BonFIRE: Bond-Selective IR Fluorescence Microscopy

• BonFIRE is a bond-selective imaging technique that combines IR vibrational excitation with fluorescence detection for precise chemical mapping.
• It leverages both double-resonance and photothermal modulation to selectively activate and capture vibrational modes in complex biological and material systems.
• Integrated with advanced computational models and high-resolution instrumentation, BonFIRE achieves submicron spatial resolution and rapid, quantitative spectral predictions.

Bond-Selective Fluorescence-Detected Infrared-Excited (BonFIRE) spectro-microscopy integrates vibrational excitation in the mid-infrared (MIR) or far-infrared (FIR) domain with fluorescence-detected readout, providing chemical bond selectivity, submicron spatial resolution, and high measurement sensitivity. In BonFIRE, an IR pump pulse excites targeted vibrational modes, and subsequent changes are detected via sensitive fluorescence quantum-yield modulation of coupled fluorophores. This technique leverages the orthogonality of IR vibrational selectivity and fluorescence labeling, enabling selective imaging and spectroscopy of chemical bonds within complex biological and material environments with minimal background. BonFIRE encompasses both purely photothermal approaches as well as true double-resonance (direct vibrational upconversion) modalities. The approach is supported by robust theoretical models and computational workflows that allow for in silico molecular probe design and quantitative spectral prediction (Kocheril et al., 17 Jan 2026, Zhang et al., 2021, Razumtcev et al., 2024, Ao et al., 6 Apr 2025, Zhao et al., 2023, Li et al., 2021).

1. Physical Principles and Theoretical Framework

BonFIRE exploits the interplay between IR vibrational excitation and fluorescence sensitivity. Two principal signal-generation mechanisms define the breadth of methodologies:

Double-Resonance (Vibrational Upconversion):

A pulsed MIR pump excites a specific vibrational mode (transition $S_0 \rightarrow |S_0^*,i\rangle$ at $\omega_{\mathrm{IR}}$). A near-IR probe then up-converts vibrational population into an electronically excited state ($S_0^*,i \rightarrow S_1,f$), and fluorescence is collected from $S_1$ (Kocheril et al., 17 Jan 2026). The steady-state BonFIRE intensity $I_\mathrm{BF}$ is given by:

$$I_\mathrm{BF}(\omega_\mathrm{IR}, \lambda_\mathrm{Fl}) \propto \sum_{i,f} |\langle S_1, f|\mu_e(Q)|S_0^*,i\rangle|^2 \cdot |\langle S_0^*,i|\mu_v|S_0\rangle|^2$$

where $\mu_v$ is the vibrational transition dipole and $\mu_e(Q)$ the electronic transition dipole (Herzberg–Teller expanded about $Q=0$) (Kocheril et al., 17 Jan 2026).

Photothermal Modulation:

Alternatively, the absorption of IR light by molecular bonds results in localized heating ($\Delta T$), which modulates the quantum yield $Q$ of thermo-sensitive fluorophores according to:

$\frac{\Delta F}{F_0} \approx \kappa\,\Delta T$

with $\kappa$ typically $\sim1\%/$K (for Rhodamine dyes, GFP, FITC) (Zhang et al., 2021, Li et al., 2021). The thermal field $T(r,t)$ evolves as:

$$\frac{\partial T(r,t)}{\partial t} = \alpha \nabla^2 T(r,t) + \frac{Q(r,t)}{\rho c_p}$$

where $Q(r,t)$ is the IR absorption power density, $\alpha$ thermal diffusivity, $\rho$ mass density, and $c_p$ specific heat (Zhang et al., 2021).

Bond Selectivity:

Both approaches derive chemical specificity from the sharp dependence of the IR absorption cross-section $\sigma(\nu)$ on vibrational resonance. By spectrally tuning $\omega_\mathrm{IR}$, one excites distinct chemical bonds (e.g., amide I at 1650 cm$^{-1}$, nitrile at 2200–2300 cm$^{-1}$) with high selectivity (Kocheril et al., 17 Jan 2026, Zhang et al., 2021).

2. Instrumentation and Methodological Implementations

BonFIRE is realized in several sophisticated optical configurations:

Architecture	IR Source / Tuning	Detection Modality
Point-Scanning	QCL, 1,000–1,886 cm$^{-1}$	PMT, lock-in at IR pulse rate
Wide-field	QCL, chopped or pulsed IR	CMOS camera, “virtual lock-in”
FT-BonFIRE	Synchrotron broadband FT-IR	Step-scan, lock-in per mirror pos
3D BonFIRE (FBS-IDT)	QCL, 1500–1800 cm$^{-1}$	Computational phase tomography
OBF-MIP	QCL, 980–1800 cm$^{-1}$	SiPM, dual lock-in, boxcar gating

IR Excitation:

Quantum Cascade Lasers (QCLs) enable mode-selective excitation over the fingerprint region (e.g., 10 cm$^{-1}$ bandwidth, $\sim$2 ps pulses), with mechanical or electronic chopping for lock-in referencing (Zhang et al., 2021, Ao et al., 6 Apr 2025). Broadband implementations use FT-IR beamsplitters with step-scan interferometry for high spectral resolution (Razumtcev et al., 2024).

Fluorescence Readout:

Sensitive detection employs PMTs, APDs, or SiPMs. Fluorescence is synchronized via lock-in amplification at the IR pulse rate, with difference imaging acquired as “hot” and “cold” frames (corresponding to IR-on and IR-off states). Boxcar gating captures rapid thermal decay and minimizes bleaching (Ao et al., 6 Apr 2025).

Spatial Resolution:

The spatial resolution is dictated by the visible probe (e.g., $\lambda=520$–532 nm, NA=0.8–1.2). Achievable lateral resolutions are 300–700 nm (FWHM), exceeding the IR diffraction limit by over fivefold (Zhang et al., 2021, Razumtcev et al., 2024).

3. Fluorophore Chemistry and Probe Design

Fluorophores for BonFIRE are chosen for large $dF/dT$ with high photostability and targeted cell/organelle localization. Examples include:

• FITC, Cy2, Rhodamine 6G, Nile Red, Rhodamine 123, LysoSensor DND-189 (all 1%/K sensitivity)
• GFP, NucSpot dyes for genetically encoded or nucleic-acid targeting
• Probes can be conjugated for localization to lipids, proteins, mitochondria, lysosomes, or membranes (Zhang et al., 2021, Ao et al., 6 Apr 2025).

Optimal excitation/detection filters must avoid IR-absorption bands, minimize cross-talk, and ensure spectral compatibility with desired imaging depth and photostability profiles (Razumtcev et al., 2024). The use of dual-color or environment-sensitive dyes enables multiplexing and functional mapping.

4. Computational Workflows and Spectral Prediction

A fully automated computational pipeline (“AutoDFT”) enables first-principles prediction of BonFIRE spectra directly from chemical structure (ChemDraw/SMILES). The workflow consists of:

• Geometry building: 3D structure generation (OpenBabel)
• Ground-state optimization: Gaussian 16, B3LYP/6-31G(d,p), SMD (DMSO) solvation
• Vibrational analysis: DFT Hessian for normal modes $\{Q_k, \omega_k\}$, IR intensities
• Excited-state calculations: TD-DFT, S$_1$ energy/gradients, adiabatic Hessian for Duschinsky rotation
• Vibronic assembly: FCclasses3 sum-over-states, calculation of NIR spectra from pre-excited vibrational states $|v_i=1\rangle$, Gaussian broadening for experimental bandwidth matching

Key approximations include harmonic treatment of vibrations, frequency scaling (0.97–0.953 for strong anharmonics), the Condon approximation, and steady-state neglect of vibrational lifetimes or polarization. AutoDFT predictions replicate experimental BonFIRE, IR, and fluorescence spectra within 5 cm$^{-1}$ (peak positions), within a factor of 2 (relative intensities), and preserve correct bond selectivity across multiple chemical reporters (Kocheril et al., 17 Jan 2026).

5. Quantitative Performance Metrics

Sensitivity and Signal-to-Noise:

Fluorescence modulation of 1–5% per IR pulse train is typical (ΔT = 1–5 K, ΔF/F₀ ≈ 1–5%) within submicron volumes. Single-bacterium detection limits at amide I are ∼$10^{-17}$ cm$^2$ per molecule, with SNR ∼30 in ms-scale acquisition (Zhang et al., 2021). Advanced AI denoising (SPEND, 3D U-Net) achieves image SNR improvements of 26.9× and spectral SNR 5.3× (Ao et al., 6 Apr 2025).

Spectral and Spatial Resolution:

• Spectral: QCL linewidth 2–10 cm$^{-1}$, channel spacing ∼2–10 cm$^{-1}$, broadband FT step-scan Δν = 8 cm$^{-1}$
• Spatial: Lateral 300–700 nm (visible focus); axial ∼1–2 µm
• Imaging speed: Pixel dwell 30–200 µs (point-scan), 40 Hz full-frame (wide-field), ∼1–10 ms/pixel (FT/hyperspectral) (Zhang et al., 2021, Ao et al., 6 Apr 2025, Razumtcev et al., 2024)
• Volume rates: 6 Hz for 3D reconstructions in FBS-IDT (Zhao et al., 2023)

Bond Selectivity:

High-fidelity discrimination among bonds: amide I (∼1650 cm$^{-1}$, proteins), C=O (1740 cm$^{-1}$, lipids), CH stretches (2850, 2920 cm$^{-1}$, lipids), CN (nitriles 2200–2250 cm$^{-1}$). Combinatorial excitation enables multiplexed detection with minimal cross-talk.

6. Advanced Architectures and Computational Tomography

Fluorescence-guided Bond-Selective Intensity Diffraction Tomography (FBS-IDT):

Extends BonFIRE to 3D chemical imaging by coupling oblique illumination (phase tomography, 16-angle synthetic aperture) with IR photothermal modulation and fluorescence guidance (Zhao et al., 2023). This results in isotropic 350 nm (lateral) × 1.1 µm (axial) resolution, volumetric acquisition at 6 Hz, and site-specific extraction of chemical fingerprints (e.g., secondary structure composition in amyloid fibrils).

Optical Boxcar Enhanced F-PTIR (OBF-MIP):

Implements boxcar timing, AI denoising (SPEND), and spectral unmixing (MCR-LASSO) to resolve metabolic compositional heterogeneity in vivo, including lysosomal dynamics and disease states (Ao et al., 6 Apr 2025).

Broadband and Synchrotron-Driven FT-BonFIRE:

Employs step-scan Michelson interferometry with broadband IR sources for extended spectral coverage (500–4,000 cm$^{-1}$), submicron lateral resolution, and improved SNR over traditional FTIR (Razumtcev et al., 2024).

7. Applications, Challenges, and Future Prospects

Applications:

• In situ vibrational chemical imaging of metabolites, lipids, proteins, nucleic acids
• High-throughput rational design of environment sensors via in silico calculation of solvatochromism and vibrational shifts
• Directed evolution of fluorescent vibrational probes for multiplex single-molecule detection
• Inverse molecular design for 10–100 color super-multiplexed IR imaging with machine-learning guidance (Kocheril et al., 17 Jan 2026, Ao et al., 6 Apr 2025)
• 3D chemical mapping of protein secondary structure, lipid–protein interaction, and metabolic fingerprinting at the organellar level (Zhao et al., 2023, Ao et al., 6 Apr 2025)

Challenges and Solutions:

• Background heating and nonuniform thermal diffusion: mitigated by background subtraction, thermal modeling, and phase-sensitive detection
• Photobleaching: reduced by low-duty-cycle illumination, fast boxcar gating, photostable fluorophores
• Multiplexing limitations: overcome via orthogonal spectral, temporal, and molecular probe encoding

Future Directions:

• Real-time and high-throughput imaging: leveraging EOM/AOM modulation (>100 kHz), CMOS/SPAD array detection
• Super-resolution: exploiting thermal localization and computational reconstruction
• Expansion to endogenous tagging and in vivo clinical/biomedical applications

BonFIRE thus constitutes a powerful and versatile framework for bond-level chemical mapping in complex systems, integrating high-resolution microscopy, computational chemistry, advanced photophysics, and AI-enhanced analysis (Kocheril et al., 17 Jan 2026, Zhang et al., 2021, Razumtcev et al., 2024, Ao et al., 6 Apr 2025, Zhao et al., 2023, Li et al., 2021).