Alzheimer's disease is increasingly blamed on small, soluble clumps of the amyloid-beta and tau proteins, called oligomers, rather than on the large plaques that brain imaging detects. What makes an oligomer toxic is largely its shape. Many are thought to fold into β-sheets, flat ribbons of protein chain, and their arrangement differs from that of mature fibrils. Infrared light can read this shape directly. The protein backbone absorbs near 1650 cm⁻¹ in a band called Amide I, and the exact profile of that band separates helices, disordered chains, and the different kinds of β-sheet. There are two obstacles. A single oligomer absorbs very little, and liquid water absorbs strongly in the same band. The best label-free infrared images of tau aggregates therefore replace the water with heavy water (D₂O) and study only large, mature fibrils.
We propose a chip that addresses both obstacles. A pair of microrings, which are tiny ring-shaped optical resonators where light circulates thousands of times, sits on a membrane suspended over an opening etched through the silicon wafer. A mid-infrared pump laser comes from below, through the infrared-transparent silicon, so it reaches proteins captured on the ring surface before passing through more than a thin layer of water. When the proteins absorb, they warm slightly, and the ring's resonance shifts. This is photothermal spectroscopy, meaning detection of absorption through the heat it produces. One ring carries capture molecules and the other is passivated. Subtracting their signals removes the heating of the water and the chip itself. Sweeping the pump wavelength across Amide I gives an absorption spectrum of the captured proteins, without labels, in normal H₂O buffer.
Core Claim
A suspended, differentially read microring can record Amide I spectra of surface-captured amyloid-beta 1–42 species in H₂O at concentrations and sizes that free-space photothermal microscopes cannot reach. The underlying reason is physical. A free-space photothermal microscope must detect light scattered by the particle itself, so its signal falls steeply as the particle shrinks. A ring instead reads the total heat deposited in its own optical field, which scales only with how much material absorbed. We therefore predict a size crossover: below some aggregate size the cavity readout should beat the best free-space method. Locating that crossover, and showing that the resulting spectra separate oligomers from monomers and fibrils, is the central test.
Where This Sits
The closest prior work falls into three groups, and none does this.
The first group supplies the method. Single-Particle Photothermal Microscopy Using On-Chip Silicon Nitride Microring Resonators locked a probe laser to a silicon nitride ring with quality factor about 6×10⁴. A higher quality factor means a sharper resonance that is more sensitive to small shifts. With this setup they detected single carbon nanotubes at signal-to-noise ratio 282 ± 7. That work was dry and non-biological. It also reported two limits that motivate our design. Background absorption in the chip capped the signal-to-background ratio at 3.5 ± 0.1. And their simulations predicted that a fully suspended membrane would raise the signal 53-fold by stopping the silicon from draining heat away. We build that suspended geometry and add differential readout to attack the background. We also expect the gain in water to be well below 53-fold, because water conducts heat far better than air. The measured gain is one of our outcomes, not an assumption. Ultra-sensitive chemical vapor detection using micro-cavity photothermal spectroscopy modeled a 10⁴ photothermal enhancement in resonant microdisks for gas sensing and identified heat-sink thermal noise as the floor. We use that noise analysis to set expectations, but our setting is liquid and biological, which the model does not cover.
The second group establishes that infrared photothermal methods can resolve protein structure, and shows their limits.
- Mid-infrared chemical imaging of intracellular tau fibrils using fluorescence-guided computational photothermal microscopy. deconvolved Amide I inside cells to map β-sheet content in tau fibrils. It required D₂O, targeted mature fibrils, used no cavity, and reports no detection limit.
- Vibrational Spectroscopic Detection of Single Virus by a Confocal Interferometric Mid-Infrared Photothermal Microscope distinguished single ~100 nm viruses by their Amide I/II ratios (0.42 versus 0.51). It achieved about 25× gain over scattering-only detection by interfering the particle's scattered light with a reference. It remains our strongest competitor.
- Bond-selective intensity diffraction tomography and 3D Chemical Imaging by Fluorescence-detected Mid-Infrared Photothermal Fourier Light Field Microscopy extend photothermal imaging to 3D. Both target cell-scale objects rather than nanometer oligomers.
The third group motivates the biological target. A buoyant plasmonic microbubble-based SERS sensing platform for amyloid-beta protein detection in Alzheimer's disease uses surface-enhanced Raman scattering (SERS), a method that amplifies Raman light near gold nanostructures. It reaches 1 nM detection of amyloid-beta 1–40 oligomers. However, it reads the deformation of a reporter molecule, so it reports binding rather than structure. On the disease side, Characteristics of discordance between amyloid positron emission tomography and plasma amyloid-β 42/40 positivity documents that plaque imaging and soluble-peptide blood assays often disagree. That is the gap a structure-resolved oligomer measurement could help explain. We do not attempt a clinical diagnosis here.
Research Program
The chip. Each device holds two silicon nitride rings on a shared membrane, a few millimeters apart, suspended over a backside window etched through the silicon.
- Silicon is largely transparent near 6 µm, the Amide I region.
- A tunable quantum cascade laser, a compact mid-infrared source, pumps from below, through the window.
- A near-infrared probe laser is locked to each ring's resonance edge with the Pound-Drever-Hall method used by Podorova et al., which keeps the laser pinned to the steepest part of the resonance.
Surface chemistry.
- The sensing ring carries a capture layer. We use a conformation-agnostic one, such as an N-terminal antibody fragment, so that capture itself does not select for a structure.
- The reference ring carries the same passivation layer with no capture agent.
The measured quantity. The signal is the modulated resonance shift of the sensing ring minus that of the reference ring, normalized by pump power at each wavenumber $\nu$:
$$S(\nu) = \frac{\Delta\lambda_{s}(\nu) - \Delta\lambda_{r}(\nu)}{P(\nu)}$$
Here $\Delta\lambda_{s}$ and $\Delta\lambda_{r}$ are the sensing and reference resonance shifts at the pump modulation frequency, and $P$ is the pump power reaching the membrane. Water and the membrane heat both rings almost identically, so they largely cancel. Captured protein heats only the sensing ring.
Modulation frequency. We raise the modulation from the 1.1 kHz used previously into the tens of kHz. This shortens the distance heat spreads in water during one cycle from several micrometers toward the ring's own field depth, so less water heating contributes. The cost is a smaller signal, so we sweep the frequency and report the trade-off.
Spectral analysis. We fit the Amide I profile with the same Fourier self-deconvolution used for tau fibrils, separating sub-bands for helix, disordered chain, and parallel versus antiparallel β-sheet.
Samples. We first validate on calibration particles of known size and absorption: PMMA beads at 50, 100, and 200 nm, and protein-coated beads. We then move to amyloid-beta 1–42 in three forms:
- monomer, size-exclusion purified;
- oligomer, prepared by a standard low-temperature protocol;
- fibrils, with fibril morphology checked by cryo-EM using the RELION amyloid pipeline (Amyloid structure determination in RELION-3.1).
All three are prepared in H₂O PBS, and every prep is also made in D₂O PBS as a matched control.
Baselines
Naive reference point. This is the Podorova-style design run without our changes: a single ring on a solid substrate, front-side pump, no reference ring. It is measured in H₂O on the same captured samples. We expect the water and cladding background to swamp the Amide I signal. A measured signal-to-background ratio at or below their 3.5 ± 0.1 on non-biological particles, with no recoverable Amide I peak, would confirm the problem we are solving.
Strongest competitor: free-space confocal interferometric photothermal microscopy, following Vibrational Spectroscopic Detection of Single Virus by a Confocal Interferometric Mid-Infrared Photothermal Microscope. For a fair comparison:
- We build it on the same quantum cascade laser and the same pump power at the sample.
- It images the same capture-functionalized membranes, so the same particles are measured by both readouts. Particles are located by correlative atomic-force microscopy sizing.
- Each method gets the same integration time per particle and the same tuning budget for probe power and modulation frequency.
The cited paper reports about 25× signal gain over scattering-only detection for 100 nm PMMA, and single-virus spectra at 319 nm resolution. We found no reported result on Lacuna for the single-particle signal-to-noise ratio of this method on protein oligomers or in H₂O. Beating it means a higher per-particle signal-to-noise ratio at matched integration time below a measured size, with the cavity's advantage growing as size falls.
Sensitivity baseline: the SERS microbubble assay of A buoyant plasmonic microbubble-based SERS sensing platform for amyloid-beta protein detection in Alzheimer's disease. Its reported detection limit is 1 nM for amyloid-beta 1–40 oligomers. We reproduce it on our amyloid-beta 1–42 preparations. Beating it is not about raw detection limit. It means correctly classifying monomer, oligomer, and fibril at concentrations where SERS detects binding but cannot tell the species apart.
Spectral-quality reference: FBS-IDT, from Mid-infrared chemical imaging of intracellular tau fibrils using fluorescence-guided computational photothermal microscopy., with no reported detection limit found on Lacuna. We use only its D₂O Amide I fibril spectra, as a reference for what a correct deconvolution looks like.
Ablations. Four ablations isolate the key ingredients:
- suspended versus solid membrane, which isolates heat sinking;
- back-side versus front-side pump, which isolates the water path;
- differential versus single-ring readout, which isolates background cancellation;
- H₂O versus D₂O, which isolates residual water contamination of the spectrum.
Why not just…?
Why not just use the free-space interferometric photothermal microscope, which already resolves single viruses without any fabrication? This is the real threat, and we treat it as the main hypothesis to falsify. That microscope must collect light scattered by the particle itself, so its signal falls faster with shrinking size than a readout that sums deposited heat. That is an argument, not data, and for 10–20 nm oligomers the advantage may be eaten by water conducting heat away from the ring. The size-sweep experiment settles it. If the free-space method matches the cavity's per-particle signal-to-noise ratio down to the smallest oligomers we can make, we drop the single-particle claim. We would then argue only for the chip's other strength: parallel, non-scanning tracking of aggregation over time.
Why not just use SERS or an immunoassay, which already reach 1 nM? For counting amyloid-beta they likely suffice, and we do not claim better raw detection. The Ho et al. readout tracks deformation of a reporter molecule caused by copper binding, so it cannot say whether a captured peptide is disordered or folded into a β-sheet. The decisive comparison is structural classification at matched concentration. If SERS peak shifts can be trained to separate our three verified preparations as well as Amide I spectra do, the structural argument weakens. We test that directly with a classifier on SERS spectra of the same samples.
Why not just measure in D₂O, or on dried films with ordinary infrared spectroscopy that averages over the whole sample? D₂O changes hydrogen bonding and can change aggregation speed and pathway, and averaging the whole sample hides the minority of oligomers that matter. We quantify both concerns. We run matched aggregation time courses in H₂O and D₂O on chip and compare when β-sheet signal appears. If the two are indistinguishable, and bulk spectroscopy separates the species equally well, then the H₂O capability is a convenience rather than a necessity, and we would say so.
Decisive Experiments
Size-crossover test. On identical captured particles, we measure per-particle signal-to-noise ratio against atomic-force-microscopy size, for the suspended differential ring and for the interferometric microscope. Success is a crossover below which the ring wins, with the ring still producing a recognizable Amide I peak at oligomer sizes.
Water test. We record Amide I spectra of fibrils in H₂O and compare them with D₂O spectra from the same preparation, adjusting the comparison for the known deuteration shift. Success is agreement of the β-sheet sub-band position within the deconvolution's uncertainty, together with a reference-ring residual that stays flat across 1600–1700 cm⁻¹.
Structure classification test. We train a held-out classifier on monomer, oligomer, and fibril spectra, and report accuracy as concentration falls. We compare it with SERS at the 1 nM regime.
Kinetics demonstration. We track amyloid-beta 1–42 in H₂O over 48 hours, matching the window Ho et al. monitored. The question is whether antiparallel β-sheet signatures appear on the chip before parallel fibril signatures. That is the observation that would make this tool useful for Alzheimer's biology.
References
- Single-Particle Photothermal Microscopy Using On-Chip Silicon Nitride Microring Resonators
- Ultra-sensitive chemical vapor detection using micro-cavity photothermal spectroscopy
- Mid-infrared chemical imaging of intracellular tau fibrils using fluorescence-guided computational photothermal microscopy.
- Vibrational Spectroscopic Detection of Single Virus by a Confocal Interferometric Mid-Infrared Photothermal Microscope
- Bond-selective intensity diffraction tomography.
- 3D Chemical Imaging by Fluorescence-detected Mid-Infrared Photothermal Fourier Light Field Microscopy
- A buoyant plasmonic microbubble-based SERS sensing platform for amyloid-beta protein detection in Alzheimer's disease
- Characteristics of discordance between amyloid positron emission tomography and plasma amyloid-β 42/40 positivity
- Amyloid structure determination in RELION-3.1