated clinical phenotypic features that directly correlate with the likelihood of identifying a genetic etiology for CS. Thus, these may help to stratify which EOS patients are suggested to receive genetic testing and what testing should be done. Patients with younger age at presentation, fewer multi-systemic defects, and a smaller Cobb angle are more likely to be affected with TBX6 compound mutations, i.e., TACS [7–9 22]. For these patients, an aCGH analysis or clinical chromosomal microarray (CMA) CNV analysis supplemented by targeted next-generation sequencing (NGS) would be beneficial and cost-effective. In contrast, patients with younger age at presentation and more multi-systemic defects are more likely to have a molecular diagnosis in genes other than TBX6. Given the genetic heterogeneity in this patient group, exome or genome sequencing would be the first-line diagnostic tool for patients with more complex phenotypes.

The identification of a molecular diagnosis for certain genetic disorders from this study could inform clinical management and provide new clinical insights. Three pathogenic RYR1 variants were identified in three EOS patients (Table S5), resulting in a molecular diagnosis of central core disease (MIM: 117000), which increased their potential risk of developing malignant hyperthermia (MH) upon general anesthesia [28]. One of the three patients (XH696) indeed developed malignant hyperthermia during his surgery, and fortunately survived after immediate clinical response to dantrolene salvage. Under such a condition, the risk of MH could be readily avoided by alternative anesthesia if the molecular diagnosis is revealed prior to the surgery [29].

Even in patients with clinically diagnosed syndromic disorders, such as ACH and NF1, the exact molecular diagnosis, including presumed pathogenic variant allele contribution to the patient’s disease process that was revealed by ES could inform clinical management. Of the four patients with NF1, three were molecularly diagnosed with single nucleotide variants (SNV) alleles in the NF1 gene and the other was found to carry a 573 kb genomic deletion CNV spanning NF1; different alleles with potentially different clinical consequences. NF1 caused by 17q11.2 deletion is associated with early-onset of neurofibromatosis[30] accompanied with congenital heart malformation and cognitive dysfunction[31], i.e., a more severe clinical phenotype is potentially anticipated than that caused by NF1 variant alleles that were SNV mutations. Therefore, the precise molecular diagnosis has important management implications and potentially lifelong clinical (cardiovascular, skeletal, and malignancy) surveillance issues for this patient with NF. Moreover, the variant information may help contribute to family management and recurrence risk information that they seek. Clinical experience suggests up to half of NF cases may be due to new mutations. Moreover, the clinically observed phenotype of ‘segmental NF’ may potentially reflect mosaic mutations[32].

There are several limitations of our study. First, we enrolled more CS cases in the Chinese cohort, this patient selection hinders us from exploring the real diagnostic rate of other forms in EOS patients. Secondly, the use of multiple capture and sequencing platforms for ES resulted in different capture regions and exome coverage across the cohort, which might affect the identification of some small fraction of pathogenic SNVs and CNVs. Due to the small number of individual genetic loci (34 genes of about 20,000 interrogated by ES) involved in the molecular diagnoses, subtle differences in capture design, genomic sequencing or sequence analytical pipelines are not likely to affect our molecular diagnostic rate. Nevertheless, due to the small number of molecularly diagnosed Mendelian conditions by each specific gene, we are unable to perform gene-based phenotypic analysis, and thus cannot explain the mechanism of phenotypic characteristics identified from molecularly diagnosed patients.

To conclude, ES enabled the molecular classification of EOS in 84 out of 447 (18.8%) patients from the Chinese cohort and 1/13 patients from the US cohort. Specific clinical features/feature pairs are able to indicate the likelihood of gaining a molecular diagnosis through ES.