Highly Sensitive Diagnosis of 43 Monogenic Forms of Diabetes or Obesity Through One-Step PCR-Based Enrichment in Combination With Next-Generation Sequencing

Type 2 diabetes and obesity are complex disorders that are associated with several factors of genetic, epigenetic, and environmental origins (1,2). Familial aggregation of both type 2 diabetes and obesity demonstrates a high heritability (between 40 and 70%), which can make it difficult to select cases more likely to have a monogenic cause (1,3). However, deleterious coding gene mutations have been shown to cause almost totally penetrant severe forms of diabetes and obesity, including neonatal diabetes mellitus (NDM), maturity-onset diabetes of the young (MODY) and several related syndromes like Bardet-Biedl syndrome (BBS), Alström syndrome (ALMS), Wolcott-Rallison, or Wolfram syndrome (4,5). All of these monogenic forms of diabetes and obesity tend to occur at younger ages and often exhibit other clinical features (4,5).

In this context, an accurate molecular diagnosis of these extreme and often familial forms of diabetes and obesity is crucial for an optimal care of the patients and genetic counseling for their families. The most striking example is seen for the NDM patients carrying a heterozygous point mutation in the ABCC8 or KCNJ11 gene, encoding the two subunits (SUR1 and KIR6.2, respectively) of the pancreatic β-cell–expressed ATP-dependent K⁺ channels. Indeed, these patients can be optimally treated by oral sulfonylurea drugs instead of lifelong insulin therapy, leading to remarkable improvements in glucose control and quality of life (6–8). Furthermore, a recent study demonstrated that personalized genetic medicine applied to patients with NDM, which is currently based on standard Sanger sequencing of KCNJ11 and ABCC8 that is performed by clinical diagnostic laboratories (at a cost of $2,815 in the U.S. [9]), leads to high financial benefits (9). However, as KCNJ11 and ABCC8 encode a total of 40 coding exons, this genetic testing is obviously time-consuming, labor-intensive, and restricted to two genes only while NDM can be due to mutations in at least 11 genes (Table 1) (1,4). We previously showed that whole-exome sequencing (WES) was an attractive alternative as it was a comprehensive, cost-efficient, and rapid method to identify causal mutations in patients with monogenic disorders (10). However, this technology is not currently perfect for clinical molecular diagnosis, as it leads to marked gaps of sequence (>5% of the target regions, even with high mean depth of sequencing coverage) (10,11), which is problematic when the investigators are looking for only one causal mutation.

In the current study, we aimed to assess a new method based on PCR enrichment in microfluidic droplets and next-generation sequencing (NGS) for the molecular diagnosis of 43 subtypes of monogenic diabetes or obesity. A total of 40 patients carrying a known causal mutation for those subtypes according to genetic diagnostic laboratories were blindly included in the study.

We selected a total of 40 patients presenting with a monogenic form of diabetes (n = 19) or obesity (n = 21) who were carriers of a known causal mutation according to diagnostic laboratories: 9 NDM patients, primarily assessed by the Department of Genetics in Robert-Debré Hospital (Assistance Publique-Hôpitaux de Paris, Paris, France); 10 MODY patients, primarily assessed by the Cruces University Hospital- Centro de Investigación Biomédica en Red de Diabetes y Enfermedades Metabólicas Asociada (n = 5; Barakaldo, Spain) and the CNRS UMR8199 unit (n = 5; Lille, France); 12 patients with an early-onset severe obesity, primarily assessed by the Department of Genomics of Common Disease in Imperial College of London (n = 6; London, U.K.) and the CNRS UMR8199 unit (n = 6; Lille, France); and 9 BBS patients, primarily assessed by the Genetic Diagnostic Laboratory in the Hôpitaux Universitaires de Strasbourg (Strasbourg, France).

The study protocol was approved by all local ethic committees, and study participants signed an informed consent. For children <18 years, an oral consent was obtained (when ≥3 years of age), and the parents provided written informed consent.

We selected all susceptibility genes for monogenic forms of diabetes or obesity, which were known at the time of the design (n = 43; Table 1 and Supplementary Table 1). All exons, including at least 40 base pairs (bp) of the flanking intron of each exon, and 1,000 bp upstream of exon 1 and downstream of the last exon were targeted. The primer library was designed using proprietary pipeline developed by RainDance Technologies (Lexington, MA). Briefly, primer selection was based upon standard PCR criteria, namely guanine-cytosine content between 25 and 80%, amplicon length between 200 and 600 bp, melting temperature between 56 and 60°C, and primer length between 15 and 33 bp. Suitable primers were then screened against the dbSNP131 database to remove all primers that hybridized to polymorphism containing targeted sequences. The 542 targeted exons were covered by 970 primer pairs (targeting a total of 336 kb of sequence) that were encapsulated into microfluidic droplets via RainDance Technologies (Table 1).

The 40 DNA samples (3 μg) were fragmented to 2–4 kb by sonication (Bioruptor NGS; Diagenode, Liège, Belgium) and purified using the MinElute system (Qiagen, Valencia, CA). The quality of both fragmentation and purification was assessed using the 2100 Bioanalyzer (Agilent Technologies, Santa Clara, CA). Subsequently, the RainDance primer library was merged with each sheared genomic DNA sample on the RDT1000 (RainDance Technologies), according to the manufacturer’s protocol. The resulting emulsion, containing ∼1 million potential PCR reactions, was subsequently amplified by PCR on the Mastercycler Pro S (Eppendorf, Hamburg, Germany). The addition of Destabilizer reagent (RainDance Technologies) resulted in the degradation of the emulsion and the separation of oil and aqueous phases. A brief centrifugation completed this separation. PCR products were then purified using MinElute columns (Qiagen) and assessed for both quality and predicted amplification profile on the 2100 Bioanalyzer (Agilent Technologies).

Two DNA samples (3 μg) were also captured using the Agilent SureSelect Human All Exon Kit (version v4; Agilent Technologies), following the manufacturer’s protocols and as previously described (11).

PCR products were repaired to blunt ends and concatenated using the NEB Quick blunting kit and the NEB Quick ligation kit (New England Biolabs, Ipswich, MA). The samples were then purified using Agencourt AMPure XP beads (Beckman Coulter, Fullerton, CA) and fragmented to ∼200 bp by sonication (Bioruptor NGS; Diagenode). The quality of fragmentation and purification was assessed using the 2100 Bioanalyzer (Agilent Technologies). Adaptors were ligated to the fragments by using NEBNext Multiplex Oligos for Illumina (New England Biolabs). Finally, the 40-amplicon enriched samples were sequenced on the HiSeq2000 (Illumina, San Diego, CA) in 76-bp paired-end reads by multiplexing 12 samples per lane, while the two whole-exome–enriched samples were sequenced on the HiSeq2500 (Illumina) in 100-bp paired-end reads by multiplexing the two samples in two lanes.

Demultiplexing of sequence data were performed with CASAVA (version 1.8.2).

Regarding the 40-amplicon enriched samples, sequence reads were mapped to the human genome (UCSC NCBI37/hg19) using the Burrows-Wheeler Aligner (version 0.6.1; algorithm "BWA-SW"; default parameters) to generate a BAM file. This aligner deals with chimeric reads generated by the library preparation method described above. Variant calling was performed with SAMtools (version 0.1.18; algorithm "pileup" with the following parameters: -u -g -B -m 3 -C 50 -d 1000000 -L 1000000 -F 0.0002 -Q 0), when depth of sequencing coverage was ≥8×.

Regarding the two whole-exome–enriched samples, sequence reads were mapped to the reference human genome (UCSC NCBI37/hg19) using the pipeline CASAVA (version 1.8.2). Variant detection was performed with the same pipeline and filtered to reach depth of at least 8×. In both samples, we obtained a mean depth of coverage of at least 200×.

All detected variants were annotated using the Ensembl Perl API (version 66).

For each participant, after the detection of all variants in the targeted regions, we focused on rare variants of interest (i.e., nonsynonymous variants, essential splice-site variants, or indels leading to a gain of STOP codon and nonsense variants) with minor allele frequency <1% (or “not available”) according to the Single Nucleotide Polymorphism database #135 (dbSNP135), the 1000 Genomes project (12), and the National Heart, Lung, and Blood Institute Grand Opportunity Exome Sequencing Project (ESP) (13). We primarily looked for variants of interest located in genes related to the main phenotype of the patient (Table 1). Furthermore, we checked if there were additional variants of interest in the other genes of the panel, according to the model of inheritance leading to the disease (presence of homozygous mutations in recessive models or presence of heterozygous mutations in dominant models; Table 1).

When we found additional variants with a putatively deleterious effect, we confirmed them by Sanger sequencing on a 3730xl DNA Analyzer (Applied Biosystems, Foster City, CA). A standard protocol was performed. Primer designs and PCR conditions can be provided upon request. Sequencing reads were assembled and analyzed using Variant Reporter software (Applied Biosystems).

In genomic DNA samples from 40 patients presenting with an elucidated monogenic form of diabetes or obesity, we sequenced a total of 43 genes (Table 1) using PCR enrichment in microfluidic droplets followed by NGS. The overall median depth of sequencing coverage per gene was 427× (Table 1). In each DNA sample, 98.6 ± 0.8% of the targeted regions were successfully sequenced with a depth of coverage ≥8× (Table 2). We successfully sequenced 96.1 ± 1.9% and 95.1 ± 2.6% of the targeted regions when we fixed more stringent depths of coverage ≥20× and ≥30×, respectively. When we considered coding exons only, 508 exons out of a total of 520 were perfectly sequenced for all participants (Fig. 1), while nine exons were partially sequenced (in CEL, BBS9, CEP290/BBS14 LEPR, and TTC8/BBS8; Fig. 1) and three exons were poorly sequenced in all participants (in BBS7 and CEP290/BBS14; Fig. 1). Therefore, the sequencing problems using the present technology were not random but seemed to be specific to few exons. Of note, the failed sequence regions were not rich in guanine and cytosine (between 20 and 40% of guanine-cytosine). Two DNA samples were sequenced through RainDance Technologies in combination with Illumina sequencing and through Agilent SureSelect All Exon kit (Agilent Technologies) in combination with Illumina sequencing. In both samples, 100% of coding variants that were identified by WES were also found using RainDance Technologies in combination with Illumina sequencing.

As a blind test, the investigators only knew the main phenotypes of the 40 patients but not the known causal mutation carried by these patients (according to the different genetic diagnostic laboratories). Except for a patient presenting with BBS (patient 21), we reidentified the putatively causal mutation in every participant, which had primarily been detected by genetic diagnostic laboratories (Table 2). The list of these causal mutations included: 23 nonsynonymous variants, 7 nonsense variants, 4 splice-site variants, and 10 indels (involving 1–5 nucleotides) (Table 2). Our algorithm failed to detect a highly complex homozygous mutation in patient 21, BBS5-c.572_594inv{ins567_568};595_603del / p.His180Tyrfs*2, which was previously described (14). However, when we investigated the BAM file for this patient, we found a dramatic gap of coverage at the locus of the mutation (Supplementary Fig. 1).

Importantly, in three individuals, we surprisingly detected other mutations with a putatively deleterious effect, in addition to those previously reported by the genetic diagnostic laboratories. Indeed, in patient 1 presenting with NDM (and diagnosed with a KCNJ11 mutation), we also found a missense mutation in ABCC8 (p.Leu225Pro), also known as rs1048095 (Tables 2 and 3). According to PolyPhen-2 (15), SIFT (16), LRT (17) and MutationTaster (18) software, the ABCC8-p.Leu225Pro was predicted to be damaging (Table 3), and several studies reported this mutation in patients presenting with NDM (6,19,20). Furthermore, Masia et al. (19) showed that the mutation contributed to overactivity of the ATP-dependent K⁺ channel. In patient 22 diagnosed with a BBS6 mutation, two additional missense mutations were detected in genes associated with monogenic diabetes: ABCC8-p.Thr695Ala and HNF1B-p.Val61Gly (Table 2). The ABCC8-p.Thr695Ala mutation was novel and predicted to be damaging according to both LRT and MutationTaster software (Table 3). The HNF1B-p.Val61Gly (or rs147816724) showed a minor allele frequency of 0.07% in European-American participants from the ESP project (Table 3). Previous studies reported that this mutation led to kidney disease and uterine abnormalities (21,22). In patient 29 presenting with severe early-onset obesity and diagnosed with an MC4R homozygous mutation, we detected a novel missense heterozygous mutation in ABCC8 (p.Arg755Gln), which had not been detected by the 1000 Genomes or ESP project and was predicted to be damaging according to the LRT software (Tables 2 and 3). Unfortunately, we did not have access to additional clinical features for patients 1, 22, and 29.

Currently, the methods of molecular diagnosis based on Sanger sequencing are restricted (when performed) to only a few loci as this technology is quite expensive, long, and tedious. Since an accurate etiological diagnosis of severe forms of diabetes and obesity has been proven to lead to marked improvements in patient care and in family counseling, the development of cost-efficient, fast, and high-throughput methods for an accurate DNA sequencing is of major medical interest in metabolic diseases and beyond. In the current study, we assessed the PCR-based technology in microfluidic droplets in combination with NGS so as to achieve these goals.

We found this method very accurate as we were able to reidentify almost all known mutations in each sample (44 of 45). Furthermore, most of the targeted coding regions were perfectly sequenced in all participants (Fig. 1). To achieve even better results in gene screening, the design of the current panel of primers can easily be reoptimized, designing new primer pairs for failed (or partly failed) amplicons. Furthermore, when a small set of regions fails (as in our study), it is quite feasible to fill in these rarely occurring gaps using a standard protocol of Sanger sequencing. A nearly perfect diagnosis protocol can therefore be defined for monogenic diabetes and obesity genes. Of note, we only failed to call a highly complex indel in one BBS patient, although we identified a dramatic drop of coverage at the indel locus (Supplementary Figure 1). While waiting for improvements in NGS algorithms of indel calling and detection (that should come soon), it is necessary to double-check the uniformity of coverage throughout the sequence targets in negative samples, and it would be helpful to perform a standard multiplex ligation-dependent probe amplification to surely dismiss any presence of large indels.

The cost of this method is very reasonable compared with Sanger sequencing or WES. When considering reagents only and one DNA sample, the RainDance PCR-based enrichment in combination with NGS was five times less expensive than WES ($330 vs. 1,630; based on the sequencing of 12 RainDance amplicon-enriched libraries per lane and two whole-exome–enriched libraries per lane, respectively, using the Illumina HiSeq2000) and 20 times less expensive than Sanger sequencing with forward and reverse sequence for 970 amplicons ($330 vs. 6,783). Of note, the RainDance panel is conceived for 1,500 samples. Therefore, if the number of samples is restricted, then the experiment is less cost-efficient. However, the library can include 20,000 amplicons (the present library includes 970 amplicons only). Thus, additional genes involved in other rare disorders could be easily added to design a master panel applicable to several dozens of rare diseases. Furthermore, it is noteworthy that the number of patients presenting with monogenic diabetes or obesity is totally underestimated. Shields et al. (23) reported that >80% of MODY cases were undiagnosed. The present methodology will be able to improve this poor diagnosis.

Moreover, the present method is very fast: for one sample, the microdroplet-based PCR enrichment, the sequencing, and the analysis can be performed in 2 weeks only.

Finally, we were able to find more than one variant with a putatively deleterious effect in three patients, which should help us to understand better the range of phenotypes found in families with early onset forms of diabetes or obesity. Indeed, given current sequencing progress, it becomes clearer that the phenotypes associated with so-called monogenic diseases can be due to a set of penetrant mutations in several causative genes. In a recent study, through RainDance PCR-based enrichment in combination with NGS, Schrauwen et al. (24) also identified two causal mutations in two different genes in a patient presenting with hearing loss. We believe that the assessment of all known susceptibility genes for a disease in only one step will bring a less-biased analysis of the causes of the disease and an obvious benefit for the patient (and possibly his/her family if several putatively deleterious mutations are present in the same pedigree).

The present panel could primarily be tested in patients presenting with: 1) NDM (with an onset of diabetes <1 year of age); 2) a potential MODY (according to known guidelines and algorithms); 3) a potential BBS or ALMS; 4) severe early-onset or familial obesity (in at least three generations); or 5) severe early-onset obesity in a family with no other obesity case in siblings and parents (to find a putative de novo mutation). After detection of all variants, the clinical geneticist could pick up the potential causal rare variants by: 1) focusing on variants previously described in the literature or on novel variants not found in dbSNP, the 1000 Genomes project, or ESP project; 2) assessing extended family members to verify inheritance pattern and/or segregation of the mutation with the phenotype; 3) using in silico damage prediction and conservation analyses; and, if possible, 4) performing functional studies in cell lines or animal models.

A limitation of our assay may be the use of a coverage threshold of 8× for variant detection. A recent study has reported that raising the coverage threshold to 13× markedly limited the number of missing variant calls (25). However, in the current study, a concordance rate of 100% was found between WES and PCR-based technology in microfluidic droplets in combination with NGS (in target genes that are present in the panel) using a coverage threshold of 8×. Therefore, the differences are at worst quite limited.

In conclusion, we demonstrated that this novel sequencing method is accurate and that any rare failures of amplicon analysis could easily be detected and solved by Sanger resequencing or a redesign of the panel of primer pairs. This approach therefore has the potential to considerably simplify the current genetic diagnosis of severe forms of diabetes and obesity. As cost and time to deliver results have been key barriers to uncovering a molecular cause in the many undiagnosed cases likely to exist, the present methodology should be considered in patients displaying features of monogenic diabetes or obesity.

Acknowledgments. The authors thank all of the patients who participated in this study.

Funding. Our study was supported by the French Agence Nationale de la Recherche (ANR-10-LABX-46 and ANR-10-EQPX-07-01), the European EurOCHIP obesity FP7 Consortium, and the transnational European research grant on Rare Diseases (ERANET-09-RARE-005).

Duality of Interest. No potential conflicts of interest relevant to this article were reported.

Author Contributions. A.B. designed the study and wrote the manuscript. J.P. performed the PCR-based enrichment and NGS and wrote the manuscript. E.D. performed the PCR-based enrichment and NGS. J.M., M.A., R.M., H.C., L.C., and M.V. managed the primary genetic analyses on the participants of the current study, reviewed the manuscript, and contributed to discussion. S.S., M.P., and J.-L.M. managed the primary genetic analyses on the participants of the current study. F.D.G. and I.R. performed the bioinformatics analyses. V.D. reviewed the manuscript and contributed to discussion. O.S. performed the bioinformatics analyses, reviewed the manuscript, and contributed to discussion. P.F. designed the study, wrote the manuscript, and managed the primary genetic analyses on the participants of the current study. All authors read and approved the final version of the present draft. P.F. is the guarantor of this work and, as such, had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.