Main

Optical imaging is important to biology and medicine as it offers exceptional spatiotemporal resolution for non-invasive in vivo imaging, with potentially diffraction-limited or sub-diffraction-limited spatial resolution in real time, and it therefore complements X-ray computed tomography, magnetic resonance imaging and ultrasound imaging. However, the spatial resolution and penetration depth of in vivo fluorescence imaging into live tissues is limited by the combined effects of absorption, scattering, tissue autofluorescence, the quantum yields (QYs) of probes, the optical configuration and detector sensitivity/efficiency. High-resolution fluorescence imaging relies on ballistic and slightly scattered snake-like photons transmitting through tissues, whereas multiple-scattered diffusive photons contribute to noise or background and worsen the diffraction-limited resolution1,2. Light scattering within tissues is dominated by Rayleigh and Mie scattering3, both of which decrease at longer wavelengths as λ−α(ref. 4; Fig. 1a), where λ is the imaging wavelength and α = 0.2–4 for tissues5. Reduced light scattering can afford deeper and higher-contrast fluorescence imaging with less diffusive noise at longer wavelengths.

Fig. 1: Motivation for NIR-II fluorescence imaging.
figure 1

a, Reduced scattering coefficients of different biological tissues and the Intralipid phantom with concentrations of 0.8% and 4.6% in the 400–3,000-nm window. b, The attenuation length (1/(1/ls + 1/la)) of brain tissue mimicked by 5% Intralipid solution in the NIR-I and NIR-II windows, where ls and la are the scatter and attenuation lengths, respectively12. The stars in the graph denote the effective attenuation lengths of mouse brain that have been reported previously78,119. c, Fluorescence imaging of a 50-μm capillary submerged at various depths in 5% Intralipid by a wide-field system with a 2D InGaAs camera or a confocal microscope with SNSPDs. An 808-nm and a 1,319-nm laser were used for NIR-IIb wide-field imaging and confocal microscopy, respectively. NIR-IIc confocal microscopy was performed with 1,540-nm or 1,650-nm excitation12. d, Non-invasive NIR-I, NIR-IIa and NIR-IIb fluorescence imaging of cerebral vasculatures of mice17. Credit: panels b and c adapted with permission from ref. 12, Springer Nature Ltd. Panel d reproduced with permission from ref. 17, Wiley.

Full size image

For decades, near-infrared imaging in the 700–900-nm NIR-I window has been pursued for in vivo biomedical imaging6 so as to benefit from the suppressed light scattering by tissues compared to visible light as well as lower absorption by haemoglobin7. NIR-I fluorescence imaging became widely accepted as a result of the successes of fluorophores such as indocyanine green (ICG) and the advent of digital imaging technology in the early 2000s8. However, in vivo imaging in NIR-I still suffers from feature blurring caused by light scattering, shallow penetration depth and a high background due to both scattering and autofluorescence from endogenous chromophores or pigmented components in the body9,10.

The Dai group, in 2009, demonstrated the first 1,000–1,700-nm NIR-II preclinical fluorescence imaging of mice using hydrophilic-polymer-coated single-walled carbon nanotubes (SWNTs) and a liquid-nitrogen-cooled indium gallium arsenide (InGaAs) camera11. In 2022, the group performed in vivo imaging in the 1,700–2,000-nm range and refined the definition of NIR-II to 1,000–3,000 nm (ref. 12), largely overlapping with the 900–3,000-nm short-wave infrared (SWIR) range. Subsequently, the group further improved the imaging spatial resolution, imaging depth and signal/background ratio (SBR) and diminished tissue autofluorescence with NIR-II over NIR-I imaging5,10,11,12,13,14,15,16,17,18,19,20,21,22,23,24,25,26,27,28,29. Light absorption by water in biological tissues and light scattering by tissues limit the penetration depth of light into a living body. Given water absorption by vibrational overtone modes, tissue scattering of light and the detection range of ~900–1,700 nm of InGaAs cameras, one can divide the NIR-II range into several sub-windows with local maxima in light penetration depth versus wavelength (Fig. 1b)12; these sub-windows include the NIR-IIa (1,300–1,400 nm), NIR-IIb (1,500–1,700 nm), NIR-IIc (1,700–2,000 nm) and NIR-IId (~2,100–2,300 nm) windows. Moderate light absorption by water in the 1,400–1,500-nm range was shown to enhance NIR-II image contrast due to the fact that multiply-scattered diffusive light decays faster than ballistic light30,31. Beyond NIR-IId, light absorption by the vibrational normal modes of water is too strong, making through-tissue fluorescence imaging impossible12. Fluorescence imaging of a fluorophore-filled capillary through different thicknesses of tissue phantoms made of Intralipid solutions (Fig. 1c)12 and mouse brain (Fig. 1d)17 showed clearly improved resolution and SBR when transitioning from NIR-I and IIa to the IIb and IIc sub-windows. This is because high resolution and low feature smearing hinge on low scattering, which is provided with longer-wavelength light. NIR-II imaging beyond a tissue depth of 1 cm has also been demonstrated32.

Notably, although NIR-I light exhibits much lower light absorption, its tissue penetration depth is much shallower than for NIR-II light due to scattering (Fig. 1b), making NIR-II imaging a step-out technology over the traditional NIR technique. The large Stokes shift between excitation (typically 808 nm) and emission for wide-field imaging in the >1,500-nm NIR-IIb, IIc and IId sub-windows eliminates any tissue autofluorescence background (Fig. 1d), even in mouse liver10. The NIR-IIb and IIc regions are the highest-performing sub-windows for in vivo NIR-II imaging.

Recent progress in NIR-II fluorophores and nanoprobes

Fluorescent probes for biological imaging should exhibit high brightness (high quantum yield (QY)) and molar absorptivity/extinction coefficient) and biocompatibility. The QY of NIR-II fluorophores is lower than those for their visible or NIR-I counterparts. In molecular fluorophores, non-radiative relaxation between the zero-vibrational level of excited states and the higher isoenergetic vibrational levels of the ground state can quench molecular fluorescence33, and this effect, called the ‘energy gap law’, becomes more pronounced as the energy gap shrinks at longer wavelengths. In aqueous environments, the NIR-II molecular fluorophores, typically with larger π-conjugated backbones, suffer from stronger intermolecular interactions, which lead to further non-radiative decay of the NIR-II emission34. The abundance of hydroxyl groups in aqueous solution is also reported to be a serious quencher of NIR-II luminescence of rare-earth nanoparticles (RENPs)19. The nonpolar conjugated backbones of molecular fluorophores and the hydrophobic capping layers of inorganic nanoparticles require effective hydrophilic modification for biocompatibility, a process that decreases the fluorescence QY drastically. Despite these challenges, recent years have witnessed an outpouring of promising high-performance NIR-II probes.

Inorganic nanostructured NIR-II probes

The first NIR-II imaging11 utilized photoluminescent SWNTs in the 1,000–1,700-nm range (Fig. 2a)13, depending on the nanotube chirality and diameter (0.7–1.4 nm; Fig. 2b)5,15,17,35. NIR-II quantum dots (QDs) such as silver sulfide (Ag2S; 1,100–1,400 nm)16,36, lead sulfide (PbS; 1,000–2,000 nm; QDb: 1,500–1,700 nm and QDc: 1,700–2,000 nm; Fig. 2c)12,21 and indium arsenide (InAs; 900–1,600 nm; Fig. 2d)37 were shown to exhibit a higher fluorescence QY than SWNTs. These QDs were typically overcoated with a passivation shell to avoid oxidation, affording core–shell QDs with bright NIR-II emission in aqueous solutions12,21,37. Downconversion RENPs have shown fascinating optical properties, such as narrowband emission spanning the NIR-II range39, long luminescence lifetimes (on the scale of up to tens of milliseconds)40 and Auger-effect-based persistent luminescence after X-ray irradiation38. To enhance the NIR-II emission, Ce3+-doping19 and cubic-phase23 strategies can suppress upconversion while boosting Er3+ downconversion luminescence at 1,550 nm by approximately nine- and eightfold, respectively. More recently, cubic-phase RENPs based on a Tm3+ emitter have been developed (Fig. 2e) that exhibit 1,600–1,700-nm sub-NIR-IIb fluorescence amplification41. Finally, gold molecular clusters with ultrasmall size (Au25, ~1.6 nm; Fig. 2f) exhibited luminescence in the 1,000–1,400-nm range27,42,43.

Fig. 2: NIR-II fluorescence probes.
figure 2

a, Emission spectral ranges of different inorganic nanostructured NIR-II probes. b, Schemes of SWNTs with proteins anchored on the surface via pyrene π-stacking35. c, Emission spectrum and schematic design (inset) of NIR-IIc core–shell PbS/CdS QDs12. d, Schematic overview of InAs-based QDs coated with a passivation shell37. e, Schematic of the Tm3+-doped cubic-phase core–shell-structured RENPs41. f, Crystallographic representation of Au25 nanoclusters27. g, Schematic illustration of the hydrophilic RENPs with P3 coating crosslinking polymeric layers23. h–l, Example structures of the NIR-II molecular fluorophores polymethine (JuloFlav744 and FD-108045; h), donor–acceptor molecules (IR-FEP46 and CPTIC-4F47; i), BODIPY (FBPEG 91248; j), rhodamine (NIR-II-HD5-ALP49; k) and metal–macrocycles complexes (EB76650; l). Credit: panel b adapted with permission from ref. 35, American Chemical Society. Panel c reproduced with permission from ref. 12, Springer Nature Ltd. Panel d adapted with permission from ref. 37, Springer Nature Ltd. Panel e reproduced with permission from ref. 41, Springer Nature Ltd. Panel f adapted from ref. 27 under a Creative Commons licence CC BY 4.0. Panel g adapted with permission from ref. 23, Springer Nature Ltd. Panel h (left) adapted with permission from ref. 44, Springer Nature Ltd. Panel h (right) adapted with permission from ref. 45, American Chemical Society. Panel i (left) adapted with permission from ref. 46, Wiley. Panel i (right) adapted with permission from ref. 47, Springer Nature Ltd. Panel j adapted with permission from ref. 48, Wiley. Panel k adapted with permission from ref. 49, Wiley. Panel l adapted with permission from ref. 50, Springer Nature Ltd.

Full size image

Inorganic nanostructured NIR-II probes are often synthesized in organic solvents and coated with hydrophilic, crosslinked polymer layers (‘P3 coating’; Fig. 2g) to impart high biocompatibility for preclinical use23,25. The P3 crosslinked surface coating enables rapid biliary clearance and reduces long-term retention-induced side effects of a wide range of nanomaterials, including RENPs, PbS QDs and superparamagnetic iron-oxide nanoparticles, enhancing their in vivo pharmacokinetics and their potential use for nanomedicine25.

Molecular fluorophores

Molecular fluorophores are important NIR-II probes owing to their well-defined structures, rich chemical and structural tunability and generally high biocompatibility and favourable pharmacokinetics. Thus far, NIR-II molecular fluorophores include polymethine (Fig. 2h)44,45, donor–acceptor molecules (Fig. 2i)18,46,47, boron-dipyrromethene (BODIPY; Fig. 2j)48, rhodamine (Fig. 2k)49 and metal–macrocycles complexes (Fig. 2l)50. High-performance in vivo NIR-II fluorescence imaging has been demonstrated using organic fluorophores exhibiting advanced properties such as a long-wavelength peak absorption up to 1,400 nm (ref. 45), large absorption coefficient of 105 mol−1 cm−1 (ref. 51), high QYs of >5% (ref. 47) or long emission in the NIR-IIb window52. These molecules generally have large conjugated backbones with high hydrophobicity. For aqueous solubility, researchers typically encapsulate the molecules in amphiphilic polymer matrices21,53 or functionalize them with hydrophilic side chains54.

The absorption/emission wavelength of NIR-II molecular fluorophores can be redshifted by increasing the conjugated backbone length52, increasing the donor/acceptor unit strength55 or forming J-aggregates45. Video-rate multicolour imaging with NIR-II fluorescence under a multiplexed excitation wavelength has been demonstrated with flavylium polymethine dyes with finely tuned heterocycle modification44. Owing to the energy gap law, redshifted fluorophores generally show lower QYs, especially for molecular fluorophores with peak absorption over 1,000 nm. Strong interactions between water molecules and conjugated backbones cause substantial non-radiative decay for NIR-II molecular fluorophores54,55. Protecting the conjugated backbones from water molecules is thus vital for high fluorescent QYs under aqueous conditions.

An advantage of organic NIR-II fluorophores over inorganic nanoparticle probes is their smaller size, which favours renal excretion upon administration to a body. Renal excretion of NIR-II probes is preferred when considering potential clinical use, because excretion through the kidney/urinary pathway is fast, making them safer and less likely to cause toxic effects than probes remaining in the body for extended times. So far, only a few NIR-II molecular fluorophores have been reported with renal-excretion ability18,20,48,56. A caveat is that attaching highly hydrophilic side groups such as β-cyclodextrin (β-CD)56 and poly(oligo(ethylene glycol) dimethacrylate) polymer brushes48 to NIR-II fluorophores can facilitate renal excretion, but this typically lowers the fluorescence QY due to non-radiative relaxation of excited states by strong fluorophore–water interactions.

Fluorescent proteins

Genetically encoded fluorescent proteins (FPs) have been widely deployed for long-term visualization and tracking of molecules, cells or structures with high specificity in cells or organisms. Recently, there has been growing interest in the emission tails of NIR-I FPs into the >1,000-nm NIR-II window, thanks to their ability to achieve reduced light scattering, improved imaging depth/resolution and reduced diffused noise57. Bacterial phytochrome photoreceptors (BphPs), cyanobacteriochromes (CBCRs) and allophycocyanins (APCs) have been employed as a source to design NIR-I FPs58. It has been demonstrated that FPs engineered from BphPs (for example, iRFP670, iRFP682, iRFP702, iRFP713 and iRFP720)59 and CBCRs (for example, monomeric miRFP670nano and miRFP718nano)60 exhibit fluorescence emission tails in the NIR-II window. iRFP713 has been knocked into the mouse genome for long-term monitoring of liver regeneration models and imaged at >900 nm (ref. 59). The NIR-II fluorescence of miRFP718nano was three times brighter than miRFP670nano and 1.5- and two times brighter than the fluorescence of miRFP709 and miRFP703, respectively60. The performance of miRFP718nano has been evaluated for liver inflammation models beyond 1,050 nm, using 50 mW cm−2 excitation and a 30-ms exposure time60.

NIR-II imaging modalities

NIR-II 2D wide-field imaging

NIR-II wide-field fluorescence imaging employs an excitation source such as an expanded laser beam11, a light-emitting diode (LED)23, an X-ray beam61 or Cerenkov radiation62 (Fig. 3a) to illuminate an entire 3D object (for example, a mouse), and projects the generated NIR-II fluorescence to a 2D image captured by a camera. Non-coaxial excitation is commonly employed to avoid the use of dichroic mirrors and background signals due to intense reflections from the samples. Other NIR-II wide-field modes requires no excitation source, including chemiluminescence63, bioluminescence64 and afterglow fluorescence imaging38. Wide-field imaging of phantoms or tissues has shown NIR-II imaging penetration depths and resolutions that are ~1.7 (refs. 5,17,32,59,64,65,66) and ~2.1 (refs. 15,17,66,67,68,69) times better than NIR-I imaging, respectively, though these are significantly influenced by the imaging conditions.

Fig. 3: NIR-II imaging modalities.
figure 3

a, Top: schemes of NIR-II wide-field imaging systems for small-animal imaging and image-guided surgery. Bottom: sensitivity range of the different camera materials. b, Left: scheme for the NIR-II confocal microscope. Middle: non-invasive confocal microscopy of blood vessels (red) in the NIR-IIa window and HEVs labelled by aMECA-79-QDc (green, QDc: NIR-IIc PbS QDs) in the NIR-IIc window in an inguinal LN. Right: confocal microscopy of CD169+ macrophages (aCD169-QDa, QDa: NIR-IIa PbS QDs) and CD3+ T cells (aCD3-QDc) in an inguinal LN12. c, Top: schemes of NIR-II light sheet microscopes with normal and oblique configurations and NIR-II structured-illumination LSM. Bottom left: time-course recording of PD-1+ cells (white circles) in a CT26 tumour labelled by anti-PD-1-PbS QDs, at 20 frames per second using oblique LSM24. Bottom right: a higher-resolution NIR-II structured-illumination LSM of aCD4-ErNPs (red) and aOX40-QDb (green) in a CpG-treated tumour26. Credit: panel b (middle, right) reproduced with permission from ref. 12, Springer Nature Ltd. Panel c (bottom left) adapted with permission from ref. 24, Springer Nature Ltd. Panel c (bottom right) reproduced with permission from ref. 26, National Academy of Sciences.

Full size image

NIR-II imaging in various sub-windows requires a suitable camera and optical filters on the imaging side. For NIR-IIa and NIR-IIb fluorescence imaging, a cooled InGaAs camera (900–1,700 nm) with wide dynamic range and low read noise and dark current is used. For NIR-IIc and NIR-IId wide-field imaging, cameras based on photosensitive semiconductors with small bandgaps, such as ‘extended InGaAs’ (900–2,600 nm), indium antimonide (InSb, 960–5,000 nm) and mercury cadmium telluride (HgCdTe or MCT, 800–14,000 nm) are required. Recently, NIR-IIc wide-field imaging was explored utilizing an MCT camera70, but this has higher cost, higher noise and lower sensitivity than the commonly used InGaAs cameras. The NIR-II wide-field imaging acquisition speed has reached 300 fps (frames per second) using a fast InGaAs camera71. The resolution of NIR-II wide-field imaging with a field of view covering the entire mouse is ~100 µm, limited by the small numbers of pixels of available cameras.

For NIR-II imaging-guided surgery, a multispectral system is essential, allowing concurrent visible photographic and NIR-II fluorescence/luminescence imaging under bright surgical-room light conditions (Fig. 3a). To avoid imaging parallax, the colour camera and NIR-II camera could share the same chromatic aberration-corrected lens set28 or use two separate lens sets sharing a portion of the same coaxial optical path72, allowing both cameras to capture the same location from the same angle.

NIR-II 3D confocal microscopy

NIR-II confocal microscopy employs a laser beam tightly focused to a point raster scanned in x–y–z to excite fluorophores, point by point, in a sample. At each point the emitted fluorescence is detected after passing through a pinhole to reject out-of-focus signals, and the signal is used to construct a three-dimensional (3D) image (Fig. 3b). Confocal NIR-II fluorescence imaging increases the tissue penetration depth limit by approximately tenfold compared with visible confocal microscopy (<100 µm in the visible). The penetration depth of confocal microscopy can be optimized by employing both long excitation and emission wavelengths, high-QY fluorophores and detectors with high sensitivity and low noise.

Initially, NIR-II confocal microscopy was realized by using NIR-I excitation, NIR-II emission and an InGaAs photomultiplier tube (PMT) detector21,73. For example, NIR-IIb confocal imaging of QDb in mouse blood vasculature under 785-nm excitation resolved blood vessels at ~700 µm in intact tumours on mice with sub-10-µm resolution21. Confocal imaging of aggregation-induced emission (AIE) dots-filled cerebral blood vessels after craniotomy was performed using 793-nm excitation and >1,000 nm emission, and achieved an 800-µm penetration depth in mouse brain with ~9-µm resolution74. Recently, we exploited a superconducting nanowire single-photon detector (SNSPD) for NIR-II confocal microscopy and found it superior to InGaAs PMTs, with shorter timing jitter, higher sensitivity and lower noise12. A home-built SNSPD with a timing jitter of ~109 ps was employed for NIR-II lifetime imaging using an 800-nm femtosecond laser for excitation75.

NIR-II confocal microscopy with 1,310-nm excitation, QDb probes and an SNSPD enabled the imaging of cerebral blood vessels in vivo at a depth of ~1.7 mm into the hippocampus region after craniotomy76, close to the ~1.6-mm imaging depth achieved by 1,280-nm-excited two-photon microscopy77. The tunable spectrum-response range of SNSPDs presents opportunities for confocal imaging in the NIR-IIc and NIR-IId windows, beyond the detection limit of InGaAs PMTs. To push the penetration depth limit of in vivo non-invasive one-photon imaging, NIR-IIc confocal microscopy with 1,650-nm excitation was demonstrated using QDc and an SNSPD, achieving an imaging depth of ~1.1 mm into an intact mouse head. It also allowed non-invasive through-tissue molecular imaging of mouse inguinal lymph nodes (LNs) with single-cell and single-vessel resolution (Fig. 3b, middle and right)12. This was the first time that both excitation and emitted light in the >1,500-nm regime were utilized for in vivo confocal imaging to minimize light scattering and maximize imaging depth.

The longest excitation wavelength of 1,650 nm for NIR-II one-photon confocal microscopy is close to that (~1,700 nm) used for multiphoton microscopy78, with the excitation light intensity decaying similarly upon travelling through tissues. One-photon fluorescence emission of the excited probes scales linearly with excitation light intensity, whereas two-photon and three-photon fluorescence scale with the second and third power of the excitation, respectively12,79. This suggests a slower emission intensity decay and deeper imaging depth of confocal microscopy than multiphoton microcopy with a similar excitation wavelength. In vivo NIR-IIc confocal microscopy can be conducted non-invasively through intact tissues, in contrast to multiphoton intravital microscopy. Multiphoton imaging is advantageous in terms of its higher SBR79 and the availability of genetically engineered probes. By combining NIR-IIc confocal microscopy with multiphoton microscopy, both of which use ~1,650–1,700-nm excitation, one could maximize the capability of multichannel molecular-specific and cellular-specific imaging to investigate complex biological systems in vivo.

NIR-II 3D light sheet microscopy

Light sheet microscopy (LSM) utilizes orthogonally arranged illumination and wide-field detection to afford high-speed optical sectioning and 3D volumetric imaging. This approach minimizes phototoxicity and improves subcellular resolution80 and enables sub-diffraction-limited resolution by using lattice illumination and adaptive optics81. However, the imaging depth of LSM for in vivo imaging of live tissues in the visible window is shallow (~200 µm for mouse brain after craniotomy24) due to light scattering. Two-photon LSM at 1,040 nm allows for deeper imaging into mouse brain (up to ~300 μm), with high resolution, due to reduced scattering of the NIR-II excitation82. The penetration depth can be further extended by using a Bessel83 or Airy beam84 for excitation, but this is still limited by scattering of the visible emitted light.

An oblique NIR-IIb LSM with ~1,319-nm excitation and ~1,500–1,700-nm detection was developed for in vivo mice imaging with cellular resolution24 (Fig. 3c, middle). This NIR-II LSM avoided the shadows and stripes caused by tissue scattering and absorption, problems common to visible LSM. NIR-IIb LSM enabled non-invasive imaging/sectioning of an intact mouse head with a total penetration depth of ~750 μm, resolving vascular channels connecting the skull and brain cortex of mice. These channels are used by immune cells trafficking between the skull bone marrow and cortex for immune protection of the mouse brain85. In another application, PD-1+ cells migrating irregularly in tumour vasculatures were monitored by NIR-IIb LSM at a frame rate of 20 fps (Fig. 3c, bottom left).

The wavelength of NIR-II light is two to four times longer than that of visible light, and the diffraction-limited spatial resolution (Rayleigh criterion86, 0.61λ/NA) of NIR-II LSM is lower than that of visible LSM. In addition, NIR-II imaging of deep tissues still experiences light scattering, causing an increase in background and reducing spatial resolution. A scanning Airy beam with self-healing or attenuation-compensation properties84 has been employed for excitation87 in NIR-II LSM to improve the SBR, tissue penetration and z-direction resolution.

Structured illumination has also been introduced into NIR-II LSM to improve the spatial resolution by extracting high-frequency details embedded in low-resolution moiré fringes88 imaged under a scanning Gaussian-beam comb pattern with several shifted phases26. The NIR-II structured-illumination LSM can minimize background interference, increase the SBR and increase the spatial resolution by up to two times26 (Fig. 3c, right), and has been utilized for the longitudinal imaging of immune cells in response to immunotherapy in the tumour microenvironment of a mouse model26. The resolution can be further enhanced by using objectives with a higher numerical aperture (NA).

Applications in preclinical imaging

NIR-II imaging has been extensively performed preclinically since 2009 for (1) visualizing blood vasculature structures and measuring haemodynamics and perfusion for cardiovascular diseases; (2) LN imaging; (3) molecular imaging; and (4) functional imaging.

Vascular and haemodynamic imaging

The dynamic NIR-II imaging speed for haemodynamics has increased from the initial ~5 fps using SWNT probes89 to ~90 fps using ErNPs23 more recently. Cardiovascular disease models have been investigated by NIR-II imaging using circulating carbon nanotubes5, QDs37, AIE nanodots90 and gold clusters91, respectively. Dynamic monitoring of blood perfusion and haemodynamics in individual blood vessels for disease models of peripheral arterial disease (PAD)15,89, middle cerebral artery occlusion (MCAO) stroke5 (Fig. 4a) and traumatic brain injury (TBI)92 have been performed. Vascular regeneration has been imaged longitudinally with PbS/CdS QDs in the NIR-IIb window in a mouse model of PAD. Blood flow was also imaged with InAs QDs in disordered vasculatures in glioblastoma multiforme tumour to observe the impact of brain tumour growth on cerebral vasculatures (Fig. 4b)37. The tumour, arterial vessels and venous vessels were identified by dynamic contrast-enhanced imaging through principal component analysis (PCA)6,18,23,25,38.

Fig. 4: NIR-II preclinical imaging.
figure 4

a, NIR-II PCA overlaid images showing arterial and venous vessels of a heathy mouse and a mouse with MCAO5. b, NIR-II imaging of a brain tumour, arteries and veins37. c, NIR-II imaging of sentinel LNs (inguinal LN, iLN) after injection of ICG and Au-PC27. A 1,300-nm longpass filter (1300 LP) was used. d, Two-plex NIR-IIb imaging of a mouse bearing a CT-26 tumour at 24 h post intravenous injection of ErNPs-aPDL1 and PbS QDs-aCD823. e, Time-course recording of vaccine trafficking pathways after injection of pErNP-OVA–CpG B29. f, NIR-IIb imaging of mice bearing E.G-7 tumours with intratumoral injection of pErNP-OVA–CpG B or PBS29. g, Treatment efficacy corresponding to treatment in f29. h, NIR-IIb imaging of sO2. Top: relationship between sO2 value and the ratio of fluorescence signals excited at 650 nm and 980 nm. Bottom: NIR-IIb sO2 imaging of the perfusion of pErNP into a 4T1 tumour 23.54 s post injection (p.i.)105. Credit: panel a (left, right) adapted with permission from ref. 5, Springer Nature Ltd. Panel b reproduced with permission from ref. 37, Springer Nature Ltd. Panel c adapted from ref. 27 under a Creative Commons licence CC BY 4.0. Panel d adapted with permission from ref. 23, Springer Nature Ltd. Panels e,f,g adapted from ref. 29 under a Creative Commons licence CC BY 4.0. Panel h adapted with permission from ref. 105, Springer Nature Ltd.

Full size image

Lymph node imaging

Sentinel LNs are the initial drainage nodes of a primary tumour where cancer metastasis first occurs. Locating the sentinel LNs for biopsy is important for assessing metastatic spread to the LN basin93. NIR-II fluorescence imaging provides accurate localization of LNs and lymphatic vessels, with better contrast and resolution than in the NIR-I window25,94,95. Recently, ‘super-stealth’ Au-phosphorylcholine (Au-PC) nanocluster probes were developed for imaging the draining LNs of cancer tumours after intratumoral administration, with minimal interference from surrounding tissues in vivo (Fig. 4c)27.

Molecular imaging

NIR-II molecular imaging of tumour biomarkers has been pursued with targeted NIR-II probes conjugated with antibodies or other ligands, and is capable of high spatial resolution and high contrast differentiation of tumour from normal tissue, with high tumour-to-normal tissue ratios (T/NT)22,23,24,25,26,27,28,29,32 in the range of 8–20 in the 1,000–1,400-nm spectral range and up to ~40 in the NIR-IIb window23,25,28, higher than the T/NT = 1.1–4.4 in the NIR-I window28,96,97. Recently, ICG conjugated with bevacizumab has been used to target rat orthotopic colorectal cancer and has been imaged by a white-light and NIR-II endoscopy system98.

High endothelial venules (HEVs) in LNs are small postcapillary venules responsible for mediating the entry of immune cells from the blood circulation into LNs99. Recently, using targeted antibody-NIR-II probes, in vivo NIR-IIc confocal microscopy has been used to perform non-invasive through-tissue molecular imaging of HEVs, CD169+ subcapsular sinus macrophages and CD3+ T cells in the inguinal LNs of mice (Fig. 3b)12.

We employed NIR-II molecular imaging to assess the immune responses of mice to immunotherapy. The different fluorescence lifetimes of ErNPs (~4.6 ms, emission ~1,600 nm) and QDb (~46 μs, emission ~1,600 nm) were exploited for in vivo two-plex NIR-IIb molecular imaging of PD-L1 and CD8, revealing the accumulation of CD8+ cytotoxic lymphocytes (CTLs) in the CT26 tumour following treatment by anti-PD-L1 conjugated to ErNPs (Fig. 4d)23. Wide-field imaging and structured-illumination LSM were used for multiplex and multiscale molecular imaging of the CT26 tumour microenvironment in mice26, for longitudinal tracking of CD4, CD8 and OX40 at the single-cell level in response to immunotherapeutic cytosine–phosphate–guanine (CpG) and OX40 antibody treatment by intratumoral injection.

Recently, a cancer nanovaccine was developed by conjugating ovalbumin (OVA) covalently and class-B CpG (CpG B) electrostatically to pErNP29. Upon subcutaneous injection, NIR-IIb imaging revealed trafficking of the nanovaccine, rapidly migrating to inguinal LNs (iLN) through the lymphatic vessels (Fig. 4e). Two doses of vaccination led to tumour eradication and cure/survival of mice. Wide-field imaging and structured-illumination LSM revealed abundant OVA antigen-specific CD8+ CTLs recruited to the tumour in the treated mouse (Fig. 4f). This was the first time that in vivo imaging of antigen-specific CTLs was performed to correlate with the immunotherapeutic effects of cancer vaccines (Fig. 4g).

NIR-II functional imaging

Functional imaging to probe the environmental parameters and cellular responses to a stimulus is another exciting direction for in vivo NIR-II imaging, with examples including NIR-II fluorescent molecules responding to an external stimulus or environment, such as pH65, redox species100, nitroreductase101, Aβ plaques102 and cell endocytosis103. A unimolecular NIR-II chemiluminescence probe for H2S was constructed by conjugating Schaap’s dioxetane with a donor–acceptor core104. A more recent advancement is NIR-IIb imaging of oxyhaemoglobin saturation (sO2) in blood vessels, based on the absorption difference between oxyhaemoglobin and deoxyhaemoglobin at specific excitation wavelengths (650, 808 and 980 nm) of pErNPs, enabling visualization of the sO2 levels in tumour-associated vessels (Fig. 4h)105. Atomically precise NIR-II Au22 clusters with strong NIR-II fluorescence exhibit potent enzyme-mimetic activities, which is promising for early intervention regarding oxidative stress43.

NIR-II imaging-guided surgery

Preclinical NIR-II imaging for intraoperative navigation is an active area of research with potential clinical translations. Surgical removal of tumours (for example, glioblastoma18, pancreatic tumour106, colorectal tumour20, ovarian tumour107 and breast tumour94) navigated by NIR-II imaging holds great promise. NIR-IIb molecular imaging of tumours using ErNP-TRC105 targeting tumour vasculature angiogenesis has afforded a tumour-to-muscle signal ratio of up to ~300, allowing high-precision image-guided tumour resection down to the few-cell level28. A recent work has shown successful surgical removal of LNs labelled with QDb, achieving high LN-to-muscle ratios of ~200 (ref. 94). In another work, NIR-II imaging-guided surgery led to complete resection of severe inflammatory bowels and ensured a secure surgical anastomosis by using AIE nanoprobes108.

Towards clinical imaging

For any successful clinical translation of NIR-II fluorescence imaging, it is imperative to develop contrast agents that are safe for use in humans. Several groups found that traditional NIR-I organic dyes such as ICG and IRDye800CW exhibited emission tails into the NIR-II window34,109 and can be utilized for NIR-II imaging to benefit from the reduced light scattering and high imaging contrast and resolution. Because ICG is a Food and Drug Administration (FDA)-approved fluorophore, clinical trials of NIR-II imaging with ICG in human patients is of relatively low risk, but requires switching to an InGaAs-based imaging system, which has not gone through rigorous regulatory approval. Along this line, NIR-II fluorescence-guided surgical resection of liver tumours in human patients was successfully performed after intravenous injection of ICG at a dose of 0.5 mg kg−1, demonstrating a higher tumour detection sensitivity and rate than imaging in the NIR-I region (Fig. 5a)96. However, the ICG is a non-targeted probe, and thus gives false positives of tumours as it accumulates in other tissues96. Active tumour targeting for imaging-guided surgery has clinically tested bioconjugates of IRDye800CW in the NIR-I window34. IRDye800CW conjugates, exhibiting a tail emission beyond 1,000 nm, have potential for the better determination of tumour margins28.

Fig. 5: NIR-II clinical imaging.
figure 5

a, Top: schematic showing example equipment for visible, NIR-I and NIR-II image-guided surgery for clinical tumour resection. Bottom: intraoperative visible, NIR-I and NIR-II imaging of a tumour before and during tumour resection96. The tumour-to-normal-tissue ratio (TNR) of NIR-II imaging was higher than that of NIR-I imaging after the first resection, and the TNR of NIR-I and NIR-II imaging became the same after the second resection. b, Preoperative NIR-I (top) and NIR-II (bottom) fluorescence imaging, revealing four possible perforators after intravenous injection of ICG. NIR-II imaging resolved more perforators than the traditional Doppler method. One of them overlapped with the Doppler location110. c, Colour image of a human heel after flap transplantation (left), and NIR-I (middle) and NIR-II (right) fluorescence imaging of flap perfusion and revascularization. NIR-II imaging enhances the visualization of revascularization compared to NIR-I imaging110. Credit: panel a adapted with permission from ref. 96, Springer Nature Ltd. Panels b and c adapted from ref. 110 under a Creative Commons licence CC BY 4.0.

Full size image

Another promising direction for clinical translation is NIR-II imaging of perfusion. NIR-II imaging of ICG-tagged blood has been used to observe anastomotic vessels and salvaged distal limbs110. It has allowed the observation of skin perforator vessels at the deep fascial level (Fig. 5b) and revascularization (Fig. 5c) before and after flap transplantation, respectively, with higher contrast, better resolution and a longer duration of observation than with NIR-I imaging110.

Clinical trials of SWIR imaging using a label-free approach by exploiting the absorption properties of water have been reported111. One example is the otoscope, which uses the negative contrast of the water absorption band at 1,480 nm to detect fluid in the middle ear111.

Outlook and future directions

Probes and fluorophores

Currently, organic NIR-II fluorophores with high absorptivity, QY and aqueous solubility, as well as the ability to be excreted renally and conjugated to target ligands, are still rare. Molecules emitting predominantly in the NIR-IIb sub-window are also desired to compete with nanoprobes based on inorganic QDs and rare-earth nanoparticles. Another major challenge is the synthesis of functional NIR-II fluorophores with optical properties sensitive to the environment and stimuli, especially for imaging-based sensing of pH, gas molecules, Ca2+ and other ions, and voltages and action potentials across the ion channels of neurons.

The wide emission spectra of NIR-II fluorophores (full-width at half-maximum of ~75–290 nm) have limited multiplexed imaging18,19,20,21,27,36,37,42,112. Multiplexed molecular imaging can be expanded by employing multicolour probes with different narrowband emissions, probes with different excitation wavelengths, and probes with different fluorescence/luminescence lifetimes, all conjugated to molecular-specific ligands to target different molecules in a body. For inorganic-based nanoparticles, it is desirable to have narrow emission widths across the 1,000–2,300-nm range, with little spectral overlapping. Currently, aqueous soluble, biocompatible NIR-IId probes (emission peak at ~2,200 nm) are lacking. Developing NIR-II probes with tunable fluorescence lifetime is another important approach to increase multiplexed molecular imaging23 in vivo and should be pursued further. Thus far, three-plex NIR-IIb imaging has been realized by combining continuous-wave and lifetime imaging using QDs and rare-earth nanoparticles29.

Developing genetically engineered NIR-II fluorescent proteins has been a daunting challenge so far, but the success of this would mirror the green fluorescent protein (GFP) revolution, and undoubtedly boost the NIR-II field and lead to much broader adoption of this imaging modality by biologists and medical scientists. NIR-II fluorescent proteins exist in nature, and several purple photosynthetic bacteria, including Blastochloris tepida, Blastochloris viridis and Halorhodospira halochloris, possess bacteriochlorophyll b-based light-harvesting complexes exhibiting absorption and fluorescence in the NIR-II range113. Among these, the light-harvesting 1–reaction centre (LH1–RC) complex from Blastochloris viridis has been observed to emit fluorescence with a peak in the NIR-II window113, opening up opportunities for the development of NIR-II fluorescent proteins, but these very large protein complexes are still very difficult to use in genetic labelling strategies in mammalian cells114.

NIR-II imaging devices and methods

New camera technologies with high sensitivity, low noise, a broad spectral range spanning 1,000–2,300 nm and greater pixel numbers are important to enhance NIR-II imaging performance and capability. High-quantum-efficiency image intensifiers in the NIR-II range are needed for time-resolved/ultrafast imaging and for the detection and imaging of weak fluorescence. Better cameras for NIR-IIc and NIR-IId imaging beyond InGaAs are required to optimize the benefit of in vivo fluorescence imaging in 2D wide-field and 3D LSM modes. Point detectors such as SNSPDs have enabled high-resolution, deep-tissue confocal microscopy in the NIR-IIc sub-window, but remain a challenge for the ~2,200-nm NIR-IId range, with low dark noise.

To push the resolution limit, it is desirable to introduce optical super-resolution methods to microscopic imaging in the NIR-II window, similar to the approaches developed for the visible range, such as nonlinear structured illumination microscopy (SIM), stochastic optical reconstruction microscopy (STORM), photoactivated localization microscopy (PALM) and stimulated emission depletion (STED) microscopy. To realize these, specially designed NIR-II fluorescent probes and low-light-sensitive detectors are required.

An interesting and exciting direction is to use deep learning and artificial intelligence (AI) to enhance NIR-II fluorescence imaging. Recently, the cycle generative adversarial network (CycleGAN) was used to transform a blurred in vivo NIR-I or NIR-IIa image into a much higher-clarity image resembling a NIR-IIb image115. Training with experimental data in a higher sub-window (for example, NIR-IIc) could be used for machine learning, and then applied to transform and improve images acquired in the lower sub-window (for example, NIR-IIb). AI approaches could address the problems of a scarcity of probes115 and the affordability of high-end expensive cameras in the higher sub-windows, enabling noise reduction and sensitivity enhancement.

Clinical translation

Preclinical in vivo NIR-II fluorescence imaging has produced a large body of promising results for potential clinical translation. However, a major hurdle is the lack of clinically proven high-performance NIR-II fluorophores or nanoprobes that are safe and have favourable pharmacokinetics for human use. Although the FDA-approved ICG has a high safety track record and exhibits tail fluorescence into the NIR-II window, the emission is mostly in the <1,200-nm range, and imaging still suffers from substantial light scattering and high background. ICG also lacks the functional groups required for conjugation to target ligands and cannot be used for molecular imaging. Alternative dyes or probes are needed that have a safety profile similar to that of ICG, with longer wavelength emission ideally in the NIR-IIb sub-window.

Among the inorganic probes, rare-earth downconversion nanoparticles are bright emitters for the high-performing (low scattering, low autofluorescence) NIR-IIb imaging window, and have afforded excellent molecular imaging agents. Similar-composition upconversion nanoparticles have proven highly safe in mice, preclinically. However, clinical translation is uncertain owing to probe scaling-up issues and the lack of safety data from clinical settings. QDs are even more challenging because of the toxic elements used. Another highly promising NIR-II probe comprises molecular gold clusters such as Au25GSH and Au25PC, as Au is widely accepted to be a safe element, the clusters are rapidly excreted renally, exhibiting little non-specific tissue binding/uptake, and have shown higher performance in NIR-II LN imaging compared to ICG27. Regardless, the clinical translation of any NIR-II agent must undergo rigorous phase I to III clinical trials for pharmacokinetics, toxicity, stability, side effects and risks to humans, and proof of benefits116,117.

Standardization of NIR-II imaging systems is another key step towards clinical translation. A set of characteristics for image devices for clinical use have been suggested to meet the requirements of the operating-room environment and clinical workflow118. Although originally intended for the evaluation of NIR-I fluorescence-guided surgery systems, these criteria can provide a guide for future clinical NIR-II imaging devices, including (1) the overlay of white-light and fluorescence images in real time, (2) operation within surgical lighting, (3) high sensitivity, (4) in situ quantitative capabilities, (5) concurrent multiplex fluorescence imaging and (6) maximized ergonomic utility for surgery118. The standardization of NIR-II imagers and contrast agents will accelerate regulatory approval, optimize device development, guarantee product quality, standardize clinical trials and reduce risk116,117.

The NIR-II imaging has enabled tumour resection down to the few-cell level with zero background28. Under this resolution, manual resection/surgical operation by hand could be challenging. We envisage that the combination of NIR-II imaging and surgical robots could become a powerful tool for precision medicine.