Keywords

These keywords were added by machine and not by the authors. This process is experimental and the keywords may be updated as the learning algorithm improves.

1 Introduction

Robust, sensitive, and easy-to-use biosensors for the detection and quantification of rare biomarkers will have significant applications in both basic research and clinical practice. If made available, these platforms could aid in understanding of fundamental biology, in accurately detecting diseases at their early stage, and in evaluating and monitoring the efficacy of therapy [1–3]. To realize such sensors, the underlying detection technology should ideally (1) enable high sensitivity and accuracy, with minimal false positives and negatives; (2) support short assay time with minimal sample processing; and (3) allow for multiplexed detection in a single parent sample [4]. Different types of sensing platforms, fulfilling some of these requirements, have been developed based on optical [5, 6], electronic [7, 8], or mass-based [9] detection. These systems, however, often require lengthy sample purification, large sample volumes, or long assay time, which can potentially limit their clinical utility and adaption.

Biosensors based on magnetic detection have recently emerged as a promising diagnostic platform. Due to the intrinsically negligible magnetic susceptibilities of biological entities, magnetic detection experiences little interference from native biological samples; even optically turbid samples will often appear transparent to magnetic fields. Biomarkers of interests, when magnetically labeled, however, can attain a high contrast against the biological background. Recent progresses in the synthesis of magnetic nanoparticles (MNPs) have further advanced the magnetic detection technology. With their size scale similar to that of biological molecules, MNPs can efficiently and abundantly bind to biological targets, amplifying analytical signals [10–13]. Various detection technologies have been developed based on this magnetic-tagging concept. These include techniques that use magnetometers, such as superconducting quantum interference device (SQUID) [14–16], magnetoresistive sensors [17–20], and Hall sensors [21], all of which directly measure the magnetic fields arising from the magnetically labeled targets.

We have recently developed a new magnetic sensing platform, diagnostic magnetic resonance (DMR) [22]. Contrary to directly measuring the magnetic moments of the labeled targets, the DMR uses nuclear magnetic resonance (NMR) as the detection mechanism. When placed in NMR magnetic fields, MNPs create local magnetic fields and change the relaxation rate of surrounding water molecules [23]. The detection offers an intrinsic signal amplification mechanism, as more than millions of water molecules can be affected by a single MNP. Moreover, since the signal is generated from the entire sample volume, the assay procedure is significantly simpler than the direct magnetic detection in which MNP-labeled targets have to be closely positioned to the sensing elements.

By optimizing MNPs and miniaturizing NMR detectors, the DMR detection sensitivities for various target types have been considerably improved over the last few years. These developments enable rapid and multiplexed detection on a wide range of targets in microliter sample volumes, including nucleic acids [24], proteins [22], drugs, bacteria [25], and tumor cells [26–28]. With the recent integration of bioorthogonal targeting strategies [27, 29] as well as accurate and real-time control of device temperature [23], the DMR platform has become more robust and sensitive, allowing operation in clinical settings [26]. This chapter reviews the latest development of the DMR technology, focusing on its three major components: magnetic nanoagents, miniature NMR systems, and optimized assay protocols. Specific biomedical and clinical DMR applications will also be discussed.

2 Principle of DMR Detection

The DMR detection of MNP-labeled cells is realized by exploiting the “T 2-shortening” effect of MNPs in NMR measurements [30]. When placed in static, polarizing magnetic fields for NMR detection, MNPs produce local dipole fields with strong spatial dependence, which efficiently destroy the coherence in the spin-spin relaxation of water protons. MNP-labeled objects consequently cause faster decay of NMR signal, or shorter transverse relaxation time T 2, than nontargeted ones (Fig. 9.1).

Fig. 9.1
figure 1

Principle of DMR detection. Biological objects (e.g., cells) tagged with MNPs can accelerate the transverse relaxation of water protons. Compared to the nontagged samples, the NMR signal will decay faster in time domain, therefore providing a sensing mechanism

Full size image

The capability of MNPs to induce T 2 changes is defined as transverse relaxivity (r 2) [31]. With MNPs in solution, the relaxation rate (\({R}_{2} = 1/{T}_{2}\)) can be expressed as [28]

$${R}_{2} = {R}_{W} + {r}_{2} \cdot \frac{{N}_{P}} {V } ,$$
(9.1)

where R w is the relaxation rate of the background (usually water), V is the NMR detection volume, and N P is the total number of MNPs in V. If each biological cell has n MNPs and the total number of cells is \({N}_{C}(= {N}_{P}/n)\), the net change of \({R}_{2}(\Delta {R}_{2} = {R}_{2} - {R}_{w})\) is given as

$$\Delta {R}_{2} = {r}_{2} \cdot \frac{{N}_{P}} {V } = \frac{n \cdot {r}_{2}} {V } {N}_{c} = {r}_{2}^{\mathit{cell}} \cdot \frac{{N}_{C}} {V } ,$$
(9.2)

where \({r}_{2}^{\mathit{cell}}(=\,n \cdot {r}_{2})\) is defined as the cellular relaxivity (transverse relaxivity per given cell concentration). Note that \({r}_{2}^{\mathit{cell}}\) is indicative of the abundance of relevant surface biomarkers. NMR thus can be used effectively for molecular profiling of target cells [27, 28]. Equation 9.2 provides valuable insights into how to increase the sensitivity and specificity of NMR-based sensors:

  • MNPs with high r 2 relaxivity. Pronounced R 2 changes will occur when cells are labeled with MNPs of high r 2 relaxivity [23]. Because r 2 is proportional to the magnetic moment (μ p ) of particles [32, 33], making magnetically stronger MNPs will benefit the measurements.

  • Maximal MNP labeling on cell. The R 2 changes are also directly proportional to the number of MNPs loaded onto cells. In addition to increasing the r 2 potency of individual MNPs, it is equivalently important to establish a labeling protocol to maximize and/or amplify MNP loading on cells.

  • Miniaturized NMR probes. Higher sensitivity can be achieved on the device level by decreasing the NMR detection volume (V ). This approach can effectively increase the analyte concentration (N C  ∕ V ), leading to large \(\Delta {R}_{2}\). Furthermore, smaller NMR probes assume higher SNR (signal-to-noise) ratio due to the increased sample filling factor. It can be shown that the sensitivity of NMR coils (with a typical dimension of d) scales as \({d}^{-1/2}\) [34].

These recognitions motivated us to explore three major activities in DMR development: synthesis of new MNPs, optimization of such MNPs for cellular labeling, and miniaturization of NMR systems. The following sections will describe these accomplishments.

3 New Magnetic Nanoparticles

We have developed many different types of MNPs (Table 9.1) and engineered the particle size and composition to enhance the transverse relaxivity. According to the outer-sphere model of transverse relaxation, the r 2 value of an MNP is proportional to \({\tau }_{d} \cdot {M}^{2}\), where τ d is the residence time of water molecules around the particle and M is the particle magnetization [35]. The efforts to enhance r 2 were thus focused on synthesizing larger MNPs using magnetically stronger material. We herein introduce two types of such particles: metal-doped ferrite and iron (Fe)-based MNPs.

Table 9.1 MNPs developed for DMR applications
Full size table

3.1 Ferrite-Based MNPs

With their excellent stability and biocompatibility, cross-linked iron oxide (CLIO) nanoparticles have been widely used for DMR applications [22, 36]. CLIO nanoparticles contain a superparamagnetic iron oxide core (3–5-nm monocrystalline iron oxide) composed of ferrimagnetic magnetite \(({\mathrm{Fe}}_{3}{\mathrm{O}}_{4})\) and/or maghemite (\(\gamma {\mathrm{Fe}}_{2}{\mathrm{O}}_{3}\)). The metallic core is encased with biocompatible dextran, which is cross-linked and functionalized with primary amine. Amine-terminated CLIO nanoparticles have an average hydrodynamic diameter of 38 nm, and about 60 amine groups are available for bioconjugation per nanoparticle. The r 2 of CLIOs is \(\sim 50\,{\mathrm{s}}^{-1}{\mathrm{mM}}^{-1}\) [Fe] [24, 37] as measured at \(4{0}^{\circ }\mathrm{C}\) and at the external field of B 0 = 0. 5 T.

Two main strategies have been employed to further improve the magnetization of ferrite nanoparticles and thereby the r 2 relaxivity: magnetic doping and nanoparticle sizing. Magnetic doping with ferromagnetic elements such as manganese (Mn), cobalt (Co), or nickel (Ni) has been known to modulate the overall magnetization of MNPs [38, 39]. Among these doped ferrite MNPs, \({\mathrm{MnFe}}_{2}{\mathrm{O}}_{4}\) nanoparticles have the highest magnetization, as Mn2 +  ions have the highest spin quantum number (5/2). Moreover, larger nanoparticles are also known to have increasing magnetization [40]. Spin canting, a feature that decreases the overall magnetic moment of small nanoparticle due to tilted surface spins, can be reduced in bigger nanoparticles to increase the overall magnetization. Concurrently, larger particle size further enhances the particle r 2 by increasing τ d .

We employed both magnetic doping and sizing strategies to produce \({\mathrm{MnFe}}_{2}{\mathrm{O}}_{4}\) nanoparticles with superior r 2 relaxivity [28]. These particles were synthesized in the organic phase by reacting iron (III) acetylacetonate [Fe(acac)3], manganese (II) acetylacetonate [Mn(acac)2], and 1,2-hexadecanediol at elevated temperature (300 ∘ C). Through a seed-mediated growth approach, the particle size was stepwise increased from 10 nm to 12, 16, or 22 nm. \({\mathrm{MnFe}}_{2}{\mathrm{O}}_{4}\) nanoparticles with diameter ≤ 16 nm were found to be highly monodisperse and superparamagnetic at 300 K (Fig. 9.2). Because of their larger magnetic core, these \({\mathrm{MnFe}}_{2}{\mathrm{O}}_{4}\) nanoparticles assumed high relaxivities with r 2 values approaching \(420\,{\mathrm{s}}^{-1}.{\mathrm{mM}}^{-1}\)[metal] (equal to \(6 \times 1{0}^{-14}L \cdot \,{\mathrm{s}}^{-1}\) per particle), more than eight times greater than CLIO nanoparticles in metal basis (\(50\,{\mathrm{s}}^{-1}{\mathrm{mM}}^{-1}\)[metal] or \(7 \times 1{0}^{-16}L\,{\mathrm{s}}^{-1}\) per particle) [28].

Fig. 9.2
figure 2

Mn-doped ferrite for DMR detection. (a) Transmission electron microscope (TEM) images of \({\mathrm{MnFe}}_{2}{\mathrm{O}}_{4}\) MNPs prepared in our laboratory. The particles have a narrow size distribution and consisted of a single crystal (insets). (b) The particles in (a) showed superparamagnetic behavior at room temperature (Adopted from [28]. Copyright 2009 National Academy of Sciences, USA)

Full size image

3.2 Fe-Core MNPs

Ferromagnetic metals, instead of their oxides, have been suggested as an ideal constituent of MNPs for their superior magnetization. However, while Fe-core MNPs can achieve high r 2 relaxivities [41], these monometallic MNPs are extremely reactive and require protective layers to prevent rapid oxidation.

Recently, a unique 16-nm Fe-core/ferrite-shell MNP (Fe@ferrite) has been developed for DMR applications (Fig. 9.3) [42]. The particle consists of an elemental iron core (not iron oxide) and a protecting oxide shell. A novel synthetic route was established, as summarized below, that allows for the preparation of large yet monodisperse Fe MNPs and the growth of protective ferrite shells around existing Fe MNPs (Fig. 9.3a).

Fig. 9.3
figure 3

Fe@ferrite MNPs. (a) New synthetic route. The size of the Fe core is increased by elevating the reaction temperature. The core is then overcoated with a ferrite shell. To further enhance the particle magnetization, the shell is metal doped. (b) TEM image of Fe@ferrite MNPs. Fe MNPs (dotted circles) were preserved during the coating process. (c) Fe@ferrite MNPs showed stable magnetic properties over time with small changes ( < 10 %) in saturation magnetization (M s ). (d) The field-dependent magnetization (M) of Fe@MFe\({}_{2}{\mathrm{O}}_{4}\) MNPs at 300 K showed an unusual feature: negligible remanent moments but the presence of hysteresis (Adopted from [42]. Copyright 2011 John Wiley and Sons, Inc.)

Full size image

  • Larger Fe cores. Fe MNPs were formed by thermally decomposing metal complexes [Fe(CO)5] in the presence of surfactant (oleylamine) under air-free condition. By increasing the reaction temperature during Fe MNP synthesis, the particle size could be proportionally increased [42]. The phenomenon can be attributed to the higher reactivity of Fe ions at elevated temperatures during particle formation [43]. Applying this approach, Fe MNPs with diameters up to 18 nm could be prepared while maintaining the relative size variations < 5%.

  • Ferrite-shell overcoat. Native Fe MNPs undergo rapid oxidation, which necessitates the presence of protective shells. To prepare such shells, we carried out the reaction for ferrite synthesis in the presence of as-prepared Fe MNPs, based on the hypothesis that Fe MNPs could serve as nucleation sites for ferrite formation. Indeed, the resulting particles (Fe@ferrite, Fig. 9.3b) had Fe cores whose sizes were similar to that of the initial Fe MNPs (dashed circles in Fig. 9.3b). Importantly, Fe@ferrite MNPs maintained their shape and magnetic properties over time (Fig. 9.3c), verifying the sturdy protection against oxidation by the shell. Note that we further enhanced the overall magnetic moment and r 2 relaxivity by doping the ferrite with Mn2 +  during the shell formation (Fe@MnFe\({}_{2}{\mathrm{O}}_{4}\) MNPs).

The Fe@ferrite MNPs assumed higher saturation magnetization (796 kA m − 1) and r 2 (\(7 \times 1{0}^{-14}\,L\,{\mathrm{s}}^{-1}\) per particle, \(430\,{\mathrm{s}}^{-1}{\mathrm{mM}}^{-1}\)[metal]) than similarly sized ferrite MNPs, primarily due to the large Fe cores. It is noteworthy that the Fe core is in a thermally stable ferromagnetic state with nonzero coercivity. The ferrite shell, which is superparamagnetic, however, effectively reduces the overall coercivity of particles by leading the magnetization processes at small external magnetic fields [42]. The resultant Fe@ferrite MNPs thus displayed a unique magnetic feature, namely, the presence of hysteresis with negligible coercivity (Fig. 9.3d). This property is crucial in preventing interparticle aggregations from magnetic interactions. When applied for DMR assays, these Fe@ferrite MNPs achieved superior performance, capable of detecting picomolar avidin and single cancer cells in whole blood samples.

4 Optimizing MNPs for DMR Applications

In addition to the above-mentioned strategies to improve nanoparticle relaxivities through inorganic chemistry, postsynthesis modifications such as better particle surface chemistry and new labeling approaches have also been developed for DMR applications. These novel postsynthesis modifications not only improve the detection sensitivities but also simplify the targeting assays, making the DMR platform easily applicable to detect a wide range of biological entities and translatable for effective clinical utility.

4.1 Biocompatible Coating on Hydrophobic MNPs

Most MNPs, synthesized via the thermal decomposition method, are suspended in nonpolar solvents and coated with hydrophobic surfactant. For biological applications, these particles should be transferred into aqueous phase. We have traditionally used a small bifunctional molecule DMSA (meso-2,3-dimercaptosuccinic acid) to replace hydrophobic capping layers (e.g., oleic acid or oleylamine) on MNP surfaces [44, 45]. The resulting particles, however, displayed short-term stability ( < 3 month), gradually precipitating out in physiological buffers [44].

Overcoming the issue, we have established a new, polymer-based surface coating that can render MNPs hydrophilic with superb stability under varying pH and ionic strength (Fig. 9.4a) [46]. As a coating substrate, we selected polyvinyl alcohol (PVA) since the material is synthetic, inexpensive, hydrophilic, and biodegradable [47–49]. The polymer was further modified into carboxymethyl polyvinyl alcohol (CMPVA); we hypothesized that multidentate carboxylic (-COOH) groups would allow CMPVA to strongly bind to the metal oxide surface. Indeed, when hydrophobic \({\mathrm{MnFe}}_{2}{\mathrm{O}}_{4}\) MNPs were coated with CMPVA, the particles showed excellent, long-term solubility ( > 12 months) in aqueous buffers, whereas PVA-coated particles precipitated spontaneously (Fig. 9.4b). CMPVA coating also provided free amine ( − NH2) groups for bioconjugation.

Fig. 9.4
figure 4

Highly stable polymer coating on MNPs. (a) Hydrophobic layers on MNPs were first replaced with TMAH (tetramethylammonium hydroxide) and then further coated with carboxymethyl polyvinyl alcohol (CMPVA). TEM confirms that the particles are well dispersed in water after the CMPVA coating. (b) The presence of carboxylic acids on polyvinyl alcohol (PVA) is crucial to the stability of the coating; PVA-coated MNPs aggregated in aqueous solutions. (c) Cancer cells (SkBr3) were labeled with fluorescent CMPVA-MNPs by targeting HER2/neu surface receptors. CMPVA coating displayed remarkably low nonspecific binding to the cells (Reproduced from [46]. Copyright 2010 John Wiley and Sons, Inc.)

Full size image

When tested for cellular labeling, the CMPVA coating exhibited extremely low nonspecific binding. Figure 9.4c shows an example of cellular labeling with CMPVA-MNPs. Cancer cells (SkBr3) were first labeled with biotinylated anti-HER2/neu monoclonal antibodies, followed by an incubation with neutravidin-modified, fluorescent CMPVA-MNPs. Control samples were prepared in the same way but without antibody injection. Strong fluorescence signal could be visualized on the surface of targeted cells, whereas the signal from control samples was negligible. This obvious difference between the control and positively labeled samples is highly important for DMR assay in reducing false positives in biosensing.

4.2 New Labeling Method for High MNP Loading

Besides improving the physical properties of MNPs, equally important for the DMR assay is to develop an efficient method for MNP labeling on target cells. A conventional way is to use MNPs preconjugated with target-specific affinity ligands [22], which often requires extensive optimization of the affinity ligands and the conjugation method for each new target.

We have developed a new targeting strategy, BOND (bioorthogonal nanoparticle detection), that is modular and broadly applicable and can amplify MNP binding to biological objects [27]. The BOND is based on [4 + 2] Diels-Alder cycloaddition, especially between tetrazine (Tz) and trans-cyclooctene (TCO; Fig. 9.5a) [50]. The reaction is fast and irreversible (covalent) and can be performed at room temperature without using any catalyst (copper). We have adapted the chemistry for MNP labeling of cells (Fig. 9.5b), wherein cells are pretargeted with TCO-modified antibodies and subsequently incubated with Tz-loaded MNPs (Tz-MNPs). Multiple TCO tags (usually ∼ 20) could be incorporated onto an antibody, and such modified antibodies were found to maintain their affinity. Consequently, the antibodies functioned as a scaffold to promote multiple attachments of Tz-MNPs. Indeed, flow cytometry measurements showed that the BOND yielded ∼ 15-fold improvement in MNP loading on cells, compared to labeling with antibody-MNP direct conjugates (Fig. 9.5c). The trend was further confirmed in NMR-based cell detection (Fig. 9.5d); the BOND method yielded more pronounced T 2 changes and improved the cellular detection limit.

Fig. 9.5
figure 5

Bioorthogonal nanoparticle detection (BOND). (a) The method is based on the Diels-Alder cycloaddition between trans-cyclooctene (TCO) and tetrazine (Tz). (b) Cells are prelabeled with TCO antibodies and targeted with Tz-MNPs. The antibody provides sites for mutliple MNP couplings. (c, d) Compared to the direct targeting with MNP-antibody conjugates, BOND method enabled higher MNP loading on target cells as confirmed by fluorescent (c) and μNMR (d) measurements (Reproduced from [27]. Copyright 2010 Nature Publishing Group)

Full size image

BOND has been successfully adapted for DMR molecular profiling of experimental cellular samples [27] and clinical fine-needle aspirate samples [26]. Recently, the BOND technology was further generalized by developing newer two-step detection schemes based on complementary oligonucleotide approaches [29], alternative cycloaddition chemistries [51], and cyclodextrin/adamantine supramolecular interactions [52].

5 Miniaturized NMR System

The development of miniaturized nuclear magnetic resonance (μNMR) systems [22] represents a key milestone in sensitive detection in DMR. Device miniaturization brings several distinctive advantages for sensitive detection and clinical translation. First, it provides a promising way to improve the detection sensitivity. Smaller NMR systems reduce the detection volume, which in turn effectively increases the concentration of MNP-targeted cells for large R 2 changes (Eq. 9.3) Second, miniaturized NMR probes (coils) produce much stronger radio-frequency (RF) magnetic fields per unit current, leading to higher signal-to-noise per unit sample volume [53]. Third, with smaller RF coils, the requirement for spatial homogeneity of static magnetic fields becomes less stringent, making it possible to use small, portable magnets [22]. The entire measurement system can be realized as a portable device for point-of-care operations.

5.1 System Concept

NMR-based cell detection has been previously performed on benchtop relaxometers (e.g., minispec, Bruker) [36, 54–56]. The systems are equipped with permanent, low-field ( < 1 T) magnets for field generation, which simplifies operation and housing of the equipments; however, the main drawbacks of the benchtop system include the use of relatively large sample volumes (\(\sim 100\,\mu \mathrm{L}\)) and the lack of capability for parallel measurements.

Overcoming these limits, the first μNMR prototype was designed and tested for the feasibility of miniaturization. Figure 9.6 shows the main features of the μNMR system: planar microcoils, microfluidic networks, onboard NMR spectrometer, and a portable magnet. The microcoils are used for NMR detection and are arranged in an array format for parallel measurements. The microfluidic networks facilitate the handling and distribution of small volumes of samples. A small, portable magnet (NdFeB, B 0 = 0. 5 T) was employed to generate NMR field. The system measured the T 1 relaxation time using inversion recovery pulse sequences; for T 2 measurements, Carr-Purcell-Meiboom-Gill (CPMG) spin-echo pulse sequences were used to compensate for the inhomogeneity of the polarizing magnetic field. To generate versatile pulse sequences while using minimal electronic parts, we devised a new circuit schematic for NMR electronics that has served as a blueprint for subsequent NMR systems.

Fig. 9.6
figure 6

Schematic of microNMR system. (a) The system consists of an array of microcoils for NMR measurements, microfluidic networks for sample handling and mixing, embedded NMR electronics, and a permanent magnet for polarizing magnetic field generation. The whole setup can be packaged as a handheld device for portable operation. (b) Micrograph of a microcoil array. The microcoil (inset) generates RF magnetic fields to excite samples and receives the resulting NMR signal. (c) Schematic of the NMR electronics. The circuit is designed to perform T 1 and T 2 measurements via inversion recovery and CPMG (Carr-Purcell-Meiboom-Gill) pulse sequences, respectively. (d) Example of a microfluidic network for effective mixing between magnetic nanoparticles and the samples (Reproduced from [22]. Copyright 2008 Nature Publishing Group)

Full size image

5.2 Optimal NMR Probe Design

Reducing sample volume requirements can lead to the effective increase of cell concentrations. However, it can also lead to degradation of the signal-to-noise ratio (SNR), as the absolute level of the NMR signal is proportional to the sample volume. System miniaturization thus should be accompanied by measures to maintain or enhance SNR to truly improve the detection sensitivity. In the second generation of μNMR system, we focused on improving SNR by engineering the NMR probes.

The SNR of a NMR probe can be expressed as

$$\mathit{SNR} = \kappa {M}_{0} \cdot \sqrt{\frac{{\mu }_{0 } Q{\omega }_{0 } {V }_{c } } {4{k}_{B}T\Delta f} ,}$$
(9.3)

where κ is the fraction of the coil volume (V c ) occupied by the samples (filling factor), M 0 is the nuclear magnetization of the sample, μ0 is the vacuum permeability, ω0 is the Larmor frequency, Q is the quality factor of an NMR coil, and \(\Delta f\) is the bandwidth of a receiver electronics. For a given NMR setup (i.e., the same magnets and electronics), SNR could be improved by increasing κ and Q, which are properties of the NMR probes. Indeed, we have demonstrated a new probe design that achieves both maximal \(\kappa (\approx 1)\) and high Q ( ∼ 25) [28]. In this design, the probe consisted of a solenoidal microcoil embedded in a microfluidic structure (Fig. 9.7a). Solenoidal coils were chosen for their higher SNR than planar or birdcage coils [53]. To increase κ, we adopted the cast-molding technique in device fabrication. First, the coils were wound around polyethylene tubes and subsequently immersed into a polymer (PDMS). After PDMS cure, the tubes were withdrawn to open up the fluidic channels. With this design, the entire bore of the coil was available for samples; in one example, the new probe displayed > 350 % larger SNR than a similar coil wrapped around a tube (Fig. 9.7b). Compared to the lithographically patterned planar coils in our previous systems [22], the improvement in SNR was much more significant ( > 20-fold enhancements); the solenoidal coil excited larger volumes of samples by producing more homogeneous radio-frequency magnetic fields and had smaller electrical resistance (e.g., 0.3 vs. \(3.2\Omega \) at 20 MHz).

Fig. 9.7
figure 7

New NMR coils for higher sensitivity. (a) A solenoidal coil is embedded along with a microfluidic channel. The entire bore of the coil is available to samples, maximizing the filling factor ( ≈ 1). (b) The new coil, due to its high filling factor, offered > 350 % enhancement in SNR compared to a similar coil wrapped around a tubing. (c) The R 2 changes can be amplified by concentrating MNP-labeled samples inside the microcoil. Based on the coil design in (a), we incorporated a filter that captures target objects. The photo shows a prototype device containing 100-nm pore filter (Adopted from [28] and [25]. Copyright 2009 National Academy of Sciences, USA and John Wiley and Sons, Inc.)

Full size image

We have further modified the embedded coil to improve the overall SNR and to streamline the assay procedure. In the new probe design, a cell-capturing membrane filter was incorporated at the outlet side of the coil (Fig. 9.7c). The filter serves two essential functions. First, it size selectively captures cells and concentrates them inside the NMR detection coil, which leads to more pronounced R 2 changes (Eq. 9.2). The filter also provides a way to detect a small number of targets from large sample volumes. Second, the filter enables on-chip separation of cells from unbound MNPs, therefore obviating the need for separate off-chip purification steps (e.g., centrifugation). Together with the highly magnetic Fe@ferrite MNPs, this new probe has been applied to diagnose tuberculosis [25]. Using the attenuated bacillus Calmette-Guérin (BCG) as a surrogate for Mycobacterium tuberculosis, we detected as few as 20 colony-forming units (CFUs) in sputum (1 mL). Importantly, the entire detection procedure was performed in a single-chip format, minimizing sample loss and making the assay simple and fast (in less than 30 min).

5.3 Clinical System

Our third-generation μNMR system is designed for practical applications in a clinical environment (Fig. 9.8) [57]. Its small size, easy accessibility, and high robustness provide end users with fast and stable measurements of biological samples. The new μNMR system is composed of three core parts: a newly designed probe for clinical samples, NMR electronics, and a user-friendly software.

Fig. 9.8
figure 8

Overview of the clinical μNMR system. The entire system was redesigned for robust operation in clinics. The NMR probe accepts samples sealed in disposable tubes to prevent probe contamination (left). NMR coils are embedded into the polymer support to open up the entire coil bore, and a thin-walled (thickness ∼ 65 μm) tube is used to minimize loss in filling factor. Block diagram of the NMR electronics for sample excitation and NMR signal reception (middle). The electronics is highly programmable and can be controlled via external terminals. Functional view of NMR software that processes data and shows graphical output of NMR measurements (right)

Full size image

The system design is based on that of previous generations of μNMR and consists of a small portable magnet (B 0 = 0. 5 T) and a solenoidal coil for higher SNR (Fig. 9.8 left). A custom-made PMMA (polymethyl methacrylate) housing cages the magnet, the microcoil, and RF matching circuit. With its major improvement focused on clinical translatability, this new system uses disposable thin-walled polyimide tubes to load biological samples for measurement, thereby eliminating potential contamination of the NMR probe. The disposable tubes are filled with samples ( ∼ 5 μL) and are inserted into the coil bore for NMR detection. Modular coils made in a variety of sizes can be plugged into the system to optimally accommodate available sample volumes (\(1\mbox{ \textendash }100\,\mu \mathrm{L}\)).

A small form-factor (20 ×12 ×5 cm) NMR electronics is implemented using off-the-shelf integrated circuit (IC) chips to achieve cost-effective (\(<\$200\)) and highly programmable NMR platform (Fig. 9.8 middle). The NMR electronics generates the NMR pulse sequences, acquires the NMR signal, and communicates with external terminals (computer, mobile devices). It has three main parts: a microcontroller unit (MCU), an RF transmitter, and a signal receiver. The MCU (TMS320F28235, Texas Instruments) controls overall RF transceiver operations as well as data communication with external terminals. As an RF transmitter, a direct digital synthesis chip (AD9954, Analog Devices) is employed to generate two RF signals with 90 ∘  phase difference that are modulated by voltage-controlled switches (ADG1419BRMZ, Analog Devices). The RF heterodyne system is implemented to process the NMR signal. First, the signal is amplified by low-noise amplifier (AD604, Analog Devices) and down-converted to baseband (1–10 kHz) by a mixer (ADE-6, Mini-Circuits). Baseband signals subsequently pass a low-pass filter and are digitized by an analog-to-digital converter (AD7625, Analog Devices). Mobile devices (e.g., iPhoneTM, iPadTM), which are connected to the MCU via a standard communication channel (e.g., USB, Bluetooth), receive digital data and show graphical outputs.

The NMR software incorporates graphical user interface (GUI), data acquisition/process, data logging/sharing, and a temperature compensation engine, which deliver a user-friendly interface and contribute to the robustness of the new μNMR system (Fig. 9.8 right). The software is programmed with Objective-C using Cocoa and Cocoa Touch frameworks and operates on iOSTM and OS XTM. In order to implement a graphical representation of R 2 relaxation curve, open-source plotting framework (Core Plot) is cross-linked with Cocoa and Cocoa Touch frameworks. Time domain NMR data are acquired at the negative edge of spin echoes and processed real time. Negative edge-triggered data acquisition reduces the size of the data to be transferred and consequently decreases the data acquisition time via USB/Bluetooth connections by more than 10-fold. Transferred NMR data are then processed to obtain T 2 relaxation time. The overall signal pattern is displayed, and the data is stored in the terminal device. In addition, wireless communication module is embedded in the software for data logging/sharing over an encrypted wireless network, promoting usability in a remote clinical site.

To ensure stable and reliable T 2 relaxation time measurements, a temperature compensation engine has been implemented to run independently from the main thread. This temperature compensation engine transforms NMR data from time domain to frequency domain using fast Fourier transformation (FFT) and detects any changes in the NMR frequency due to environmental temperature fluctuation (described below).

5.3.1 Temperature Compensation

The NMR frequency f 0 is the parameter that requires most frequent adjustments and affects the measured NMR signals most significantly. The frequency f 0 changes as the magnetic field (B 0) from the permanent magnet drifts with temperature [58]. For example, with a 1 ∘ C increase in temperature, B 0 field from a NdFeB magnet will drop ∼ 0. 1 % from its initial value, and f 0 will proportionally decrease by 0.1 %; when the initial f 0 is 20 MHz (B = 0. 47 T), the frequency change is then ∼ 20 kHz. Such changes can place down-converted NMR signals near or beyond the low-frequency cutoff in the amplification stage, distorting the measured signal. Commercial benchtop NMR systems address the problem by housing the entire magnet block inside a heated container. This solution, however, significantly undermines the portability of the system due to the use of bulky and power-consuming parts. In the new μNMR system, we employed a dynamic control approach. Namely, programmable hardware in the NMR electronics and temperature compensation engine in the NMR software are designed to track and compensate for temperature dependency of the system. These implementations ensure optimal measurement settings for reliable and robust performance.

Figure 9.9a and b show the algorithm for temperature compensation. The feedback loop tracks the Larmor frequency f 0 and reconfigures the frequency f of NMR excitation. Coarse-tuning mode starts with initial NMR excitation frequency f i and increases \(f\) by \(\Delta f\). When the spectral power (P) of NMR spin echo reaches a predefined threshold P th , a fine-tuning mode takes over to measure the frequency offset \({f}_{d}(=\vert f - {f}_{0}\vert )\). The fine tuning iterates until f d reaches a target value. The target f d value is carefully selected to keep down-converted NMR signal within the passband of the low-pass filter. Once the new NMR excitation frequency f has been established, CPMG pulse sequence is used to measure the T 2 relaxation time of the sample.

Fig. 9.9
figure 9

Automated temperature tracking in the new portable μNMR system. (a) A flow diagram of the temperature compensation engine. The tracking routine searches for the Larmor frequency f 0 at a given environmental setting, using the successive tracking routines (coarse and fine). Once f 0 is determined, a full NMR measurement is performed. (b) Temperature tracking algorithm. For a given RF excitation with the frequency f, a spin echo signal is measured and transformed into the frequency domain via fast Fourier transform (FFT). The coarse search mode checks whether the amplitude (P) of a peak in the spectral power is larger than the predefined value (P TH ). In the next stage, the fine search mode iteratively tunes f until the frequency offset \({f}_{d}(=\vert f - {f}_{0}\vert )\) reaches the optimal offset value ( ∼ 3 kHz). (c) The effectiveness of temperature compensation was evaluated using an MNP solution (\({T}_{2} \approx 100\,\mathrm{ms}\)). When the tracking routine was turned off (square), the T 2 variation was > 200 % of its initial value. With the tracking routine turned on (circle), the fluctuation was significantly reduced ( < 1 %). (d) The robustness of μNMR measurements across a broad range of temperature (\(4\mbox{ \textendash }5{0}^{\circ }\mathrm{C}\)) was demonstrated. The dotted line indicates a theoretical prediction (Reproduced from [57]. Copyright 2011 RSC Publishing)

Full size image

Figure 9.9c demonstrates the effectiveness of the developed temperature compensation method. When f 0 was allowed to drift but the RF frequency (f) for sample excitation was fixed, T 2 values varied up to 200 % relative to its starting value with typical fluctuation of room temperature \((\Delta T \sim {2}^{\circ }\mathrm{C})\). When the temperature compensation engine was activated, however, T 2 variations were significantly reduced to < 1 %. We further tested the system in environmental settings with a wide range of temperature differences (\(4\mbox{ \textendash }5{0}^{\circ }\mathrm{C}\)). To determine the measurement accuracy, the linear dependence of T 2 on temperature was utilized. Figure 9.9d shows the T 2 values of an MNP solution monitored at different temperature. For a given environment setting, the μNMR system was operated with the temperature tracking activated to compensate for minute temperature variations (\(\sim {1}^{\circ }\mathrm{C}\)). The results show a linear relationship (R 2 > 98 %) as theoretically predicted, demonstrating the capacity for reliable T 2 measurements in various settings.

6 Biological Applications

6.1 Cancer Detection and Profiling

Sensitive detection and rapid characterization of tumor cells in minimally processed biological samples will have significant impact on both biomedical research and clinical practice. Using the first-generation DMR device (μNMR-1) [22], DMR molecular profiling of cancer markers (Her2/neu, EGFR, and EpCAM) on human cells was demonstrated with CLIO nanoparticles directly conjugated to monoclonal antibodies. With the development of the second-generation μNMR-2 and the highly magnetic \({\mathrm{MnFe}}_{2}{\mathrm{O}}_{4}\) nanoparticles, subsequent cellular detection sensitivity was remarkably improved to approximately single-cell level, far surpassing the sensitivity of other conventional clinical methods (Fig. 9.10a). Furthermore, a new assay protocol was established that reports the expression level of a specific biomarker and the target cell density.

Fig. 9.10
figure 10

Cellular detection via DMR. (a) The detection limit by DMR is nearly at single cell level ( ∼ 2 cells), which is superior to the current clinical methods (cytology and histology). (b) The uptake of unmodified MNPs by cells was exploited to estimate the cell population in the samples. The measured \(\Delta {R}_{2}\) was linearly proportional to the cell concentrations. Importantly, the linear trends were statistically identical in different cell types. (c) The analytical accuracy of the μNMR was benchmarked against flow cytometry and Western blotting by measuring HER2/neu expression on breast cancer cells (Adopted from [28]. Copyright 2009 National Academy of Sciences, USA)

Full size image

To independently measure cell numbers by μNMR, we exploited a phenomenon of low-grade phagocytosis of nontargeted MNPs by tumor cells [59]. When mammalian cells were incubated (15 min at \(3{7}^{\circ }\mathrm{C}\)) with unmodified MNPs (MNP-\(\Phi \)), linear and cell-number-dependent R 2 changes (\(\Delta {R}_{2}^{\Phi }\)) were observed (Fig. 9.10b). Interestingly, these changes were similar across a wide variety of cell types. The results were fitted to Eq. 9.2; \(\Delta {R}_{2}^{\Phi } = {r}_{2}^{\Phi } \cdot {n}_{C}\), where \({r}_{2}^{\Phi }\) is the cellular relaxivity for MNP-\(\Phi \) and n C is the cell concentration (N C  ∕ V ). The cellular relaxivities (\({r}_{2}^{\Phi }\)) were statistically identical (p > 0. 99) among different tumor cell lines, suggesting that the method may provide a universal measure for estimating n C .

Using Eq. 9.2 and the cell density information (above), the expression level (ξ) of a select marker was defined as

$${\xi }^{\mathit{Ab}} = \frac{\Delta {R}_{2}^{\mathit{Ab}}} {\Delta {R}_{2}^{\Phi }} = \frac{{r}_{2}^{\mathit{Ab}} \cdot {n}_{C}} {{r}_{2}^{\Phi } \cdot {n}_{C}} = \frac{{r}_{2}^{\mathit{Ab}}} {{r}_{2}^{\Phi }} ,$$
(9.4)

where \(\Delta {R}_{2}^{\mathit{Ab}}\) and \({r}_{2}^{\mathit{Ab}}\) are R 2 changes and the cellular relaxivity, respectively, with a marker-specific MNP. In this normalized form, ξ now reports the cellular expression level of a targeted marker, providing a way to molecularly profile target cells regardless of cell numbers in a sample. The method was extensively verified by comparing μNMR measurements to other standard methods (flow cytometry, Western blotting) [22, 27, 28]. In one set of experiments, we measured the expression level of HER2 in breast cancer cell lines (Fig. 9.10c). The measured \({\xi }^{\mathit{HER}2}(= \Delta {R}_{2}^{\mathit{HER}2}/\Delta {R}_{2}^{\Phi })\) from μNMR (requiring \(\sim 1{0}^{3}\) cells) showed good agreement (R 2 > 98 %) with both flow cytometry (requiring ∼ 105 cells) and Western blotting (requiring ∼ 107 cells), validating the analytical capability of DMR. Note that DMR detection was much faster ( ∼ 15 min) and performed using > 102 times fewer cells.

6.2 Clinical Trial

Through the integration of the complementary DMR and BOND technologies, this chip-based NMR detection platform has been applied in clinical trials of cancer cell profiling [26]. A total of 50 patients with suspected abdominal malignancies were enrolled. Each patient underwent fine-needle aspiration (FNA) using a 22-G needle, followed by routine core biopsies (17-G needle) for conventional analysis. The FNA samples were aliquoted and profiled for 11 predefined cellular markers: 9 cancer-related markers (EpCAM, MUC-1, HER2, EGFR, B7H3, CK18, Ki-67, p53, Vimentin), CD45 for leukocyte counts, and control (MNP-\(\Phi \)) to measure the total cell density. A priori selection of cancer markers was based on current practice (e.g., EpCAM, CK18) [60, 61] or on reports of clinically relevant overexpression [62–64]. On average, 3,850 cells were obtained per patient via FNA. Of these, approximately one-third were CD45-positive leukocytes (mean: 1,273 cells), and the remaining were nonleukocytic, primarily tumor cells as determined by flow cytometry. The aliquots (containing ∼ 350 cells) were labeled using the BOND method and measured by the clinical μNMR system. The R 2 data with CD45 was used to account for the contribution by leukocytes; for each cancer marker, its cellular expression level (ξ) was obtained based on Eq. 9.3.

Figure 9.11 shows the profiling results of 9 cancer markers for 50 patients. When plotted for each patient, the expression level of the markers showed considerable heterogeneity across samples, which reconfirmed the importance of multiple-marker screening in cancer detection. Indeed, when analyzed for the diagnostic sensitivity, specificity, and accuracy (Table 9.2), a single marker showed accuracy of < 70 %. The highest accuracy for cancer diagnosis with our cohort was obtained with a quadruple marker combination (\(\mathrm{MUC}1 + \mathrm{EGFR} + \mathrm{HER}2 + \mathrm{EpCAM}\), 96 % accuracy), followed closely by weighted triple markers (\(\mathrm{MUC}1 + \mathrm{EGFR} + \mathrm{HER}2\), 92 % accuracy). Interestingly, the EpCAM and CK18 combination, which is routinely used in detecting circulating tumor cells in peripheral blood, achieved an overall diagnostic accuracy of 80 %, wherein high sensitivity (84 %) was offset by low specificity (50 %).

Fig. 9.11
figure 11

Expression levels of different biomarkers arranged by patient number. Note the high degree of heterogeneity of marker expression per person. Patients 5, 12, 17, 18, 21, and 42 had benign lesions. x-axis, patient number; y-axis, expression level (ξ) of a marker (Reproduced from [26]. Copyright 2011 American Association for the Advancement of Science AAAS)

Full size image
Table 9.2 Sensitivity, specificity, and accuracy with DMR
Full size table

The DMR results were also compared to those by standard-of-care methods. Conventional cytology on FNA specimen was performed in 49 of 50 cases and was diagnostic in 36 cases with 11 misdiagnoses (accuracy 74 %). Conventional histology on all 50 biopsy samples correctly diagnosed 37 cases out of 45 diagnostic samples (accuracy 84 %), with the remaining results nondiagnostic (5 cases). DMR consistently outperformed the other methods, with 2 misdiagnoses in all 50 samples (accuracy up to 96 %). Also note that DMR permitted fast detection (in less than 60 min for each patient), whereas the mean clinical turnaround time (from sample submission to final report) was 3 days for cytology (1–8 days) and 4 days for surgical pathology (1–11 days).

7 Conclusions and Outlook

As a novel biosensing platform, DMR offers many synergistic advantages over traditional approaches, such as high detection sensitivity, multiplexed capability, rapid measurement, and small sample volume requirement with minimal sample processing. Indeed, DMR thrives through the complement of several cutting-edge technologies, namely, magnetic nanomaterials, bioconjugation chemistry, and microfabrication. With new developments such as chip-based μNMR devices, optimized magnetic nanoparticles, and advanced labeling techniques, the DMR technology has proven itself as a robust and sensitive approach for quantitative and molecular analyses for biomedical research. Moreover, our recent clinical trial has confirmed that the DMR is capable of detecting and molecularly profiling cells with minimal false negatives. Its remarkable performance and potential impact on clinical disease management would no doubt accelerate the advance of personalized treatment by providing valuable information on molecular signature of individual patients.

We further envision broader application of the DMR in global healthcare. The DMR technology does not require extensive sample purification and can be packaged as a portable device. The system thus is well suited for rapid and point-of-care (POC) testing, especially in resource-limited primary clinics where majority of diagnoses are made based on physical symptoms only. Indeed, we plan to evaluate the developed system for TB (tuberculosis) detection in fields. Successful completion of this research will be a cornerstone for realizing POC technology for TB detection, which will bring significant societal benefits worldwide.