The present invention relates to systems and related methods for the quantification and identification of nucleic acid target sequences in a heterogeneous sample after performing a universal nucleic acid amplification and, in particular, relates to the analysis of universal polymerase chain reactions using High Resolution Melting (HRM), and machine learning.
In clinical diagnostics and pathogen detection, profiling of complex samples for low-level genotypes represents a significant challenge. Advances in speed, sensitivity, and extent of multiplexing of molecular pathogen detection assays are needed to improve patient care.
The rapid and accurate profiling of pathogen genotypes in complex samples remains a challenge for existing molecular detection technologies. Currently, the identification of bacterial infections relies primarily on culture-based detection and phenotypic identification processes that require several days to weeks to complete. The practical application of molecular profiling technology is limited by several factors. To replace culture, molecular approaches must capture an equally wide array of pathogens while also providing specific and sensitive identification in a turnaround time fast enough to impact clinical decision making1-3. Studies also suggest that quantification of pathogen load may offer added benefits beyond what culture can offer4. However, the number of microbial genomes present in a clinical sample may be extremely low and/or the sample may be comprised of several different microbes. Current bacteria-targeted rapid screening technologies suffer from non-specific hybridization (e.g. microarrays, FISH), non-specific protein signals (e.g. protein mass spectrometry), or limited resolution of species (e.g. nucleotide mass spectrometry)5-7. Sequencing with conserved primers targeting the 16S or rpoB genes is the most useful molecular approach for detecting a wide range of bacteria with broad sensitivity, but is a time-consuming process that requires non-trivial technical expertise, computational resources, and analysis time. Moreover, recent studies report that several NGS platforms for microbial detection approach the analytical sensitivity of standard qPCR assays3. For applications where turnaround time is critical, high-level multiplexing of PCR-based identification strategies remain an active area of research.
Detection of microbial DNA in clinical samples like blood poses unique challenges. In this case, human DNA often vastly outnumbers that of the infecting microbe, necessitating targeted microbial DNA amplification. Since a vast array of pathogens can cause infection, a broad-based approach for amplifying all microbial DNA and a high level of specificity in post PCR sequence characterization techniques are required. Highly related organisms may have few DNA sequence differences, requiring single nucleotide specificity. Moreover, the number of microbial genomes present in the sample may be extremely low and/or the sample may be comprised of several different microbes, a polymicrobial infection, necessitating single cell level analysis. These requirements challenge the abilities of current molecular detection technologies, including next generation sequencing (NGS).
In one aspect, the disclosure provides a method comprising combining a sample comprising a target nucleic acid sequence with universal amplification primers, one more DNA intercalating dye and an amplification mixture to form a reaction mixture; partitioning the reaction mixture into 10,000 or greater partitions (e.g., 10,000 or greater partitions, 20,000 or greater partitions, 50,000 or greater partitions, 100,000 or greater partitions, or 1,000,000 or greater partitions; amplifying the target nucleic acid sequence; detecting the amplified nucleic acid by simultaneously heating and imaging the partitioned reactions; and performing melt curve analysis using machine learning.
In one aspect, a method of profiling the sequence of a target nucleic acid in a sample is provided, the method including combining a sample including a target nucleic acid with one or more universal amplification primers, one or more DNA intercalating dyes, and reagents for amplifying a target nucleic acid to form a reaction mixture; partitioning the reaction mixture into 10,000 or greater partitions; amplifying the target nucleic acid sequence to produce an amplicon; performing melt analysis of the amplicon by simultaneously heating and imaging the partitions, whereby a melt curve for the amplicon is produced; and profiling the sequence of the target nucleic acid by comparing the melt curve for the amplicon to a plurality of melt curves from different nucleic acid sequences.
In some embodiments, the method can optionally include one or more of the following features. The reaction mixture can further include a reference dye. The method can further include detecting the reference dye and excluding noise values based on the detection of the reference dye. The reaction mixture can include a surfactant in an amount of from about 0.01% to about 0.8% v/v. The reagents for amplifying a target nucleic acid can include a high fidelity polymerase. Performing a melt curve analysis can include excluding melt curves below a threshold melt temperature. The reaction mixture can include at least two, at least three, at least five, at least ten unique universal amplification primers, or between two and six unique universal amplification primers. The one or more universal amplification primers can be selected from bacterial primers, fungal primers, viral primers, and combinations thereof. The target nucleic acid sequence can have an amplicon having a size greater than 1000 base pairs.
In some embodiments, the heating can be performed at a heating rate of 0.005° C./s to 0.5° C./s. Performing melt analysis of the amplicon by simultaneously heating and imaging the partitions can include performing imaging at an imaging rate that is synchronized with the heating rate. The imaging rate can be synchronized with the heating rate to maintain a resolution of from about 0.005° C. to about 0.1° C.
In some embodiments, the intercalating dye can be a fluorescent dye selected from the group consisting of EvaGreen SYBR Green I, LC Green, LC Green Plus, ResoLight, Chromofy and SYTO 9. In some embodiments, the reference dye is not an intercalating dye. The reference dye can be ROX.
In some embodiments, the amplicon originates from less than two target nucleic acid molecules in the sample. In some embodiments, the amplicon originates from a single target nucleic acid in the sample.
In some embodiments, the method can optionally include one or more of the following features. A plurality of melt curves can include at least one reference melt curve from a known nucleic acid sequence. Profiling the sequence of a target nucleic acid can be automated by use of a computer model or algorithm that classifies the target nucleic acid based on the comparison to the reference melt curve. Profiling the sequence of the target nucleic acid can include comparing the melt curve for the amplicon to a plurality of melt curves from different nucleic acid sequences. The computer model algorithm can be a machine learning algorithm for automated melt curve identification classification. In some embodiments, the algorithm can identify a target sequence versus junk, identify a novel pathogen, or classify a known sequence. The machine learning algorithm can use a probabilistic model for detection of nucleic acid signatures and for classification into a known pathogen or a novel pathogen unknown to the database. The classification model selected from a generative classifier and a discriminative classifier. The generative classifier can be a Naïve Bayes classifier. The discriminative classifier can be a Support Vector Machine (SVM) or Logistics Regression. In some embodiments, the detection of unknown pathogen can use Dynamic Time Warping. Profiling the sequence of the target nucleic acid can include comparing melt curve shape or melting temperature (Tm) of the amplicon. In some embodiments, profiling of nucleic acid can also be associated with a confidence score, such as, e.g., a confidence score generated using entropy measures.
In some embodiments, partitioning the reaction mixture can include fractionating the sample to contain zero to one nucleic acid molecules prior to the combining step. In some embodiments, the sample can include mammalian DNA.
In some embodiments, the method can include subjecting the sample to a first heating rate to obtain a first melt curve signature; subjecting the sample to a second heating rate to obtain a second melt curve signature; and performing a heating rate-dependent melt curve analysis using a computer model or algorithm that classifies the target nucleic acid based on the comparison to the reference melt curve. The computer model algorithm can optionally be a machine learning algorithm for automated melt curve classification. The machine learning algorithm can optionally use a probabilistic classification model selected from a generative classifier and a discriminative classifier.
In some embodiments, the method can further include identifying a pathogen or pathogen marker present in the sample, wherein the pathogen or pathogen marker has a known nucleic acid melt-curve signature. In some embodiments, the method can further include identifying a pathogen or pathogen marker present in the sample, wherein the pathogen or pathogen marker has an unknown nucleic acid melt-curve signature. In some embodiments, identifying a pathogen or pathogen marker present in the sample can have an error rate of less than 0.2%.
In some embodiments, the method can further include assigning confidence scores to sequence profiles obtained from profiling the sequence of the target nucleic acid. In some embodiments the machine learning algorithm can use a probabilistic model to assign confidence scores on classification. In some embodiments the algorithm can use an entropy measure, such as, e.g., Shannon entropy.
In some embodiments, partitioning the reaction mixture includes loading the reaction mixture onto a digital PCR chip. The method can further optionally include controlling a temperature difference between any two points on the chip during heating to 0.6° C. or less when performing melt analysis of the amplicon.
In some embodiments, partitioning the reaction mixture can include fractionating the reaction mixture into droplets using a ddPCR platform.
In some embodiments, the method further includes partitioning the reaction mixture into 20,000 or greater partitions, into 50,000 or greater partitions, into 100,000 or greater partitions, or into 1,000,000 or greater partitions.
In another aspect, a method of treating a patient suspected of having sepsis is provided, the method including identifying one or more pathogens having a high probability of mediating the patient's sepsis by performing the method of any one of claims 1 to 37; and administering to the patient one or more antimicrobials effective against the one or more identified pathogens.
In another aspect, a system is provided, including a digital sample partitioning platform; a thermoelectric heating device; one or more temperature feedback sensors; a temperature controller configured to control the thermoelectric heating device based on temperature feedback from at least one of the one or more temperature sensors; an imaging device; and a device including a processor configured to perform machine learning for automated melt curve classification.
In some embodiments, the system can optionally further include one or more of the following features. The digital sample partitioning platform can include a reaction mixture including one or more nucleic acids and a high fidelity polymerase. The reaction mixture can further include a reference dye. The reaction mixture can further include a surfactant in an amount of from about 0.01% to about 0.8% v/v. The reaction mixture can include at least two, at least three, at least five, or at least ten unique universal amplification primers. The reaction mixture can include between two and six unique universal amplification primers. The one or more universal amplification primers can be selected from bacterial primers, fungal primers, viral primers, and combinations thereof. The target nucleic acid sequence can have an amplicon having a size greater than 1000 base pairs. The temperature controller can be a proportional-integral-derivative (PID) controller. The system can further include a conductive plate positioned between the digital sample partitioning platform and the heating device. The conductive plate can have a thickness of from about 0.02 mm to about 6 mm. The system can further include a thermal paste positioned between the digital sample partitioning platform and the conductive plate. The system can further include thermal paste positioned between the thermoelectric heating device and the conductive plate. The system can further include a heat sink. The system can further include a fan. The imaging device can have a high numerical aperture, a low magnification, and a large field of view. The imaging device can include a multi bandpass filter.
In some embodiments of the system, the digital sample partitioning platform can be a digital PCR chip. In some embodiments of the system, the digital sample partitioning platform can be a digital droplet PCR platform.
In some embodiments, the universal PCR primers are designed to bind to conserved regions within the sequences of interest. Variable regions within the targeted sequences are required to generate unique melt curves.
In some embodiments, the PCR amplification mix is an optimized PCR master mix. The reaction mixture is loaded onto a digital PCR chip by smearing the reaction mixture across the chip and allowing hydrophobic/hydrophilic interactions to draw the reaction mix into the wells.
In some embodiments, amplification of the target nucleic acid sequence is performed by placing digital PCR chip on flatbed thermocycler and cycling reaction temperature using a standard 3-step PCR cycle. This amplifies the signal given by the DNA dye intercalated within the double-stranded DNA.
In some embodiments, simultaneously heating and imaging the digital PCR chip is performed. The chip is heated in a controlled manner using a Peltier element. The control system can, in some embodiments, use a feedback system to provide temperature feedback to control the heating or cooling of the system. Simultaneously a microscope with a powerful LED and highly sensitive CMOS sensor continuously images the fluorescence of the chip.
The images from the microscope are analyzed using code that automatically reads the intensity from each well and relates it to the temperature of the chip, generating the high resolution melt curves. After normalizing the curves they are identified using a supervised machine-learning code based on the shape and temperature of the melt curves.
Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs. Methods and materials are described herein for use in the present invention; other, suitable methods and materials known in the art can also be used. The materials, methods, and examples are illustrative only and not intended to be limiting. All publications, patent applications, patents, sequences, database entries, and other references mentioned herein are incorporated by reference in their entirety. In case of conflict, the present specification, including definitions, will control.
Other features and advantages of the invention will be apparent from the following detailed description and figures, and from the claims.
The patent or application file contains at least one drawing executed in color. Copies of this patent or patent application publication with color drawing(s) will be provided by the Office upon request and payment of the necessary fee.
The present disclosure is based in part of the development of an integrated platform enabling the identification of nucleic acid sequences, such as, e.g., bacterial, viral, or fungal pathogen DNA or RNA sequences, in complex samples, in some cases in less than four hours. In some aspects, the platform incorporates a microfluidic chip or digital droplet platform and instrumentation to accomplish universal nucleic acid amplification, High Resolution Melting (HRM), and machine learning (e.g., profiling of the resultant melt curves with a machine learning-assisted classification algorithm) within at least 20,000 picoliter scale reactions simultaneously. Clinically relevant concentrations of pathogen nucleic acid molecules (e.g., bacterial DNA molecules) are separated by digitization across 20,000 or more reactions and amplified with universal primers. Amplification is followed by HRM sequence fingerprinting in all reactions simultaneously. The resulting pathogen-specific melt curves are identified by one or more machine learning-assisted classification algorithms (e.g., by Support Vector Machine (SVM) learning), and individual pathogen loads can be quantified.
As used herein “nucleic acid amplification” or “amplification” are used interchangeably and refer to reaction in which replication of a nucleic acid sequence occurs repeatedly over time to form multiple copies of at least one segment of a template nucleic acid molecule. Amplification may generate an exponential or linear increase in the number of copies as amplification proceeds. Typical amplifications produce a greater than 1,000-fold increase in copy number and/or signal. Exemplary amplification reactions for the assays disclosed herein may include the polymerase chain reaction (PCR) or ligase chain reaction (LCR), each of which is driven by thermal cycling. Thermal cycling generally involves cycles of heating and cooling a reaction mixture to perform successive rounds of denaturation (melting), annealing, and extension.
Amplification may be performed with any suitable reagents. Amplification may be performed, or tested for its occurrence, in an amplification mixture, which is any composition capable of generating multiple copies of a nucleic acid target molecule, if present, in the composition. An amplification mixture may include any combination of at least one primer or primer pair, at least one probe, at least one replication enzyme (e.g., at least one polymerase, such as at least one DNA and/or RNA polymerase, and/or at least one ligase), and/or deoxynucleotide (and/or nucleotide) triphosphates (dNTPs and/or NTPs), among others.
As used herein, the term “amplicon” refers to a product of an amplification reaction. An amplicon may be single-stranded or double-stranded, or a combination thereof. An amplicon corresponds to a segment or or the entire length of a nucleic acid target molecule.
The disclosed platform combines the ability of digital PCR to quantitatively determine sample concentration with the sequence identification of high resolution melt curves enabling counting and analysis of specific sequences within a mixed sample. In a first part of one assay embodiment, the sample is spread across 20,000 PCR reaction wells within a digital PCR chip. All sequences of interest are amplified using universal primers. The high resolution melt curves are generated by heating the amplified sequences in the presence of a DNA intercalating dye and simultaneously imaging the chip to determine fluorescence loss as a function of temperature. Using a machine-learning algorithm the sequences are identified and counted based on their fingerprint melt curves.
High resolution melt alone is capable of sequence identification based on fingerprint melt, but does not provide any quantitative information. Digital PCR is unable to provide quantitative information about mixed samples and is limited by the requirements put in place by TaqMan probes. Sequencing can provide sequence-specific quantitative information, but is limited by short sequence reads making it inappropriate for studying alternative splice variants and by the necessity for a large starting amount of material to give accurate reads.
The platform provides several advantages including, for example, reducing reaction volumes by 99.995% and achieving a greater than 200-fold increase in dynamic range of detection compared to traditional PCR HRM approaches. Type I and II error rates are reduced by 99% and 100% respectively compared to intercalating dye-based digital PCR (dPCR) methods. The platforms described herein reduce errors resulting from temperature gradients across sample plates. The platform allows for sensitive melt curve analysis, including heating-rate-dependent analysis. With the platforms described herein, identification and quantification of multiple species and even multiple variants within same species organisms can be achieved from heterogeneous samples, including samples contaminated with host or other non-target DNA and RNA. Thus, the methods and systems disclosed herein impact a number of quantitative profiling applications, especially infectious disease diagnostics. Additionally, some embodiments of the platform provide improved uniformity and linearity of thermal control during melting analysis in general.
The combined capabilities of universal dHRM offer significant advantages over qPCR, microarray, and even NGS profiling approaches, especially for infectious disease detection where breadth of detection, timing, and cost are critical factors. The ability to quantify even low level organisms in polymicrobial samples within hours significantly impacts clinical microbiology diagnostic practice, where patients suspected of bloodstream infection suffer from an hourly increase in mortality risk due to lack of diagnostic information and inaccurate treatments. Moreover, retrospective studies suggest that absolute quantification of bacterial genomic load in patients may useful to assess severity of infection and to predict prognosis in sepsis cases.
The term “high resolution melt” or “HRM” refers to a technique for determining a sequence variation in a nucleic acid by analyzing a melting curve of the nucleic acid. The nucleic acid can be double-stranded or single-stranded. In some embodiments, a signal representing a double-stranded nucleic acid can be measured in real time. In some embodiments, real time measurement (Ct) of the PCR can be used as a quality control value for melt analysis such as HRM analysis. In some embodiments, end point analysis can be employed. In some embodiments, single-stranded nucleic acids can be analyzed, for instance, single-stranded nucleic acids can fold to form duplexes or other higher ordered structures that can then be analyzed. Herein, the term “high resolution melt” or “HRM” can also refer to a technique for determining sequence variations between two different nucleic acids by analyzing the shape of the melting curve including the melting temperature and the slope.
HRM has been validated as a rapid, inexpensive, closed-tube DNA sequence characterization technique. Precisely heating and unwinding post-PCR DNA amplicons in the presence of a fluorescent intercalating dye8-10 or sloppy molecular probes11,12 loss-of-fluorescence melt curves are generated, providing unique DNA sequence signatures. The expansion of HRM into a broad-based profiling technology by preceding it with universal PCR13 has been proposed. Priming conserved DNA regions flanking genetic variation sites or mutations, genetic locus sequence differences can be identified by changes in the gene amplicon melt curve signature. This universal HRM technique replaces the need for targeted primers or probes and relies only on the intrinsic melting properties of the amplified sequence. Universal HRM methods have been developed for several applications, including identification of oncogenic mutations14, gene methylation patterns15,16, and bacterial identification17-22. The inventors previously advanced universal HRM to enable single nucleotide specificity for the discrimination of microRNA in the Lethal-7 family and for species-level identification of bacteria using the 16S gene23,24. However, if multiple sequence variants are present, as often occurs in clinical samples, individual sequences cannot be identified in the conventional universal HRM format consisting of a single bulk reaction13,25. Likewise, although generally reproducible melt curves are obtained, in-run template standards are typically required to overcome run-to-run variability and enable curve matching by user intensive curve identification procedures. These shortcomings have restricted the application of universal HRM to primarily pure homogeneous samples, constrained the breadth of profiling to only a few sequence variants, and limited the technique's specificity, since single nucleotide changes often manifest as very slight temperature or curve shape changes.
The inventors previously developed an approach called universal digital high resolution melt (U-dHRM) by integrating universal amplification strategies and temperature calibrated HRM with limiting dilution digital PCR (dPCR) in a 96-well plate format23. The inventors demonstrated that this approach, in principle, could overcome many limitations of current profiling technologies to achieve single nucleotide specificity, broad-based detection, single molecule sensitivity, and absolute quantification simultaneously. Separately, the inventors developed machine learning approaches using nested, linear kernel, One Versus One Support Vector Machines (OVO SVM) to automatically identify sequences by their melt curve signatures despite inherent experimental variability24,26. Through these approaches, the inventors have shown that U-dHRM is capable of automatically identifying multiple distinct genotypes in a mixture with single molecule sensitivity and single nucleotide specificity. Others have also demonstrated the ability of U-dHRM to sensitively detect rare mutants/variants27,28 and also novel variants29. These findings suggest that U-dHRM has the potential to offer desirable features for several profiling applications that require a combination of speed, sensitivity, quantitative power, and broad profiling ability. However, no platform exists for accomplishing U-dHRM in a high-content format required to reach a clinically relevant dynamic range of detection.
The sensitivity and quantification power of U-dHRM profiling relies on full digitization of the sample, i.e. spreading the sequence mixture across many reactions so each target molecule is isolated from others. Since the process of loading DNA into wells is stochastic at limiting dilutions, the dynamic range of single molecule detection follows a Poisson distribution, requiring the total number of reactions to be approximately 10 to 100 times the number of sequence molecules. That is, the average occupancy (λ) across all reactions must be 0.1 to 0.01 copies of DNA per well. The probability of DNA occupancy in any well, i.e. the fraction of wells having 1, 2, 3, etc. copies, is given by the Poisson probability distribution P=(e−λ*Xn)/n!, where n is the total number of wells. U-dHRM is currently performed in traditional PCR multi-well plates using HRM enabled qPCR machines. In this format, only about 9 molecules in a sample can be profiled at the single molecule level per 96-well plate (
In one aspect, the present disclosure provides a digital melt analysis platform that integrates sample partitioning, along with heating and imaging components. The sample can include one or more target nucleid acid sequences. In some embodiments, the sample can be a heterogeneous sample. In some embodiments, the sample can include mammalian DNA, bacterial DNA or RNA, viral DNA or RNA, fungal DNA or RNA, or combinations thereof. The methods, systems, and devices of the present disclosure can perform highly sensitive high throughput, multi-dimensional, DNA melting analysis to achieve nucleic acid sequence fingerprinting, screening, and anomaly detection with single molecule sensitivity. The methods, systems, and devices of the present disclosure achieve simultaneous high sensitivity imaging and precisely controlled heating of arrayed, small volume, low thermal mass, single molecule nucleic acid amplification reactions partitioned according to a digital assay platform, such as digital PCR (dPCR) or droplet digital (ddPCR). The methods, systems, and devices can include pairing the integrated platform with classification systems (e.g., machine learning algorithms) to correct for errors and analyze the melt curves for identification and quantification.
In one aspect of the methods, assays, and systems described herein, a sample, e.g., a heterogeneous sample, is fractionated or partitioned into many small volume PCR reactions, such that each reaction contains two or fewer target molecules, or zero target molecules, or one target molecule. Thus, amplicons resulting from each reaction will, in some cases originate from two or fewer target nucleic acid molecules in the sample, less than two target nucleic acid molecules in the sample, or one or fewer target nucleic acid molecules in the sample. Each partitioned reaction can have a discrete volume that is the same or different from other reactions originating from the same sample. In some embodiments, the partitioned reactions are contained within, for example, a droplet in a ddPCR platform or a well on a dPCR chip. Each reaction can have a volume of one nanoliter or less, one picoliter or less, or less than about one picoliter. dPCR chips are microfluidic devices that allow partitioning and manipulation of small volumes or droplets of samples and/or reaction mixtures using electrowetting or electrophoresis. ddPCR platforms divide reaction mixtures into water-in-oil droplets using a combination of microfluidics surfactant or detergent.
In some embodiments, partitioning a sample or reaction mixture includes partitioning the sample or reaction mixture into at least 10,000 partitions, droplets, or wells. In some embodiments, partitioning a sample or reaction mixture includes partitioning the sample or reaction mixture into at least 20,000, at least 50,000, at least 100,000, at least 200,000, at least 500,000, or at least 1,000,000 partitions, droplets, or wells. Partitioning platforms useful in the methods and systems described herein include droplet digital platforms and digital microfluidic chips.
The term “droplet” refers to a volume of liquid. Droplets may, for example, be aqueous or non-aqueous or may be mixtures or emulsions including aqueous and non-aqueous components. Droplets may take a wide variety of shapes; nonlimiting examples include generally disc shaped, slug shaped, truncated sphere, ellipsoid, spherical, partially compressed sphere, hemispherical, ovoid, cylindrical, combinations of such shapes, and various shapes formed during droplet operations, such as merging or splitting or formed as a result of contact of such shapes with one or more surfaces of a droplet actuator (i.e., a device for manipulating droplets). Moreover, a droplet may include a reagent, such as water, deionized water, saline solutions, acidic solutions, basic solutions, detergent solutions and/or buffers. Other examples of droplet contents include reagents, such as a reagent for a biochemical protocol, such as a nucleic acid amplification protocol, an affinity-based assay protocol, an enzymatic assay protocol, a sequencing protocol, and/or a protocol for analyses of biological fluids.
In some embodiments, the sample or reaction mixture is partitioned on a microfluidic chip platform, including chips having an array of reactions configured such that the reactions can fit into a single field of view for imaging. For example, in some embodiments, a dPCR chip may be used. Exemplary dPCR chips can include, for example, a silicon substrate etched with nano-scale or smaller reaction wells. In some embodiments, a dPCR chip has a low thermal mass. For example, the chip may be constructed of thin, highly conductive materials that do not store heat energy. In some embodiments, a dPCR chip has a surface area of from about 50 mm2 to about 150 mm2. In some embodiments a dPCR chip has a surface area of about 100 mm2. Limiting the surface area can allow for greater uniformity of heating of the chip during melt analysis and a reduction in run-to-run variation in the melt cure analysis, a reduction in errors in melt curve generation, and increased discrimination of melt curves in the analysis.
In some embodiments, the sample or reaction mixture is partitioned in a droplet digital platform such as ddPCR. Continuous-flow droplet systems can, in some embodiments, provide enhanced processing speed through limiting heat inertia mostly to the droplet sample's thermal mass.
The inventors have discovered that, in some embodiments, smaller reaction volumes can enable improved sensitivity as compared to typical PCR volume formats by, e.g., reducing the effective level of background DNA and inhibitors that interfere with nucleic acid amplification and DNA melt analysis for successful amplification and HRM from single molecules. It has further been discovered that smaller reaction volumes allows for a higher packing density of reactions such that all reactions can be imaged simultaneously using a large field of view camera. Imaging all reactions simultaneously includes collecting all melt curves for individual target sequences simultaneously, resulting in faster throughput of HRM and more efficient error reduction. Additionally, it has been discovered that increasing the number of reactions analyzed enables a higher dynamic range of detection and higher throughput of HRM.
One or more samples having a target nucleic acid are combined with other components to generate a reaction mixture. Reaction mixture components can include, in addition to the sample, e.g., one or more universal amplification primers, one or more DNA intercalating dyes, and reagents for amplifying a target nucleic acid. In some embodiments, the reaction mixture can be generated prior to loading the sample contained in the reaction mixture onto the dPCR chip or ddPCR platform. In some embodiments, the sample can be loaded onto the chip or ddPCR platform and partitioned prior to mixing in the remaining components of the reaction mixture. In some embodiments, different components of the reaction mixture can be added at different times or stages. For example, any of the reaction mixture components described herein can be added simultaneously before or after partitioning. In another exemplary embodiment, some reaction mixture components may be added to the sample prior to partitioning, while other components may be added later, such as after partitioning.
The reaction mixture can include one or more universal amplification primers. Exemplary universal amplification primers useful in the methods and systems described herein include bacterial primers, fungal primers, viral primers, and combinations thereof. Some examples of useful primers include universal primers targeting the bacterial 16S gene, Flavivirus NS1, NS3, NSS, and 3′ NC gene regions, and fungal internal transcribed spacer 1 (ITS1), internal transcribed spacer 2 (ITS2), 5.8s rDNA, 18S, and 28S gene regions may be targeted. In some embodiments, the reaction mixture can include combinations of at least two, at least three at least five, or at least ten unique universal amplification primers. In some embodiments, the reaction mixture can include between two and six unique universal amplification primers. In some embodiments, combinations of primers allow multiplexing for broad-based detection of multiple taxonomic units of organisms in a single sample.
In some embodiments, detection of bacterial, fungal, and viral pathogens can utilize different fluorescence channels. For example, a well-characterized universal fungal primer set targeting the ITS-1 region can be redesigned with a FRET quenched blue fluorophore (LUX primer technology, Thermo Fisher) such that upon incorporation into an amplicon, blue fluorescence will be generated. Similarly, a primer set targeting the HSV I and II DNA polymerase gene can be labeled with a Cy5 probe. Reactions positive for only green fluorescence can be analyzed according to the bacterial and resistance gene databases, while reactions positive for blue and green fluorescence will be identified as fungal and reactions positive for Cy5 fluorescence can be identified as HSV. Such multiplexed system can allow sorting of fungal and viral melt curves from bacterial and resistance melt curves, and can, in some embodiments, effectively triple the temperature range of melt curve output. In some multiplex embodiments, melt curve can be generated in a single fluorescence channel for all DNA, while fluorescence in other channels determines which database will be used for pathogen identification.
In some aspects, the amplification reactions described herein amplify an amplicon having a size greater than 500 base pairs, 750 base pairs, 1000 base pairs or 1500 base pairs in length. While not being bound by theory, it is believed that an amplicon having a size greater than 500 base pairs, 750 base pairs, 1000 base pairs or 1500 base pairs in length allows for improved sequence discrimination and/or identification, thus minimizing background detection of degraded environmental/non-relevant DNA, which is shorter in length.
The reaction mixture can include reagents for amplifying a target nucleic acid, such as one or more polymerases. In some embodiments, the polymerase is a high fidelity polymerase, such as a Phusion or similar high fidelity polymerase. The use of high fidelity polymerase can improve yield and allow for reduced reaction volume, as well an ensure accuracy of amplification to result in more accurate melt curve analysis.
The reaction mixture can also include one or more buffers. In some embodiments, the buffer is a high fidelity buffer. In some embodiments, buffer hydrophilicity is optimized by addition of detergents or surfactants to enable reliable loading of master-mix into reactions on a digital PCR chip or in droplet digital PCR reactions. One or more detergents or surfactants can be present at a concentration of from about 0.005% to about 0.85% v/v, or from about 0.01% to about 0.8% v/v. In an exemplary embodiments, from about 0.01% to about 0.8% v/v Brij L4 can be included in the reaction mixture.
The reaction mixture can include one or more DNA intercalating dyes. During melt the fluorescence of the DNA intercalating dye is lost as a function of temperature, allowing for imaging and measurement of the melt curve of the amplicon. In some embodiments, low-inhibitory DNA intercalating dyes can be used. For example, in some embodiments, EvaGreen dye (Biotium) can be included in the reaction mixture. Other exemplary dyes useful in the methods and systems described herein include SYBR Green, Chai Green, LC Green, BoBo, ToTo, and the like.
In some embodiments, the reaction mixture can include one or more reference dyes. Reference dyes are non-DNA-interchalating dyes that allow for excluding noise values in the melt curve analysis. In some embodiments, the reference dye is used to normalize the melt dye (DNA interchalating dye) at each temperature point. Any passive water soluble dye that does not interfere with PCR or melt analysis can be used. Some exemplary reference dyes useful in the methods and systems described herein include Tamra, TexasRed, AlexaFluor, Rox, and the like.
Following loading of the reaction mixture onto a dPCR chip or ddPCR platform, universal PCR is conducted, then a melt analysis is conducted for all reactions simultaneously. Since each reaction amplifies from a single target molecule, each digital melt curve is a sequence fingerprint of only one sequence within a given heterogeneous sample. Some methods and systems described herein use an integrated dPCR chip or ddPCR platform, heating device, and imaging device.
One exemplary embodiment of a system described herein and useful in the methods described herein is shown in
An embodiment of a reaction and heating portion 900 of an exemplary system described herein and useful in the methods described herein is shown in
A dPCR reaction chip 921 holds the reaction mixture adjacent to an opening 917 in the enclosure 910 for imaging the reactions with an imaging device. In some embodiments, a surrogate dPCR chip 922 may also be present within the enclosure. The surrogate chip 922 can, in some embodiments, include a temperature feedback device or sensor 925 (e.g., a thermocouple or a resistance temperature detector (RTD)) to provide data for estimating the temperature inside the reaction dPCR chip and determining when the dPCR chip reaches the target temperature. In some embodiments, a temperature feedback device or sensor (e.g., a thermocouple or a resistance temperature detector (RTD)) can be placed within the dPCR reaction chip rather than a surrogate chip. In some embodiments the temperature feedback sensor can be situated in the center of a dPCR chip.
The one or more dPCR chips 921, 922 are positioned adjacent, or embedded in a first conductive block or plate 931. The conductive block or plate can be any conductive material, such as, for example, gold, silver, copper, copper polished with gold, or any other material that allows for efficient heat transfer. In some embodiments, the first copper block can have a thickness of between about 1 mm and about 10 mm, about 2 mm and about 8 mm, or between about 0.02 mm and about 6 mm. In some embodiments, the conductive block can be u-shaped, with side walls ranging from about 1 mm to about 5 mm in thickness and a base ranging from about 1 mm to about 10 mm in thickness. In some embodiments, a thin conductive film can be used. The conductive block 931 is position between the one or more dPCR chips 921, 922 and a thermoelectric device 940. The thermoelectric device 940 is used to heat the dPCR chip to melt target nucleic acid molecules for melt curve analysis.
The thermoelectric device 940 can be controlled by a controller (not shown), such as, e.g., a proportional, integral, derivative (PID) controller. The controller can also control optional fan 960, such as by turning the fan on or off and/or controlling fan speed. Fan 960 that has a surface 968 exposed to air at room temperature. In some embodiments, the controller receives feedback from one or more sensors, such as, e.g. a sensor 935 embedded in the conductive block 931, or sensor 955 embedded in heat sink 950. Control of the fan can be based on, e.g., data from one or more sensors such as the heat sink sensor 955. In some embodiments, sensor 935 is embedded in the center of conductive block 931. In some embodiments, sensor 955 is embedded in the center of heat sink 950. The heat sink 950 and fan 960 can provide enhanced and controllable heat dissipation, allow for minimization of between-run times, and improve the ability to heat precisely to higher temperatures.
The sensors can provide temperature feedback to the temperature controller to allow for fast and sensitive control of the system temperatures through control of the thermoelectric device 940 and the fan 960. Sensors useful in the system include, e.g., RTD sensors, negative temperature coefficient (NTC) thermistors, and the like. In some embodiments, a second conductive block or plate 932 is positioned between the thermoelectric device 940 and heat sink 950. Sensors can provide feedback control of the heating and cooling system. In some embodiments, thermal paste or thermal grease (not shown) may be present at an interface between the dPCR reaction chip and the first conductive block, the dPCR surrogate chip and the first conductive block, the thermoelectric device and the first conductive block, the thermoelectric device and the second conductive block, or combinations thereof. Thermal paste can improve heat transfer between the various components of the system, e.g., between the conductive block and the dPCR chip. Exemplary thermal pastes or greases that can be used include, e.g., Arctic Silver 5 Thermal Compound and OmegaTherm 201. In some embodiments, the inclusion of sensors and/or thermal paste can allow for repeatable heating between runs, limit instantaneous error and fluctuations, and reduce the need for visual inspection of the melt curve data or curve shifting based on between-run heating fluctuations. In some embodiments, temperature of the first conductive block can be controlled precisely with a maximum error of 0.004° C. difference between the expected temperature and the actual temperature.
The exemplary embodiment of
The systems can further include an imaging device for imaging the loss of fluorescence of the DNA interchalating dye during heating of the platform to conduct melt curve analysis. In some embodiments, the imaging device can include a microscope. The imaging device can, in some embodiments, have a high numerical aperture, low magnification, a large field of view, or combinations thereof. The imaging device is configured to simultaneously capture all wells/melt curves without compromising signal quality. In some embodiments, the imaging device is configured to simultaneously image a dPCR chip having a size of 10 mm by 10 mm and containing 20,000 reactions in a single field of view (FOV). In some embodiments, the imaging device can provide fast triggering of the fluorescent light source. In some embodiments, the imaging device includes a polychroic or multi bandpass filter. In some embodiments, the imaging device is configured to enables near simultaneous imaging of DNA interchalating dye and reference dye in two channels (e.g., red and green fluorescence) while maintaining a rapid heating and imaging rate. In some embodiments, the imaging device is configured to maintain high temperature resolution in the generated melt curves (images of the melt dye captured per degree rise in temp). In some embodiments, the imaging device is configured to capture two fluorescent channels at each temperature point by triggering the light source at from about 50 cycles per second to about 100 cycles per second (about 50 Hz to about 100 Hz. In some embodiments, the imaging device is configured to trigger the light source at from about 50 cycles per second to about 100 cycles per second (about 50 Hz to about 100 Hz) when the target heating rate of the dPCR chip or ddPCR platform is from about 0.25° C. to about 1° C. per second such that the heating and imaging are synchronized.
In some embodiments, the imaging device is configured to simultaneously image in multiple light channels while the dPCR chip or ddPCR platform is being heated. In some embodiments, the system can achieve multiple channel fluorescence (i.e., a two-channel fluorescence, a four-channel fluorescence, a six-channel fluorescence, etc.) necessary for bacterial, fungal, and viral detection in a simplified and rapidly-triggered optical configuration. In some embodiments, a motorized filter wheel may be used. In some embodiments, a filter slider can be used host a single custom exciter-emitter filter cube, which can be multibandpass with a polychroic mirror (Chroma) to reflect 2, 3, 4, 5, 6, 10, or more excitation spectra and allow the respective number of emission spectra to pass through. For example, a custom filter slider can be used to enable simultaneous imaging of EvaGreen melt curves in FITC channel, ROX loading control in TRITC channel, and LUX primers in DAPI and Cy5 channels while avoiding slow motorization speeds of a filter wheel.
The systems can further include an analysis device having a processor configured to perform machine learning for automated melt curve classification. In performing the melt analysis of the amplicon by simultaneously heating and imaging the digital PCR chip or ddPCR platform, a melt curve for the amplicon is produced. The analysis device processor can be configured to profile the sequence of the target nucleic acid by comparing the melt curve for the amplicon to a plurality of melt curves from different nucleic acid sequences. The plurality of melt curves can include at least one reference melt curve from a known nucleic acid sequence. The analysis device processor can be configured to automate profiling of the sequence of a target nucleic acid by use of a computer model or algorithm that classifies the target nucleic acid based on the comparison to the reference melt curve. The computer model algorithm can be a machine learning algorithm for automated melt curve classification. In some embodiments, the machine learning classification algorithm can use a Naive Bayes classifier. In some embodiments, the analysis device processor can be configure to profile the sequence of a target nucleic acid is performed using a Support Vector Machine (SVM). The analysis device processor can be configured to compare the melt curve for the amplicon to a plurality of melt curves from different nucleic acid sequences to profile the sequence of a target nucleic acid. The analysis device processor can be configured to exclude melt curves below a threshold melt temperature to profile the sequence of a target nucleic acid. In some embodiments, exclusion of melt curves below a threshold melt temperature can improve type I and II errors by correcting for short non-specific amplification products that melt at a lower temperature than the target amplicon. The analysis device processor can be configured to compare melt curve shape or melting temperature (Tm) of the amplicon.
In addition to standard chip formats (e.g., dPCR chip platforms described above), droplet-based systems, e.g., droplet high resolution melt platforms (ddPCR platforms), may be utilized to allow for continuous analysis of 1,000,000 or more droplets or reactions. The imaging components used in the droplet based systems can be the same as those described above for the dPCR chip platforms. For heating, droplet based systems can utilize PID controlled stable temperature gradients. The heating rate can be controlled by flow speed of droplets and steepness of the temperature gradient, which allows for varying the heating rate. The flow rate can determine how many droplets are analyzed per second. Since the droplet system can be operated continuously, the number of reactions analyzed is limited only by the number of droplets that are made. Droplet generation could be made to run continuously as well and a PCR cycling region could be added to the heating apparatus and fluidic chip to make a completely continuous system from sample prep, droplet making, thermal cycling for amplification, and HRM melting for sequence identification.
The systems and devices described herein can be used in the methods and assays described herein. In one aspect, a method of profiling the sequence of a target nucleic acid in a sample is provided, the method including combining a sample including a target nucleic acid with one or more universal amplification primers, one or more DNA intercalating dyes, and reagents for amplifying a target nucleic acid to form a reaction mixture; partitioning the reaction mixture into 10,000 or greater partitions; amplifying the target nucleic acid sequence to produce an amplicon; performing melt analysis of the amplicon by simultaneously heating and imaging the partitions, whereby a melt curve for the amplicon is produced; and profiling the sequence of the target nucleic acid by comparing the melt curve for the amplicon to a melt curve from different nucleic acid sequences.
In some embodiments, the method can include controlling a temperature difference between any two points on the chip or platform during heating to 0.6° C. or less when performing melt analysis of the amplicon. In some embodiments, differences in the high temperature calibrator melt temperature on the chip or platform vary from 0.22° C. to 0.6° C., after excluding the outliers due to evaporation at the corners and edges of the chips or platforms. In some embodiments, with evaporation outliers included, median absolute deviation ranging from 0.05° C. to 0.1° C. and standard deviation of 0.06° C. to 0.13° C. occur across the chip or platform. As compared to conventional PCR, the smaller reaction volumes of dPCR could, in some embodiments, lead to larger variations in melt temperature (Tm) due to evaporation. However, in some embodiments, the small form factor of the digital chip or platform may maintain a more uniform thermal gradient across the chip or platform, leading to smaller variations in Tm. A previously published study reported Tm differences ranging from 0.35° C. to 1.24° C. across 32-96 well plate melt instruments with standard deviations of 0.018° C. to 0.274° C.44 Performance could be improved by optimizing for loading errors and evaporation. For example, the application of oil onto the loaded reaction wells could be sensitive to timing and amount deposited, and automation would ensure that the corner wells are covered as quickly as the central wells to minimize evaporation.
In some embodiments, the heating of the dPCR chip or ddPCR platform is performed at a heating rate of from about 0.002° C./s to about 1° C./s or from about 0.005° C./s to about 0.5° C./s. The melt analysis of the amplicon can be performed by simultaneously heating and imaging the digital PCR chip or ddPCR platform is performed at a rate that is synchronized with the heating rate. For example, heating of the conductive block can be independently controlled by standalone software, while the proxy temperature measurement from the surrogate chip or platform can be synchronized with fluorescent imaging by the microscope control software (e.g., NIS-Elements). Synchronizing imaging with temperature measurement can be accomplished with, for example, the use of an NIS-Elements compatible temperature acquisition system (Tokai Hit Co., Japan) using a K-type thermocouple probe. However, in some instances, such acquisition systems only provide limited resolution of temperature measurement to 0.1° C. with a temperature sampling rate of ˜0.2 Hz irrespective of the imaging rate. To compensate, the conductive block temperature can be precisely controlled, a repeatable relationship between the conductive block-embedded RTD and the surrogate chip-embedded or platform-embedded thermocouple established, and integrated thermocouple temperature data and fluorescence imaging data can be used to plot melting curves.
In some embodiments, the methods can further include performing automated melt curve classification with an analysis device having a processor configured to perform machine learning. The melt analysis of the amplicon can be performed, in some embodiments, by simultaneously heating and imaging the partitions to produce a melt curve for the amplicon. The method can include profiling the sequence of the target nucleic acid by comparing the melt curve for the amplicon to a plurality of melt curves from different nucleic acid sequences. The plurality of melt curves can include at least one reference melt curve from a known nucleic acid sequence. The method can include automated profiling of the sequence of a target nucleic acid by use of a computer model or algorithm that classifies the target nucleic acid based on the comparison to the reference melt curve. The computer model algorithm can be a machine learning algorithm for automated melt curve classification. In some embodiments, the machine learning classification algorithm can use a Naive Bayes classifier. In some embodiments, the method can include profiling the sequence of a target nucleic acid by using a Support Vector Machine (SVM). The method can include comparing the melt curve for the amplicon to a plurality of melt curves from different nucleic acid sequences to profile the sequence of a target nucleic acid. In some embodiments, the method can involve excluding melt curves below a threshold melt temperature to profile the sequence of a target nucleic acid. The method can include comparing melt curve shape or Tm of the amplicon. Probability based classification models useful in the methods and systems described herein include Gaussian Naïve Bayes (NB), multinomial Logistic regression (LR), and Multi-layer Perceptron (MLP) approaches for probabilistic classification. NB allows for incorporation of prior knowledge into the prediction model. For example, in some embodiments, epidemiological prevalence can be added as prior probabilities to provide added robustness of identification. Some embodiments can use equal priors, i.e. equiprobable organism classes.
In some embodiments, the algorithm can apply OVO SVM to match a test melt curve to a melt curve contained in a pre-defined database. In some embodiments, a pre-defined database can include melt curves of one or more typicalpathogens, such as, for example, neonatal pathogens. In some embodiments, a melt curve anomaly detection algorithm using probabilistic classification is employed. Thus, training melt curves for each organism class can be used to build statistical models for those classes. Then the probability of a test curve belonging to each class can be calculated and this can be used as a measure for outlier detection. This approach is uniquely made possible by the novel high-throughput dHRM format described herein, where thousands of training curves are generated per run to robustly model each organism class.
Classification can, in some embodiments, be accomplished by combining the NB model with a decision rule to choose the most probable class as the label for a test curve using the maximum a posteriori (MAP) rule. In some embodiments, LR can be used to model the training data and compare classification performance to that of NB. LR model coefficients can be initially estimated using a squared regularizing function and then solved using maximum likelihood estimation. Goodness of fit can be measured by calculating the deviance of the model. In some embodiments, a non-linear MLP model can be constructed using the same training data and can have multiple layers of neurons that each use a non-linear logistic activation function. Back propagation can be used for training and can minimize the cross-entropy loss function to generate probability estimates for whether a test curve belongs to each organism class. The classification accuracy of each approach described above can be optimized with parameter tuning and tested using LOO, k-fold, and bootstrap cross validation on a given database of organisms. The best performer can be used to explore anomaly detection methods, such as for detecting unknown pathogens. In some anomaly detection embodiments, the Shannon entropy (SE) approach can be applied. The SE is a measure of the uncertainty in the posterior probabilities of a test curve, which can be calculated using the best performing probabilistic classification model described above. Higher magnitudes of SE are associated with more uncertainty in the classification decision made, which enables finding empirical thresholds on the entropy to minimize erroneous classifications. In some embodiments, this can enable the algorithm to abstain from making a decision for a classification with high associated entropy and instead provide information about how close to the current classes of pathogen that the curve is. At the same time, the SE can be used to generate a confidence score for all test curves, thus providing a more informative diagnostic answer. The databases of organisms can be segmented into training organisms and mock anomaly test organisms for this purpose and again LOO, k-fold, and bootstrap cross validation can be used to determine an optimal entropy threshold giving the highest accuracy in anomaly detection. In some embodiments, greater than 99% accuracy can be achieved in the identification of non-database organisms in mock anomaly datasets.
In some embodiments, dynamic temperature warping (DTW) can be used as a method to calculate a distance-from-class measure for outlier detection. DTW is typically a measure for estimating the similarity between two temporal sequences. In some embodiments of the system, temperature sequences can replace temporal sequences. By sweeping one melt curve over the other, DTW calculates the optimal match between the sequences, by trying to explain any fluctuations in the y-axis of melt curve by warping the temperature axis. A statistical model can then be built on the DTW measure for each class to identify melt curves with DTW measures that are unlikely, according to an empirically selected threshold that minimizes Type I and II errors. In some embodiments, information theoretic techniques can be used to estimate upper bounds on the number of pathogens that can be discerned.
In some embodiments, the method can further include calculating and/or assigning a confidence score or measure to the resultant classification. Thus, confidence of the determined melt curve classification can be provided and a determination can be made whether a given melt curve classification, for example a given species identification or sub species identification, can be determined to be clinically relevant.
In some embodiments, DTW can be applied first to identify any melt curves that are highly anomalous, followed by classification using LR or Multilayer Perceptron modeled on pre-defined database, then calculation of SE as a confidence measure on the prediction. In some embodiments, uncertainty in classification can be established using the LR model. In some embodiments, the Shannon entropy can be calculated across the posterior probabilities. Misclassified melt curves have a higher entropy in comparison to the correctly classified melt curves. Thus Shannon entropy measurement of the posterior probability can be used along with the prediction class as a measure of confidence of the classification.
In some embodiments, the method can further include subjecting the sample to a first heating rate to obtain a first melt curve signature; subjecting the sample to a second heating rate to obtain a second melt curve signature; and performing a heating rate-dependent melt curve analysis using a computer model or algorithm that classifies the target nucleic acid based on the comparison to the reference melt curve, such as by using a machine learning algorithm for automated melt curve classification. In some embodiments, subjecting the sample to a first heating rate and a second heating rate includes cooling the sample and re-melting the same sample. The number of re-melt cycles that can be performed can include, e.g., at least one re-melt cycle, at least two re-melt cycles, at least five re-melt cycles, ten or fewer re-melt cylces, five or fewer re-melt cycles, or two or fewer re-melt cycles. In some instances, the number of re-melt cycles that can be performed may be limited by photobleaching effects. In some embodiments, subjecting the sample to a first heating rate and a second heating rate includes subjecting a first portion of the sample to a first heating rate to obtain a first melt curve signature, and subjecting a second portion of the sample to a second heating rate to obtain a second melt curve signature. It has surprisingly been discovered that some nucleic acid molecules exhibit heating rate-dependent melt profiles and that the rate-dependent melt profiles can be used for more precise target identification. It was previously generally thought that heating rate changes only result in shifts in melt curve Tm, whereas the dynamic melting characteristics of a PCR product are thought to be primarily determined by GC content, sequence length, and nucleotide order.44,45 However, the inventors have discovered that some long amplicons are highly sensitive to melting ramp rate, which not only shifts their Tm, but also changes the number and size of distinct melting transitions present. The ability of the methods and systems described herein to identify a heating rate dependence of melt curve shape is in large part due to the tunability, uniformity, and throughput of the digital melt platform. For the long amplicon sequences studied, slower heating rates resulted generally in a single melt transition, whereas faster rates generated multiple melting domains. The inventors have surprisingly discovered that this response to heating rate is highly sequence dependent. Some long amplicon sequences maintained the same melt curve shape for multiple heating rates, while others do not. Thus, the response of a long amplicon to heating rate changes provides additional sequence-specific information that could enhance the specificity of melt curve-based sequence profiling. That is, where one ramp rate cannot discriminate two sequences by their melt curve, a combination of multiple ramp rates may reveal distinct melt responses. Without wishing to be bound by theory, it is believed that the mechanism underlying these differences may involve kinetic sampling of transition states. For example, slower rates would be expected to enable amplicons to sample a wider range of transition states, where shifting, re-organized binding, or secondary structure formation could effectively average out the fluorescence decay across the bulk population of amplicons. Faster rates may induce more uniform transition behavior involving abrupt local DNA “bubbles” that melt separately at a different temperature than the remainder of the sequence. Indeed, faster rates of melting have previously been associated with higher Tm accuracy in homozygous melt analysis.34 Alternatively, since heteroduplex melting has been found to be more apparent at faster heating rates, the multiple melt domains we observe at faster ramp rates may be the result of distinct heteroduplex binding transition states induced in homoduplex molecules.34
In some embodiments, the methods described herein may further include identifying a pathogen or pathogen marker present in the sample. As used herein, a pathogen marker can include a nucleic acid molecule or sequence derived from a living or recently deceased pathogen present in a sample. In some embodiments, the methods described herein may further include identifying a pathogen or pathogen marker present in the sample, wherein the pathogen or pathogen marker has a known nucleic acid melt-curve signature. For example, the melt curve signature obtained from processing the sample containing the pathogen or pathogen marker according to the methods described herein can match one or more reference melt curve signatures in a classification database. In some embodiments, the methods can further include identifying a pathogen or pathogen marker present in the sample, wherein the pathogen or pathogen marker has an unknown nucleic acid melt-curve signature. For example, the melt curve signature obtained from processing the sample containing the pathogen or pathogen marker according to the methods described herein may not match one or more reference melt curve signatures in a classification database. In such case, the machine learning algorithm can identify a pathogen based on anomaly detection ______. In some embodiments of the methods, identifying a pathogen or pathogen marker present in the sample has an error rate of less than 0.2%.
In another aspect, a method of treating a patient suspected of having sepsis, such as, for example, neonatal sepsis is provided. The method includes identifying one or more pathogens having a high probability of mediating the patient's sepsis by performing melt curve analysis by any of the methods described herein, and administering to the patient one or more antimicrobials effective against the one or more identified pathogens. For example, up to one million or more reactions, on either droplet-based or chip-based platforms can be used for single pathogen detection. The methods described herein provide rapid identification of pathogens, which can provide life-saving improvements in the timing and accuracy of treatment regimens. Further, these techniques can extend to other applications in agriculture, defense, and environmental monitoring of organisms.
The invention is further described in the following examples, which do not limit the scope of the invention described in the claims.
The inventors developed a proof-of-concept U-dHRM platform for the clinical application of neonatal bacteremia diagnosis. Clinically relevant bacterial loads are estimated from culture techniques to be between 1 to ˜2,000 colony forming units (cfu) per blood sample (1 ml), where 76% of samples have <50 cfu30,31. This load requires 20,000 reactions to provide a dynamic range of detection up to 1,810 bacterial genomic DNA molecules at the single molecule level (
The challenge of generating high quality U-dHRM curves in picoliter-scale reactions was first approached by tuning fluorescent intercalating dye concentrations to maximize signal-to-noise ratio. An EvaGreen dye concentration of 2.5× was found to be the highest concentration that did not inhibit amplification on-chip. Next, the simultaneous imaging and heating process of melt curve generation (
Next, bacterial DNA from clinical isolates of Listeria monocytogenes and Streptococcus pneumoniae, two common pathogens causing neonatal bacteremia33, were used to further optimize signal-to-noise ratio and melt curve shape resolution (i.e. temperature resolution). First, HRM optimization was carried out on a standard qPCR HRM machine. In this format, melt curve shape, a key discriminating feature of bacterial 16S melt curves24, was found to be highly dependent on imaging rate. A low imaging rate of 1 image per 0.3° C. smoothed melt curve shape features (
S. pneumoniae
L. monocytogenes
We then integrated our automated OVO SVM melt curve identification approach with our U-dHRM platform to enable automated identification of bacteria based on their melt curve signatures. A training database of bacterial melt curves was generated on-chip to enable automatic curve identification. Bacterial DNA from L. monocytogenes and S. pneumoniae were loaded onto separate chips in excess, □ of 223 and 141, respectively, as calculated from spectrometer readings. This ensured each of the 20,000 reactions would be positive for amplification and would generate a training melt curve for the bacterial isolate. Each sample underwent U-dHRM using the optimized ramp and imaging rates described above.
Digital quantitative power relies on the ability to specifically identify true positive amplification from non-specific background amplification. To assess the absolute quantitative power of our platform, we compared U-dHRM melt curve quantification to intercalating dye-based endpoint dPCR quantification. A chip was loaded with a monomicrobial DNA sample of L. monocytogenes according to the concentrations described in the lower panel of Table 1 and U-dHRM was conducted. Then, true positive amplification was quantified two ways. For the first quantification method, we followed the typical endpoint PCR enumeration approach (top graph in
Next, we compared the number of curves quantified by our melt curve Tm enumeration method with the sample DNA concentrations calculated from spectrometer readings and qPCR standard curve methods (
To begin to test the specificity and breadth of profiling of our U-dHRM platform, mock polymicrobial samples were generated to represent challenging detection scenarios where one organism vastly outnumbers another. Defined mixtures of S. pneumoniae and L. monocytogenes DNA were prepared at two different ratios, 1:1 and 3:1, respectively (Table 2). These mixtures were applied separately to two chips at concentrations nearing the low and high end of a typical clinical pathogen load for neonatal bacteremia (50-2,000 copies). Importantly, this dynamic range cannot be assessed by any current HRM format (
SVM analysis.
A mock experiment was conducted to test whether the large amount of human DNA associated with a clinical blood sample would inhibit U-dHRM pathogen identification. Human DNA, extracted directly from a clinical blood sample of a healthy patient, was mixed with DNA from L. monocytogenes in the range of a typical pathogen load (<2,000 bacterial genomes/ml blood). This mixture was loaded onto the chip and U-dHRM was performed using our integrated platform. A Tm threshold value was calculated (
S. pneumoniae
L. monocytogenes
S. pneumoniae
L. monocytogenes
The integrative U-dHRM platform described herein advances HRM profiling by enabling the absolute quantification and identification of multiple genotypes in heterogeneous samples and at clinically relevant concentrations. By achieving HRM curve generation in 0.005% of the traditional HRM volume, and by massive parallelization of HRM across 20,000 reactions simultaneously, we achieve over a 200-fold increase in the dynamic range of detection compared to current HRM formats. Reduction in the size of reactions allows smaller volumes of reagents to be used while maintaining optimal reagent concentrations. Partitioning heterogeneous mixtures across 20,000 picoliter-scale reactions is also expected to overcome environmental microbial DNA contamination that may occur in real-world samples by spatially diluting, i.e. contaminating DNA and target DNA are partitioned from each other for discrimination and quantification23. An increased number of reactions also permits rapid generation of a large training curve database for each organism. Incorporating reference temperature calibrator sequences into each reaction helps normalizes against reaction condition variations for improved reliability. Automated melt curve identification is accomplished by removing non-specific melt curves by Tm thresholding and subsequently matching the remaining melt curves to a training database using our OVO SVM machine learning algorithm24,26. Together, these approaches comprise our microfluidic U-dHRM system and enable the quantitative characterization of complex samples containing multiple bacterial organisms.
Intercalating dye-based dPCR is typically used to detect a single, specific amplification product from one bacteria. Probe-based dPCR can be used to specifically identify up to four bacteria by multiplexing fluorescent probes, or a universal probe can be designed to detect the presence of bacteria non-specifically. By incorporating HRM and universal amplification into dPCR, our platform enables probe-free differentiation of multiple bacteria in a single sample. In our previous work, we showed that 37 clinically relevant organisms could be distinguished by general intercalating dye-based melt curves24. We anticipate that our U-dHRM platform will achieve at least this level of multiplexing and potentially more, since we were able to accomplish a signal to noise ratio and temperature resolution on-chip that matched standard qPCR HRM machines.
While a direct comparison of our U-dHRM detection method to a universal probe-based dPCR detection method was not feasible, due to different polymerase and reaction chemistry requirements, a comparison to typical intercalating dye-based dPCR techniques suggested that our platform and automated analysis approach may offer specificity and sensitivity improvements. Standard intercalating dye-based dPCR relies on thresholding total fluorescence intensity of digital reactions to determine whether they are positive or negative for amplification. Inhibitors that reduce amplification efficiency or non-specific background amplification could result in fluorescence intensities that are misclassified, giving rise to false positives and false negatives. However, melt curve analysis may offer a more reliable way to resolve these two conditions. For our reaction chemistry, we found that the typical dPCR approach of applying an intensity threshold to remove false positives left a significant number of reactions misclassified. Bacteria-specific melt curves were observed in several reactions classified as negative by this technique, and non-specific melt curves were observed in several reactions classified as positive. Our platform enabled Tm thresholding, which improved accuracy by 99% and 94%, respectively, in the Type I and II error rates based on manual observation of melt curves. Our approach could help to ensure that true single molecule sensitivity is attained for optimal lower limit of detection. One reason dPCR total fluorescence thresholding performed poorly in our study could be that we thermocycled significantly longer than most dPCR protocols recommend. A typical dPCR cycle number is kept to ˜35, but we find that 70 cycles ensures full endpoint amplification from single molecules23. While this extended cycling improves accuracy of single-molecule target detection, it also allows off-target amplification to fluoresce more prominently in negative reactions.
Indeed, U-dHRM experiments showed evidence of two kinds of non-template amplification: non-template bacterial DNA amplification (contamination) and off-target amplification. Bacterial contamination produced distinct melt curves within the Tm range of 84-90° C. (
Off-target amplicons were observed to melt at lower temperatures (Suppl.
Importantly, it also highlights a strength of U-dHRM. Because digital reaction partitioning (1) reduces the effect of inhibitors, (2) reduces the effective concentration of contaminating DNA molecules that give rise to off-target amplification, and (3) allows for extended cycling to overcome low efficiency of amplification, since quantification is an endpoint measurement, it is not surprising that we achieve greater sensitivity in the dHRM format (
However, some foreseeable limitations exist. Improvements to the temperature ramp reliability will be critical to ensure a larger database of melt curves are reliably resolved by U-dHRM. Here, calibrator sequences were used to align curves for initial Tm thresholding, but subsequently aligned to their derivative fluorescence value of 0.1 for shape comparison. This second alignment had the effect of ignoring Tm differences in bacteria-specific amplicons, and was required due to fluctuations in the temperature ramp from run-to-run. Insulation from environmental temperatures, an improved chip design with lower thermal mass, and incorporation of a PID controller are expected to overcome this issue. These improvements could also to lead to reduced background noise in the melt curve signal. This would improve our ability to resolve small changes in melt curve shapes generated on the U-dHRM platform, which are occasionally removed by our curve processing algorithms due to background noise.
The capabilities of our microfluidic U-dHRM system could impact infectious disease detection applications like neonatal bacteremia, where speed, breadth of detection, and sensitivity are critical factors. Clinical microbiology relies on lengthy culture-based assays to diagnose bacteremia, which has a high mortality rate that increases with every hour a patient goes undiagnosed and imprecisely treated. Polymicrobial bacteremia is associated with an even higher mortality rate than monomicrobial infection, highlighting the need to detect multiple organisms sensitively, and simultaneously. Immediate conservative treatment with broad-spectrum intravenous antibiotic therapy is typically initiated without any diagnostic information, leading to inaccurate and overtreatment as well as misuse of multiple antibiotics giving rise to the emergence of drug resistant pathogens. The ability to identify bacterial organisms in a blood sample within hours could change clinical practice by providing diagnostic information in time to alter treatment decisions. Retrospective studies also suggest that absolute quantification of bacterial genomic load in patients may be useful to assess severity of infection and to predict prognosis4. The detection of microbial DNA in clinical samples is typically challenged by the excess of human DNA compared to pathogen DNA, which can contribute to PCR reaction inhibition4,40-42. DPCR has been shown to decrease the impact of inhibitory substances43. Likewise, we find that U-dHRM detection of microbial DNA in mock blood samples is not inhibited by high human DNA background or inhibitors carried over in the DNA extraction from blood. This suggests that our device could have exciting implications in the clinical setting. Future work will focus on optimizing and validating our U-dHRM technology on patient-derived clinical samples.
Finally, computational approaches for anomaly detection are being explored by our group to identify bacterial melt curves that are not represented in our database. Currently, a 16S amplicon that melts above the Tm threshold will be automatically classified by our OVO SVM as the organism to which the melt curve is most closely matched. For undefined samples, where significantly more organisms may arise and unexpected emerging pathogens could be present, it will be crucial to identify whether a melt curve is a poor match to the database curves. Indeed, other groups have discovered new species of bacteria by observing alterations in bulk HRM curves by eye29. Automation of this ability would represent a significant advancement for HRM profiling technology and is under development by our group.
In order to achieve high-content digital partitioning, the sample is loaded into a commercially available QuantStudio 3D Digital PCR 20K Chip v2 (Applied Biosystems, Foster City, Calif.). The chip contains 20,000 picoliter-scale wells manufactured from silicon with a hydrophilic treatment that allows high efficiency sample loading. A PCR-grade oil is deposited onto the loaded chip to prevent sample evaporation during cycling. The chip is sealed with an adhesive lid containing an optical window, which allows for imaging and the generation of melt curves. We chose to use a commercially manufactured chip for performance reliability. We coupled the dPCR chip to our custom designed master mix. The master mix is optimized to consistently amplify full length ˜1,000 bp templates of the 16S gene, hypervariable regions V1-V6, and produce high fluorescence signal intensity for melt curve analysis while maintaining optimal surface tension for easy loading. An MJ Research PTC-200 Thermal Cycler (MJ Research Waltham, Mass.) is used for endpoint amplification. The thermal cycler is tilted at a 30-degree angle to collect the bubbles generated at high temperatures in the PCR-oil. These bubbles are trapped in an air pocket located outside of the chip's sample region, preventing sample evaporation from the small volume reactions.
Wizard Genomic DNA Purification Kit (Promega Corporation, Madison, Wis.) was used to isolate DNA from an overnight culture of bacteria, and diluted in PCR water to the desired concentration. Absorbance measurements were made on stock DNA at concentrations within the working range of the spectrophotometer. Then, the DNA was serially diluted, and the expected concentration of the dilution used for dHRM was reported in the Tables and Figures for direct comparison of the different measuring modalities. The optimum PCR master mix for chip amplification, contained in a 14.5 μL reaction, was found to be 1× Phusion HF Buffer containing 1.5 mM MgCl2 (Thermo Fisher Scientific, Waltham, Mass.), 0.15 uM forward primer 5′-GYGGCGNACGGGTGAGTAA-3′ (Integrated DNA Technologies, Coralville, Iowa), 0.15 uM reverse primer 5′-AGCTGACGACANCCATGCA-3′ (Integrated DNA Technologies, Coralville, Iowa), 0.2 mM dNTPs (Invitrogen, Carlsbad, Calif.), 2.5× EvaGreen (Biotium, Freemont, Calif.), 2×ROX (Thermo Fisher Scientific, Waltham, Mass.), 0.02 U/μL of Phusion HotStart Polymerase (Thermo Fisher Scientific, Waltham, Mass.), 1 μL of sample, and ultra pure PCR water (Quality Biological Inc., Gaithersburg, Md.) to bring the total volume to 14.5 μL. The dPCR chip was cycled on a flatbed thermocycler with the following cycle: an initial enzyme activation (98° C., 30 s), followed by 70 cycles (95° C., 30 s, 59° C., 30 s, 72° C., 60 s). Temperature calibrator sequences with varying GC content used for system optimization are as follows:
The U-dHRM device consists of a thermoelectric heating/cooling device (TE Technology, Unc. Traverse City, Mich.) controlled via an Arduino-based interface that uses pulse width modulation (PWM) to generate a temperature ramp (
Fluorescent imaging is accomplished using a Nikon Eclipse Ti platform customized for our dHRM system. A Nikon Plan/Fluor 4× objective with a numerical aperture of 0.13 and a working distance of 16.5× minimizes the number of images and time required to scan the entire chip. A Lumencor SPECTRA X LED Light Engine capable of producing 3-4W of visible light from 380 nm to 680 nm is used as a light source. Images of the loading control dye, ROX, and melt curve intercalating dye, EvaGreen, are captured with 488/561 nm and 405/488 nm excitation/emission filters using an exposure time of 100 milliseconds. Images are captured every five seconds using a Hamamatsu digital camera, C11440 ORCA-Flash4.0. NIS-Elements software is programmed to automatically image the chip as the heating device ramps using the following workflow: define the capture settings for the ROX and EvaGreen channels, set the stage area to the chip's sample area, generate points within that stage area to image, and run time lapse to image each location for every time point. A Prior Scientific NanoScanZ (Rockland, Mass.) motorized stage is used to scan and image the entire chip automatically via the software. For every image, the microscope automatically records the temperature registered by our temperature probe within the metadata of the image. This allows for continuous scanning of the chip and recording of the fluorescence intensity in each well while concurrently heating the chip to generate 20,000 melt curves.
Fluorescence and Tm Thresholding for Negative Reaction Removal
Two approaches to thresholding reaction fluorescence for the identification and removal of negative reactions were compared. The typical dPCR approach of thresholding total reaction fluorescence was accomplished by first plotting a histogram in MATLAB of the total fluorescence intensities at room temperature of all chip reactions. The probability density function (PDF) for a mixture of two normal distributions was then applied to identify negative and positive reaction distributions. A threshold was identified at the lowest point of intensity where the two distributions intersected (
A second approach was developed to identify a Tm threshold that separated off-target amplified reactions from true positives more accurately. First, raw melt curves were converted to derivative melt curves. On fully loaded chips where all reactions were positive (2 training chips) all reactions contained 16S amplicons, which were observed to melt with an average Tm of 89° C. On digitized chips (3 chips, testing data), off-target amplicons were observed to melt at much lower temperatures, average Tm of 81° C., while positive 16S amplicons melted reproducibly in the same range as the training chips. For thresholding analysis, the maximum peak height (Tm) above −d(Fluorescence)/dT=0.01 was found for each derivative melt curve between the range of 75° C. and 93° C. Then a histogram of the Tms was plotted in MATLAB, and the PDF for a mixture of two normal distributions was applied (
In order to generate melt curves from the acquired fluorescent images of the dPCR chip, we implement an automated image processing algorithm in MATLAB. The algorithm generates a binary mask for each temperature point to identify the centroid corresponding to each digital reaction well in the field. Then records the pixel intensity of the 441 neighboring pixels from the images of both the EvaGreen channel and the ROX channel. Each well's average pixel intensity is plotted against the measured temperature to generate the raw melt curve. Each melt curve is normalized to the ROX channel to account for any differences due to unequal loading. The Tm threshold described above is then applied to remove negative reactions and all incorrectly identified centroids. A Gaussian filter is then used to smooth the curves and the derivative is taken with respect to temperature to obtain −dF/dT. Finally, the curves are aligned via a temperature independent melt curve alignment at 0.1−dF/dT. This allowed the differences in melt curve shape to be maximized for later identification using a previously developed OVO SVM algorithm24. Briefly, an OVO SVM creates a maximal margin separating hyperplane between two data classes (i.e. melt curve signatures) using the Least Squares Method (linear kernel). OVO SVMs were created for all binary combinations of organisms with the training data generated from melt curves of known origin. During classification, a scoring method is applied and the most frequently called classification is chosen as the final melt curve identity.
DNA from a clinical blood sample, which was known to be negative for bacterial infection, was extracted and purified using a High Pure PCR Template Preparation Kit (Roche Diagnostics Corporation, Indianapolis, Ind.). The purified blood DNA was eluted in a 20 μL volume. DNA of L. monocytogenes was isolated using the protocol described above in the methods section. Approximately 2,000 genomes of L. monocytogenes were added to the purified blood extraction. The maximum amount of the blood and bacterial DNA mixture (8.63 μL) was added to the PCR master mix. The final mass ratio of human DNA to bacterial DNA on chip was 12,172:1. The master mix was then loaded onto the chip and U-dHRM was performed following 70 amplification cycles. The full chip was imaged as four tiles. Changes were made to the Tm thresholding script to account for the increased and oscillatory noise introduced by the blood DNA extract. First, the peak height for the derivative melt curve was raised to −dF/dT=0.015 to threshold noisier non-specific melt curves from true bacterial amplicons. Second, a lower threshold for melt curve troughs was added at −0.004−dF/dT, which aided in removing highly oscillatory and anomalous curves.
Clinically isolated S. pneumoniae and L. monocytogenes were grown separately overnight in Luria-Bertani (LB) broth. Sterile conditions were used to ensure uncontaminated growth of each bacteria.
To optimize the heat ramp control of the microfluidic chip for identification of sequence-specific melt curve signatures, the thermal control system of the above described digital high resolution melt platform was redesigned. Previously, the chip was housed in a copper block, which was heated or cooled by a thermoelectric (TEC) device. To precisely control the TEC, a proportional-integral-derivative (PID) controller was added with temperature feedback from the copper block. The feedback was provided by a highly accurate resistance temperature detector (RTD) sensor that was embedded in the middle of the copper block. As described previously, a thin layer of thermal grease was added between the chip, copper block, and the TEC device to ensure efficient heat transfer. Heating dissipation from the reverse side of TEC was enhanced by attaching a fan to the aluminum heat sink. The speed of the fan was also controlled by the PID controller, commensurate with the sink temperature, using a negative temperature coefficient thermistor (NTC). The addition of the fan improved the ability to precisely heat to higher temperatures and allowed rapid cooling of the chip back to room temperature, decreasing the wait-time between two consecutive runs. The use of an off-the-shelf digital PCR chip did not allow placement of a temperature probe inside the chip in use. Therefore, to ensure that the thermal control achieved for the copper block efficiently transferred to the chip, a surrogate chip with a temperature sensing thermocouple embedded at its center was placed next to the test chip on the copper block. The entire chip-heating device assembly was held in place inside a custom designed 3D printed stage adaptor to securely mount the device on a microscope for imaging (
The PID controller provided temperature control of the copper block as per the desired ramp rate of 0.1° C./s. The maximum difference in the expected block temperature and observed temperature recorded using the RTD embedded within the block was measured as 0.004° C. (
Next, the relationship between the block temperature and surrogate digital chip temperature was investigated.
Melt Characterization with Temperature Calibrator Sequences
In theory, if heat ramping is linear and heat transfer is efficient, any inaccuracies in the absolute temperatures measured during the melting process could be reliably removed using temperature calibrator sequences. Such control DNA oligos of known Tm can be included in all reactions and designed so that they melt outside of the amplicon's melting temperature range47. As long as the heat ramping rate is linear and of a constant slope, the distance between the calibrator Tm and the amplicon Tm is reliably maintained. Therefore, temperature offset errors occurring from run-to-run or well-to-well because of imperfections in temperature control or uniformity can be removed by simply shifting each melt curve to align the calibrator Tm peaks to their correct melting temperature. Therefore, to further characterize the reliability of melting behavior on the chip, we used three synthetic oligo sequences with predicted Tm ranging from ˜57-93° C. to generate melt curve data for run-to-run and well-to-well variability analysis. These temperature calibrator sequences varied in GC content and length to achieve a high(˜92.9° C.), mid(˜62.8° C.) and low(˜57.3° C.) melting temperatures. For this analysis, melt curves were generated on-chip at a heating rate of 0.1° C./s across seven replicate runs over several days.
First, we applied a peak detection algorithm to the derivative melt curves (
Once we established that the repeatability of Tm-intervals between runs were within expected error limits of the system, we sought to characterize the intra-run variability. This variability does not depend on the heat ramp, but rather the thermal gradient, loading differences, and evaporation anomalies across the chip. Interestingly, when we then investigated variability in Tm peak values across the chip by plotting the difference of the Tm peaks of each well to the corresponding mean Tm for the chip, we found a lower intra-run variability in Tm-High as compared to Tm-Mid and Tm-Low (
Having fully characterized the reliability of our melting device hardware and validated its performance biochemically, we next sought to investigate the dependence of the melt curve characteristics on ramp rate. Since our goal is to use melt curves to profile heterogeneous samples for multiple genotypes, we wanted to understand whether ramp rates could be optimized to enhance the effect of sequence differences on melt curve features. As a model genotyping task, we performed these experiments with bacterial DNA amplified from the 16S rRNA gene that included the hypervariable regions 1-6 (˜1 kb in length). We chose Acinetobacter, Moraxella, and Salmonella genomic DNA as our templates because we have previously observed that their 16S sequences melt uniquely with either one or two transitions at ramp rate of 0.1° C./s in bulk qPCR reactions23,24.
First, we generated 16S amplicons on 3 independent chips for each bacterium. These chips were then used to generate melt curves at ramp rates of 0.01° C./s, 0.05° C./s, 0.1° C./s and 0.2° C./s. We adjusted our imaging settings to maintain 0.1° C. resolution in fluorescence measurement by matching the imaging rate to heating rate (Table 3). To ensure that our thermal control was accurate for varying ramp rates, we observed the temperature profile in the proxy chip and in the copper block. These profiles were similar to what was seen for a ramp rate of 0.1° C./s, reported above. The slopes of the block temperature and fitted chip temperature were repeatable across all ramp rates, and there were negligible deviations in instantaneous rates throughout the runs (copper block and surrogate chip behavior for all ramp rates are shown in
The 16S amplicons for Acinetobacter baumanii revealed multiple melt domains for higher ramp rates of 0.1° C./s and 0.2° C./s. However, a single melting domain was observed for 0.01° C./sec as seen in
We have successfully designed and characterized a high resolution digital melt platform. Our design achieves highly repeatable temperature profiles for a range of melt rates and commensurate imaging frequencies. The run-to-run errors we observed were similar to the expected theoretical error limits of our system, approximately 0.14° C. Precision in temperature control is an important factor in being able to resolve melt curves. However, imaging systems can contribute to melt errors as well. The resolution of our imaging system is one image per 0.1° C. Thus, the minimum total RMSE error due to imaging (+/−0.05) and heating (+/−0.05) for our system is expected to be +/−0.07° C. These error limits can be improved with the development of a custom optical system with integrated and tunable image and temperature acquisition capabilities. Such design improvements would allow us to integrate a chip temperature probe with better accuracy and resolution into the imaging system. This, in addition to increasing our sampling rate for fluorescent data acquisition, would enhance the device's ability to resolve smaller Tm differences. Beyond hardware limitations, other potential sources of variation could arise due to interactions in AT-GC content and the length of amplicons9,48, slight differences in salt environments48,49 or DNA concentrations47,50-51, concentration and stability of DNA saturating dye52 and reference ROX dye53, pipetting errors, differential evaporation across reactions, and data processing methods54.
The reliability and performance of our heating and hardware system enabled us to identify and characterize other sources of melt variation. Understanding well-to-well variation across the chip is crucial for absolute load quantification and sequence profiling at single genome sensitivity. We observed that wells producing outlier Tm peaks are typically present at the corners and along the edges of the chip, as shown in
As compared to conventional PCR, the smaller reaction volumes of digital PCR could be expected to lead to larger variations in Tm due to evaporation. On the other hand, the small form factor of the digital chip is expected to maintain a more uniform thermal gradient across the chip, leading to smaller variations in Tm. A previously published study reported Tm differences ranging from 0.35° C. to 1.24° C. across 32-96 well plate melt instruments with standard deviations of 0.018° C. to 0.274° C.44. In comparison, after excluding the outliers due to evaporation at the corners and edges of our chips, differences in the high temperature calibrator Tm on our digital melt platform were observed to vary from 0.22° C. to 0.6° C. This represents a significant improvement in heating uniformity compared to the standard well-plate format. Even with evaporation outliers included, median absolute deviation ranging from 0.05° C. to 0.1° C. and standard deviation of 0.06° C. to 0.13° C. were observed across the chip, which is less than that of the well-plate format. Optimizing for loading errors and evaporation would be expected to further improve performance. For example, the application of oil onto the loaded reaction wells could be sensitive to timing and amount deposited, and automation would ensure that the corner wells are covered as quickly as the central wells to minimize evaporation.
Our observation that the Tm-Low and Tm-Mid for the temperature calibrator sequences were more variable than the Tm-High calibrator is somewhat expected. Unfortunately, our high-throughput platform revealed that the Tm variations of these low and mid calibrators were large enough to prevent their utility for melt curve normalization across runs. It is important to note that their variability was not correlated within individual wells or regions, indicating that it is not a function of location on the chip or differences in reaction conditions from well-to-well. Kinetic binding rates of DNA are known to vary based on GC content, as association rates of GC-rich oligomers are higher than rates of AT-rich equivalents55. Here, the low and mid calibrators have lower GC content (0% and 12% respectively) and are also shorter in length compared to the high calibrator (76% GC). DNA dissociation rates are also known to increase exponentially with temperature55,56. Further, physical models of DNA melting behavior predict that AT duplexes go through several cycles of hydrogen bond breakage and reformation, often involving an overall shift by one or more bases along the helix, before fully and finally disassociating. In contrast, the corresponding GC duplexes usually come apart only once57. Taken together, this suggests that it may be possible to design more reliable low and mid temperature calibrators by using very short GC-rich sequences.
It is generally thought that heating rate changes only result in shifts in melt curve Tm, whereas the dynamic melting characteristics of a PCR product are thought to be primarily determined by GC content, sequence length, and nucleotide order44,45. However, our study revealed that some long amplicons are highly sensitive to melting ramp rate, which not only shifts their Tm, but also changes the number and size of distinct melting transitions present. Our ability to identify a heating rate dependence of melt curve shape is in large part due to the tunability, uniformity, and throughput of the digital melt platform. For the long amplicon sequences, we studied, slower heating rates resulted generally in a single melt transition, whereas faster rates generated multiple melting domains. Interestingly, though, this response to heating rate was highly sequence dependent. Some long amplicon sequences maintained the same melt curve shape for multiple heating rates, while others do not. Thus, the response of a long amplicon to heating rate changes provides additional sequence-specific information that could enhance the specificity of melt curve-based sequence profiling. That is, where one ramp rate cannot discriminate two sequences by their melt curve, a combination of multiple ramp rates may reveal distinct melt responses. The mechanism underlying these differences may involve kinetic sampling of transition states. For example, slower rates would be expected to enable amplicons to sample a wider range of transition states, where shifting, re-organized binding, or secondary structure formation could effectively average out the fluorescence decay across the bulk population of amplicons. Faster rates may induce more uniform transition behavior involving abrupt local DNA “bubbles” that melt separately at a different temperature than the remainder of the sequence. Indeed, faster rates of melting have previously been associated with higher Tm accuracy in homozygous melt analysis47. Alternatively, since heteroduplex melting has been found to be more apparent at faster heating rates, the multiple melt domains we observe at faster ramp rates may be the result of distinct heteroduplex binding transition states induced in homoduplex molecules47.
In conclusion, our novel digital melt analysis platform with well-controlled and well-characterized heating across 20,000 reactions advances the concept of digital melt curve-based sequence profiling and could also support fundamental studies of DNA dissociation kinetics.
Three temperature calibrator sequences with varying GC content and known melting temperatures were used to optimize the heating of our system:
(Integrated DNA Technologies, Coralville, Iowa). The master mix containing the three temperature calibrators was created as follows: 1× Phusion HF Buffer containing 1.5 mM MgCl2 (Thermo Fisher Scientific, Waltham, Mass.), 4 μM of equal mixtures of the three temperature calibrator sequences, lx ROX (Bio-Rad Laboratories, Hercules, Calif.), 2× EvaGreen (Biotium, Freemont, Calif.) and Ultra Pure PCR water (Quality Biological Inc., Gaithersburg, Md.) to bring the total volume to 15 μL. A volume of 14.5 μL of the 15 μL master mix was then loaded onto a commercially available dPCR chip containing 20,000 picoliter-sized reaction wells, the QuantStudio™ 3D Digital PCR 20K Chip V2 (Applied Biosystems, Foster City, Calif.), as described in Ortiz et al. The chips were filled with a PCR-grade oil, QuantStudio™ 3D Digital PCR Immersion Fluid (Applied Biosystems, Foster City, Calif.), to prevent sample evaporation during thermocycling and sealed with an adhesive lid which contained an optical window for imaging (included in 3D Digital PCR 20K Chip V2 Kit, Applied Biosystems, Foster City, Calif.).
Bacterial gDNA was isolated from an overnight culture of bacteria using the Wizard Genomic DNA Purification Kit (Promega Corporation, Madison, Wis.). The stock DNA concentration was determined by the biospectrometer absorbance readings. Next, the desired DNA concentration was achieved through serial dilutions and added to the master mix which contained the following concentrations: 1× Phusion HF Buffer containing 1.5 mM MgCl2 (Thermo Fisher Scientific, Waltham, Mass.), 0.15 μM forward primer 5′-GyGGCGNACGGGTGAGTAA-3′ (Integrated DNA Technologies, Coralville, Iowa), 0.15 μM reverse primer 5′-AGCTGACGACANCCATGCA-3′ (Integrated DNA Technologies, Coralville, Iowa), 0.2 mM dNTPs (Invitrogen, Carlsbad, Calif.), 2.5× EvaGreen (Biotium, Freemont, Calif.), 2×ROX (Thermo Fisher Scientific, Waltham, Mass.), 2×ROX (Bio-Rad Laboratories, Hercules, Calif.), 0.02 U/μL Phusion HotStart Polymerase (Thermo Fisher Scientific, Waltham, Mass.), 0.3 μM temperature calibrator sequence with 0% GC-content (see above) and Ultra Pure PCR water (Quality Biological Inc., Gaithersburg, Md.) to bring the total volume to 15 μL. A reaction volume of 14.5 μL was spread onto the dPCR chip (see above). A flatbed thermocycler was used to amplify the hypervariable regions, V1 to V6, of the 16S rRNA gene using the following PCR cycle: 1 cycle of 98° C. for 60 s; 70 cycles of 95° C. for 15 s, 58° C. for 30 s, 72° C. for 60 s.
Clinically isolated Moraxella, Acintobacter, and Salmonella Enterica were grown separately overnight in Luria-Bertani (LB) broth. Sterile conditions were used to ensure uncontaminated growth of each bacteria.
The thermoelectric heating/cooling (TEC) device was purchased from TE Technology, Inc. (Traverse City, Mich.). The Proportional-Integral-Derivative (PID) controller was purchased from Meerstetter Engineering GmbH (Rubigen, Switzerland). RTD (Class 1/3B) and thermocouple (K type) sensors were purchased from Heraeus (Hanau, Germany) and OMEGA Engineering (Stamford, Conn.), respectively. Medium to high amount to thermal paste gave the most repeatable results (data not shown).
Nikon Eclipse Ti (Nikon, Tokyo, Japan) platform is customized to accomplish imaging for the dHRM system, as described in our earlier work. Fluorescent images are captured with a melt curve intercalating dye, EvaGreen, and a control dye, ROX, at 488/561 nm and 405/488 nm excitation/emission filters, respectively, with an exposure time of 100 ms at a LED intensity of 40%. The microscope is interfaced with Hamamatsu digital camera, C11440 ORCA-Flash4.0 for image acquisition at a rate commensurate with the heat ramp. The imaging rate is adjusted based on the heat ramping to maintain a resolution of 0.1° C. between images (Table 1). NIS-Elements software is programmed to automatically image the chip as the heating device ramps by running a time lapse to image for every specified time point. For every image, the microscope automatically records the temperature of the surrogate chip registered by the thermocouple temperature probe within the metadata of the image. For this experiment, we used Nikon Plan/Fluor 4× objective with a numerical aperture of 0.13 and a working distance of 16.5×, to image a corner of the chip. Hence, every section of the chip was imaged as a part of separate run with simultaneous heating of the entire chip. This allowed us to maximize the number of runs/data per chip to characterize our heating system. For an ideal use case, as described in our previous publication, we can sweep imagining location to image the entire chip for all runs.
Melt curve data generation. First the acquired fluorescence images are aligned using a template matching plugin in ImageJ. Then, melt curves are then generated using an automated image processing algorithm implemented in MATLAB. The algorithm applies median filter to remove salt and pepper noise in the images. It then generates a binary mask for each well on the chip and tracks them on all images. Pixels within 80% of the detected well radius are recorded and averaged to generate the fluorescence value in both ROX and EvaGreen channels for the specified well. The fluorescence values are tracked for each well across all images to generate curves for both EvaGreen and ROX channels. Filter (EvaGreen) curves generated are normalized against filter (ROX) values to account for any localized errors/noise due to bubbles in the chip or any abrupt change in ambient light as described in our previous publication24.
Imaging software records the temperature corresponding to each image from the surrogate chip. However, the temperature acquisition rate is limited to an approximate 0.20 Hz. Line fitting is performed using the unique temperature, time pair acquisitions to estimate temperature for each acquired image for faster imaging rates (
It is to be understood that while the invention has been described in conjunction with the detailed description thereof, the foregoing description is intended to illustrate and not limit the scope of the invention, which is defined by the scope of the appended claims. Other aspects, advantages, and modifications are within the scope of the following claims.
This application claims priority to U.S. Application Ser. No. 62/438,660, filed on Dec. 23, 2016. The disclosure of the prior application is considered part of the disclosure of this application, and is incorporated in its entirety into this application.
Filing Document | Filing Date | Country | Kind |
---|---|---|---|
PCT/US2017/068313 | 12/22/2017 | WO | 00 |
Number | Date | Country | |
---|---|---|---|
62438660 | Dec 2016 | US |