1 INTRODUCTION

Autism is recognized as a spectrum of clinically and genetically heterogeneous disorders that are highly heritable. Although the genetic etiology is not completely understood a genetic cause can be found in approximately 30–40% of autistic individuals (Schaefer et al., 2013). Associated G-band karyotype abnormalities previously identified include maternally inherited duplications of the chromosome 15q11q13 “Prader Willi/Angelman syndrome” region, terminal deletions of chromosome bands 2q and 22q, and deletions within chromosome band 7q31 (Rosenfeld et al., 2010). However, the autism genomic landscape is now recognized to also involve not only single-gene disorders resulting from single nucleotide variation (SNV) or nucleotide triplet repeat expansion (Fragile X syndrome) but also copy number variation (CNV). Both Fragile X and Chromosome Microarray Analysis (CMA) are now considered standard of care and first-tier genetic tests for autism (Schaefer et al., 2013). Genetic testing for other single-gene disorders associated with autism is not routine and typically directed by co-occurring clinical or phenotypic features.

Several previous large-scale genome-wide CNV studies have consistently shown that CNV is enriched in autism vs. controls with overall detection rates estimated in unselected cohorts at approximately 10% with higher rates ranging from 11.6% to 12.5% in more specific autism cohorts up to 27.5% in syndromic autism (Rosenfeld et al., 2010; Schaefer et al., 2013). Recent consecutive case series from diagnostic clinical laboratories have reported CNV detection rates for autism between approximately 20.0 and 24.4% (Ho et al., 2016; Roberts et al., 2014).

These early studies established that CNV associated with autism is usually rare (<1% population frequency), large in size (>500 kb), and defined by recurrent “hotspots” or “CNV susceptibility loci.” Interestingly, smaller CNVs were not found enriched in autism. “Hotspot” CNV tends to overlap the critical region of the reciprocal deletion/duplication syndromes associated with developmental delay (DD)/intellectual disability (ID). The highest CNV rates involve deletions and duplications within chromosome band 16p11.2. Other recurrent CNVs include three regions within band 15q11.2q13.3 and less frequently in bands 1p36.22, 1q21.1, 7q11.23, 13q14.3, 16p13.2, 22q11.2, and 22q13.33, among others (Girirajan et al., 2013; Menashe et al., 2013; Rosenfeld et al., 2010; Sanders et al., 2011; Schaefer et al., 2013; Sebat et al., 2007). Such CNV may harbor single or multiple genes and while studies have failed to find the elusive autism-specific genes among CNV susceptibility loci, the size of the deletion or duplication has been shown to influence autism phenotype associations. As deletion size increases nonverbal IQ decreases but not for duplications. In contrast, as duplication size increases autism severity and repetitive or stereotypic behavior increase but not for deletions. This supports the hypothesis that increased gene dosage enhances autism severity while haploinsufficiency is associated more strongly with DD and ID (Girirajan et al., 2013). Most of these large-scale studies involved simplex autism families (single proband) and were found enriched for de novo rare and large (>500 kb) CNVs. In contrast, studies of multiplex families (more than one affected sibling) found that rare and large CNVs were more likely inherited with a lower rate of de novo CNV consistent with multigenic and/or polygenic risk loci (Leppa et al., 2016). Intriguingly, very few autism-specific genes were refined among all these early studies likely due in part to their low (rare) frequency, genome-wide distribution, and variable penetrance and expressivity of an associated phenotype (Girirajan et al., 2013; Menashe et al., 2013; Sanders et al., 2011; Sebat et al., 2007). This failure is consistent with the notion that an imbalance of multiple genes contributes to the autism phenotype (Girirajan et al., 2013). Research efforts continue, however, with a recent report that integrated CNV and single nucleotide variant (SNV) whole-exome data and found novel de novo CNVs as well as rare inherited CNVs, encompassing single genes that were proposed as potential candidates associated with autism: VPS13D, CNTNAPS, PRDK2, RBFOX1, WWOX, MACROD2, NCNLS, ABPBAS and CTNNA3 (Bacchelli et al., 2020).

DNA extracted from peripheral blood yields high molecular weight (HMW) DNA of high purity, which has been the gold standard for genetic testing. However, it requires invasive venipuncture and special specimen handling requirements, cold storage, and transport. Other disadvantages include avoidance of blood-draw due to needle-stick phobia, reluctance for repeat blood draws, difficult vein access, and hesitation of the family/care givers or the phlebotomist to draw blood from elderly and feeble individuals, or young children/infants. Blood sampling from autistic patients has additional challenges involving stereotypic behavioral movements, which may require patient restraint or sedation for collection. In contrast, saliva offers an alternative biospecimen for DNA that can be collected noninvasively. It contains ~4.3 × 10⁵ cells/mL comprised of epithelial and leucocyte cell types derived from embryonal ectoderm and mesoderm, respectively. High-quality DNA can be obtained from saliva even when stored at room temperature prior to processing from 6 months to several years (Sun & Reichenberger, 2014).

Saliva has been recognized previously as a source of DNA for genetic and genomic studies using PCR-based methods, next-generation sequencing whole-exome and whole-genome studies, including genome-wide association studies (GWAS) using single nucleotide polymorphism (SNP) genotyping (Sun & Reichenberger, 2014). The majority of early GWAS studies showed no difference between DNA from saliva or blood for SNP genotyping with call rate concordance >97%. However, saliva has not been the specimen of choice for CMA and there are early controversial reports over whether saliva DNA is comparable to blood DNA for CNV detection. In one small study, saliva was as good as blood (Dellinger et al., 2010), however, another showed significant differences (Fabre et al., 2014) and yet another study showed better detection using DNA from the blood than buccal swab (Marenne et al., 2011). Very few reports have been published using saliva DNA for CMA testing within clinical diagnostic laboratories (Francis et al., 2022; Reiner et al., 2017), and as far as we are aware, there are none investigating autism cohorts.

Therefore, we performed SNP CMA using saliva DNA with the main focus being to evaluate the clinical utility of this biospecimen for identifying CNV among children with autism.

2 METHODS

3 RESULTS

3.3 CNV pathogenicity and burden

SNP CMA results for all saliva DNA samples can be found and summarized in Table S2, which comprised 48 probands and parents analyzed as either trios (n = 38), duos (n = 9), or singleton (n = 1). Trio or duo genotype analysis confirmed expected familial relationships. Excluding the two samples that failed, no CNV was found in 17/46 (37.0%) probands, however, two of these showed ROH at 4.8% and 10.4%. In 29/46 (63.0%) probands with CNV, there was a total count of 44 CNVs classified as either pathogenic (n = 3), VoUS (n = 3), likely benign (n = 13), or benign (n = 25); see Table S3. Among the 27 probands, a single CNV was found in 17 probands, two CNV in five, three CNV in four, and five CNV in one proband; see Figure 1. CNV was found inherited from a carrier parent in 23 probands.

4 DISCUSSION

DNA from blood is considered the gold standard for genetic testing, including for CMA. While saliva DNA has been extensively used with other genetic testing methods, there are surprisingly few reports of saliva DNA used for CMA testing, especially among clinical diagnostic laboratories. Herein, we have successfully explored the utility of saliva as an alternative biospecimen among a series of autistic probands.

We successfully obtained SNP CMA results from almost all saliva samples (98.5%), except for two (Fus_023 and Fus_047). Overall, mean saliva DNA concentration was 142.1 ng/μL but this varied widely (1–474 ng/μL) accompanied by an acceptable mean 260/280 nm ratio (1.76) but low mean 260/230 nm ratio (1.15) (Table 1). The variable DNA yield most likely reflects the collection technique and variable saliva volume while the low OD ratios suggest DNA impurities from residual solvent contaminants remaining after DNA extraction. Despite this, we did not further purify saliva DNA before use but still obtained valid SNP CMA results, which suggests the low 260/230 nm ratio did not significantly affect assay performance. Indeed, the mean beadchip quality metrics for DLR, LogR Dev, BAF Dev, and call rate overall saliva samples were exceptionally good (Table 1). This included several samples with unsatisfactory BAF Dev values (0.04 or 0.05) and the two failed samples (FUS_023 and FUS_047) with unacceptable DLR, LogR Dev, and BAF Dev values; see Table S1. Despite occasional unsatisfactory BAF Dev values, we could still successfully interpret results in almost all samples, except for the two that failed. However, laboratories should take care when interpreting suboptimal SNP CMA data and evaluate their own acceptable quality metrics before the use of saliva DNA.

These findings suggest that saliva DNA of variable yield and purity may still produce a successful result. This success may in part reflect the robustness of the Infinium Beadchip technology to produce high-quality beadchip data. Beadchip data quality might be further improved by developing an “in-house” saliva DNA clusterfile (.egt file) to reduce “noisy” genotyping data, which could particularly benefit samples with poor data quality. We note two sample failures (Fus_023 and Fus_047), which most likely reflected low DNA yield (concentrations at 1 ng/μL and 5 ng/μL) and/or poor DNA purity (Table S1). However, in our experience, DNA samples of <5 ng/μL may still produce an SNP CMA result, as exemplified by a saliva sample (001_M) with DNA at 1 ng/μL, although with a suboptimal BAF Dev metric. As part of the study, we also showed that duo and trio genotype analyses can be performed using saliva DNA and confirmed familial relationships in 47/48 probands, even for one of the failed samples (Fus_023). The other failed sample (FUS_047) could not be genotyped.

In a subgroup of 17 probands (and five parents) with CGH array and therefore both saliva and blood DNA, we confirmed that blood DNA has greater yield (mean 322.4 ng/μL) and purity than saliva, according to 260/280 nm (1.78) and 260/230 nm (1.84) ratios (Table 2). However, blood DNA can still result in CGH array test failure as evidenced by two samples (sample IDs 005_P and 007_P) with poor DLR values (see Table S1). Despite the lower DNA yield and purity in our saliva DNAs, we found saliva still yields DNA of sufficient quantity and purity for successful SNP CMA, which was even successful for the two probands with failed CGH arrays.

We next compared CNVs identified between saliva and blood DNA in this subgroup and all were concordant with a mean size difference ~70 kb (Table S4). The size discrepancy here reflects the different probe coverage/spacing within the CNVs and resolution of the different microarray platforms. We then looked at CNV as called by Bluefuse Multi using only saliva samples for 23 probands with an inherited CNV. We found the mean size difference of called CNV was ~9 kb (Table S5). Despite this size variation, we deemed this acceptable and attributable to the number of consecutive probes within the CNV as called by the software and background noise of the CytoSNP assay. Any size discrepancy typically reflected whether a single SNP probe was included in the CNV call, consistent with the mean backbone probe spacing of ~9 kb for the CytoSNP-12 beadchip.

In 46 probands (Tables S2 and S3) with successful saliva SNP CMA results, there were 29 probands (58.7%) that harbored a total of 44 CNVs. The majority were classified as either likely benign (n = 13) or benign (n = 25) with the remainder VoUS (n = 3) or pathogenic (n = 3). There were no likely pathogenic CNV. Of the pathogenic CNVs, two were duplications within 16p13.11, found in different probands. One showed a paternally inherited 1.26 Mb duplication (Sample ID 001_P) and the other a maternally inherited 1.16 Mb duplication (Sample ID 219_P). Both encompass the same critical region known as the recurrent “16p13.11 duplication neurocognitive disorder susceptibility” locus; see DECIPHER database. The remaining proband (sample ID 013_P) had a de novo 0.12 Mb deletion within 16p11.2, also known as the recurrent “16p11.2 deletion syndrome”; see GeneReviews NBK11167 and OMIM #611913. Of the VoUS, these reflected a CNV of unknown inheritance (sample 016_P) and two deletions in one proband that were de novo or maternally inherited (sample 236_P). Benign CNVs reflected those overlapping known population CNVs and likely benign CNVs were those inherited from an “apparently” unaffected parent; acknowledging that any potential phenotype in a carrier parent may be masked by variable expressivity and reduced penetrance.

We also assessed CNV burden among the 27 probands with CNV and found they harbored a single CNV (n = 17 probands) followed by two CNVs (n = 5), three CNVs (n = 4), and five CNVs (n = 1). Although the majority of these CNVs were classified as likely benign or benign, it is interesting that 10 probands had two or more CNVs. Increased CNV burden among individuals with neurodevelopmental phenotypes has been correlated with severity, including among autism. However, the effect size was reported greater in those with both autism and ID, rather than autism alone (Girirajan et al., 2011). Our study size is too small to draw any significance regarding CNV burden, however, it is interesting to note that one proband harbored five CNVs (Sample ID: 236_P).

Overall, the study findings as a whole demonstrate that saliva DNA is comparable to blood DNA for CNV detection. The successful and accurate identification of CNV in our autistic cohort is consistent with the reliability of previous studies. In this regard, in another ID/DD cohort of limited sample size (n = 20), saliva DNA was also shown to be reliable for successful CNV detection (Reiner et al., 2017). More recently, SNP CMA using saliva DNA has been reported to provide an additional ~4% diagnostic yield, particularly for detecting somatic mosaicism for chromosomal aneuploidy and terminal segmental abnormalities (Francis et al., 2022). Unfortunately, we did not observe mosaicism in our study but cannot completely exclude low-level mosaicism at, or below, the generally well-accepted lower detection limit (~10%) of SNP CMA (Conlin et al., 2010; Cross et al., 2007).

Limitations of our study include the relatively small sample size and missing saliva DNA OD readings for eight individuals, however, we considered this to have an insignificant effect on assessing DNA purity. Furthermore, due to limited resources, we did not compare CNV between blood vs. saliva DNA on the same CytoSNP-12 SNP CMA platform but instead verified CNV obtained from 17 probands with previous CGH array data, which we considered to be a reasonable orthogonal method of comparison. Except for these first 17 probands, we also did not verify CNVs detected in saliva DNA using another microarray platform due to resource constraints, but in 23 probands, they were verified as inherited from a parent. While the study was not designed or sufficiently powered to identify novel CNVs in autism, it provides promising results on the use of saliva for SNP CMA testing within this population.

Saliva has the advantage of being an alternative biospecimen that does not require special handling, storage and transport conditions, or immediate processing. It can be collected in a noninvasive manner using simple instructions, thereby avoiding phlebotomy and anxiety over needle-stick phobia. This is especially the case for young autistic children with sensory or behavioral difficulties that pose significant issues for blood collection, or for patients where it is challenging to perform blood collection. Saliva collections can even be done at home rather than in clinical or pathology settings, which has the added advantage that saliva collection kits can be mailed to remote/rural areas, where it can be challenging for families to visit a blood collection center. Furthermore, saliva collection can be additionally advantageous following comprehensive face-to-face clinical assessment in the event of an abnormal genetic test result that requires subsequent follow-up testing of other family members, which can be carried out by saliva kits being sent out and returned by routine mail without needing families (especially those from rural/remote areas) to make further visits to clinical facilities. Disadvantages include potential reduced DNA yield and purity and risk around improper specimen collection, labeling, and sample provenance due to self-collections. However, the risk can be reduced via education and awareness provided using online TeleHealth calls or demonstration videos with further direct supervision offered by local staff for specimen collection and labeling, if needed.

5 CONCLUSION

Our study has shown that saliva yields DNA of sufficient quantity and quality to perform SNP CMA with high test success rate and accuracy for calling CNV thereby demonstrating clinical utility as an alternative biospecimen for genetic testing in autism. More specifically, autistic individuals and in particular children with sensory and communication issues may experience significant physical and psychological discomfort from invasive blood collection. The option of saliva collection has several advantages over blood collection, which can be particularly challenging for population groups such as autistic cohorts where physical restraint or sedation of children can be avoided. Our findings have important and significant implications for the use of saliva in clinical practice guidelines for SNP CMA and genetic testing.

AUTHOR CONTRIBUTIONS

Dale C. Wright, Natalie Silove, and Valsa Eapen conceptualized and designed the study. Valsa Eapen acquired funding. Dale C. Wright wrote the first draft of the manuscript. Maria L. Baluyot, Johanna Carmichael, Artur Darmanian, Ngaire Jose, Con Ngo, Luke St Heaps, Katherine Holman, Sylvia Ziso, Feroza Khan, and Anne Masi acquired the data. All authors contributed to the review, editing, and approval of the final manuscript.

FUNDING INFORMATION

This work was supported by the Autism CRC Ltd. (grant/award no.: 1.059RS).

CONFLICT OF INTEREST STATEMENT

The authors declare that they have no competing interests.