The Application of Nanotechnology for Quantification of Circulating Tumour DNA in Liquid Biopsies: A Systematic Review

Technologies for quantifying circulating tumour DNA (ctDNA) in liquid biopsies could enable real-time measurements of cancer progression, profoundly impacting patient care. Sequencing methods can be too complex and time-consuming for regular point-of-care monitoring, but nanotechnology offers an alternative, harnessing the unique properties of objects tens to hundreds of nanometres in size. This systematic review was performed to identify all examples of nanotechnology-based ctDNA detection and assess their potential for clinical use. Google Scholar, PubMed, Web of Science, Google Patents, Espacenet and Embase/MEDLINE were searched up to 23rd March 2021. The review identified nanotechnology-based methods for ctDNA detection for which quantitative measures (e.g., limit of detection, LOD) were reported and biologically relevant samples were used. The pre-defined inclusion criteria were met by 66 records. LODs ranged from 10 zM to 50nM. 25 records presented an LOD of 10fM or below. Nanotechnology-based approaches could provide the basis for the next wave of advances in ctDNA diagnostics, enabling analysis at the point-of-care, but none are currently used clinically. Further work is needed in development and validation; trade-offs are expected between different performance measures e.g., number of sequences detected and time to result.


I. INTRODUCTION
C ANCER is a leading cause of mortality worldwide with an estimated 18.1 million new cases and 9.6 million deaths in 2018 [1]. Early diagnosis dramatically improves the prognosis [2]. Cancer therapy is complicated by the fact that efficacy varies due to poorly understood patient-specific factors [3]. Therefore, routine monitoring of disease progression during treatment is vital to determine how tumours respond and enable personalisation of therapeutic strategies (schedules, dose, drug choice).
Nanotechnology may enable simpler, cheaper, faster ctDNA diagnostics. For the purposes of this review, nanotechnology involves the use of objects that have at least one dimension of one to a few hundred nanometres [20].
Nanotechnology is well-suited to biosensing [21], [22] due to the properties of nanoscale materials. For inorganic nanostructures the high surface-area to volume ratio often enhances sensitivity, as the signal is usually related to interaction with the surface. Biological nanostructures can help to deliver  [71]. Gold nanoparticle image reproduced from [29] under CC-BY licence. Scale bar represents 100 nm. (b) A mechanistic overview of readout types included in this systematic review: absorbance/colour, fluorescence, electrochemical, SPR/Raman. biomimetic molecular recognition for customised diagnostics. Furthermore, nanotechnology can potentially deliver new detection modes. Electron behaviour in inorganic nanostructures may be affected either by chemical/physical effects arising from the surface or quantum phenomena that only appear in small objects [23]. Nanostructures can be prepared by topdown methods, where bulk material is eroded to leave the desired shape, or 'bottom-up' techniques, where structures are constructed piecewise from small building blocks. Bottom-up fabrication can be performed via self-assembly [24], which means that the components spontaneously arrange themselves into the desired form due to the interactions between them. Nanostructures used for ctDNA analysis may include nanoparticles, nanorods, nanowires, nanotubes, nanosheets or DNA nanostructures ( Fig. 1(a)) .
Nanoparticles [25]- [39] have various shapes, from spherical to irregular, are less than 1 micron in diameter and can be made of diverse materials, including polymers, metals, semiconductors, and magnetic substances. Gold nanoparticles are popular due to their commercial availability, optoelectronic properties, and the ease of functionalisation with thiol-modified DNA probes via gold-sulphur bonds. Some scientists also use the word 'nanoparticle' to refer to lipid-based hollow vessels, but here liposomes are regarded as a separate category. Nanoparticles and other nanostructures have also been investigated for drug delivery [40]- [43]. Nanosheet materials include graphene and molybdenum disulphide. Graphene comprises a single layer of carbon atoms arranged in a hexagonal pattern, in which electrons are confined strictly to two dimensions [44].
Nanorods, nanowires and nanotubes are linear nanostructures less than a few hundred nanometres wide. Nanotubes are hollow, whereas nanorods/wires are solid. Nanorods are generally shorter than nanowires, but the distinction is not fixed. Various materials can be used for nanorods/nanowires [45]. Gold nanorods are popular for similar reasons to gold nanoparticles. Nanotubes are often made from carbon [46] and a single-walled carbon nanotube is like a rolled-up graphene sheet.
Nanostructures can also be made using biomolecules, including phospholipids, nucleic acids, and peptides. A phospholipid molecule has a hydrophilic head and a hydrophobic tail. In aqueous solution, phospholipids will self-assemble into structures such as micelles and liposomes, to separate hydrophobic tails from water. Liposomes are used for delivery of chemotherapy (Caelyx, pegylated liposomal doxorubicin) and mRNA vaccines (Pfizer/BioNTech and Moderna COVID vaccines) but are used only rarely for ctDNA analysis [47]. DNA nanostructures are made by self-assembly of synthetic designed DNA strands. They range from simple tetrahedra [48] to gigadalton-scale assemblies [49]. It is possible to engineer their shape and chemical modifications with near-atomic precision [50], [51]. Some assays feature DNA hairpins [28], [52]- [64] which consist of a single DNA strand folded in a stem-loop configuration. Hairpin opening or conformational changes in DNA nanostructures are often induced by toehold-mediated strand displacement, the enzyme-free exchange of one strand with another via branch migration, where the input strand can be a ctDNA target [65]. PNA (peptide nucleic acid) components can be used for enhanced recognition of ctDNA targets [26], (a) A schematic representation of a multipart process to detect ctDNA -in the presence of mutated KRAS the nanoparticles assemble into a network but if only wild type KRAS is present the nanoparticles are dispersed in solution. Reproduced from [25] under CC BY-NC 3.0 licence. (b) A mechanistic representation of a colorimetric/fluorescence assay for ctDNA detection. In the presence of the appropriate ctDNA, the CRISPR-Cas12a is activated and it severs the DNA linkage between the nanoparticles, leading to a fluorescence signal. Reproduced with permission from [37], Copyright 2020 American Chemical Society. (c) A schematic representation of an electrochemical ctDNA detection assay involving gold nanoparticles, where the extent of the current is related to the amount of ctDNA bound to DNA probes. Reproduced from [39] under CC BY-NC 3.0 licence. (d) A schematic representation of a magnetoresistive nanosensor for ctDNA detection, as described in [72]. In the presence of the PCR product target, the magnetic nanoparticle is brought close to the magnetoresistive sensor, producing a signal.
[32], [36], [66]- [70]. PNA is a synthetic polymer analogue of nucleic acids, with a polyamide backbone in place of the sugar-phosphate arrangement. PNA can base-pair with nucleic acids, and this is particularly stable and specific.
Nanostructures made entirely from peptides [73] have not yet been explored widely for ctDNA analysis. Generally, they are less well-established than their DNA counterparts because interactions between peptides are more difficult to predict than DNA base-pairing.
Many techniques use multiple types of nanostructure, and some incorporate enzymes, usually associated either with readout (electrochemical reactions or colour changes) [74] or amplification (PCR) [75].
Various readout types are used ( Fig. 1(b)), many involving light. Some methods (absorbance/colorimetric) involve changes in intensity [28] or wavelength [31] of light passing through/reflected from a sample. These changes may even be visible to the naked eye [25], as with the colorimetric method shown in Fig. 2(a). This assay uses the exonuclease ExoIII, which digests properly paired duplexes but fails to digest mismatched sequences. The strand known as 'mLinker' is fully complementary to the mutated KRAS sequence and the resultant duplex is thus digested by ExoIII. In contrast, when the wild type sequence is present, mLinker is not digested and remains in solution, where it can disrupt nanoparticle assembly. The nature of the KRAS sequence (mutated/wild type) determines the extent of mLinker digestion and thus the extent of nanoparticle assembly, which dictates the observed colour.
Fluorescence methods measure the output of an excited fluorophore [59]. For example, DNA nanostructures or hairpins can be designed such that a ctDNA target induces a conformational change that modifies the separation of fluorophore-quencher pairs or two fluorophores that interact via Förster Resonance Energy Transfer (FRET), leading to a change in fluorescence signals. Colorimetric and fluorescence methods can be combined ( Fig. 2(b)). In this example, the fluorophore is initially quenched by the gold nanoparticle. The presence of ctDNA triggers Cas12a to cleave the DNA linking two nanoparticles together. This changes the arrangement of the nanoparticles, altering the colour of the solution from purple to red. Additionally, metal-enhanced fluorescence is used to quantify the ctDNA present in the sample [37].
Surface Plasmon Resonance (SPR) [76] uses light to probe surface plasmons, collective excitation of electrons occurring at some interfaces, which behave differently depending on what is bound to the surface. Raman Spectroscopy utilises light to measure vibrational, rotational and other modes of molecules. Surface Enhanced Raman Spectroscopy (SERS) exploits metallic nanomaterials (e.g., gold nanoparticles) to enhance Raman signal by ten orders of magnitude [77], [78]. This enables the chemical differences between DNA bases to be distinguished, revealing mutations and modifications [79], [80].
Electrochemical technologies involve translation of a chemical signal into an electrical one, based on measurement of potential (voltage), charge or current [26]. For example, in one approach (Fig. 2(c)) the presence of the target DNA catalyses assembly of nanoparticles tagged with the redox label methylene blue, such that the separation of redox labels is increased and there is reduced electron tunnelling to the underlying electrode, yielding lower current [39]. Other readout techniques include dynamic light scattering and magnetoresistive sensors. In one case a giant magnetoresistive nanosensor was used to detect changes in the local magnetic field arising from the binding of an iron oxide nanoparticle to a capture probe specific to a ctDNA sequence ( Fig. 2(d)) [72].
The aim of this work is to consolidate the evidence on the use of nanotechnology for ctDNA detection. In contrast to a standard review, this systematic review was performed using a robust reproducible methodology to ensure an unbiased assessment of the literature, to avoid omission of relevant works. The focus was primarily on quantitative measures, the aim being to establish how nanotechnology compares to more traditional methods and explore the benefits, challenges and limitations of nanotechnology for ctDNA analysis.
As will be seen from other examples of systematic reviews [81]- [83], the procedure for carrying out such a review is precisely defined, similar to an experimental protocol [84]. Here, our goal is to apply this methodology to elucidate the significance of nanotechnology for ctDNA analysis. We focus here on sensing systems applied specifically to ctDNA, but there are many examples of nanotechnology-enabled sensors for generic DNA sequences or other biomarkers, with some examples described in Refs. [85]- [88].

A. Search Strategy
This systematic review was conducted in line with PRISMA guidelines [84]. The first literature search was conducted on 6th March 2020, and a second search was conducted on 23rd March 2021. Subsequently it was determined that the original search strings were not ideally formulated for all the search engines used and further 'auxiliary' searches were performed using a modified search strategy (Supplementary Information). The auxiliary searches were restricted during screening to the same time window as the first and second searches.
The Web of Science TM search was conducted via https://www. webofscience.com/wos/woscc/basic-search. The Web of Science Core Collection TM was searched. For our institutional subscription (University of Edinburgh), this covers ten databases, as listed in Supplementary Table 4. The platform is provided by analytics company Clarivate.
Search strings were designed to capture records that mentioned a ctDNA phrase together with a nanotechnology phrase (Supplementary Table 1). Terms used to identify ctDNA-related papers were based on keywords used in ctDNA publications between 1999 and 2020. Nanotechnology terms were chosen based on common keywords used in papers identified by PubMed following a search using the medical subject heading (MeSH) term "nanotechnology". In the case of the Ovid search strings, our use of subject headings rather than keywords may limit the search as a result of errors or omissions in record indexing within the database.
The search criteria were adjusted for patents since terms such as 'nanotechnology' describe a field of research rather than the nature of an invention [89]. We also introduced phrases such as 'nucleic acid nanostructure' to reflect the fact that patent applications use more generic language and typically use 'nucleic acid' to avoid restricting the patent to just DNA, thus enabling the same wording to cover RNA and PNA nanostructures as well. To identify relevant patents, Google Patents (https://patents.google. com/) and Espacenet (https://worldwide.espacenet.com/) were searched.
Records were stored and de-duplicated using Endnote (both automatically and by hand). For Google Patents and Espacenet a customised python script (Supplementary Information 3) was used to convert .csv database output to .ris for importation into EndNote. Identified records were screened for relevance by two reviewers, according to the criteria described below. The first stage of screening was completed using the online Systematic Review Facility (SyRF, http://syrf.org.uk/). Two reviewers initially made independent decisions on inclusion/exclusion based on all titles and abstracts, before discussing any discrepancies and recording the agreed decisions in SyRF. In the second stage, two reviewers independently screened the full text of the identified records, logging decisions and reasons in Excel, before discussing discrepancies and achieving a consensus. During data extraction, papers were reconsidered critically by an additional reviewer to ensure all met the inclusion criteria. Records identified for elimination at this stage were examined again by the original two reviewers, and the three reviewers achieved consensus.
More information is provided in the Supplementary files.

B. Eligibility Criteria
The study included all primary research and patents that fulfilled the following criteria: 1) primary experimental research or a patent describing a nanotechnology-based technique for ctDNA analysis; 2) stated at least one pre-defined quantitative experimental measure (sensitivity, specificity, limit of detection (LOD), time to result, or cost per test). We included only work based on 'biologically relevant' samples, which could be (i) a relevant ctDNA sequence in buffer (ii) a body fluid spiked with a ctDNA sequence or (iii) liquid biopsies from cancer patients or healthy volunteers.
If the DNA sequence used was not explicitly identified as a relevant gene, the work was only included if it emphasised the applications of the method to cancer diagnosis or monitoring. Techniques that were based on nanopore sequencing technologies were excluded.
The review refers to 'ctDNA' when the term appears in the included records. In keeping with the predefined inclusion criteria, the term 'ctDNA' also refers to synthetic DNA sequences that mimicked DNA sequences derived from tumours. In some cases, the term cell-free DNA (cfDNA) may technically be more appropriate, as DNA extracted from liquid biopsies is not necessarily tumour-derived. However, many papers do not draw this distinction. Furthermore, very few studies in this review use clinical samples, and in these cases, it is difficult to define the difference between ct and cf DNA. Hence, the term 'ctDNA' is used throughout, to avoid unnecessarily complicating this review.

C. Data Extraction
One reviewer used a standardised Microsoft Excel form (Supplementary Information 2) to extract data, including:  r Assay performance: LOD, sensitivity, specificity, cost, time to result. A second reviewer checked extracted data and the two reviewers resolved discrepancies. Both reviewers checked extracted data again independently and discussed further discrepancies.
Specific categories were defined for components and readout type: nanoparticles, nanorods/wires/tubes, sheet nanomaterials, enzymes, PNA probes, single/double-stranded DNAs, and hairpins/DNA nanostructures ( Fig. 1(a)). Multiple categories were assigned to each record, where applicable. A few technologies were not captured by these categories, such as liposomes [90]. The readout types used were: absorbance/colour, electrochemical, fluorescence, SPR/Raman and other. SPR and Raman spectroscopy are different techniques, but some experiments have factors in common with both (e.g., SERS) and the overall number of assays using these methods was small, hence the combination into one group.
We endeavoured to record all genes mentioned by name. We also recorded specific mutations. Where five or more mutations were tested, the word 'many' was included in the dataset. Mutations used as negative controls to demonstrate the ability of an assay to distinguish a sequence from a very similar one were not included in the dataset. Where the gene was not named but the sequence was given, it was generally very short and not amenable to identification.
Where possible LOD values were converted to fM. In some cases, insufficient information was given for this conversion e.g., [29], [91]. Alternative units for LOD included number of copies, percentage of DNA that carries a mutation, or pg/mL. The most relevant form of the LOD depends on the context. The LOD may vary between sequences and sample types, and is generally lower in bodily fluids than in buffer. Normally the best value obtained experimentally for a ctDNA-like sample was recorded. Consequently, the recorded LOD does not necessarily correspond to the recorded 'liquid biopsy sample type'.
All extracted data is available in the supplementary Excel file. Note that numerical paper IDs have no significance.

D. Quality Assessment
Two reviewers assessed the quality of each included record. The assessment methods PRISMA recommends are for clinical trials and are inappropriate here.
Surprisingly, authors frequently stated a limit of detection below the lowest concentration measured experimentally, having extrapolated from a linear fit to their data. In these cases, the authors' quoted LOD was not used. Their data was examined to find the lowest concentration for which a convincing experimental measurement was presented.

E. Technology Readiness Level
For each included record a technology readiness level (TRL) was assigned. The TRL system was originally introduced by NASA [92], and it was adapted for the present work to illustrate the state of development and adoption of nanotechnology in this field. The definitions are shown in Fig. 3(a) and expanded in the SI.
TRL 3 corresponds to technologies that have not yet been demonstrated with a biologically relevant sample. For TRLs 4-6, the technology would be demonstrated via an assay using generic lab equipment whereas at TRL 7 or above a dedicated piece of hardware would be described. The term 'prototype' is used to refer to a specific device designed for the assay under discussion. If a piece of hardware was reported but clinical samples were not used, the TRL would be 4 or 5, depending on the nature of the experiments. Where naturally occurring cell-free DNA was detected in blood of healthy volunteers a TRL of 6 was assigned.   Fig. 3(a), using a modified version of NASA's system for assessing the state of development of technology. (d) The frequency of components ( Fig. 1(a) and Introduction) of the technologies included. (e) The frequency of readout types.

F. Protocol and Registration
Attempts were made to register this systematic review prospectively with PROSPERO (https://www.crd.york.ac.uk/ prospero/) but this was not possible because the work fell out of the scope of PROSPERO (notification email available from corresponding author on reasonable request). The studies included in this review primarily focused on in vitro experiments rather than projects with defined participants, interventions, and clinical outcomes.
As this review could not be registered prospectively, a Supplementary File provides the protocol that we prepared at the start of the systematic review process, in order to comply with the PRISMA guidelines. No amendments have been made to this file except for the correction of a few minor typographical errors and the addition of title, author names etc. to the first page. Changes to that original protocol are discussed in Supplementary Information 1; for the first and second searches the differences were minor, but significant changes were made to the formulation of search queries for the 'auxiliary searches'.

A. Literature Search
In total 4197 records were identified from database searching. After title and abstract review 2922 records were excluded in line with the inclusion criteria. The complete texts of 146 records were reviewed, with 80 being excluded based on the pre-determined inclusion criteria (Supplementary Protocol Table II). Therefore, 66 records were included in this systematic review [25]- [34], [36]- [39], [52]- [54], [66]- [70], [72], [74], [75], [90], [91], [93]- [121]. The included records consisted of two patents, one doctoral thesis, and 63 peer-reviewed journal articles ( Fig. 3(b)). No preprints were included. Although the doctoral thesis was included in the list, the data it contained was omitted because inspection suggested the results were mostly covered in a published paper that had already been included. Patents were also recorded but excluded from analysis, due to the limited information available therein.

B. General Record Characteristics
All records included in this systematic review were published in 2013 or later with an increase in publications each year. The majority of records (52/63 papers) were published in 2018 or later ( Fig. 4(a)). All but one of the papers stated LOD and/or presented data showing the LOD, and in 49/63 (78%) cases the LOD was available in a format that could be converted to fM concentration for comparison ( Fig. 4(b)). Time to result was stated in 13 of the included records and ranged from 10 minutes to 5 hours. The sensitivity and specificity of the technique were each reported in five papers (Fig. 4(b)). The most common gene investigated was KRAS (15 records) followed by PIK3CA (10 records) and EGFR (9 records). Of the records that named a gene, 34 referred to a specific mutation (details in Supplementary Excel File).

C. Evaluation of Technology Readiness
None of the included records demonstrated a technology currently in clinical use, and only four featured a prototype device being tested in a laboratory environment (TRL 7). 14 records were assigned a TRL of 6 since they used clinical liquid biopsy samples, while 33 reported the use of spiked body fluid samples, corresponding to a TRL of 5. The remaining 12 records were assigned a TRL of 4 since they used synthetic DNA samples for analysis (Fig. 4(c)). It was not possible to assign a TRL to the two patent records.

D. Methodological Characteristics
The most commonly-reported component was single stranded or double stranded DNA, found in 51 records (Fig. 4(d)). The second most common was nanoparticles, which were covered in 35 records. 29 of the included records used enzymatic components in the ctDNA assay, and 20 papers made use of hairpins or DNA nanostructures. Finally, nanosheets and nanorods/wires/tubes were found in seven and five records, respectively. It was not possible to categorise the technologies described in the patents, as the descriptions provided were vaguer and often named a range of possible components/readout.
The included records included various readout types. 27 studies analysed ctDNA using electrochemical readout methods, while 15 records utilising fluorescence readout methods and eight using SPR/Raman techniques. Absorbance and colorimetric based methods were used in seven. Six records utilised methods that did not fall into any of these categories (Fig. 4(e)).
The body fluids used were usually blood-based (whole blood, as extracted from a patient; plasma, the supernatant left after centrifuging blood with anticoagulants; serum, comprising plasma with fibrinogen and clotting factors). One study also used saliva [107] and another appeared to refer to fluids from the lung [34]. Urine can also be used, even in the case of non-small-cell lung carcinoma [122], but no urine-based assays appeared in the included records. For the blood-based samples, pre-processing techniques ranged from minimal (e.g., simple dilution of whole blood) to extensive (e.g., DNA extraction and PCR).
The LODs reported in the included studies varied dramatically, from 10 zM to 50 nM (Fig. 5(a)). All but three studies achieved an LOD below 10nM, and two achieved an LOD below 1aM. Although the lowest LOD was presented in a research article published in 2020, the LOD did not show a downwards trend over time (Fig. 5(b)). The differences observed in LOD for different readout type (Fig. 5(c)) or components were assessed (Fig. 5(d)). Electrochemical technologies tended to perform well, and absorbance/colorimetric technologies were comparatively poor. However, as the number of data points is relatively small and the LOD varies by many orders of magnitude between different implementations of the same type of readout, there is insufficient evidence to conclude that any readout type is systematically superior to any other as far as the LOD is concerned, and the overall performance may be influenced by other aspects of the technology, such as sample preparation. Previous reviews have compared various readout types for different applications [123], [124] and have suggested that fluorescence and electrochemical readout methods can achieve similar levels of sensitivity in some contexts. The ease of use for electrochemical methods may give them an advantage but alternative (often overlooked) readout types such as volumetric indicators may be even simpler to implement while providing the capability for sensitive measurements.
Analysis also showed high variability between assay components, but no significant systemic difference was evident ( Fig. 5(d)). To assess the long-term potential of the technologies, it would be important to know whether the LODs are sufficient for clinical use. There does not appear to be a consensus about the LOD required, but many of the included records demonstrate remarkably sensitive technologies. Of the 49 papers for which LOD could be converted conveniently to units of fM, 25 featured an LOD equal to or below 10fM.

E. Critical Evaluation of Papers
As described in Methods, part D (Quality Assessment), in several papers LOD was determined through the use of a linear fit to extrapolate to concentrations below those detected experimentally. In these cases we quoted the lowest concentration that had been detected experimentally, as we deem the extrapolation method fundamentally flawed and inconsistent with standards applied in clinical/industrial diagnostic testing. Although the extrapolation method is used in numerous studies, LODs determined without experimental validation may not be reliable and are inadequate for medical applications. We also note that our quantitative comparison of LODs refers only to records in which the LOD could be expressed in molar concentration (fM etc) and does not include those studies that referred to number of copies of a designated sequence or mutated percentage.
Another limitation observed surprisingly frequently was a lack of structural characterisation of the nanostructures. In many of the included technologies, the shape of nanostructures is integral to the functioning of the assay. Therefore, it is reasonable to expect that proof of nanostructure formation would be included in the manuscripts or supplementary information. For many nanostructures this should be provided via images obtained by atomic force or electron microscopy, although alternative methods can be used depending on the structure. However, such data was often not provided.
Apart from the use of extrapolation to determine LODs and omission of some characterisation data, our quality assessment indicated that the included papers were basically valid. A few of the journals/publishers were not familiar to us, and we initially suspected that they might be predatory but upon investigation we were unable to find any evidence of this.
Another limitation across the research included in this systematic review and across the field as a whole is the lack of standardisation in sample preparation. This presents challenges for the evaluation of findings. For example, researchers have shown that different commercially available tubes have different yields after delays to plasma separation while storage temperature affects DNA yield and fragmentation [125]. This makes it difficult to establish whether a difference in the performance of two assays is due to the underlying technology or the pre-analytical treatment of samples.
We also found that some assays were comparatively complex, but simpler designs could be more effective as malfunctions are less likely with fewer components. In some cases, assays should have been described more clearly and in greater detail, in accordance with the convention that authors should supply sufficient information for another group to replicate their work.
We also note that in the case of Ref [117], there appear to be some inaccuracies in the description of the strand displacement cycle and sequences involved, but there is nothing to suggest that the results are invalid.

A. Advantages of Nanotechnology for ctDNA Analysis in Comparison With NGS
In the long term, nanotechnology-based assays and devices have great potential for ctDNA analysis because they could be faster and cheaper than the alternatives. Some of the technologies reported claimed a time to result of thirty minutes or less [37], [39], [54], [67], [93], [119]. Costs were reported infrequently, but one paper suggested the price of reagents for an individual test could be as low as $31 [72]. It is worth noting that the list price for one sequencing-based commercially available ctDNA test is $949. 1 Further discussion of costs would be premature at this stage as the final price include other factors not accounted for here (including the cost of research and development).
The main limitations of NGS are the resource-intensive multi-step protocols and the cost per test. NGS involves complex protocol and analysis pipelines (Fig. 6(a)) requiring many person-hours, much computational power and vast data storage capacity [126]. For example, the Illumina TruSight Oncology 500 NGS assay requires 10.5 hours of hands-on time and the overall time from initial sample to result is 4-5 days. 2 Additionally, the cost of equipment for NGS ranges from $20000 to $985000 [127]. In contrast, nanotechnology-based assays can often be performed using low-cost hardware. In a lab context, this can include microcontrollers such as the Raspberry Pi Pico or Arduino [128]- [130]. The cost of the nanostructures themselves is not easy to quantify at this stage but it is not expected to be prohibitive. showing a variety of performance measures for a ctDNA analysis technologies, illustrating the potential trade-off between different aspects. An ideal biosensing candidate would be at the centre of the web -a cheap device, approved for clinical use (high TRL), capable of analysing thousands of biomarkers in parallel in seconds, with an ultra-low limit of detection, in a very automated manner. The red web represents cobas EGFR Mutation Test v2 (Roche), a PCR based technology to identify mutations in the EGFR gene, 4 . The blue web represents TruSight Oncology 500 ctDNA (Illumina) a pan-cancer NGS based assay that allows for genomic profiling of liquid biopsy samples. 5 The purple line represents a nanotechnology approach for ctDNA analysis that utilises gold nanoparticles, CRISPR-Cas9, and dsDNA [37]. The orange-yellow line represents a nanotechnology approach for ctDNA detection that utilises iron nanoparticles joined to ssDNA probes, coupled with magnetoresistive sensors [72]. Webs are incomplete due to missing data or parameters that vary depending on circumstances.
Furthermore, nanotechnology has the potential to go beyond simple quantification of ctDNA, by identifying DNA modifications such as methylation [105], [131]- [133] which can be an important cancer prognostic determinator [134]. This has not yet been fully explored, as most of the technologies examined focused primarily on quantifying common genes such as KRAS, which are involved in multiple cancers [135]. Although common 3  genes are suitable candidates for early testing of an assay or device, this may not be the area in which nanotechnology could have the most profound impact in the long term.

B. Limitations and Future Challenges
Nanotechnology-based ctDNA analysis techniques are presently at an early stage of development. Few included records are at TRL 7, with a device prototype tested in a laboratory environment using clinical samples, but the number of records in this category is likely to increase in coming years as the field develops.
One drawback of many of the technologies included in this review is that they analyse a limited number of sequences. NGS and PCR based techniques can be multiplexed to enable a very large number of samples and sequences to be analysed simultaneously. Some nanotechnology-based approaches will be multiplexed further through miniaturisation and automation, which will enable the assays to be performed many times in parallel, with variants that are tuned to specific sequences. It remains to be seen how far this multiplexing can be extended, but in some situations, multiplexing may not be required, perhaps when a patient already has a diagnosis and clinicians wish to monitor the response to treatment using one particular sequence.
There are many long-standing biological and clinical challenges associated with ctDNA liquid biopsies. Such challenges include pre-analytical causes of variation and a lack of clinical data on the interpretation of ctDNA analysis [136]- [139]. These issues must be addressed by all ctDNA analysis technologies, which must also eventually achieve regulatory approval and clinical acceptance in order to succeed. This will require extensive clinical trials and is likely to necessitate validation of the new methods against competing approaches.

C. Recommendations
It was found that very few records stated the expected quantitative measures such as sensitivity and specificity. Here, the words sensitivity and specificity are used in the context of clinical diagnostics, where sensitivity and specificity are framed in terms of true positives, false positives, true negatives and false negatives. In the papers examined here, the term sensitivity was often taken to be synonymous with limit of detection, while specificity usually referred to the ability to discriminate between two very similar sequences. This is because the concepts of true/false positive/negative results only make sense when the diagnostic gives a positive/negative/unclear answer. The ctDNA analysis technologies that were considered here do not simply detect ctDNA, but also quantify it, giving a numerical value rather than a positive/negative outcome. However, papers that did not state clinical sensitivity and specificity did not normally provide alternatives. Therefore, there may be a need for an alternative quantitative measure to assess assay performance. We suggest that a more appropriate measure should be reported, such as an 'accuracy' value that quantifies the difference between the 'actual' concentration (or that obtained with a reference technique) and the measured concentration, perhaps in the form of a percentage that would indicate the order of magnitude of potential errors and discrepancies. This would facilitate assessment and comparison of technologies.
Quantitative measures of performance other than LOD were rarely reported. For instance, time to result was only reported on 13 occasions while cost per test was reported only twice. There is also some ambiguity about the meaning of these terms, as time to result may or may not include sample pre-processing and the cost of test could simply refer to reagent costs or could factor in costs of staff time. Although these measures are not often considered at the early stages of technology development and will change dramatically as the technology advances, they are important factors in the translation of technologies from laboratory to clinic.
To improve reporting of these measures and prevent potential delays in clinical use of these technologies, new guidelines similar to those suggested for NGS technologies [140] should be developed by clinicians and cancer researchers, in collaboration with health authorities such as the Medicines and Healthcare products Regulatory Agency (MHRA), the U.S. Food And Drug Administration (FDA) and the European Medicines Agency (EMA), to help guide researchers in other disciplines in developing technologies with the potential to have a positive impact. Ideally, when a new technology is reported, it should be described with an clear diagram of the mechanism by which the assay is performed, the name of the gene and mutations examined, a clear statement of the LOD (which should always be found experimentally rather than determined by extrapolation), a range of quantitative performance measures, confirmation that the approach can distinguish between similar inputs, the type of bodily fluid used (where applicable), and any sample pre-processing performed.

D. How Nanotechnology-Based Assays Will Evolve for Clinical Use
As the field advances, nanotechnology-based liquid biopsy assays will move up the TRL ladder towards full deployment. Ultimately, the assays will be probably performed using dedicated stand-alone devices, enabling ctDNA analysis to happen in community settings where NGS and PCR could not readily be deployed. Clinical trials of these devices will be necessary, including quantification of sensitivity/specificity and validation of the nanotechnology approaches against existing methods (e.g., NGS). They will provide a numerical measurement of the quantity of particular ctDNA sequences in the sample and may give information on modifications such as methylation. The results would be available quickly, before patients have left the clinic, and the test could be as straightforward as a blood sugar test. The devices will probably perform multiplexed detection of different DNA sequences, but most likely not as many as with NGS, so would not provide a broad overview of the cancer. We therefore envisage two main ways in which this technology would be used at the point of care. Firstly, it could be used for routine screening for cancers with a specific ctDNA signature. Secondly, it could be used for patients who have already received a diagnosis, for treatment monitoring and planning. Here, nanotechnology could complement NGS; sequencing could be used to analyse tumour or liquid biopsies and this information would be used to select the correct variant of the nanotechnology assay.
Ultimately, there will be a trade-off between different performance measures such as cost, time to result, number of biomarkers probed, LOD etc., as this is the case with every technology (Fig. 6(b)). In due course, nanotechnology approaches are likely to demonstrate improvements over NGS techniques in term of cost per test and time to result. However, as Fig. 6(b) shows, nanotechnology approaches [37], [72] are currently at a much earlier stage of development, and at present NGS outperforms nanotechnology in the number of biomarkers assayed and degree of automation. Nanotechnology approaches are expected to improve in both these respects but even so, it is expected that will not be an outright replacement for current technologies, instead offering advantages in certain settings such as point of care and in specific situations such as disease monitoring.

V. CONCLUSION
This review is limited as it focuses on literature and patents that explicitly discuss ctDNA analysis and not DNA analysis in general. This may have resulted in omission of literature focusing on DNA in a broader sense, which could be applied to ctDNA analysis. However, the findings are likely to be representative of nanotechnology-enabled mechanisms for general DNA sensing. It is worth noting that the LODs are very variable, and it is not yet clear which components and readout methods will perform best in the long run. It is possible that a wider range of nanomaterials will be utilised in future, and radically new detection mechanisms may be introduced. The methods used for detection of single-stranded DNA will readily detect RNA as well, and this is an active area of study [68], [141].
This review provides an overview of nanotechnology-based approaches and evaluated if their performance was clinically relevant. We identified potential issues in the way current technologies were reported and suggested how performance measures could be clarified in the future. Although more development is needed to get nanotechnology-based approaches into the clinic, nanotechnology has great promise for the analysis of ctDNA in liquid biopsies and therefore has the potential to have a profound impact on patients and clinicians by enabling low-cost rapid diagnostics that provide invaluable quantitative data at the point of care.

B. Data Availability
All data generated or analysed in this review is included in the article or supplementary files.

C. Competing Interests
The authors declare no competing interests other than the funding arrangements described at the start and the affiliations indicated by the author addresses.