A dual-mode targeted Nanopore sequencing assay for comprehensive SMN1 and SMN2 variant analysis

Assay Prototyping cohort

The prototype assay was developed using genomic DNA (gDNA) from cell lines (N=97) obtained from the Centers for Disease Control and Prevention Repository (N=4), National Human Genome Research Institute (NHGRI) Sample Repository for Human Genetic Research (N=12), and National Institute of General Medical Sciences (NIGMS) Human Genetic Cell Repository (N=73) at the Coriell Institute for Medical Research (Camden, NJ; Supplemental Table 1). Additionally, cell lines (N=8; described in 21) were procured from ATCC and isolated using a precipitation-based method (Qiagen, Hilden, Germany).

Assay Optimization cohort

Human-derived, presumed normal, de-identified residual whole blood specimens (N=227) were obtained from We Are Blood (Austin, TX) under the required regulatory approvals for evaluating clinical specimens. Whole blood was purified using either silica resin/column-based method (Qiagen, Hilden, Germany) or functionalized magnetic bead (Applied Biosystems, Waltham, MA). Genomic DNA quantity (ng/uL) and quality (A₂₆₀/A₂₈₀) were assessed using spectrophotometry. Samples were diluted in nuclease-free water to the target concentration for analysis using the assay.

Assay Evaluation cohort

De-identified samples with known SMA copy number status as determined by clinical testing using a droplet digital PCR assay (see below) at the CAP-accredited genomics center, at Al Jalila Children’s Specialty Hospital (Dubai, United Arab Emirates), were used for assay evaluation and test performance characterization. This study was approved by the Dubai Health Authority Research Ethics Committee (DSREC-07/2023_06 and DSREC-SR-03/2023_08).

Droplet digital PCR

Genomic DNA extracted from peripheral blood (N = 70) were tested for SMN1 and SMN2 copy number status using a proprietary droplet digital PCR (ddPCR) assay (Bio-Rad, USA) following manufacturer’s instructions. This assay was clinically validated by the Al Jalila Children’s genomics laboratory (22).

PCR and capillary electrophoresis (PCR/CE)

Only samples determined to have a homozygous deletion of the SMN1 gene by ddPCR (N = 32) were clinically tested for SMN2 copy number status by ddPCR. For the remaining samples with ≥ 1 SMN1 copies (N = 44), SMN2 dosage status was determined by the AmplideX® PCR/CE SMN1/2 Plus Kit as previously described (23). In addition, this method was utilized to assess both SMN1 and SMN2 copy number for all cell line and presumed normal whole blood samples during assay development and optimization.

PCR and Nanopore Sequencing (PCR/Nanopore)

The prototype assay workflow includes PCR master mix setup, gene-specific amplification, sample specific barcoding, paramagnetic bead size selection and concentration, library adapter ligation, Nanopore sequencing, and assay-specific analysis pipelines (Figure 1). Roughly 40ng gDNA was amplified by multiplex PCR targeting 2.7kb regions of SMN1 or SMN2 exons 7-8 and copy number neutral endogenous control amplicons from CFTR in one mix or a larger 11.2kb amplicon encompassing exons 3-8 in a separate mix. Reactions were cleaned up using 0.6X AMPure XP bead ratio (Beckman Coulter, Indianapolis, IN). Samples were tagged with unique barcodes during a second PCR reaction, normalized by mass after Qubit quantitation (Invitrogen, Carlsbad CA), pooled into a single reaction, and concentrated using 0.6X AMPure XP bead ratio.

• Download figure
• Open in new tab

Figure 1.

The prototype PCR/Nanopore Assay design and workflow involves two separate PCR reactions that amplify either 2.7kb or 11.2kb amplicon products. Amplicons are barcode tagged for each sample in a second PCR reaction, then pooled by mass into a single sequencing library. The library is prepared by appending nanopore-specific adaptors and loaded into a MinION flowcell for sequencing on a Mk1B connected to a computer running MinKNOW software (Oxford Nanopore Technologies). Data are analyzed with off-the-shelf and assay-specific software.

Sequencing libraries were prepared using the Nanopore Ligation Sequencing Kit (LSK-110 or LSK-114; ONT). Sequencing was conducted using MinKNOW software (22.10.10) on MinION flow cells (R9.4.1 or R10.4.1) with a Mk1B sequencer connected to a computer. Sequencing proceeded for at least 5 hours to obtain a minimum of 150 fully-spanning reads per copy per region. Super-accurate base calling was performed using Guppy (6.3.9) after sequencing and fully spanning reads were aligned to GRCh38 using minimap2 (2.15+dfsg-1).

To determine copy number, we trained a gradient boosting tree model to predict copy number using the ratio of Cs and Ts at the paralog specific variant c.840C>T in exon 7 (NM_000344.3c.840C>T; Single Nucleotide Polymorphism database, https://www.ncbi.nlm.nih.gov/snp, accession number rs1164325688, build 155, last accessed August 15^th, 2023) (Supplemental Figure 1). The model incorporates read depth of fully spanning 2.7kb fragments aligned to SMN1 (c.840C) or SMN2 (c.840T) and the geometric mean of endogenous control amplicons (CFTR) to infer fold change based on normalized read depth in 2-copy calibrator samples (Supplemental Figure 2). The machine learning algorithm was trained on a subset of cell lines (N=79) and an independent set of residual whole blood samples (N=352). Hyperparameters for the decision tree model were selected using an 80:20 train:test split in a stratified randomly selected five-fold cross validation scheme.

Additionally, SMN1 and SMN2 copy number was informed with the longer 11.2kb amplicon flagged as edge cases where the machine learning algorithm predicted a copy number of 2.25 to 2.75. First, unique haplotype groups were identified and differentiated by sequence variability (e.g. SNVs) corresponding to unique copies aligning to the same region of the genome (Supplemental Figure 3). This allows for the identification of all unique copies for a gene target, and thus can be used to inform copy number calls and provide proper input to variant calling software that often assumes no more than two copies of the gene are present. Group sizes were normalized according to read depth and copy number was inferred for each group. The number of normalized haplotype groups were reported as the predicted copy number for samples that were flagged as edge cases when more than one group was identified.

Predicted copy number from the PCR/Nanopore assay was compared to data collected by orthogonal methods (ddPCR and/or PCR/CE). Small nucleotide variants and insertion/deletions were identified using Clair3 (24).

Assay Development and Optimization

We developed a complementary, two-in-one MinION-based assay utilizing amplicons ranging more than 10-fold in length, including reference loci, to identify SMN1 and SMN2 variants. Assay designs and analysis methods were optimized for copy number changes and phased SNVs and indels (Figure 1). Primers were developed to amplify ∼800-3000 base fragments (“short” amplicons) encompassing both SMN1 and SMN2 exon 7-8 and endogenous controls (Supplemental Figure 2). The assay assesses SMN1 or SMN2 copy number by aligning reads spanning exons 7 and 8 to either SMN1 or SMN2, then determining the raw read depth and ratio of reads associated with c.840C (SMN1) or c.840T (SMN2). A machine learning method infers copy number by comparing the C:T ratio in exon 7 and corresponding read depths to endogenous control amplicons in calibrator samples.

We also designed primers to amplify an 11.2kb fragment (“long” amplicon) encompassing exons 3-8 to enhance phasing SNVs in SMN1 or SMN2 (Supplemental Figure 3). This long amplicon typically captures several paralog-specific variants and facilitates sequence deconvolution of reads into haplotype groups to infer copy number. In some samples, however, there are no copy distinguishing sequence variants within one or both paralogs to independently resolve copy number.

The assay was developed and evaluated in three phases (Figure 2). In the prototyping phase, gene-specific primers, protocols, and reagents were designed and tested on a diverse set of cell line (N=79) and presumed normal whole blood (N=352) samples. This set included independently analyzed SMN1 and SMN2 genotypes ranging from 0 to ≥4 copies along with SNVs and Indels described below. Sequencing data were collected with short and long amplicons. Algorithms were trained using amplicon sequence and read depth within SMN1 and SMN2 compared to known genotypes for each sample. We also identified endogenous control amplicons and calibrator samples to convert SMN1 and SMN2 read-depth ratios into copy numbers.

• Download figure
• Open in new tab

Figure 2.

Study design for assay development consisted of three phases.

In the assay optimization phase, performance was first evaluated with the short, 2.7kb amplicon design utilizing a sample cohort of 18 additional cell lines (N=97 total) and an independent set of whole-blood samples (N=227) across 0 to ≥3 SMN1 and SMN2 copies (Figure 33). Assay copy number predictions were compared to PCR/CE data and segregated by sample type.

• Download figure
• Open in new tab

Figure 3.

Performance of the assay to identify copy number across cell lines or presumed normal whole blood samples within the optimization data set for SMN1 (orange) or SMN2 (blue) utilizing the 2.7kb fragment or both designs where call threshold flagging would prioritize the copy number call from the 11.2kb amplicon. Genotype agreement was >97% for both SMN1 and SMN2 using both designs, and 100% for the limited number of 0 and 1 copy SMN1 samples in this data set.

Copy number agreement was 97.9% (95/97) between PCR/Nanopore and comparator method for both SMN1 and SMN2 in the cell-line set. One sample (NA20232) was discordant for both SMN1 and SMN2. This cell line was expected to be 3 copy SMN1 and 2 copy SMN2 (3/2) by PCR/CE yet reported as 2/1 by the PCR/Nanopore assay. Two other samples had either SMN1 or SMN2 discordant copy number data where a 3 copy was expected, yet a 2 copy was predicted by the algorithm. In each case, haplotype deconvolution with the long amplicon supported the expected copy number.

In the whole blood sample cohort, 218/227 (96.0%) SMN1 and 224/227 (98.7%) SMN2 copy number calls agreed between PCR/Nanopore and comparator method. Of the 9 discordant samples for SMN1, 7 discordant samples had more than one haplotype. We used the long amplicon design to investigate these discordances, consistent with methods described. Six of the seven samples agreed with comparator data when both amplicon designs were utilized. Four of 9 discordant samples were flagged as edge cases at or near the call threshold with the short amplicon and 3 were resolved by haplotype deconvolution with the long amplicon design.

Consequently, we applied the flagging QC across the entire dataset to determine the effect on accuracy (Figure 33). One cell line and 12 whole blood samples were flagged as edge cases at or near the call threshold with the 2.7kb amplicon for SMN1. Of these, 4 were discordant using the short amplicon alone, but 3 were corrected by calls using the long amplicon. Similarly, of the 3 SMN2 calls flagged only 1 was discordant and long amplicon design supported the expected copy number. In no instance was a sample flagged where the long amplicon call disagreed with comparator data when at least 2 haplotype groups were identified. As a result, both amplicon designs were used for all subsequent analyses where call threshold flagging prioritized the long amplicon copy number call.

In addition to copy number evaluation, the assay was designed to identify and phase SNVs and indels without reflexing to other assays. For example, the SMN2, NM_017411.3: c.859G>C variant (dbSNP, rs121909192, build 155, last accessed August 15^th 2023) is associated with a less severe SMA phenotype (25, 26, 27, 28). Two variants linked with the SMN1 duplication haplotype have been shown to flag silent carriers and increase carrier detection rates when 2 SMN1 copies occur on the same chromosome. These variants are NM_000344.3:c.*3+80T>G (alias g.27134T>G; dbSNP, rs143838139, build 155, last accessed August 15^th 2023) and NM_000344.3: c.*211_*212del (alias g.27706_27707delAT; dbSNP, rs200800214, build 155, last accessed August 15^th 2023).

PCR/Nanopore genotyping of these variants agreed with the PCR/CE comparator method in 324/324 samples for c.*3+80T>G and c.859G>C, and in 322/324 (99.4%) for c.*211_212del (Figure 4). Investigation of the two discordant samples suggested erroneous comparator data. In HG00691, a rare non-pathogenic deletion in SMN2 (rs576032516) shifted a CE peak into the mutant CE bin, resulting in a false positive. In the second sample, the automated peak caller in the CE assay incorrectly called noise along the baseline within the c.*211_212del bin, resulting in another false positive call.

• Download figure
• Open in new tab

Figure 4.

Confusion matrix comparing the expected and predicted SNVs associated with silent carrier or disease modifier risk. The algorithm accurately identified each variant type except for two. †Inspection of the underlying trace data for these two samples in the comparator assay revealed a profile consistent with false-positive calls and a defined root cause in each case. This correction would result in 100% agreement across each variant for the PCR/Nanopore assay.

Independent Evaluation of Residual Clinical Samples

Next, we evaluated assay performance across a diverse, independent cohort of samples from 70 individuals (57% females, average age 2.17 years, range 1 day – 16 years) with known SMN1 and SMN2 copy number status originally tested at Al Jalila Children’s Specialty Hospital, Dubai, UAE. Those individuals represented 14 countries, primarily from the Middle East and North Africa (Supplemental Table 2).

Of all individuals, 30 (42.9%) had homozygous deletions in the SMN1 gene, while the remaining 40 (57.1%) had 1 to 4 copies of this gene. Sixty-seven of the 70 samples were also evaluated for SMN2 copy number. Most individuals (58%) had 2 SMN2 copies, 10% had 0 copies and 42% had 1, 3, or ≥ 4 copies of this gene (Supplemental Table 3).

The PCR/Nanopore assay correctly identified 70/70 and 65/67 of SMN1 and SMN2 copy number states for a concordance rate of 100% (95% CI, 94.80% - 100.00 %) and 97.0% (95% CI, 89.75% - 99.18%), respectively (Table 1).

Cost Effectiveness and Hand-On Time

The accessibility and broad use of an SMN1/2 screening and genotyping assay depends on numerous criteria, including performance, operational factors and cost. Assay workflow, analysis and economic considerations are especially important for laboratories in lower resource environments. We assessed these elements by performing a time-motion analysis, quantifying the scalability of the assay across different sample batch volumes, and calculating estimated costs per sample.

Time-motion analysis was estimated across 6 operators based on experience for both a 24- and 96-sample batch (Supplemental Table 4). The workflow from sample to answer requires less than 48 hours for a 24-sample batch on a Mk1B connected to a computer with a recommended GPU. Gene-specific and barcoding PCR can be completed in a single work shift. Subsequently, samples can be pooled, the library prepared, and sequencing initiated with sequencing and analysis completed overnight for review the next day. A 96-sample batch required additional sequencing and data processing time but was still completed within 72 hours.

The assay supports 12 to 96 samples per batch (Supplemental Table 5). Batch size is currently only limited by the recommended barcodes available from ONT (N=96) since flow cells can routinely process >10M reads. A single Mk1B running 12 samples once per week could screen 624 samples per year. By comparison, 480 samples could be processed per batch on one GridION with 5 independently accessible flow cells, real-time base calling, demultiplexing, and alignment. Utilizing 3 staggered run batches per week a lab could process 75,000 samples per year. This throughput can scale linearly by adding additional GridION instruments. In addition, the assay supports automated reaction setup, bead isolation, quantitation, and data analysis to further improve workflow efficiency.

We estimated less than $20 USD per sample for materials costs excluding PCR (Supplemental Table 6). Since the assay is still in development, PCR enrichment costs cannot be accurately determined. For example, the assay design must be finalized and verified, reagents and kits manufactured, and quality control and release testing established. However, PCR is well documented to be highly cost-effective at scale, and we expect that the total per-sample costs will be comparable to other SMN1/2 diagnostic kits that provide far less genotyping information and insight. These costs also do not include instrument-related expenses. Importantly, ONT instruments (e.g. Mk1B and GridION) have low capital requirements and benchtop footprints, which creates flexibility for laboratories in how they implement the technology. For either instrument, a computer is required for data analysis. For Mk1B instruments, the computer must be equipped with a high-performance GPU to utilize live basecalling, demultiplexing, and alignment during sequencing.