Abstract
The core goal of cell-free DNA-based prenatal testing (at its introduction called “NIPT”) is to provide minimally invasive, clinically accurate, and financially accessible screening for fetal chromosomal aneuploidies in the early stages of pregnancy. This goal poses certain important constraints: minimal invasiveness means the test must operate from analytes in a maternal blood sample; clinical accuracy requires that even small enrichments in placenta-derived cell-free DNA (cfDNA) must be detectable and attributable to specific genomic regions; and financial accessibility is possible only with modern, scalable genomic technologies. Next-generation sequencing satisfies each of these constraints and consequently is the chosen platform for the majority of cfDNA-based prenatal testing offerings. This chapter provides a general overview of how NGS technology works and focuses particularly on its application to cfDNA prenatal testing.
Keywords
Next-generation sequencing, Cell-free DNA, Sequencing by synthesis, Multiplexing, Cluster generation
Acknowledgments
Clement Chu and Mark Theilmann provided helpful comments on the manuscript.
Introduction
The core goal of cell-free DNA based prenatal testing (at its introduction called “NIPT”) is to provide minimally invasive, clinically accurate, and financially accessible screening for fetal chromosomal aneuploidies in the early stages of pregnancy. This goal poses certain important constraints: minimal invasiveness means the test must operate from analytes in a maternal blood sample; clinical accuracy requires that even small enrichments in placenta-derived cell-free DNA (cfDNA) must be detectable and attributable to specific genomic regions, and financial accessibility is possible only with modern, scalable genomic technologies. Next generation sequencing (NGS) satisfies each of these constraints and consequently is the chosen platform for the majority of cfDNA-based prenatal testing offerings. This chapter provides a general overview of how NGS technology works and focuses particularly on its application to cfDNA prenatal testing.
Evolution of Sequencing Technology
The term “next-generation sequencing” (NGS) begs the questions of what “first-generation sequencing” was and how NGS is both similar to and different from its predecessor. Sanger developed the first generation of DNA sequencing in the 1970s . His eponymous sequencing approach is an in vitro adaptation of the cellular replication machinery that cleverly leverages unextendable DNA bases. These modified bases are introduced at low concentration in a reaction minimally containing (1) a high concentration of extendable bases, (2) the single-stranded DNA template to be sequenced, (3) a short oligonucleotide primer complementary to the template onto which new bases could be synthesized, and (4) DNA polymerase enzymes that execute the extension reaction. Early Sanger sequencing experiments involved four independent reactions, each containing a single type of unextendable base (A, T, G, or C). Whenever a polymerase randomly incorporates one of the unextendable bases into the nascent DNA molecule (e.g., an unextendable G in the nascent strand incorporated opposite a C in the template), further synthesis would terminate, yielding a truncated copy of the template. Critically, since all nascent strands anchor from the same oligonucleotide primer, the position of extension termination—and hence the length of the nascent DNA strand—is a direct proxy for the base at the 3′ end of the molecule. By using gel electrophoresis to resolve the respective lengths of terminated molecules in each of the four reactions, it is possible to infer the sequence of the entire template.
Sanger sequencing became slightly more scalable with the introduction of unextendable bases that were uniquely dyed ( Fig. 1 ). Rather than achieving base-specific information by partitioning into four reactions, a capillary electrophoresis instrument coupled with a fluorescent dye detector could resolve both the relative sizes of fragments and the identity of their terminating bases . To criticize these machines as unscalable neglects one of their unmitigated triumphs: they were the workhorses that sequenced the very first human genome in the 1990s . However, with a cost in the billions of dollars and with a timeline on the order of years, genome sequencing would remain prohibitive in a clinical setting without a major technological leap.
NGS revolutionized genome sequencing by overcoming many of the limitations of the Sanger technique , yet the most pervasive NGS methodology shares much in common with its predecessor. As described in further detail later, NGS also leverages extension termination and fluorescent bases, and it relies upon DNA polymerases that append a single base at a time to a nascent DNA molecule. Indeed, in many respects, an NGS experiment is comparable to performing millions or billions of Sanger reactions in parallel (hence the NGS moniker “massively parallel sequencing”). This explosive increase in throughput shattered some of the barriers (e.g., cost and turnaround time) that had largely prevented the use of genomics in routine clinical care, and it paved the way for cfDNA-based prenatal testing.
How NGS Works
The role of an NGS device is to distill a specially prepared library of DNA molecules into a long text file of sequences, one line for each sequenced molecule. NGS sequencers perform this molecule-to-text mapping across many research and clinical contexts, spanning everything from RNA sequencing in broccoli to ribosome profiling in bacteria to DNA sequencing for NIPT in pregnant women . These applications of NGS are distinguished primarily by the steps upstream of DNA being injected into the sequencer, termed “library preparation.” Mirroring the diversity of upstream sample preparation methods is a comparably vast range of downstream analyses, one of which is the analysis method for NIPT, covered in detail in Chapter 3 . In addition to describing how an NGS machine sequences DNA, this section discusses NIPT-specific workflows upstream and downstream of sequencing.
Upstream of the Sequencer
DNA extraction
As the name implies, cell-free DNA is not found in the blood cells, but rather must be extracted from the plasma. cfDNA fragments are relics of dead cells : when a cell undergoes programmed cell death (“apoptosis”), a suite of enzymes is synthesized that digest the genomic DNA . These enzymes can only access DNA that is not packaged into nucleosomes , the octamers of histone proteins that regulate both gene expression and genome topology in the cell . The lack of accessibility to nucleosomal DNA means that the ~ 150 nucleotide (nt) fragments of DNA encircling each nucleosome survive the apoptosis process, and these fragments ejected from the dying cell constitute the cfDNA that is sequenced to give so-called NGS reads, described later in more detail.
To extract cfDNA from plasma, the blood must first be spun in a centrifuge to separate plasma, buffy coat (which contains the white blood cells), and erythrocytes. 1
1 Centrifugation speed must be high enough to separate the blood components but low enough to avoid hemolysis, which could dilute placenta-derived cfDNA. Hemolysis could also result from other mishandling of the sample prior to extraction, for example, by storage at extreme temperature.
Approximately 55% of whole blood is plasma. When aspirating the plasma from centrifuged blood, care must be taken to avoid the buffy coat because the high concentration of maternal DNA in the white blood cells would dilute the scarce placenta-derived cfDNA and reduce or altogether eliminate sensitivity for fetal aneuploidy detection.Standard and commercially available DNA extraction techniques can purify sufficient cfDNA from an aspirated plasma sample to power the analysis . The typical concentration of cfDNA fragments in plasma is only 5–50 ng/mL, and this low level of cfDNA in plasma is noteworthy because it imposes a lower bound on the volume of blood required for cfDNA-based prenatal testing. If blood volume is too low—or if the extraction is inefficient—then the number of copies of the genome in the extracted sample is so low that the subtle changes in fetal chromosomal abundance may not be detectable. For instance, if only 10 copies of the genome are present in the extracted sample, it is likely not possible to detect a 2% change in abundance of chr21. Conversely, an efficient extraction should yield enough genome equivalents to empower detection of fetal chromosomal aneuploidies even at low fetal fraction. The number of genome equivalents required from extraction depends on the subsequent NIPT analysis to be performed. Whole-genome sequencing (WGS) NIPT requires very few cfDNA fragments at any given site, so the volume of blood drawn from the patient can be quite low , and multiple extraction attempts can be made from a single blood collection tube. By contrast, targeted techniques like the single-nucleotide polymorphism (SNP) method require hundreds of genome equivalents at each interrogated site, such that allele balance can be measured with high precision (more information on WGS and SNP methods will be given in Chapter 3 ) . Therefore blood collection volumes for SNP method NIPT testing are typically higher than for WGS.
Since cfDNA concentrations are so low, it is not trivial to measure whether enough cfDNA was extracted to power NIPT. Extracted DNA is typically amplified by PCR prior to NGS; therefore even an inefficient extraction can yield plenty of DNA for sequencing, meaning that an inefficient extraction is not reflected by the depth of NGS sequencing. Fortunately, it is possible to detect inefficient extraction via the “complexity” of the sequencing data. For instance, in the context of WGS, it is expected that, with efficient extraction, genomic positions will have either 0 or 1 aligned sequenced fragment (with the majority having 0) because the sequencing will be Poisson sampling from a very rich initial pool of genomic material . Following an inefficient extraction, however, the pool of original genomic material is very sparse, meaning that sites will tend to have 0 or ≫ 1 mapped fragments, resulting in data with low “complexity.” If extraction efficiency is very high, then PCR may not be required to yield sufficient DNA for sequencing; such “PCR-free” library preparation is expected to have high library complexity. It is important to monitor the sequencing complexity of NGS data to ensure that claims of fetal ploidy have adequate statistical power.
Library preparation
An NGS machine can only sequence an ensemble of DNA molecules—called a “library”—that have been properly prepared. Specifically, for the Illumina devices that predominate in the clinical genomics setting, each DNA molecule in the library must have a common set of flanking adapters (sequences typically ~ 50 nt and specified by the manufacturer), with all 5′ ends sharing one sequence that differs from the one sequence shared by all 3′ ends 2
2 DNA is a directional polymer with a backbone that is a series of bonded sugar molecules. Each sugar has five carbons (referred to as 1′, 2′, 3′, 4′, and 5′, with each pronounced “1 prime,” “2 prime,” etc.), and the bonding of one sugar to the next occurs via a phosphodiester bond between the 5′ and 3′ carbons.
. Flanking all molecules with common adapters permits efficient amplification and extension of the entire library using only a single pair of primers. Such amplification and extension occur (1) upstream of the sequencer to attain a sufficient input concentration (this step is optional), (2) inside the NGS machine immediately prior to sequencing during the “cluster amplification” procedure (described later), and (3) during the sequencing reaction itself (also described later).The most common process by which all 5′ ends have one adapter and all 3′ ends have a different common adapter involves clever molecular biology . First, all double-stranded cfDNA molecules—which may have short single-stranded overhangs at the termini—are incubated with a polymerase enzyme that both trims back 3′ overhangs, fills in 5′ overhangs, and appends an adenine (A) base to the 3′ of all molecules, that is, creating an A overhang ( Fig. 2 ). Finally, the DNA fragments are mixed with “Y adapters” and a ligase. The Y adapters contain two single strands of DNA that are complementary at one end (the trunk of the “Y”) but not at the other (the branches of the “Y”). The double-stranded portion of the Y adapter has a T overhang, which means that it will hybridize to the cfDNA fragments’ A overhangs. The asymmetrical design of the two strands in the Y adapter ensures that, subsequent to ligation of Y adapters at each end of the cfDNA molecule, each of the resulting single strands has a common 5′-specific adapter and a common 3′-specific adapter.
Sequencing biases are minimized when the fragments undergoing NGS are of similar length . In most NGS applications, an in vitro fragmentation reaction—followed possibly by a size-selection step—produces fragments of comparable and acceptable size. For NIPT, however, no such steps are needed, as the in vivo DNA fragmentation that occurs during apoptosis yields fragments with highly uniform length near 150 nt . In fact, the length of cfDNA-fragment processing in vivo is so precise and reproducible that even subtle differences between placental nucleosomes and those from other tissues cause cfDNA-length disparities that can be analyzed to give an estimate of the fetal fraction (placental fragments are systematically shorter than nonplacental) .
cfDNA fragment length can be captured via WGS-based NIPT but not via SNP-based NIPT due to differences in their library preparation approaches. WGS-based NIPT simply appends Y-adapters to cfDNA molecules that are unmodified (other than the blunt ending and A-tailing described previously). This simple library preparation workflow is preferable for WGS-based NIPT because the goal is to have an unbiased sampling of all cfDNA extracted from the plasma. For SNP-based NIPT, however, only the cfDNA fragments that overlap informative SNP sites provide insight into possible fetal aneuploidy. Therefore SNP-based NIPT requires a molecular enrichment of fragments of interest, which can be achieved via a multiplex PCR reaction . In multiplex PCR, hundreds to thousands of different primer sets can be mixed together in a single PCR tube with the cfDNA extracted from a given sample; with proper primer design and reaction conditions, fragments from each of the targeted locations can be massively enriched in preparation for sequencing. Adapter sequences can either be appended to the multiplex primers themselves—in which case multiplex PCR alone yields an NGS-competent library—or Y-adapter ligation can occur subsequent to multiplex PCR. The reason length-based information is lost in such a reaction is that the amplicon length is dictated by the designed primers, not by the template cfDNA fragment off of which they amplify.
The form factor of NGS machines necessitates one more critical step during NIPT library preparation, the barcoding of samples . Illumina sequencing data is sold in units of flowcells, where each flowcell yields hundreds of millions or billions of reads. The number of reads per flowcell exceeds what is needed for a single sample. Therefore it is economically advantageous to load many samples onto a single flowcell, a process called “multiplexing” ( Fig. 3 ).
However, unlike other lab devices that maintain physical separation of samples throughout the assay workflow—for example, qPCR machines, ELISA devices, and capillary sequencers—an NGS flowcell eliminates physical separation between samples during sequencing. Thus a “demultiplexing” mechanism is needed by which NGS data can be parsed into sample-specific cohorts again after sequencing. Demultiplexing is enabled by sample-specific barcodes, short (typically ~ 6–8 nt) DNA sequences included in the set of Y adapters used to create a library for a given sample. Critically, the barcode will differ from sample to sample, but all molecules from a given sample will have the same barcode. The NGS machine emits one text file for barcode sequences and a separate text file of cfDNA fragment sequences, where rows in each file correspond to the same molecule (i.e., the sequence in the first line of the barcode file is the barcode of the cfDNA molecule whose sequence is the first to appear in the cfDNA-fragment sequence file). Using these files, it is possible to parse the whole-flowcell sequencing files into sample-specific files, thereby recapitulating in silico the physical separation of samples that existed when the samples were being processed in multiwell plates upstream of sequencing.
The Act of NGS: From Molecular Library to Text File
Though multiple post-Sanger sequencing technologies exist that could each claim the title of “NGS”, in the clinical genomics setting of NIPT, the term “NGS” effectively implies Illumina-style sequencing, as it is the predominant platform. Therefore, below we describe the “sequencing-by-synthesis” process by which Illumina sequencers execute NGS .
Cluster generation
To gain intuition into the Illumina-style NGS workflow, recall from the earlier comparison of Sanger sequencing and NGS that both methods involve the determination of DNA sequence by measuring a fluorescence signal one base at a time. Therefore at the most fundamental level, an NGS machine must be able to resolve single molecules and capture the series of fluorescent signals that correspond to the molecules’ respective DNA sequences. The process of cluster generation ensures that single molecules are resolvable and that the fluorescence signal for a single molecule can be adequately captured.
The first step of cluster generation is the loading of a DNA library (chemically denatured into single strands) into a glass chamber called a flowcell. The surfaces of the flowcell are coated with oligonucleotides homologous to the adapter sequences appended to cfDNA fragments during library preparation. Single-stranded fragments anneal to the surface of the flowcell at random positions. The position to which a fragment anneals is critical, as that fragment will remain in the same position of the flowcell throughout the entire NGS procedure. The concentration of the loaded library must be carefully calibrated: if the concentration is too high, multiple library fragments could occupy the same location in the flowcell, obscuring the ability to detect fragment-specific fluorescence; if the loaded concentration is too low, then the sequencing capacity of the flowcell is underutilized, which could cause the sequencing depth per sample to be too low for confident aneuploidy detection.
The second step of cluster generation is called bridge amplification and is a PCR reaction that occurs on the surface of the flowcell. This localized amplification of DNA hybridized to the flowcell surface is needed because the intensity of a single fluorescent base incorporated into a single molecule undergoing sequencing is too weak for the NGS machine’s camera to capture. To boost the fluorescence signal to a detectable level, the original library fragment is proximally copied hundreds to thousands of times to yield a dense clonal population of fragments called a “cluster.” Bridge amplification—schematized in Fig. 4 —utilizes the oligonucleotides affixed to the flowcell surface. These oligos serve as primers for each successive round of amplification. Because they are bonded to the glass slide, each single-stranded molecule becomes primed for a subsequent round of extension by bending over to make a bridge with a flowcell-bonded oligo. The final step of cluster generation is to use enzymatic cleavage and chemical denaturation to remove the same single strand in each duplex (e.g., remove the one with a pink primer bound to the flowcell), which leaves single-stranded DNA molecules that all have an identical sequence (see top panel of Fig. 5 , where all single-stranded molecules have pink primer on top and blue primer bonded to the flowcell).
