Importance of copy number variants in childhood apraxia of speech and other speech sound disorders

Abstract

Childhood apraxia of speech (CAS) is a severe and rare form of speech sound disorder (SSD). CAS is typically sporadic, but may segregate in families with broader speech and language deficits. We hypothesize that genetic changes may be involved in the etiology of CAS. We conduct whole-genome sequencing in 27 families with CAS, 101 individuals in all. We identify 17 genomic regions including 19 unique copy number variants (CNVs). Three variants are shared across families, but the rest are unique; three events are de novo. In four families, siblings with milder phenotypes co-inherited the same CNVs, demonstrating variable expressivity. We independently validate eight CNVs using microarray technology and find many of these CNVs are present in children with milder forms of SSD. Bioinformatic investigation reveal four CNVs with substantial functional consequences (cytobands 2q24.3, 6p12.3-6p12.2, 11q23.2-11q23.3, and 16p11.2). These discoveries show that CNVs are a heterogeneous, but prevalent, cause of CAS.

Copy number variations identified in families with childhood apraxia of speech and other speech sound disorders provide a deeper understanding of the genetic basis for naturally acquired speech.

Introduction

Human speech is a complex phenomenon engaging >72 muscles with synchronized control, modulated by the nervous system. Speech, the most common form of human communication, is naturally acquired by babies and toddlers during development. Defined developmental language milestones link acquisition of individual speech sounds to connected, mature speech, and eventually, literacy (see Fig. 1). In some children this natural cascade of developmental events is disrupted leading to communication disorders, a heterogeneous grouping of conditions that are “an impairment in the ability to receive, send, process and comprehend concepts of verbal, nonverbal and graphic symbol systems”¹.

Fig. 1. Model of speech and language development.

The model illustrates the relationship between speech and language development and reading acquisition. As depicted in the model, speech and oral language precede literacy acquisition. The green boxes represent stages in speech sound development³, where acquisition of some speech sounds are earlier than others. Preliteracy skills develop during preschool, including phonemic awareness and letter-sound correspondences that draw from early speech sound development. Linguistic skills (in the light blue boxes) are shared by spoken and written language. Literacy development is represented by the dark blue and gray boxes. Double arrows represent the reciprocal relationship for spoken and written language. Based on a simple view of reading¹¹³, there are two components to skilled reading, decoding (dark blue) and comprehension (gray).

Communication disorders are highly prevalent in the United States, with approximately one in twelve children ages 3-17 years in the population exhibiting any disorder². The most common types of difficulty are either a speech (5%)² and/or a language problem (3.3%)², present in children both with or without intellectual disabilities. While early work showed that the neural basis of speech and language lies in two critical regions of the left hemisphere of the brain, Wernicke’s area and Broca’s area, recent advances have shown that many other neural pathways as well as other regions of the brain are also involved^3–5. Therefore, babies and children with cognitive deficits and delays due to brain malformations and other frank developmental problems and are often nonverbal, as the basic neural systems for speech are disrupted. After omitting children with intellectual disabilities, cleft palate and hearing loss, a subtler class of deficits emerge in a broad grouping of children with speech sound disorder (SSD), who have normal IQ.

SSD include errors of articulation or phonetic structure⁶ (i.e., errors due to poor motor abilities associated with the production of speech-sounds) and phonological errors (i.e., errors in applying linguistic rules to combine sounds to form words). SSD are highly prevalent in preschool children, approximately 16% at three years of age⁷, with an estimated 3.8% of children continuing to present with speech delay at six years of age⁸. One of the most severe forms of SSD, excluding those with known developmental syndromes, is childhood apraxia of speech (CAS). The American Speech-Language-Hearing Association (ASHA) defined CAS as a “…speech disorder in which the precision and consistency of movements underlying speech are impaired in the absence of neuromuscular deficits (e.g., abnormal reflexes, abnormal tone)”^1,9. CAS shows a range of severity¹⁰ and is rare, with a prevalence 1 to 2 per 1000^11,12. Recent literature^13–16 has demonstrated the wide variability of CAS in terms of severity, persistence, and presence of neuropsychatric comorbidities. Along with the observation that CAS is often comorbid with language impairment, CAS in isolation is uncommon.

CAS typically presents in a single child in a family or in families with less severely affected siblings with SSD; large cohorts of this diagnostic entity are not available for genetic analysis. Despite attempts to identify sib pairs and families with CAS, very few such families have been reported to be able to perform heritability calculations, as this method requires that multiple family members be affected (or a quantitative trait be available). The KE family that led to identification of FOXP2 as an important locus for CAS exceptionally had multiple affected family members^17–19. This is not the norm and CAS is generally present in the population as a sporadic event, not inherited by multiple siblings/family members in a recessive or dominant manner. In our full cohort of 115 CAS families spanning a 20+ year recruitment time period (data not shown), only two families with a proband and another sib affected with CAS were ever reported.

An alternate genetic model that has gained significant traction in the genetic literature is emergence of de novo mutations as a major cause for several syndromes and diseases²⁰. The de novo mutation model has been applicable because of advancement in whole exome and whole genome sequencing methods, leading to de novo mutations being reported for Kabuki, Schinzel–Giedion, Bohring–Opitz, Baraitser–Winter and Coffin–Siris syndromes²⁰. CNVs were first reported for CAS by Raca et al. ²¹ and Fedorenko et al. on 16p11.2²². Further extending this concept, Morgan et al. ²³ recently reviewed a large number of case reports and noted the sporadic nature of CAS. In addition, variable expressivity is known to occur with some genetic variants^24–26, leading to more severe consequences in certain individuals.

A number of recent studies of CAS performed whole genome sequencing (WGS) or whole exome sequencing (WES) to identify associated structural, copy number, and rare coding variants^{16,21,27–33}. Other studies examined individuals with a well-characterized deletion on chromosome 16p11.2, and found that the majority of subjects had a highly penetrant and severe form of CAS^22,34, with one study showing 77% of individuals with this deletion to have CAS³⁴; albeit other neurodevelopmental conditions are also associated with this deletion. Another recent study identified CNVs in children with CAS, but focused their search on variants ≤ 5 Mb in size³⁵. Infrequently, CAS has been reported in cytogenetic case reports^36–38, but most of the cases remain unresolved, and the majority of these children are not referred for genetic testing. A similar approach has been taken in dyslexia and language impairment³⁹. Many of these reports illustrate varying severity and variability in clinical manifestations in children with CAS that carry these copy number variants. Given the considerable phenotypic variability associated with copy number variants like 16p11.2^40,41, additional clinical characterization of these variants is important for clinical care.

Approximately 1 in 7000 live births produces a cytogenetically visible deletion⁴² with larger deletions causing more severe consequences. In general, deletions lead to greater morbidity and mortality than duplications⁴³, and larger CNVs are often associated with higher burden of developmental disorders⁴⁴, but the effect of CNVs on SSD phenotypes has not been well described. As data from whole genome sequence or exome sequencing becomes more widely available, it will be feasible to quantify the penetrance and phenotypic correlations of CNVs that are not associated with particular syndromes, but do lead to subtle phenotypes. Recently this effort has been bolstered by bioinformatic methods that predict the effect of the deletion/duplication based on the gene content, regulatory profiles and size or position of the event on the chromosome using large cohorts^45,46.

The Cleveland Family Speech and Reading Study (CFSRS) has accrued a large number of children with CAS, providing us a unique opportunity to identify variants associated with CAS and other communication disorders. We hypothesize that typical clinical caseloads of CAS/SSD seen by speech and language pathologists are genetically heterogeneous in pathology, although they likely share common pathways that perturb normal speech and language phenotypes. Prior studies have discovered some structural variants associated with CAS^{16,21,27–33}, but we posit that many more remain to be discovered as the evidence thus far suggests that rare variants predominate in CAS. One possibility is that some CAS cases are actually milder forms of known neurodevelopmental conditions that were not ascertained via routine medical exams because their symptoms did not warrant clinical genetic or neurologist consults. Very likely, pediatricians referred these children to SLPs, OT and PT care following normal care procedures in the US. In a population, individuals who bear the same genetic variant may not express an identical phenotype due to alterations in a range of factors²⁴. The best chance of seeing common phenotypes with the same genetic variant is in a family, particularly siblings, as they frequently share background genetics (i.e. the rest of the genome) as well as exposures to similar demographic and lifestyle factors. Many neuordevelopmental genes display considerable variable expressivity in phenotype, and we hypothesize that some of the variants identified in CAS probands will be shared in children with SSD even if they don’t show as severe a phenotype. Additional discovery efforts in this area are important for a number of reasons: 1) As we hypothesize above, there are likely as yet undiscovered rare genetic variants associated with CAS; 2) To translate these findings to clinical care, extensive clinical characterization of the implications of these variants is needed; and 3) As causal variants are discovered and clinically characterized, future efforts can focus on personalized medicine^47,48.

We performed WGS in 27 families containing a child with CAS. Discovery of CNVs was performed using WGS, followed by validation using B-allele frequencies (BAF), and Log R Ratio (LRR) from high dimensional microarray data. We focused on families where multiple children were affected with SSD (but not necessarily CAS), to increase the likelihood of finding CNVs that segregated with these traits. We discovered 19 large copy number variants across ~50% of the CAS probands, most of which were not previously reported. These discoveries could give us greater insight into the mechanisms of this rare disorder and SSD in general.

Material and methods

Subject ascertainment

This study was approved by the Institutional Review Board of Case Medical Center and University Hospitals. All parents provided written informed consent for their children to participate and children older than five years provided assent. All ethical regulations relevant to human research participants were followed. Families were ascertained through a proband identified from caseloads of speech-language pathologists (SLPs) in the Greater Cleveland area and referred to this study. Initial binary trait classifications were based on parental report of a diagnosis from a referring SLP. These diagnoses were confirmed within our study by direct testing by an SLP, using an extensive battery of standardized communication measures (Fig. 1, Supplementary Fig. 1) as well as noting specific diagnostic criteria of CAS from recordings of conversational speech, as described below (encapsulated in Supplementary Table 1).

All probands in the overall study met inclusion criteria based on direct assessment and information provided by a parent in an interview, or via questionnaire, including: a score ≤ 16th percentile on the Goldman-Fristoe Test of Articulation⁴⁹ upon entry into therapy, at least three phonological process errors, normal hearing and fewer than six middle ear infections prior to age three, intact oral structures, monolingual English speaker, no neurological diagnosis such as autism or developmental delays other than speech and language, performance IQ ≥ 80 on the Wechsler Preschool and Primary Scale of Intelligence (WPPSI)⁵⁰, and a diagnosis of a SSD, or suspected CAS, by a local SLP.

Detailed CAS Phenotyping

Diagnostic criteria for CAS were based on the speech characteristics reported in the literature at the time of diagnosis or the diagnostic criteria described in the CAS Technical Report of the American Speech Language and Hearing Association⁵¹ (Supplementary Table 1). While children could be directly tested because of the age-appropriate nature of these communication measures, the majority of SSD resolves by adulthood and self-report lacks reliability. Thus, we did not include parent historical self-report of SSD in our data because our SLPs could not confirm the diagnosis. We examined measures that were appropriate to the age of the child (Supplementary Fig. 1, Supplementary Table 2) covering the domains of oral motor control, speech, language, and literacy. In sum, from the CFSRS^52–55, we examined 101 individuals from 27 CAS families who had both DNA for WGS and endophenotype data available (Table 1). We also examined whether the child demonstrated persistent speech errors after age 8.5 years¹³.

Table 1.

Sample characteristics

Total sample size	101
Total number of CAS families	27
Parents not directly tested	43
	Total sample	Children only
Sample size	101	58
Sex N (proportion Female)	37 (0.37)	15 (0.26)
Binary Traits N (proportion)
CAS	28 (0.28)	28 (0.48)
SSD + LI	8 (0.08)	6 (0.10)
SSD only	25 (0.25)	11 (0.19)
LI only	5 (0.05)	5 (0.09)
Unaffected	35 (0.35)	8 (0.14)

Other phenotype data

Language impairment (LI), recently described as Developmental Language Disorder (DLD)⁵⁶, was also reported based on diagnosis by an SLP, and confirmed by a score < 1 SD below the mean on the Clinical Evaluation of Language Fundamentals (CELF-3) (Supplementary Table 2). In addition to classifying children in the study according to the presence or absence of CAS, SSD, and LI as described above, we also conducted a wide range of age-appropriate assessments of articulation, phonological awareness, speeded naming, vocabulary, receptive and expressive language, and literacy (Supplementary Table 2, Supplementary Fig. 1), as in our previous work^{10,52,53,57–59}. These quantitative endophenotypes were used to characterize the deficits in children identified with CNVs, as described below. Additional clinical manifestations (issues with feeding, delayed language onset, gross/fine motor incoordination, and ADHD) were recorded in the database based on parent interview. In addition, we characterized severity of CAS based on membership in clusters identified through our earlier analysis¹⁰; in that analysis, we utilized measures of articulation, vocabulary, and reading to group children with CAS into subgroups of varying severity. Cluster membership is indicated in Supplementary Table 3.

WGS methods

DNA was extracted from buffy coats or saliva samples as previously described⁶⁰. Purified DNA was sent to the Broad Institute or the University of Michigan for sequencing on the Illumina HiSeq platform, where DNA was processed into Illumina sequencing libraries using the TruSeq DNA PCR-Free prep kit. DNA was sheared using a Covaris ultrasonicator to approximately 300-500 base pairs. Sheared DNA fragments were blunt end repaired and the 3’ ends were adenylated. Illumina indexed paired-end adapters were ligated to the fragments, purified, and pooled for sequencing. Library preparation quality and fragment size was assessed using an Agilent Bioanalyzer.

Raw sequence data were transferred to the High Performance Computing (HPC) cluster at Case Western Reserve University (CWRU) for analysis. The reads were trimmed for quality (phred score > 20) as well as any residual Illumina adapter sequences at end of reads using TrimGalore!⁶¹. Reads that passed quality control were then aligned to the human reference genome (GRCh19) using bwa-mem (v0.7.17) [https://bio-bwa.sourceforge.net]. The raw sequence data for part of the dataset were not available to realign to the new reference (GRCh38). All reads were aligned to GRCh19 due to the limitations of the legacy data. Aligned reads were reported by bwa and formatted to BAM and processed as individual sample files using the GATK (v4) pipeline using “Best Practice” guidelines⁶². Assessment of the average coverage across the genome was done using Mosdepth on the aligned BAM files⁶³. The average coverage was 33X across all samples (17.2 - 49.1X). Included in the GATK process; deduplication using Picard tools, local realignment using GATK:RealignerTargetCreator and IndelRealigner and recalibration for quality scores using GATK:BaseRecalibrator. Optimized BAM files were used as input into GenomeSTRIP⁶⁴ (v2.0) and processed using SVPreprocess followed by CNVDiscovery and SVDiscovery. Variants were reported by GenomeSTRIP in vcf format and subsequently filtered and annotated using VCFTools (v0.1.12b) [https://vcftools.sourceforge.net] and ANNOVAR (v2017Jun01) [https://annovar.openbioinformatics.org].

Filtering and cross-validation of copy number variants

We used GenomeSTRIP to discover 10,185 copy number variants across the entire cohort. Two main filters were used to winnow the CNVs for further examination: size of the deletion/duplication and its presence in multiple affected individuals. 5,361 CNVs were identified in at least one proband. We focused on CNVs >50 kb since these could be replicated in our independent microarray dataset, as described below. Filtering for CNVs that spanned >50 kb resulted in 48 CNVs, with 3 CNVs falling into annotated genomic regions with low mappability or poor reference genome assembly. These annotations included regions compiled by ENCODE and 10X Genomics^65,66. One additional CNV was found in 77/101 sequenced individuals and was removed from consideration. After merging the remaining 44 CNVs for overlap, we were left with 21 candidate CNVs. These 21 CNV regions were pulled from the BAM files and manually examined using the Integrated Genome Viewer (IGV)^67–69. Regions were examined for stretches of coverage showing increased or decreased coverage, and reads spanning deletion or duplication junctions.

We cross-referenced the WGS data with already available microarray data⁵⁸. CNV characterization was performed using Genome Studio data analysis software, which uses the signals generated from fluorescence intensity of both polymorphic and nonpolymorphic markers, and the cnvPartition Plug-in. The goal of the cnvPartition algorithm is to identify regions of the genome that are aberrant in copy number using two outputs: the LRR, a measure of normalized total signal intensity, and the BAF, a measure of allelic intensity ratios. The expected LRR is 0 for every SNP, while the expected BAF is to equal 0, 0.5, and 1⁷⁰. There is evidence that CNVs are present when samples deviate from the expected LRR and BAF. The copy numbers with confidence scores are generated to produce a map of the CNV regions. Data were loaded into GenomeStudio [www.illumina.com/genomestudio] from the raw. idat files, a sample sheet, the Illumina provided manifest file for the Infinium Omni2.5 BeadChip, and the cluster file. Plots for LRR and BAF were generated for every autosomal SNP calculated for each sample. For CNV calling, a minimum of 3 consecutive SNPs was required with a minimum region size of 1 Mb and confidence >35. All CNVs that were confirmed using both methods (IGV and GenomeStudio) were then examined for segregation with phenotype.

Inheritance of CNVs in families

First, we examined whether CNVs were inherited by determining if the CNV was present in either parent as well as the proband’s siblings. Table 2 shows the presence/absence of the CNV in all individuals in the family who had WGS data. In some cases, parents did not have sufficient DNA available for WGS, but were part of our GWAS study⁵⁸.

Table 2.

List of families with copy number variations in 17 unique genomic regions

If parent not listed, parent was not genotyped (DNA unavailable).

^aPredicted duplication in this family, was confirmed in another family.

^bPhenotypes determined by direct testing of children. Parents were not directly tested. CAS childhood apraxia of speech, Lang language impairment.

Colors indicate overlapping variant across families (one per each color).

Inheritance pattern: I = inherited from one parent and/or the other, U = unknown (one parent missing DNA), and DN = de novo. Inheritance determined by examination of microarray data.

Second, we considered a CNV to be inherited with the phenotype if it was present in the proband and in siblings with CAS, speech, and/or LI, and if the variant was absent in siblings who were not affected with any of these 3 disorders. There were three variants that were excluded from further consideration based on these criteria (Supplementary Table 4). If the variant was present in children with speech or LI, we closely examined the children’s respective values on speech, language, and literacy measures given as part of the study, in order to further characterize the potential implications of these variants. We considered a child to have a deficit in these quantitative measures (Supplementary Table 2) if they demonstrated a standard score of ˂ 85 on measures of articulation, motor speech ability, receptive and expressive language, vocabulary knowledge, single word decoding, phonological processing, reading comprehension and spelling. Articulation was additionally evaluated by previously recorded audio samples (Supplementary Tables 1, 3, and 5).

Functional annotation

We used multiple sources of information to functionally annotate the deletions and duplications, synthesizing evidence when possible. We examined regulatory disruption, population frequencies, evolutionary constraint, as well as literature reviews and validation in our microarray analysis.

The regulatory disruption score is the sum of the predicted expression scores for each gene affected by the CNV, weighted by the gene’s intolerance metric (Supplementary Methods). A CNV regulatory disruption score annotation was generated using the CNV_FunctionalAnnotation program and methodology⁴⁵ that relies on a simple linear model to predict expression effects of rare CNVs. The model was developed using brain samples from the Common Mind Consortium (CMC) study and incorporates information about genes contained in, or affected by, the CNV, including exon overlap, promoter overlap, enhancer overlap, topologically-associating domain (TAD)-gene overlap⁷¹, CNV length and CNV type. The overall regulatory disruption of a CNV is determined by weighting the predicted expression consequences by the relative deleteriousness of the gene affected and summing across all genes affected. The mean regulatory disruption scores, where normative data are reported in Han et al. ⁴⁵ were used as a point of reference for predicting the pathogenicity of our CNVs.

As an estimate of population frequency for our CNVs we report the maximum frequency for CNVs observed in the Database of Genomic Variants (DGV)⁷¹, focusing on the DGV gold standard variants. The DGV gold standard variants are a curated set of variants from a select number of studies in DGV. DGV CNVs considered showed a 30% or greater overlap with our 266 CNV and were of the same type regarding copy number gain or loss.

As an estimate of evolutionary constraint for the genes affected by our CNVs, we supply gene specific probability of loss-of-function intolerance (pLI) scores⁷², extracted from University of California at Santa Cruz genome browser tables and the gnomAD database. We report gene and corresponding pLI if the metric was greater than 0.9. The pLI score reflects the tolerance of a given gene to the loss of function on the basis of the number of protein truncating variants, that is, the frameshift, splice donor, splice acceptor, and stop-gained variants referenced for this gene in gnomAD database weighted by the size of the gene and the sequencing coverage⁷². Lastly, we incorporated data on dosage sensitivity⁷³ from recent data by Collins et al.³⁰ and CNV-ClinViewer⁴⁶.

Additional details on all annotation methods are provided in the Supplementary Methods, and the annotation described above is provided in Supplementary Data 1 and Table 3.

Table 3.

Detailed summary of CNVs

Variant location (hg19)	Size (bp)	Cytoband	Families with variant	Children carrying variant with persistent CAS	Number of children (probands and siblings) with variant	Replication in our independent SSD sample	LoF intolerant	Likely pathogenic or causal based on gene regulation disruption score	pHaplo(max)	pTriplo(max)	CNV ClinViewer (based on region)	Case reports or support from prior literature for CAS, SSD or other neuro diseases	Frequency
chr11:113059924-115271926	2212002	11q23.2-11q23.3 (E)	8139	1394	1		X	X	0.97	0.91	Uncertain	X	NR
chr6:49976443-52091444	2115001	6p12.3-6p12.2 (E)	33	135	1		X	X	0.99	0.99	Pathogenic	X	NR
chr2:163855968-165518209	1662241	2q24.3 (E)	8047	1124	1		X	X	0.96	0.96	Uncertain	X	NR
chr16:29581236-30197928	616692	16p11.2 (E)	12, 25	50, 107	2	E	X	X	0.6	0.99	Pathogenic	X	0.05%
chr22:16340236-17295468	955232	22q11.1 (E)	8107	1302a, 1304a	3				0.16	0.49	Uncertain		0.03%
chr2:97735774-97804498	68724	2q11.2 (E)	8055	1159	2	E			0.12	0.33	Uncertain	X	0.53%
chr16:20545972-20596528	50556	16p12.3 (E)	8049	1128	1				0.08	0.18	Uncertain		0.10%
chr4:3892602-4168684	276082	4p16.3 (E)	8060		1				-	-	Uncertain	X	0.44%
chr4:189241227-189512452	271225	4q35.2 (E)	74	410	2	E			-	-	Uncertain	X	0.02%
chr4:168808644-168993082	184438	4q32.3 (E)	8008		1	U			-	-	Uncertain	X	0.40%
chr8:137680264-137735693	55429	8q24.23 (E)	8055	1159	2	E			-	-	Benign		5.42%
chr2:164641142-164646533	5391	2q24.3 (E)	8107	1302a, 1304a	3				-	-	Benign	X	7.01%
chr12:85939000-86285088	346088	12q21.31 (U)	8055	1159	2				0.16	0.11	Uncertain	X	NR
chr10:45208524-45336137	127613	10q11.21 (U)	8066	1188	1	U			0.28	0.17	Uncertain	X	1.11%
chr2:133176490-133188129	11639	2q21.2 (U)	8061	1174	1				0.19	0.07	Uncertain	X	0.05%
chr5:59710570-59769304	58734	5q12.1 (U)	33	135	3	U			-	-	Uncertain		0.44%
chr7:17165709-17222842	57133	7p21.1 (U)	8130		1				-	-	Uncertain	X	0.05%
chr2:132718104-132979733	261629	2q21.2 (U)	8055	1159	1	E,U			-	-	Benign	X	0.98%

E Deletion, U Duplication, * Unknown persistence, NR Not reported in dbGV; - under pHaplo and pTriplo indicates unavailable from Collins paper.

Results

Group characteristics

Our study included 27 families with at least one affected child with CAS, totaling 101 individuals. The average family size was 3.7 individuals, with a range of 1 (singleton) to 6. Affected children presented with CAS (28, including one family with two cases of CAS), as well as other communication disorders (six had SSD + LI, 11 had SSD alone, five had LI alone), eight siblings were unaffected for CAS, SSD, and LI (Table 1). One family was African-American, the rest were Caucasian. Expectedly, the majority of the children were male (74%), as CAS and SSD predominantly affect males.

Children with an early diagnosis of CAS were followed and assessed from preschool to adolescence (Fig. 1). Direct testing of participants at the preschool assessment revealed that scores ˂1 SD below the test mean were observed in the majority of CAS subjects on speech sound measures of single word articulation (75%), multisyllabic real word repetition (93%), multisyllabic nonword repetition (82%), and oral motor function (82%) (Supplementary Table 3). All participants scored 1 SD below the mean on at least one speech sound or motor-sequencing measure. Per parent report, the following developmental difficulties were noted on a parent questionnaire: 14 participants had fine motor difficulties, 10 participants had gross motor difficulties, 10 participants had oral motor/feeding difficulties, and three participants were born prematurely (Supplementary Table 6). Comorbid ADHD was reported for 16 participants, with five of these participants reportedly on medication for ADHD.

Children with CAS were also directly assessed for comorbid LI at preschool and reading disorders (RD) at school-age. Six participants did not demonstrate LI based on test scores. The remaining participants scored < 1 SD below the mean on standardized language measures of receptive language (43%), expressive language (68%), total language composite (46%), receptive vocabulary (25%), and expressive vocabulary (29%). On literacy measures, nine participants did not demonstrate literacy difficulties based on test scores. The remaining participants scored < 1 SD below the mean on standardized measures of single real word decoding (46%), single nonword decoding (39%), reading comprehension (43%), spelling (54%), and measures of phonological awareness (39%) and rapid automatized naming (36%) (Supplementary Table 3).

Identification of CNVs

Using GenomeSTRIP, we identified 5,361 CNVs within the 28 children with CAS from the 27 families. We focused on CNVs that were >50 kb in size, and additionally excluded genomic regions with low mappability or poor reference genome assembly, leaving 48 CNVs. Among these, 19 were validated using both microarray data and IGV manual inspection (Table 2). The deletions/duplications spanned 17 unique genomic regions. Two of the CNVs occurred in different families with different breakpoints, and one of these overlapping CNVs was much shorter (5.kb). The largest CNV identified was ~2.2 Mb. In our replication analysis, we discovered large CNVs (13 deletions and six duplications) present in 14 unrelated families. Some regions did not contain genes, but the 16p11.2 locus borne by two probands encompassed 47 genes (Supplementary Data 1). In total of the 27 probands examined, 14 (51.9%) had at least one CNVs that was associated with CAS and/or other speech sound disorders.

There were three CNVs that were shared across three families. The first, on chromosome 16p11.2 at 29-30 Mb, indicated with green in Table 2, is the aforementioned locus that has been previously observed in CAS. In both cases, this is a de novo deletion, which is consistent with the literature^21,34. The child in family 12 had a wealth of deficits, including all speech and language measures, some literacy measures, oral motor issues, fine motor incoordination, and ADHD, that demonstrated persistence beyond age 8.5, and was present in the most severe cluster from our earlier cluster analysis¹⁰. The child in family 25 had difficulties with multisyllabic word repetition, expressive and receptive language, reading, fine motor incoordination, ADHD, had persisting problems beyond age 8.5, and was identified as being moderately severe from an earlier cluster analysis¹⁰. Second, the deletion on chromosome 2q24.3, indicated with blue in Table 2, is inherited in two families. The child in family 8047 had some difficulties in articulation and literacy, as well as oral motor problems and ADHD, was also persistent, and a member of the high severity cluster¹⁰. The child in family 8107 had issues with all speech measures but was too young for the literacy assessment and did not have data on persistence and severity, but did have oral motor problems and delayed language onset. Finally, the duplication on chromosome 2q21.2 is inherited in both families. The child with CAS in family 8055 had difficulties with all speech measures as well as expressive language, expressive vocabulary, and most literacy measures, as well as oral motor problems, delayed language onset, gross motor incoordination, and ADHD, and was also in the most severe cluster¹⁰ and had persistent errors after age 8.5. The child with CAS in family 8061 also had difficulties with all speech measures, most language measures, and had delayed language onset and ADHD, as well as persistence after age 8.5, but was in the mildest severity cluster¹⁰; this CNV in family 8061 was smaller (~5 kb) than all of the others, but overlapped with the large CNV in family 8055. In children affected with CAS with deletions or duplications, difficulties with speech measures were present in all children, difficulty with language and literacy measures were present in most, and other clinical manifestations varied (Supplementary Table 3 and 6).

Functional implications of CNV regions

We have provided detailed annotation of each of these CNV regions in Supplementary Data 1, including the regulatory disruption score described in the Methods, whether there was potential intolerance due to loss of function in the region according to gnomAD, and links to these regions in Decipher and the Database of Genomic Variants. In addition, Table 3 contains data on dosage sensitivity^30,46. There were four regions that showed potentially significant effects on function based on their regulatory disruption score: chromosome 16p11.2, chromosome 11q23.2-11q23.3, chromosome 6p12.3-6p12.2, and chromosome 2q24.3, and all four of these were likely pathogenic based on either their pHaplo scores and/or CNV-ClinViewer (Table 3). The deletion on chromosome 16 has been associated with CAS in several studies^{21,22,29,34,74} and there is potential loss of function of CORO1A, TAOK2 and MAZ genes within this region (Supplementary Data 1). The deletion on chromosome 11 has been associated with autism, schizophrenia, neocortical development, and other social cognitive behaviors^75–78. Moreover, two genes within the chromosome 11 deletion boundaries, ZBTB16 and NCAM1, are loss of function intolerant (Supplementary Data 1). The region on chromosome 6 has been associated with brain-face shape, gross and fine motor control, lack of expressive speech, as well as sleep quality^79,80. Finally, the deletion on chromosome 2 has been previously associated with speech development, lack of social interactions, and intellectual disability in a case report⁸¹.

Replication via prior microarray data

In order to reproduce our results in an independent cohort, we additionally examined 131 cases of SSD for whom we had high dimensional microarray data but not WGS data, using the same bioinformatic approach we previously applied to validate WGS results. This analysis was limited to regions where there was sufficient marker density, thus excluding the chromosome 2 locus ~164 Mb, and two additional regions. In the extended SSD cohort, we identified 19 additional subjects carrying 8 CNVs in the same physical locations as those described in our discovery cohort (Supplementary Table 7), a rate of 47.0% (8 CNVs / 17 unique CNV regions from Table 2). Two children had CAS, but the majority had SSD and LI or SSD alone. While most loci had duplications or deletions, one locus showed deletions and duplications. One child with SSD and LI had two duplications. The deletion on chromosome 8 is fairly common in the population at ~5%, and this replication cohort showed a prevalence of 4.5% (6/131) (Table 3, Supplementary Data 1) in this dataset. These results demonstrate that variants discovered in children with CAS are also present in children with less severe SSDs, suggesting that these CNVs have variable expressivity and that these findings are reproducible.

Determination of likely causal variants

We synthesized the results from functional annotation, replication analyses, severity of CAS based on persistence, and the literature in Table 3, in order to posit which CNVs among those discovered were likely causal. First, the top of Table 3 lists the four aforementioned CNVs with likely deleterious effects. The fourth of these, on chromosome 2q24.3, was observed in one family as a large deletion, and a deletion of a smaller, overlapping region in another family. The family with the smaller deletion, observed in the population at ~7%, also carries a large CNV on chromosome 22q11.1, which we posit to be the causal variant in this family. Two additional variants, on chromosomes 22q11.1 and 2q11.2, are likely causal based on their nonzero pHaplo scores and literature suggesting a link to speech phenotypes (Table 3, Supplementary Data 1).

Two other variants, on chromosomes 2q11.2 and 4q35.2, are also potentially causal based on their existence in children with persistent CAS, replication in our independent dataset, and previous case studies in the literature in children with communication and/or other neurological traits^15,16,82. CNVs on chromosomes 4q32.3, 4p16.3, 10q11.21, 2q21.2 ( ~ 133 Mb), 7p21.1, 5q12.1, and 12q21.31 are not as well characterized, but are potentially of interest based on two or more of the following criteria; their existence in children with persistent CAS, replication in our independent sample, or previous case reports in the literature^36,82–91. The CNV on 8q24.24.3 was replicated in our independent cohort, however, the common frequency of this variant within the population weakens support for causality. The CNV on chromosome 2q21.3 ( ~ 132 Mb) is of unknown significance, as both deletions and duplications observed in our data have also been reported in the literature^92,93. One family in the WGS discovery cohort has four inherited variants, three of which are present in a sibling with LI. The 2q21.2 duplication restricted to the child with CAS of this family is of unknown significance, because duplications as well as deletions were observed in our replication dataset. However, other studies of learning disabilities have demonstrated genetic interactions between multiple CNV can have causal effects^29,94,95, so it is possible that the 2q21.2 duplication may be additive in the case of CAS, while the other CNVs are only associated with severe LI in this family.

Phenotypic heterogeneity

By comparing families where children without CAS but with SSD or LI carried CNVs, we were able to characterize the other phenotypic implications of the variants using our extensive database of communication measures (Supplementary Table 3, 5, 6). In family 33, the variant on chromosome 5 was present in children with poor scores in articulation, fluency, and oral-motor skills. The CAS affected child in this family had a variety of other clinical manifestations as well (Supplementary Table 6). In family 74, children with the variant demonstrated poor scores in reading, elision, and rapid naming; this CAS patient also had problems with feeding and a limited repertoire of sounds. In family 8055, there were three variants, all present in children with severe language difficulties, and the CAS patient had broad clinical manifestations. In family 8107, children with the variant on chromosome 22 demonstrated poor scores on articulation measures, and the CAS affected child demonstrated delayed language onset.

Discussion

CAS is a rare SSD, reportedly found in only 3.4%–4.3% of children who are clinically referred for SSD⁹⁶. Given the infrequent occurrence of the disorder, we nevertheless successfully ascertained a relatively large group of individuals who presented with CAS in early childhood, and who we followed longitudinally into adolescence and early adulthood. The CFSRS included participants whose primary impairment was CAS, eliminating participants with syndromes and other known neurodevelopmental disorders that can affect speech, such as autism spectrum disorder, sensorineural hearing impairment, and intellectual disability. Nonetheless, nearly 72% of the current participants demonstrated oral language difficulties related to language processing or verbal expression, congruent with previous reports of deficits in these areas^55,97,98. Furthermore, 55% of the participants demonstrated difficulties with single real words, or nonword decoding and spelling, signaling that a diagnosis of CAS confers a higher risk for literacy difficulties than other forms of idiopathic SSD⁹⁹. A diagnosis of CAS is also associated with persistent speech sound errors. This was apparent in our study, as 62% of the participants continued to demonstrate speech errors beyond the age of typical speech normalization of 8.5 to nine years of age^11,100–102. Put together with our genetic data, these observations suggest that the loci discovered have an effect on multiple phenotypes across an individual’s lifespan.

We identified 17 CNV regions containing 19 different genomic regions with CNVs associated with CAS, speech, and language. This high rate of families where CAS is inherited with CNVs (51.9%) is consistent with other studies²⁹. Most of these loci were unique across families, demonstrating significant genetic heterogeneity in CAS and speech disorder. We were able to replicate many (47.0%) of these loci in an independent sample with microarray data, and through that analysis, demonstrate that these variants are not exclusively associated with CAS but are also present in children with less severe speech disorders. Using bioinformatic resources and the extant literature, we have strong evidence for an association between CAS and deletions on chromosomes 2q24.3, 6p12.3-p12.2, 11q23.2-q23.3, and 16p11.2, and probable evidence with deletions 2q11.2 and 22q11.1.

CNVs in several regions have been previously associated with CAS, as well as speech and language. The deletion on chromosome 2q11.2 is associated with speech delay, ADHD, and developmental disability⁸². We observed a deletion on chromosome 4q35.2, and deletions and duplications at this locus in our replication study. This genomic region has been previously associated with an unbalanced dislocation and duplication in children with CAS^15,38. A deletion on chromosome 7p21.1 has been previously associated with severe speech and language disorder, autism, and developmental delay⁸⁷, here, we observed a duplication of this region. Finally, the duplication that we observed in two families on chromosome 2q21.2 that we replicated in our independent microarray data, has been previously associated with reading and language³⁹. While we did not replicate all of the previously identified CNVs that been associated with CAS in other WGS and/or microarray studies^{21,27–29,35}, this further suggests that CAS is genetically heterogeneous and that there may be additional, as yet undiscovered, causal variants. This in turn implies that multiple genes controlling naturally acquired speech and language are distributed across many locations in the human genome, and that evolution of speech and language did not occur in a single evolutionary step.

Additional CNVs in the literature have been associated with other neurological traits, such as autism, ADHD, and intellectual and developmental disability, as well as syndromes. Given that the CAS-associated CNVs also are present in children with developmental phenotypes, these findings are relevant for our work. The deleted region on chromosome 22q11.1 has been associated with autism, intellectual disability, ADHD, and DiGeorge syndrome^103,104. While we observed a duplication on chromosome 10q11.21, other studies reported microduplications in this region associated with developmental delay and seizure⁸⁵ and facial dysmorphism⁸⁴. A deletion on chromosome 4p16.3, has been associated with Wolf-Hirschorn syndrome⁸³. We observed a duplication on chromosome 12q21.31, while another study found a deletion associated with intellectual disability⁸⁸. These developmental disabilities are often associated with delays/disorders of speech and language^105–107.

Our work builds on previous literature^{21,22,29,34,74} demonstrating that the 16p11.2 deletion is a major cause of CAS. Previous literature also asserts that this deletion demonstrates variable expressivity^21,40,41,104 with variable deficits seen in language²¹, speech production²¹, gross motor issues or delay^21,22,74, severe articulation issues^22,34, and speech delay⁷⁴. The severity of these traits varies dramatically, even within study when evaluated by the same research team. Our work adds to these phenotypes by reporting all but one of the children in our cohort with this deletion have persistent speech errors beyond the age 8.5. Notably, the one nonpersistent child demonstrated relatively mild speech difficulties at baseline. Consistent with previous work³⁴, these findings imply that children with a 16p11.2 deletion should be evaluated for CAS and other communication disorders, especially since some of the comorbidities of CAS could be mistaken as autism spectrum disorder³⁴.

Three CNVs were seen in multiple families, but the rest were unique. Replication of these CNVs in our independent microarray data, consisting of children manifesting speech sound disorders other than CAS, suggests there are multiple CNVs that are common causes of CAS and other communication disorders. For example, the deletion on chromosome 16p11.2 was previously associated with both CAS and autism; our study extended this clinical characterization to milder forms of SSD. SSD are heterogenous with regard to etiology and clinical presentation, and the diversity of CNVs identified support that conclusion.

As in previous studies of CAS²⁹, we also observed a subset of families with multiple CNVs inherited with CAS. Other authors have hypothesized a “two-hit hypothesis” among pathogenic CNVs with variable penetrance and expressivity^29,94,95. Our data supports this hypothesis; among the three apraxics with multiple CNVs, variable severity in terms of persistence, severity, deficits in language and literacy, and existence of other clinical manifestations is observed, as one of the three children is “mild” based on our metrics.

This work is not without limitations. Only a portion of the CFSRS sample had WGS data available, so we were unable to identify whether these CNVs were present in a larger cohort of children affected with CAS or SSD. Since some children were lost to follow-up, we do not have complete data on literacy measures for some children. Clinical manifestation data were based on self-report and were not verified in all cases with medical records, so they may be incomplete or inaccurate. Despite the strict inclusion criteria, our participants presented with a range of co-occurring developmental delays, LI, and RD, in addition to persistent speech difficulties. These findings are consistent with the view of CAS as a multi-domain motor-speech disorder, affecting processes related to auditory-perceptual encoding, memory, and motor planning¹⁰⁸. Comparison of our findings with the literature is challenging because of variable expressivity seen with CNVs¹⁰⁹, so some of our observed CNVs may not have been previously reported in their association with speech phenotypes. DNA changes like CNVs may be caused by a variety of environmental factors^110–112, but unfortunately, we did not collect data on such factors, so we are unable to explore this hypothesis. Lastly, we focused our search on CNVs ≥ 50 kb in size, so we were unable to replicate recent findings³⁵. Smaller CNVs are more common in the population, necessitating a different study design and analytical approach, for which this sample was underpowered. Thus, this will be the focus of future research, in an independent cohort.

In sum, our analysis of WGS data in 28 children with CAS revealed 17 CNV regions associated with CAS or other SSDs. The children with these CNVs demonstrated a wide variety of severity and speech, language, and literacy deficits, as well as other clinical manifestations. These data support CNVs as a major cause of CAS and other communication disorders in our sample, and that these CNVs have variable expressivity. Additional work is needed in new cohorts to discover novel potentially causal CNVs, as well as identify whether some of these CNVs are shared in other CAS subjects. Such future work will be invaluable for assisting practitioners in diagnosing disorders, screening potentially affected siblings, and allowing patients to seek targeted therapies^47,48.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Author contributions

ERC, CMS, and SKI conceptualized and designed the study, drafted the initial manuscript, and reviewed and revised the manuscript. ERC, PB, KB, and BT conducted the statistical analyses. AS and BAL helped conceptualize the study and critically reviewed the manuscript for important intellectual content. GM, LF, JT, and BAL collected the data and revised and reviewed the manuscript. All authors approved the final manuscript as submitted.

Peer review

Peer review information

Communications Biology thanks Beate Peter and the other, anonymous, reviewer(s) for their contribution to the peer review of this work. Primary Handling Editors: Julio Hechavarría and Benjamin Bessieres.

Data availability

The datasets generated and/or analyzed during the current study are not publicly available due to Institutional Review Board (IRB) restrictions on the data, but are available from the corresponding author (Sudha Iyengar, ski@case.edu) on reasonable request and will require an IRB application.

Code availability

All software versions are identified within the Methods. If there is no version number, then that software package only has one (current) version. There were no custom scripts created for the analyses conducted in this paper.

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at 10.1038/s42003-024-06968-y.