The diagnosis starts with specimen collection and incubation with bacteria specific MNP (Supporting Information Figure S1). MNP bind to the bacterial wall, rendering the bacteria superparamagnetic. In a subsequent step the spin-spin relaxation time (T₂) of the whole sample is measured by NMR. As the magnetic fields from MNP dephase the precession of nuclear spins in water protons[6, 7], each MNP-tagged bacterium can shorten the T₂ of billions of surrounding water molecules. To increase detection sensitivity, we have incorporated signal amplification schemes that made it possible to detect small quantities of bacteria in relatively large sample volumes. At the nanoparticle level, the detection signal has been enhanced by synthesizing Fe-based MNP with the high transverse relaxivity (r₂). At the device level, the signal was amplified by concentrating bacteria into a microfluidic chamber where the NMR signal was measured. To provide portable, on-chip bacterial detection, the NMR signal was read out using a miniaturized NMR system we have recently developed.[8].

First, we developed hybrid MNP with a large Fe core and a thin ferrite shell (“cannonballs”; CB), that have very high r₂ per particle (Table S1). With the limited number of binding sites per bacterium, the T₂ of samples will be shorter and thereby the detection will be more sensitive when the individual MNP have higher r₂. Since r₂ is M²•d², where M and d are the particle magnetization and the diameter respectively[6], we focused on making larger MNP using highly magnetic material (Fe). Figure 1a shows an example of highly mono-disperse CB (d = 16 nm). Initially, we made Fe-only MNP by thermally decomposing Fe(CO)₅, followed by controlled air-oxidation to grow the ferrite shell.[9] Compared to chemical oxidation[10], our method produces a thinner shell and thereby leaves a larger Fe core (Figure S2), leading to higher M. The shell showed high crystallinity (Figure 1b) and X-ray diffraction revealed a typical pattern for a spinel structure (Figure 1c), confirming the ferrite nature of the shell.[11] The shell protected the Fe core from oxidation to maintain the magnetic properties of CB (Figure S3). CB showed high magnetization (139 emu g⁻¹ [Fe]) and yet were superparamagnetic at room temperature (Figure 1d). Most importantly, CB assumed high r₂ (= 6.1×10⁻¹¹ m L s⁻¹, 1.5 T; Figure S4), due to their high magnetization and large diameter.

To render CB specific for BCG, we conjugated anti-BCG monoclonal antibodies to their surface (CB-BCG; see Methods in Supporting Information). Bacterial samples were then incubated with CB-BCG for 10 min, followed by a wash step to remove excess particles. Optical microscopy with fluorescent CB-BCG showed excellent targeting (Figure 2a). Binding of CB-BCG was further verified by the element mapping (Figure 2b), which showed high Fe signal on the bacterial membrane. The number of CB-BCG per bacterium, quantified by inductively coupled plasma atomic emission spectroscopy (ICP-AES; Figure S5), was ≈ 10⁵. Magnetic tagging thus made the bacteria highly efficient T₂-shortening agents (Figure 2c). The binding of CB-BCG to BCG could be inhibited by antibody blockade against bacterial epitopes. For control CB or when CB-BCG were used against other bacterial strains, the CB binding was minimal with T₂ changes (ΔT₂) < 5%. The number of CB-BCG tested against different bacteria species were < 600 per bacterium (Figure S5), confirming the specificity of CB-BCG.

To further enhance detection sensitivity and to streamline assay procedures, we next developed a chip-based filter system with NMR compatibility (Figure 3). A key component is a microfluidic chamber enclosed by a membrane filter and surrounded by a microcoil. The membrane filter performs two functions. First, it captures bacteria and concentrates them into the microfluidic chamber for NMR detection, providing a way to detect a small number of bacteria from large sample volumes. Second, the filter enables on-chip separation of bacteria from unbound CB, obviating the need for separate off-chip purification steps. Figure 4a illustrates the operating principle of the NMR-filter system. An unpurified sample is introduced into the microfluidic channel. Bacteria are retained by the membrane (pore size ≈ 100 nm), while excess CB permeate. Subsequently, a buffer solution is repeatedly injected to wash off unbound CB and the membrane is backwashed to redisperse the captured bacteria. Figure 4b shows an operation example with a sample containing BCG and CB-BCG. After sample loading, the microfluidic channel was flushed with PBS to remove unbound CB-BCG. Finally, the flow direction was reversed to resuspend the captured BCG for NMR measurements. The number of washing steps to remove unbound CB-BCG was determined by measuring T₂ after each wash (Figure 4c). T₂ values plateaued in 4–5 washing steps with the whole washing procedure complete in < 5 min. The electron micrograph of the membrane after complete wash (Figure 4d) verified bacterial capture on the filter. Note that by removing unbound particles, CB-induced T₂ changes are much more pronounced than the previously described homogeneous assay[8], achieving >10³ times higher detection sensitivity.

To compare the effect of MNP relaxivity on detection sensitivity, we first used a NMR system without a membrane filter. Two types of nanoparticles, CB and cross-linked iron oxide[12] (CLIO; r₂ = 7.0×10⁻¹³ mL s⁻¹), were used (Figure 5, black and blue curves). CB-BCG showed much higher mass sensitivity, detecting as few as 6 CFU (in 1 µL sample volume), whereas the detection limit was ~100 CFU with CLIO-BCG. Using the NMR-filter system, we achieved substantially high concentration detection sensitivity (Figure 5, red curve). When BCG samples (100 µL) targeted with CB-BCG were filtered, the concentration detection limit was ~60 CFU mL⁻¹. It is important to note that the concentration limit is theoretically unlimited with the filtering, because bacteria can be concentrated into the NMR detection chamber from large sample volumes.

suppl