TBX6-associated congenital scoliosis (TACS) as a clinically distinguishable subtype of congenital scoliosis: further evidence supporting the compound inheritance and TBX6 gene dosage model

Abstract

Purpose

To characterize clinically measurable endophenotypes, implicating the TBX6 compound inheritance model.

Methods

Patients with congenital scoliosis (CS) from China (N=345, Cohort 1), Japan (N=142, Cohort 2), and the USA (N=10, Cohort 3) were studied. Clinically measurable endophenotypes were compared according to the TBX6 genotypes. A mouse model for Tbx6 compound inheritance [N=52] was investigated by micro-CT. A clinical diagnostic algorithm (TACScore) was developed to assist in clinical recognition of TBX6-associated CS (TACS).

Results

In Cohort 1, TACS patients [N=33] were significantly younger at onset than the remaining CS patients (P=0.02), presented with one or more hemivertebrae/butterfly vertebrae (P=4.9×10⁻⁸), and exhibited vertebral malformations involving the lower part of the spine (T8–S5, P=4.4×10⁻³); observations confirmed in two replication cohorts. Simple rib anomalies were prevalent in TACS patients (P=3.1×10⁻⁷), while intraspinal anomalies were uncommon (P=7.0×10⁻⁷). A clinically usable TACScore was developed with an AUC of 0.9 (P =1.6×10⁻¹⁵). A Tbx6^{−/mh (mild-hypomorphic)} mouse model supported that gene-dosage effect of TBX6 underlies the TACS phenotype.

Conclusion

TACS is a clinically distinguishable entity with consistent clinically measurable endophenotypes. The type and distribution of vertebral column abnormalities in TBX6/Tbx6 compound inheritance implicate subtle perturbations in gene dosage as a cause of spine developmental birth defects responsible for about 10% of CS.

INTRODUCTION

Recent progress in elucidating the genetic contributions to disease and molecular etiology of clinical phenotypes presents opportunities to further subclassify human disease traits such as scoliosis according to their underlying genetic etiologies. Such genetic/genomic subclassification can result in a virtuous cycle from bench to bedside to better understand the biological perturbations underlying disease traits and pathophysiological bases of disease.^1,2 To further increase the specificity of molecular diagnosis and the clinical application of genetic testing and clinical genomics, precise correlations between genotype and phenotype, allelic series, and the clinical consequences of combinations of biallelic variants at a locus need to be established.^3,4 The recent report of a compound inheritance model in congenital scoliosis (CS) provides a genetic biomarker with which to potentially clinically subclassify scoliosis and define a specific clinical disease entity (TACS, TBX6-Associated CS) which is associated with a particular combination of variant alleles – biallelic variants consisting of a Loss-of-Function (LoF) TBX6 lesion and a common risk hypomorphic allele in trans– as an underlying genetic etiology and further explore the hypothesized TBX6 gene dosage effect.⁵

Congenital scoliosis is a form of spinal curvature that can be caused by vertebral malformations potentially resulting from defects of formation, defects of segmentation, or a combination of the two.⁶ The prevalence of CS is approximately 0.5–1 per 1,000 live births.⁷ As a major contributor to childhood and adolescent disability, CS affects patients’ lives and activities of daily living both physically and psychologically.⁸ CS can arise from developmental spine defects that result from perturbations in somitogenesis.⁷ Successful somitogenesis requires the spatial and temporal regulation of a complicated gene-interaction network in which TBX6 plays an essential role. Our previous work elucidated the TBX6 compound inheritance model accounts for approximately 7.9%−10.6% of sporadic CS in the Chinese population.⁵ Similar analyses of independent Japanese and European CS cohorts, in multiple patients from different world populations and genetic backgrounds, supported the proposed TBX6 compound inheritance model.^9,10

We previously showed the CS phenotype could result from compound inheritance⁵: consisting of null variant alleles at the TBX6 locus, either a deletion copy number variant (CNV) of the 16p11.2 region^11,12 or LoF induced by single nucleotide variants (SNVs) and indels in TBX6, which maps within the 16p11.2 region.¹³ With the delineation of the hypothesis of the compound inheritance and TBX6 gene dosage model,⁵ we were motivated to more precisely characterize the genotype-phenotype relationships of the specific genotypic combination of alleles at the TBX6 locus observed in CS cases with TBX6 variant alleles [i.e. the compound inheritance model] and the clinically observed endophenotypes characterizing the CS. In addition, we sought to explore the compound inheritance and gene dosage hypothesis in both humans and mice. Moreover, further investigations are required to test the potential utility of molecular diagnosis and clinical genomics in the ‘precision medicine directed’ clinical evaluation and management of CS.

MATERIALS AND METHODS

Participants

We recruited two independent cohorts and a multicenter case series from China, Japan, and the United States. The discovery set (Cohort 1) consists of unrelated sporadic CS patients of Chinese Han descent in the DISCO (Deciphering disorders Involving Scoliosis and COmorbidities, http://discostudy.org/) study from Peking Union Medical College Hospital (PUMCH) between October 2010 and January 2016, in which clinical diagnoses of CS were confirmed by radiological imaging.⁵ The first replication cohort (Cohort 2) comprised Japanese patients with CS recruited from Japan, in compliance with the selection criteria as reported previously.¹⁰ The second replication case series (Cohort 3) comprised patients with 16p11.2 deletion/TBX6 LoF variants collected from the United States. Genomic DNA was extracted from the peripheral blood or saliva samples. Informed consent was obtained from each participant or the corresponding guardian. The study was approved by the institutional review boards of PUMCH, RIKEN, Baylor College of Medicine, and the other participating hospitals.

Genetic Analyses

A genome-wide copy number variant analysis was performed in 20 sporadic CS patients in Cohort 1.⁵ Quantitative polymerase chain reaction (qPCR) analysis was conducted to screen for 16p11.2 deletions in the remaining 325 patients from Cohort 1 and in the 142 patients from Cohort 2. An orthogonal experimental approach to measure copy number alteration, customized comparative genomic hybridization (CGH) microarrays or digital droplet PCR (ddPCR), was used to independently confirm the deletion CNV candidates. The 16p11.2 deletion case series (Cohort 3) was identified by chromosomal microarray analysis (CMA) (Table S1).^12,14,15

The entire TBX6 gene and its upstream region were amplified and analyzed by Sanger sequencing in all patients from Cohort 1⁵ and 121 patients from Cohort 2¹⁰. TBX6 variants were also detected by exome sequencing (ES) and verified by Sanger sequencing in the remaining 21 patients from Cohort 2 and one patient (BH8084) from Cohort 3 (Table S1). Based on our previous findings,⁵ we analyzed the hypomorphic allele in patients with 16p11.2 deletions and TBX6 LoF variants.

Phenotype Evaluation

We conducted detailed phenotypic analyses of the spine, ribs and intraspinal anomalies in Cohort 1 and Cohort 2. The age of onset indicated the age at which an individual first came to medical attention, exhibiting or presenting with features such as asymmetric shoulder height or shoulder blade prominence. Morphologically, the vertebral malformations were classified as defect of formation, defect of segmentation and mixed malformations.¹⁶ Variations in the number of vertebral bodies were also described. The incidence of rib defects, along with the rib structure and the number of alterations were also analyzed. The rib defect subtypes were classified as simple or complex.¹⁷ Intraspinal anomalies were defined as any defect involving the spinal cord, such as a tethered cord or syringomyelia.

Generation and Evaluation of the Tbx6 Gene Edited Mice

Zygotes from FVB mice were edited by CRISPR/Cas9¹⁸ to generate a Tbx6 frameshift allele and in independent experiments a Tbx6 hypomorphic allele.¹⁹ Matings of strains with different Tbx6 genotypes were performed to derive the Tbx6 locus biallelic variant mouse model with a Tbx6 LoF variant in exon 2 in trans with the hypomorphic haplotype in the T binding site in the promoter of Tbx6 on the second allele (Tbx6^{wild-type (wt)/-} X Tbx6^{wt/mild-hypomorphic(mh)} to derive Tbx6^−/mh).¹⁹ Multiple adult animals [N = 92, 35–45 days] were evaluated by micro-CT (Bruker, Belgium) to assess the phenotypes; this included 52 with compound inheritance, Tbx6^−/mh. No randomization was used and no blinding was done in the animal study.

Development of the TACS Predictive Model

We developed a multivariate model of the risk score (TACScore) to clinically predict TACS. The TACScore was derived from Cohort 1 (training dataset) by logistic regression²⁰, in which all reliable variables that were significantly associated (P<0.05) with TACS were entered into a multivariate model to discover the optimal predictors for distinguishing TACS from all the enrolled subjects with CS using binary logistic regression analysis. A points system was developed for making these complex statistical models useful to clinicians and to evaluate the clinical utility of the TACScore, in which the point of each predictor was assigned according to the product of the corresponding β coefficient and value of the predictors with point totals corresponding to the risk estimate.²⁰ The receiver operating characteristic (ROC) was used to assess the effectiveness. Cutoff points for the TACScore were determined with the Youden index (J=maxc[sensitivity+specificity-1]²¹) to define the maximum potential effectiveness of the predictive models. Then the TACScore was validated independently in Cohort 2 (testing dataset).

Statistical Analysis

The Mann-Whitney U test was used to analyze the ages, the numbers of vertebral and rib malformations. The sex association and prevalence of different types of vertebral malformations were compared using the Pearson χ² test or the Fisher exact test. The odds ratio (OR) with 95% confidence interval (CI) was used to assess the influences of TBX6 LoF variants on vertebral, rib and intraspinal malformations. Statistical analyses were performed with SPSS version 15.0 (SPSS, USA). P<0.05 was considered statistically significant. The variance was similar between the groups that are statistically compared. (Methodological details are provided in the Supplementary Materials.)

RESULTS

We have enrolled 345 Chinese patients with sporadic CS (156 males and 189 females, 12y [7–15] as the median age at enrollment [the interquartile range, IQR], 3y [1–9] at onset, Figure S1), including 237 previously reported⁵ and 108 new CS patients (Figure 1). For Cohort 2, 142 Japanese patients with CS were recruited in Japan (79 previously reported¹⁰ and 63 new patients, Figure 1). Ten patients with evidence for compound inheritance, 16p11.2 deletion/TBX6 LoF variants, who were systematically evaluated for vertebral phenotypes were enrolled in Cohort 3 (Figure 1 and Table 1). Collectively, twenty-six patients with proximal 16p11.2 deletions, five patients with frameshift variants (c.1250_1251insT, c.266_267insC, c.704_705insG, c.1169_1170insC, and c.1179_1180delAG) and two patients with stop-gain variants (c.844C>T [p.R282*], c.933C>A [p.C311*]) in the TBX6 gene were identified in Cohort 1 (Table 1, Figure 1, Figure S2–4, and Table S2). Similarly, ten 16p11.2/TBX6 deletions, two TBX6 frameshift variants (c.156delG and c.935_936insGA), one nonsense variant (c.699G>A [p.W233*]), one splice-site variant (c.119–1G>A) and one novel missense variant (c.333G>T [p.M111I]), which had been confirmed as a LoF variant,¹⁰ were identified in Cohort 2 (Figure 1, Table 1, Figure S4B and Table S2). Further haplotype analysis showed that all patients with TBX6 LoF variants had the ‘T-C-A’ risk haplotype (the co-occurrence of three common SNPs, namely, rs2289292, rs3809624 and rs3809627^5,22) in trans on the opposite allele in Cohort 1 and Cohort 2 (Table 1). Thus, the percentage of new CS patients explained by the TBX6 compound inheritance model (10/108 [9.3%] in Cohort 1 and 15/142 [10.6%] in Cohort 2) was consistent with our previous findings (23/237 [9.7%]).⁵ In Cohort 3, ten patients with TBX6 deleterious variants (nine 16p11.2 deletions and one frameshift variant [c.469_470insCGGC, p.R157fs], Figure S4C, Table 1) were identified. Furthermore, we identified thirteen patients with fourteen TBX6 missense variants and one patient with an in-frame insertion variant from Cohort 1; these 14 subjects were excluded from the in-depth systematic phenotypic and clinical endophenotypic data analyses described below as the functional effects of these variants were uncertain.

Figure 1. Workflow for TBX6-Associated Congenital Scoliosis (TACS) Study Participants in the Multicenter Cohorts and the Animal Model Genotype-Phenotype Analyses.

The discovery set (Cohort 1) consisted of 345 unrelated Chinese patients with sporadic CS. The status of the TBX6 variants was screened, and the clinical characteristics were reviewed and compared according to the TBX6 genotypes, including vertebral, rib, and intraspinal anomalies. To recapitulate and investigate the phenotypic consequences of the compound inheritance and TBX6/Tbx6 gene dosage genetic model, a mouse strain with a specific combination of Tbx6 alleles was constructed. Tbx6 alleles were individually engineered and a mouse strain for the compound inheritance model was constructed by mating, introducing a truncated allele in trans with a mild-hypomorphic (mh) allele; Tbx6^−/mh. Furthermore, the TBX6-associated CS score (TACScore) was developed to guide and increase the efficiency of diagnosing TACS from clinically observed endophenotypes.

Abbreviations: CS, congenital scoliosis; VUS, variants with unknown significance; LoF, loss-of-function; TACS, TBX6-associated CS; Non-TACS, non-TBX6-associated CS.

^a Cohort 1 comprised the 237 sporadic CS patients enrolled between October 2010 and June 2014 in our previous study.⁵

^b Cohort 2 comprised 79 Japanese CS patients from 94 previously described CS patients,¹⁰ after excluding 15 patients without complete clinical data.

^c No information regarding whether the risk haplotype was in trans or in cis with the TBX6 frameshift variant.

Table 1.

Genotype and Phenotype Information of TACS Patients from Multiple Centers Worldwide.

Center	Subject	Gender	Age (year)a	LoF Mutation at First Allele	Risk haplotype at Second Alleleb	Phenotype
Cohort 1 (10 newly recruited TACS patientsc)
Peking Union Medical College Hospital	XH139	M	3	16p11.2 del	T-C-A	T12 hemi (r) and T11 wedge vertebra (r); Missing left 12th rib
	XH330	M	4	16p11.2 del	T-C-A	T10 hemi (l) and T12 hemi (r); Missing right 10th and left 12th rib
	XH468	M	3	16p11.2 del	T-C-A	T8–T11 butterfly vertebrae and T12 hemi (l); Missing right 12th rib
	XH480	F	11	16p11.2 del	T-C-A	C2–C3 blocks and T10 hemi (right, unsegmented to T9&T11);  Missing left 10th rib
	XH522	M	8	16p11.2 del	T-C-A	Hemi (r) between T9–T10;  Additional rib between right 9th-10th
	XH529	M	14	16p11.2 del	T-C-A	T12 hemi (l); Missing right 12th rib
	XH605	F	11	16p11.2 del	T-C-A	T12 hemi (r) and six lumbar vertebrae; Missing left 12th rib
	XH623	M	1	16p11.2 del	T-C-A	L4 hemi (r)
	XH636	F	9	16p11.2 del	T-C-A	L1 butterfly vertebra
	XH625	M	7	c.933C>A (p.C311*)	T-C-A	T12 hemi (r) and six lumbar vertebrae; Missing left 12th rib
Cohort 2 (6 newly recruited TACS patientsd)
RIKEN and participating hospitals in Japan	A1042	F	15	16p11.2del	T-C-A	T10 hemi, L5 butterfly vertebra
	A1076	F	9	16p11.2del	T-C-A	L2 hemi, Bilateral 13th ribs
	A9022	M	7	16p11.2del	T-C-A	T10-L1 hemi
	S1275	M	14	16p11.2del	T-C-A	T4 hemi, T2–3&L4 butterfly vertebra
	S1325	F	13	16p11.2del	T-C-A	L4–5 butterfly vertebra
	A1107	F	15	c.119–1G>A	T-C-A	L3 butterfly vertebra
Cohort 3
Baylor College of Medicine	BCM01	F	Caucasian	16p11.2del	NA	T9 hemi
	BCM02	F	Hispanic	16p11.2del	T-C-A	T9, T10 and T11 hemi
	BCM03	F	Caucasian	16p11.2del	T-C-A	T10 hemi
	BCM04	M	Hispanic	16p11.2del	c.853C>Te	T8 and T9 butterfly vertebrae, Hemi between T10 and T11
Drexel University College of Medicine	BH8084	F	Caucasian	c.469_470insCGGC p.R157fs	T-C-Af	Hemi between L4 and L5
Boston Children’s Hospital	BS01	F	Caucasian	16p11.2 del	T-C-A	T6 and T8 butterfly vertebrae
	BS19	M	Asian	16p11.2 del	T-C-A	T11 hemi (l)
Washington University School of Medicine	PT04	M	Caucasian	16p11.2 del	T-C-A	T11 hemi L3 butterfly vertebra
	PT05	M	Caucasian	16p11.2 del	T-C-A	T11 vertebral anomaly
Children’s Hospital Central California	PT08	F	Hmong	16p11.2 del	T-C-A	T11 hemi

Abbreviation: CS, congenital scoliosis; TACS, TBX6-associated CS; LoF, Loss-of-Function; del, deletion; hemi, hemivertebra; hemi (l), hemivertebra on the left side; hemi (r), hemivertebra on the right side; C, cervical vertebra; T, thoracic vertebra; L, lumbar vertebra; NA, not available.

^aAge at the time of enrollment.

^bThe risk haplotype is defined by three TBX6 SNPs (wild-type/mutant): rs2289292 (C/T) - rs3809624 (T/C) - rs3809627 (C/A).^5,22

^cThe detailed genotypes and phenotypes of the remaining 23 TACS patients in Cohort 1 were reported previously⁵ and are listed in supplementary Table S1.

^dThe detailed genotypes and phenotypes of the remaining 9 TACS patients in Cohort 2 were reported previously¹⁰ and are listed in supplementary Table S1.

^eThe variant is novel in European ancestry, and is predicted to be deleterious (Polyphen: possibly_damaging, SIFT: deleterious).

^fNo information regarding whether the risk haplotype was in trans or in cis with the TBX6 frameshift mutation.

Distinct Endophenotype of Patients with TACS

We divided patients in Cohort 1 into two groups based on the TBX6 genotype. The 33 CS patients containing TBX6 LoF variants and the in trans ‘T-C-A’ risk haplotype were classified as those with compound inheritance at TBX6 and their form of CS as TACS, and the remaining 298 patients were classified as non-TACS (Figure 1).

The TACS patients were significantly younger than the non-TACS at the age of onset (TACS, 2y [1–3]; non-TACS, 3y [1–9]; P=0.02 by the Mann-Whitney U test; Table 2). A higher proportion of male than female was found in the TACS group in Cohort 1 (TACS, 21/33 [63.6%]; non-TACS, 128/298 [43.0%]; P=0.03 by the Pearson χ² test; Table 2). Overall, the vertebral anomalies were less complex in the TACS group as defined by the number of vertebral malformations (TACS, 2 [1–2]; non-TACS, 4 [2–6]; P=2.8×10⁻⁹; Table 2 and Figure 2). Remarkably, all spinal deformities in the TACS group originated from the defect of vertebral formations, in which 29 (87.9%) had a simple type (P=9.2×10⁻¹⁴, Table 2 and Figure 2). Specifically, all TACS patients exhibited hemivertebrae or butterfly vertebrae (TACS, 33/33 [100%]; non-TACS, 171/298 [57.4%]; P=4.9×10⁻⁸; OR 1.7; 95% CI 1.6–1.9; Table 2 and Figure S6A). More segmented hemivertebrae/butterfly vertebrae (TACS, 1 [1–1.5]; non-TACS, 0 [0–1]; P=1.2×10⁻¹²; Table 2 and Figure 2) and fewer block vertebrae (TACS, 0 [0–0]; non-TACS, 0 [2–4]; P=8.8×10⁻¹⁰; Table 2 and Figure 2) were noted in the TACS group.

Table 2.

Comparison of Clinical Characteristics between TACS and Non-TACS Patients in Cohort 1.

Clinical characteristics	TACS (n=33)	Non-TACS (n=298)	P valuea	OR	95% CI
Age of onset, median (IQR), yr	2 (1–3)	3 (1–9)	0.02b
Male, No. (%)	21 (63.6)	128 (43.0)	0.03c	2.3	1.1–4.9
Vertebral malformation
Classification of origins, No. (%)
Defect of formation	29 (87.9)	69 (23.2)	9.2×10−14 c
Defect of segmentation	0 (0)	69 (23.2)
Mixed defects	4 (12.1)	160 (53.7)
Incidence of hemi-/butterfly vertebrae, No. (%)	33 (100)	171 (57.4)	4.9×10−8 c	1.7	1.6–1.9
Number of vertebral malformations, median (IQR)	2 (1–2)	4 (2–6)	2.8×10−9 b
Hemi-/butterfly vertebrae	1 (1–1.5)	1 (0–2)	0.02b
Segmented hemi-/butterfly vertebrae	1 (1–1.5)	0 (0–1)	1.2×10−12 b
Non-segmented hemi-/butterfly vertebrae	0 (0–0)	0 (0–1)	8.4×10−4 b
Block vertebrae	0 (0–0)	0 (2–4)	8.8×10−10 b
Incidence of malformations at each location, No. (%)
Cervical (C1–C7)	2 (6.1)	30 (10.1)	0.8c	0.6	0.1–2.5
Thoracic (T1–T12)	28 (84.9)	261 (87.6)	0.6c	0.8	0.3–2.2
Upper thoracic (T1–T4)	2 (6.1)	117 (39.3)	7.3×10−5 c	0.1	0–0.4
Middle thoracic (T5–T8)	6 (18.2)	181 (60.7)	3.0×10−6 c	0.1	0.1–0.4
Lower thoracic (T9–T12)	24 (72.7)	180 (60.4)	0.2c	1.8	0.8–3.9
Lumbar (L1–L5)	15 (45.5)	111(37.3)	0.4c	1.4	0.7–2.9
Sacral (S1–S5)	0 (0)	4 (1.3)	1.0c	1.0	1.0–1.0
Lower half of the spine (T8–S5)	33 (100)	247 (82.9)	4.4×10−3 c	1.2	1.1–1.3
Rib defect
Incidence, No. (%)	26 (78.8)	194 (65.1)	0.1c	2.0	0.8–4.7
Number of rib defects, median (IQR)	1 (1–1.5)	2 (0–3)	0.1b
Rib missing	1 (0–1)	0 (0–1)	4.9×10−2 b
Fused rib	0 (0–0)	0 (0–2)	2.6×10−3 b
Typed, No. (%)
Simple	25 (75.8)	88 (29.5)	3.1×10−7 c	7.5	3.2–17.2
Complex	1 (3.0)	106 (35.6)	4.2×10−5 c	0.06	0.01–0.4
Intraspinal defect
Incidence, No. (%)	1 (3.0)	131 (44.0)	7.0×10−7 c	0.04	0.05–0.3

Abbreviation: CS, congenital scoliosis; TACS, TBX6-associated CS; Non-TACS, non-TBX6-associated CS; CI, confidence interval; OR, odds ratio; IQR, interquartile range; No., number; C, cervical vertebra; T, thoracic vertebra; L, lumbar vertebra; S, sacral vertebra.

^aP <0.05 is considered significant.

^bMann-Whitney U test.

^cPearson’s chi-square test or Fisher’s exact test.

^dThe rib anomalies were divided into simple or complex types following the criteria described by Tsirikos and McMaster.¹⁷

Figure 2. Comparison of the Distribution Regarding Abnormal Vertebrae in Non-TACS and TACS Patients and in Tbx6^−/mh Mouse Compound Inheritance and Gene Dosage Model.

The X-axis shows the frequency of each malformation in each vertebra and Y-axis shows the vertebral distribution in the spine. The vertebral malformations in the non-TACS group (N=298) were normally distributed. There were 58 TACS patients in total in three worldwide cohorts (Cohort 1 from PUMCH in China, N=33; Cohort 2 from the multiple centers in Japan, N=15; Cohort 3 from multiple centers in the USA, N=10). The Tbx6^−/mh engineered mice [N=52] exhibited a distinct phenotype, namely, a defect of formation involving the lower part of the spine. In addition, malformations in the upper and middle thoracic spine were significantly less involved (P =7.3×10⁻⁵ and 3.0×10⁻⁶, respectively) in the TACS group than in the non-TACS group.

Abbreviation: CS, congenital scoliosis; TACS, TBX6-associated CS; Non-TACS, non-TBX6-associated CS; mh, mild-hypomorphic; C, cervical vertebra; T, thoracic vertebra; L, lumbar vertebra; S, sacral vertebra.

Of note, the vertebral segment location and spine level distribution regarding the type of abnormal vertebrae in the TACS group were distinct from that of the non-TACS group (Figure 2). In the TACS group, the lower part of the spine (T8–S5) was more frequently involved (TACS, 33/33 [100%]; non-TACS, 247/298 [82.9%]; P=4.4×10⁻³; Table 2 and Figure 2). While the incidence of rib anomalies was comparable between the two groups, more simplified rib anomalies (e.g. a localized fusion of two ribs or an absence of one rib) versus complex abnormalities (e.g. multiple extensive rib fusions or adjacent large chest wall defects) were more frequent in the TACS patients (TACS, 25/33 [75.8%]; non-TACS, 88/298 [29.5%]; P=3.1×10⁻⁷; Table 2 and Figure S6B). Furthermore, fewer intraspinal malformations (e.g. tethered cords or syringomyelia) were observed in the TACS group (TACS, 1/33 [3.0%]; non-TACS, 131/298 [44.0%]; P=7.0×10⁻⁷; Table 2).

Tbx6 Gene Edited Mouse Model for Compound Inheritance

We engineered the Tbx6 LoF variant (Tbx6⁻) by introducing a 1-bp insertion in exon 2. The Tbx6^mh allele was generated as a ‘functional equivalent’ to the human mild hypomorphic allele (Figure S7).¹⁹ Gene expression in vitro was down-regulated by the Tbx6^mh mutant to approximately 65% of the wild-type gene, which was close to the 70% dosage level of the human TBX6 mild-hypomorphic allele (Figure S7C).¹⁹ These engineered Tbx6 alleles were used to construct strains with different genotypic combinations.

As predicted, and consistent with literature observations, no homozygotes for the LoF variant were identified in liveborn animals and expected Mendelian ratios of particular genotypic combinations were distorted in liveborns;¹⁹ consistent with embryonic lethality of the Tbx6^−/− null animals.²³ We acquired through genetic matings and phenotypically assessed mice with five specific genotypes: Tbx6^wt/wt (N=10; wt, wild-type), Tbx6^wt/- (N=10), Tbx6^wt/mh (N=10), Tbx6^mh/mh (N=10), and Tbx6^−/mh (N=52). Consistently, only the Tbx6^−/mh mice exhibited vertebral malformations (48/52 [92.3%], P=9.3×10⁻⁹ [OR, 13.0; 95% CI, 5.1–33.3], Figure S8 and Table S3). Intriguingly, as observed in human TACS patients, all Tbx6^−/mh mice had vertebral malformations involving the lower part of the spine (Figure 2 and Table S3). Defects of vertebral column formation were present in most of the Tbx6^−/mh mice (36/52 [69.2%], P=4.9×10⁻⁵ [OR, 3.3; 95% CI, 2.2–4.9], Figure 2 & S8 and Table S3). The recapitulation of the type, extent, and distribution of vertebral malformations in the engineered compound inheritance model in mice (Figure 2) further supports the compound inheritance and gene dosage model for TACS and implicates biological perturbations in vertebral column malformations in this type of CS.

Worldwide Multicenter Replications

For the first replication cohort (Cohort 2), we recruited 142 CS patients from Japan and identified fifteen TACS patients (Figure 1, Table 1 and Table S2). As expected, all TACS patients had one or more hemivertebrae/butterfly vertebrae (TACS in Cohort 2, 15/15 [100%]; non-TACS, 99/127 [78.0%]; P=0.04 [OR, 1.3; 95% CI, 1.2–1.4]). All TACS patients in Cohort 2 exhibited malformed vertebrae at the lower part of the spine (T8–S5), which is consistent with the patterns observed in Cohort 1 (Table 1, Table S2 and Figure 2). This pattern, of vertebral malformations and distribution of defects in the vertebral column, was observed only in the mice with compound inheritance and gene dosage perturbations below that of haploinsufficiency (i.e. Tbx6^wt/-), but not equivalent to Tbx6^−/− combination of alleles.

Furthermore, we collected another replication case series (Cohort 3) containing 10 TACS patients in the United State. Intriguingly, hemivertebrae/butterfly vertebrae involving the lower half of the spine were again observed. The replication of our observations for this distinct genotypic combination in two additional independent cohorts from distinct world populations provide substantial evidence in support of the compound inheritance and TBX6 gene dosage model and speaks to the universality of TACS and the compound inheritance model in human and medical genetics.

TACScore to Clinically Predict TACS

We developed a model to predict TACS from the phenotypic data, and clinically measurable endophenotypes, obtained from Cohort 1 by logistic regression.²⁰ The final multivariate risk model, the TACScore, integrated variables including (1) segmented hemivertebrae/butterfly vertebrae involving the lower half of the spine (T8–S5), (2) the number of vertebral malformations, (3) the presence of intraspinal defects and (4) the type of rib defect (Figure 3A and Table S4). The calculated score of each variant parameter was linearly correlated with the corresponding risk in the regression model and defined from −8 to 4 (Figure 3B). The cutoff point was selected as ≥3 to achieve the highest Youden index (J=84.8%), with sensitivity, specificity, and accuracy levels higher than 90% (Table S5). The area under the curve (AUC) of the ROC curve was 0.9 (P=1.6×10⁻¹⁵; 95% CI, 0.9–1.0; Figure 3C). Importantly, we validated the TACScore in Cohort 2 with an AUC of 0.8 (P=1.5×10⁻⁴; 95% CI, 0.7–0.9; Figure 3C and Table S6). Therefore, we introduced a clinical assessment pipeline for CS to efficiently identify high-risk patients for TACS and potential compound inheritance at the TBX6 locus (Figure 3D).

Figure 3. Developing and Validation of the Risk Prediction Model and Diagnostic Pipeline of TACS. (A) TACScore: risk predictive model for TACS.

The final multivariate risk model was developed through a binary logistic regression analysis of detailed phenotypic data from Cohort 1. (B) The predictive efficacy of the TACScore. The X-axis shows the spectrum of TACScore and Y-axis shows the predicted TACS frequency and the percentage of TACS patients in all CS patients with each calculated score in Cohort 1 and Cohort 2. The TACScore presented excellent predictive efficacy by comparing the predicted TACS risk to the real TACS frequency. The cutoff point was selected as 3 to achieve the highest accuracy. (C) ROC Curve for the TACScore in Cohort 1 and Cohort 2. AUCs were 0.9 (P=1.6×10⁻¹⁵; 95% CI, 0.9–1.0) for the discovery cohort (Cohort 1) and 0.8 (P=1.5×10⁻⁴; 95% CI, 0.7–0.9) for the validation cohort (Cohort 2). (D) A proposed guideline for predicting and evaluating TACS. The risk of TACS evaluated by TACScore is suggested to perform prior to genetic testing. After the detection of TACS, a systemic evaluation, with early interventions and genetic consultation were recommended.

Abbreviation: CS, congenital scoliosis; TACS, TBX6-associated CS; ES, exome sequencing; CMA, chromosomal microarray analysis; GS, genome sequencing.

^a TBX6 compound variant contains a 16p11.2 deletion/TBX6 loss-of-function variant in the compound heterozygous configuration with the risk haplotype providing a hypomorphic variant.

^b VUS, variants with unknown significance.

^c Evidence from large-scale case-control studies, pedigree analysis, and functional studies are needed.

DISCUSSION

In this study, we expanded our knowledge of the contribution of the TBX6 compound inheritance to CS by providing international patient cohort studies and animal model-based evidence that: i) 9.6% (33/345) and 10.6% (15/142) of the patients in Cohort 1 and Cohort 2, respectively, could be parsimoniously explained by the TBX6 compound inheritance gene dosage model (Table 1 and Figure 1); ii) all patients in the three cohorts had the risk haplotype in trans with a TBX6 LoF variant, with the exception of one patient [BCM01] for whom haplotype information was not available and one patient [BH8084] for whom allelic phasing of the identified variants was not available (Table 1); and iii) only the Tbx6^−/mh genotype presented with vertebral malformations in the gene-edited mice (Figure 2 & S8 and Table S3). Moreover, we elucidated a novel genetically and phenotypically defined disease entity, which carries a TBX6 LoF variant in trans with the hypomorphic allele and manifests a distinguishable constellation of clinical features and CS endophenotypes. Furthermore, we developed the TACScore system and proposed a clinical practice algorithm (Figure 3) for evaluating CS and predicting TACS. Finally, we show qualitatively the same kind (hemivertebrae/butterfly vertebrae) and quantitatively the same distribution (lower spine) of vertebral malformations with the compound inheritance combination of variant alleles and TBX6/Tbx6 gene dosage model in human and mice.

TBX6 is a member of the T-box family.²⁴ Variants in Tbx6 lead to phenotypes of vertebral and rib defects in mice.^23,25 Tbx6 is important for cell fate in the paraxial mesoderm structure²⁶ and formation of the segmental boundary²⁷, which is the precursor of the vertebral column.²⁸ Therefore, the genetic and developmental malfunction in Tbx6 perturbation of gene dosage model may parsimoniously explains the hemivertebrae/butterfly vertebrae presenting in all TACS patients (Table 1 and Figure 2). Interestingly, Tbx6^−/− produces an embryonic lethal phenotype with expanded tailbud and lack of posterior somites,²³ suggesting that Tbx6 functions to maintain the specification of the posterior paraxial mesoderm.⁷ In zebrafish, tbx6 is required in the posterior of the embryo during the posterior part of primary neurulation and all of the secondary neurulation.^29,30 Therefore, this developmental process optimally explains the observation that all TACS patients and Tbx6^−/mh gene-edited mice show vertebral malformations anatomically localized at the lower part of the spine.

TACS does not follow ‘conventional’ rare variant Mendelian inheritance expectations likely reflecting the embryonic lethality of Tbx6/TBX6 homozygous null alleles. Instead, TACS is caused by biallelic variants at a locus, consistent with autosomal recessive trait manifestation, but the presence of a single rare LoF variant in trans with one common hypomorphic allele; a gene dosage that is less than haploinsufficiency but not zero as with homozygous null alleles. This genetic model (compound inheritance and TACS) may provide an explanation for the scenarios when monoallelic variants and phenotypes are perceived to be following a dominant inheritance pattern but for a disease trait with incomplete penetrance or the disease pattern observed is pseudodominance as found when the carrier state occurs at a high frequency in the population. Because the ‘T-C-A’ haplotype is common worldwide (44% among Asians and 33% among Europeans, but <1% among Africans³¹), the haplotype is likely to be parsed and filtered out by the current genomic analytical pipelines. Therefore, for some single-locus genetic models, the functional effects of individual variant alleles, as well as the combination of alleles should be considered when interpreting the genetic data, and specific phenotype observed, for the ‘unsolved’ cases. Moreover, the compound inheritance model may be particularly relevant to recurrent rearrangement CNV³² loci with relatively high mutational frequency rates³³ wherein deletion CNV can result in haploinsufficiency for many gene loci (e.g. the 0.6-Mb 16p11.2 deletion CNV contains 27 annotated genes).^5,15 It is important to know the functional effects of individual variant alleles, but also the functional consequences of combinations of variant alleles at a locus.

A significantly small proportion of the TACS patients in Cohort 1 were female (Table 2, 12/33 [36.4%] in the TACS patients and 170/298 [57.0%] in the non-TACS patients in Cohort 1), which might be due to the interference of the 16p11.2 deletion and TBX6 variants in the formation of the female reproductive system.³⁴ Notably, pleiotropic effects of 16p11.2 deletion CNV include other traits related to obesity³⁵, and cognitive phenotypes including intellectual disability and psychiatric disorders^15,36, birth defects such as CAKUT (congenital anomalies of the kidney and urinary tract)³⁷, and may even be involved in other undiscovered phenotypes or diseases. Thus, TACS patients are recommended to be evaluated systematically with long-term follow-up (Figure 3D).

More importantly, from an orthopedic surgeon’s clinical perspective, hemivertebrae/butterfly vertebrae are the most remarkable features of TACS (Figure 2). Surgical intervention and treatment at an early age results in a satisfactory prognosis for the patient and ameliorates scoliosis.³⁸ For example, patient XH101 was diagnosed as having CS with a L2 hemivertebra at 3 years of age, and molecular testing indicated that she had a frameshift variant in TBX6 with the in trans ‘T-C-A’ risk haplotype. The TACS diagnosis was made by integrating the molecular testing and clinical phenotypes. We performed L2 hemivertebra resection with short segment internal fixation. We removed the instrument at 3-years post-operation when she had a balanced spine. The last follow-up, which was at 9 years post-operation, showed an almost completely normal growth of the spine and normal curvature of the spinal column (Figure S9).

To realize clinical applications from bench to bedside, we developed a TACScore prediction algorithm as an objective adjuvant clinical measure to guide clinical management for patients with CS. The TACScore was validated in two independent cohorts and further integrated into the proposed guidelines for evaluating the risk of TACS (Figure 3). Considering the possibility of phenotypic expansion,⁴ the prediction algorithm should include multiple variables associated with TACS comprehensively (i.e. endophenotypes) instead of relying on one particular phenotype (e.g. scoliosis). This clinical phenotypic assessment provides efficient information to physicians and families regarding whether the patient is suitable for 16p11.2/TBX6 variant detection or a genome-wide test. Clinical ES³⁹ or genome-wide analysis are reserved for cases with low-risk TACScores or high-risk TACScores that are negative for 16p11.2/TBX6 pathogenic variants. Furthermore, the cutoff point of the TACScore could be adjusted according to medical practice experience and continued future observations as well as the local medical resources available to achieve the maximum health economic benefits.

Conclusions

We have defined a new subtype of CS; namely TACS, containing a TBX6 LoF variant compound with a hypomorphic allele. TACS is a clinical entity defined by consistent clinically measurable endophenotypes: i.e., younger age at onset, hemivertebrae/butterfly vertebrae involving the lower part of the spine, simple rib anomaly, and fewer vertebrae and intraspinal defects. TACScore can guide clinical management and genetic and clinical genomic testing. Human and mouse studies further confirm the compound inheritance and gene dosage model, and provide insights into potential biological consequences for spine development of perturbations in TBX6/Tbx6 gene dosage and expression. Such genetic models may be important to other birth defects.