The role of tissue fluorescence in in vivo optical bioimaging

The advances in modern medicine would not have been possible without the amount of bioimaging techniques present in the clinic and the vast range of diagnostic capabilities they offer. Next to the routine analysis of blood samples, there is probably nothing more common with a visit to a physician than having an x ray or ultrasound ordered. In more intricate cases, modern medicine in its most complex technological (imaging) forms like magnetic resonance imaging (MRI) or positron emission tomography (PET) comes into play.

Biomedical imaging began with the advent of the microscope. At the beginning of the 20th century, fluorescence microscopy added drastically to its usability enabling the study of complex processes in cells and tissues.^1–3 The optical imaging techniques were overtaken by the more complex and/or more expensive radiation-based (x ray, PET, etc.) or radiowave-based techniques (ultrasound, MRI) in the second half of that century. The main reason for this is the opacity of most biological tissues in the visible part of the electromagnetic spectrum, limiting most optical methods to cell microscopy or topical applications in animal bioimaging. Additionally, it was discovered that animal tissues already contained fluorophores that could emit under typical conditions in fluorescence imaging.² The term autofluorescence is used to distinguish this intrinsic fluorescence of cells and tissues from the fluorescence obtained by treating specimens with exogenous fluorescent markers that bind to cell and tissue structures or act as contrast agents. The role of the former, in turn, will be the topic of the present Tutorial. Sections II and III will present the phenomenon of tissue fluorescence and its underlying reasons and Sec. IV will describe imaging techniques based on autofluorescence. In Secs. V–VI, practical aspects and experimental designs will be discussed and techniques on the overcoming of tissue fluorescence in bioimaging will be presented, and finally in Sec. VII, conclusions and perspectives will be offered.

Imaging in the visible region (400–680 nm) of the electromagnetic spectrum is an easy way for human eyes to observe phenomena in biomedical investigations.^4–6 However, the phenomenon of autofluorescence in the visible spectral region, known since the first studies of cells and biological specimens conducted in fluorescence microscopy,² was readily recognized as a drawback of this wavelength range for imaging applications. After all, the reduction of the signal-to-noise ratio significantly affects the quality of the images. When excited with radiation of a suitable wavelength, some cell and tissue components behave as endogenous fluorophores, decaying to the ground state under fluorescence emission. This intrinsic property of all forms of life is due to the fact that most of their building blocks contain various endogenous organic molecules that also behave as fluorophores under the right conditions (e.g., when excited with strong microscope lamps, UV lamps, and/or lasers). The emitted autofluorescence depends on the concentration and spatial distribution of these fluorophores as well as the properties of the surrounding tissue including homogeneity and anisotropy. Consequently, autofluorescence spectra would enclose information about the content and molecular structure of the emitting tissue. Variations in the health status of the tissue could then, in principle, be evidenced by autofluorescence measurements and verified by histological or histochemical studies.

The most important endogenous fluorophores are molecules widely distributed in most tissues, starting from the pyridine nucleotides in DNA and RNA, over aromatic amino acids like tryptophan, going to energy transporters like NAD(P)H, flavins, and structural proteins such as collagen and elastin. Importantly, some characteristic molecules are present in different tissues (e.g., lipopigments) ultimately endowing them with functional identity. In animal cells, for instance, flavins and pyridine nucleotides are the main emitters of autofluorescence around 500 nm under ultraviolet (UV) excitation.⁷ A series of cellular proteins is excited by UV light and emits in the UV to blue, because of the presence of tryptophan, tyrosine, and histidine.⁸ Additionally, tissues present more intense visible autofluorescence compared with isolated cells due to the aggregation of abundant collagen and elastin, possessing higher quantum yields, and adding to its autofluorescence.⁹ Melanin in the skin absorbs and emits in a wide range from the visible into the near-infrared (NIR) region.¹⁰ Finally, the brain is enriched in lipids and proteins such as myelin and lipofuscin, the latter being one of the major fluorophores responsible for the brain autofluorescence emitted in the infrared spectral range.¹¹ Over the years, more and more endogenous fluorophores have been identified and their excitation and emission properties have been studied and documented. For a good overview, see the review by Richards-Kortum and Sevick-Muraca.¹² Most of these compounds emit in the blue and only a few in the green or above, allowing the majority of fluorescent dyes in cellular experiments to work without problems in the green or red.

Depending on the properties of the light beam (e.g., wavelength, pulse duration, and power density) the interaction of light and tissue can result in the modification of the incident excitation beam (while the tissue properties remain unaltered) or in the modification of the tissue (e.g., laser ablation or thermal therapy). Considering the purpose of this Tutorial, we will focus on the first case. We will describe the basic principles of the phenomena that occur when light encounters a tissue: reflection, refraction, absorption, and scattering.

Reflection (the returning of the light to the incidence medium) and refraction (change of direction of the light that is transmitted into the second medium) are strongly related phenomena that occur when light hits the interphase of two media with different refractive indices. When the irregularities of the interphase are smaller than the wavelength of the light specular reflection takes place, meaning that the angle that forms the incident beam with the normal to the separation surface is the same as the angle of the reflected beam with the normal. This situation is very rare when dealing with biological tissues so, in general, diffuse reflection is observed, meaning that the reflected light does not possess a preferential direction. The main interaction between light and tissues is experienced by the refracted beam, which will be attenuated due to the existence of absorption and scattering phenomena. The absorbed light can be reemitted by certain tissue compounds as fluorescence as mentioned above.^12,13

There are two strategies to describe the propagation of the light in the tissue: analytical and transport theory. The analytical approach consists in solving Maxwell's equations, which is usually unpractical due to the mathematical challenge of obtaining exact analytical solutions when introducing the complexities of biological tissue. However, the electromagnetic theory is useful to introduce the processes occurring when light is traveling across a tissue, by analyzing the energy conservation principle, which is summarized by the Poynting theorem for electromagnetic waves,¹⁴ 

Here, W represents the energy density, P_abs represents the absorbed power, and S represents the Poynting vector, which is related to the flux of energy to or from the medium, and it is related to the intensity (the magnitude that is measured in the experiments). From this conservation law, the radiative transfer equation (RTE) can be deduced. RTE in differential form has the following expression:

where c is the speed of light, j_ν is the emission coefficient, k_ν,s is the scattering opacity, k_ν,a is the absorption opacity, I_ν is the spectral radiance at a frequency ν, and Ω is the solid angle that the radiation is crossing. The last term represents radiation scattered from other directions onto a surface. In summary, the meaning of the RTE equation is that as a beam of radiation travels, its energy is lost due to absorption, emission, and redistributed by scattering.

Solving this equation is the objective of the transport theory, which is based on introducing certain assumptions on the behavior of the absorbed and scattered radiation, in particular, in which one is the dominant process. These methods are known as first-order scattering, Kubelka–Munk theory, diffusion approximation, Monte Carlo simulations, or inverse adding–doubling.¹⁵ We will illustrate this by discussing the behavior of absorption, scattering, and Monte Carlo approaches in more detail.

The most important hindrance for in vivo bioimaging in the visible when moving away from microscopy is not tissue fluorescence but absorption and scattering effects, affecting both the excitation light and the emitted fluorescence inside the tissues [see Figs. 1(a)–1(c) for a schematic representation of all three effects].^12,13 To introduce these phenomena, the Beer–Lambert law (BLL) can be used for didactic purposes. In its common form,¹⁶ BLL assumes that the attenuation of the detected light is completely explained by the absorption effects. Hence, if a beam of light passes through a tissue, the log of the ratio between the transmitted intensity I and the initial intensity of light I₀ is supposed to be equal to the additive inverse of the product of the tissue's absorption coefficient μ_a with the total path length L of light (which in a first approximation coincides with the thickness of the tissue). This product, in turn, is defined as the absorbance, A, of the tissue. However, since the absorption coefficient μ_a itself is the sum of the product between the molar extinction coefficients ε_i of the ith chromophore in the tissue with their respective concentrations [C_i], this relation can be summarized by the following equation:

Nevertheless, Eq. (3) is an approximation that is only valid when the tissue and the distribution of absorbing chromophores inside it are assumed to be homogeneous and scattering is considered to be negligible. The latter condition is hardly true in tissues. After all, the presence of scattering is determined by changes in the refractive indices due to media variation and inhomogeneities in tissues. It is expected to occur at borders, e.g., between cellular and extracellular spaces and other divisions inside the tissue due to structural or density reasons. Therefore to consider scattering effects, one needs to take into account a modified version of Eq. (3) known as the modified Beer–Lambert law (MBL). As opposed to the simplified rendering,¹⁷ it (1) includes a function f that accounts for the scattering effects and (2) multiplies L by a parameter called differential path length factor (DPF). The DPF accounts for increases in the optical path due to scattering, the source–detector configuration, and the geometry of the medium.¹⁸ The resulting MBL is mathematically represented by

where $μs$ is the scattering coefficient of the tissue and V is a suitable parameter accounting for the volumetric region defined by the tissue. As one can readily observe from Eq. (4), both the excitation and tissue fluorescence will be attenuated by the presence of the absorption and scattering effects. At this point, it is worth pointing out that even though the descriptions given by diffusion theory or experimentally fed Monte Carlo simulation present a higher level of accuracy when compared to Eq. (4), the Beer–Lambert models present an easy approach to recognize the dependence of the detected light on internally occurring effects.¹⁸ 

The Monte Carlo (MC) method deals with experiments on random numbers. It is applied to situations where a thorough analytical description is either lacking or too unmanageable to yield a solution. This includes problems directly concerned with the performance and aftermath of inherently random processes and problems where a causal link between events exists but the analytical treatment does not lead to tractable numerical solutions. Light propagation through a scattering medium (such as biological tissues) is considered to be probabilistic in nature.¹⁹ First, because the scattering of individual photons is governed by the laws of quantum mechanics. Second, because the nature of the medium in which they propagate is also random (i.e., there is a large variety in the properties of the scattering centers). The underlying assumption in MC is that one is dealing with a sequence of random non-correlated events.²⁰ For the case of light transport in biological tissue, this would imply that the probability of a photon changing from a state to another is independent of its previous states. In other words, it has no knowledge of its own history. In a direct MC simulation of the transport of light in a biological tissue, the method generally follows the following sequence of steps:

1. Set a starting point for a photon. This point does not need to be random as it can be determined from the known distribution of sources.

2. Trace its history as it propagates in the medium until a terminating condition is fulfilled. To do this, random samples are taken from the probability distributions (derived from physics laws or measurements), which govern the various interactions the photon undergoes.

3. Extract the desired parameters from the photon history, i.e., the exit point, the total optical path length, and the deposited dose.

4. Repeat steps (1) to (3) until sufficient statistics are achieved. The required number of photon histories depends on the extracted parameters, the properties of the scattering medium, and the desired accuracy of the simulation.²¹ Depending on the problem being studied, there might be techniques to reduce the number of photons required.

In the study of autofluorescence, the MC method has been successfully employed in the explanation of the autofluorescence decay dynamics of several tissues,^22,23 in the reconstruction of the fluorophore distribution and intrinsic fluorescence spectra of skin structures.²⁴ This in turn could yield valuable information on the composition of the tissues (for instance, the content of melanin).²⁵ MC's proper application therefore results beneficial for the study of autofluorescence in the biomedical context.

Based on these considerations on light-tissue interactions and empirical data, specific regions for optical in vivo imaging where attenuation and scattering are minimized were identified within the near-infrared region of the electromagnetic spectrum.²⁶ In 2001, Weissleder coined the term NIR biological windows (BWs) for them.²⁷ Traditionally, three NIR transparency windows have been defined: the first ranges from 680 to 950 nm, the second from 1000 to 1350 nm (NIR-II), and the third from 1450 to 1950 nm [Fig. 1(d)].^28,29 The 3rd BW is also known as NIR-IIb.³⁰ Recently, some researchers reported a new transparent NIR BW for in vivo imaging, located at 2100–2300 nm, it is known as NIR-IV BW.¹⁸ Compared with visible light imaging, where blood is one of the main absorbers in tissue due to hemoglobin, whether in the oxygenated or deoxygenated state [Fig. 1(e)],³¹ in vivo bioimaging in the NIR BWs presents three obvious advantages. First, tissue autofluorescence is much lower. The autofluorescence of skin, fat, and blood (oxygenated and deoxygenated) is significantly lower in the NIR BWs than that in visible light range, which, in turn, notably increased the quality of in vivo imaging.²⁶ Second, the increased penetration depth (>1 cm) makes deep-tissue investigations possible. Third, the significant decrease of photon scattering in the NIR is beneficial for the light propagation within the tissue [Fig. 1(d)] as it allows improved spatial resolution.^28,32 Though the NIR BWs can avoid the majority of tissue-induced intrinsic problems in fluorescence imaging, autofluorescence still can present a certain nuisance even in these regions.

The above-described concept of the near-infrared transparency windows together with a vast technological improvement in InGaAs CCDs brought the ideas for optical whole-body in vivo imaging and clinical optical tomography techniques a big step closer to their applicability. Additionally, advances in the development of fluorophores with emission in the NIR made it possible to begin proving the concepts for NIR infrared fluorescence imaging.^3,28,29,33 But at this point, researchers in the infrared imaging field discovered that, depending on the intended application, autofluorescence in the NIR was not negligible, contradictory to what had been assumed based on the tissue fluorescence observed in the visible.^34–36 Up to that point, very few mentions of tissue fluorescence in the infrared or NIR emissions of endogenous fluorophores had appeared in the literature. Generally, this has not changed a lot and so far, only a few possible sources of tissue fluorescence in the infrared have been identified unequivocally or at a molecular level.^37,38 Additionally, as recently demonstrated, biological tissues can significantly distort the shape of the spectra of NIR-emitting infrared luminescent probes.³⁹ This, in turn, could lead to erroneous conclusions concerning the temperature, health status, or composition of the tissue. Such a problem is aggravated by the thermal dependence of light absorption and scattering processes in tissues and is not avoided even within the biological windows. So although one could say that, relative to the whole electromagnetic spectrum, autofluorescence, scattering, and absorption are all minimized in the NIR, they still play an important role in this particular wavelength range.

The primary source of observable fluorescence in animal infrared imaging is a consequence of their diet, even though the special feedstuff for laboratory mice or rats is often labeled as autofluorescence free. Figure 2 shows infrared fluorescence images of different types of laboratory animal feedstuff and in Fig. 2(e) their spectra obtained via hyperspectral imaging (the emitted light is separated into different images corresponding to distinct wavelengths). All five imaged food pellets present a declining infrared fluorescence signal that reaches into the NIR-II up to 1200 nm under excitation at 808 nm with a power density of 50 mW/cm² (15 s integration time, 5 nm spectral resolution), while the feedstuff rich in fat also presents a sharp peak at 1050 nm. The origin of this sharp band is unknown, but the broad signal is most likely due to chlorophyll molecules stemming from the principle plant ingredients (e.g., alfalfa) in the food pellets and the feedstuff’s fluorescence has been observed from the food pellet through the gastrointestinal tract and in the feces of the animals.^40,41 Alfalfa contains a high amount of chlorophylls and both chlorophylls a and b have been identified to present fluorescence emissions in the NIR-I,^42,43 although the relevant emissions stem from chlorophyll a due to an energy transfer from b to a.⁴⁴ The first fluorescence spectra centered the principal emission band of chloroplasts at 682 nm and a shoulder at 740 that extends further into the IR [see Fig. 2(f)] under visible excitation,⁴⁵ which was later confirmed by Ustin et al. to be chlorophyll a.⁴⁶ These autofluorescence emissions in plants are employed to study their activity among other things.^47–49 Krasnovsky had already been studying phosphorescence of chlorophylls under red to NIR light excitation since the 1980s showing strong NIR emissions around 950–995 nm and a shoulder around 1100 nm.⁵⁰ 

A more recent study by Hartzler et al. compared the fluorescence emission of chlorophylls with its phosphorescence,⁵¹ demonstrating that the tail of the NIR-I fluorescence emission is more relevant than the orders of magnitude smaller phosphorescence [see Fig. 2(f)]. Weagle et al. associated the chlorophyll-rich alfalfa with the observed autofluorescence in animals,⁵² and König et al. showed that the autofluorescence in nude mice stemmed from chlorophyll a.⁵³ Later, the group of Weagle developed chlorophyll-free feedstuff, which surpassed common autofluorescence-free food,⁵⁴ but the improving IR detectors and experiments by other groups showed a remaining signal and hence that autofluorescence-free food was not completely free in the NIR.⁴¹ Villa et al. visualized the NIR-fluorescence up to 1200 nm of food pellets with an InGaAs camera, and del Rosal et al. measured the emission spectrum for a regular food pellet, confirming emissions into the NIR-II when exciting at 808 nm in the NIR-I.⁵⁵ Although technically the fluorescence of chlorophylls is not endogenous fluorescence of the animal, they and products of their metabolism end up in animal tissues.⁵² In practice, diet-based autofluorescence is one of the strongest signals, competing easily with a range of NIR fluorophores or contrast agents.

Hair color and skin pigmentation of an animal also play a big role in infrared tissue fluorescence. This was very effectively illustrated by del Rosal et al. in their study of various mice strains (see Fig. 3).³⁷ The skin's tissue fluorescence can basically be attributed to the main skin pigment: melanin. Melanin is a copolymer based on indole subunits that are obtained in pigment cells through biochemical polymerization reactions of the amino acid tyrosine. Due to the large size of the polymer, its aromaticity, and its lipophilicity, the separation from peptides and fatty tissues is difficult. Therefore, its exact structure is unknown. Nevertheless, three main melanins are differentiated:^56–58 (1) eumelanin exists in brown and black variants and is the most common, (2) pheomelanin has sulfur-containing heteroaromatics derived from incorporation of the amino acid cysteine, resulting in a red pigment, and (3) neuromelanin, the least common is present in some human neurons while most other species seem to lack it in their brain tissue.^59,60 Recently, it was also observed in murine brains.⁶¹ Melanin autofluorescence was first investigated in the 1990s but, due to the lack of IR-detectors, only weak autofluorescence in the visible was observed.⁶² The first works linking NIR tissue fluorescence to melanin were published in 2006.^63,64 Its NIR emission was observed under 785 nm excitation [Fig. 3(c)]. Additional studies demonstrated the in vivo fluorescence in the NIR-I, especially in the studies of moles.^65,66 The demonstration of NIR-II autofluorescence of pigments was shown in the already mentioned work of del Rosal et al. Not only do darker colored hair and tissues absorb more light and therefore heat up faster (reducing the safely employable excitation powers), but they also demonstrate a drastical increment in their autofluorescence in comparison with white hair or hairless skin [Fig. 3(b)]. According to the spectral analysis provided in Ph.D. thesis of del Rosal, this could be correlated with the melanin content.³⁷ As a consequence, together with observations of autofluorescence that Villa et al. had made when analyzing the supposedly autofluorescence-free feedstuff,⁴¹ it is necessary to rely on fluorophores with emissions in the NIR-II as agents in in vivo imaging when relying on a wavelength-based approach. It is important to note that this problem can be minimized through the use of specific animal strains that are hairless or transparent as described in more detail in Sec. V.

Pigmentation is not just related to melanin but also to aging. The accumulation of intracellular autofluorescent material or “aging pigments” has been observed in several cell types over their lifetime. Interestingly, in some diseases autofluorescent material is also stored as pigment in cells. But the aging pigments present distinct characteristics from the ones associated with diseases. Lipofuscin is the commonly employed term for aging pigments,⁶⁷ while ceroid is employed for autofluorescence storage material related to pathologies, for example, in the neuronal ceroid lipofuscinoses (NCLs),⁶⁸ a family of neurodegenerative diseases that accumulate autofluorescent material (ceroid) in the lysosome. Lipofuscin shows an autofluorescence band from 600 to 800 nm reaching into the NIR-I under excitation by a 633 nm laser,⁶⁷ while NCLs have been reported to present a visible emission band between 480 and 670 nm.^68,69

Interestingly, these naturally occurring pigments could potentially be used as accurate in vivo biomarkers to ascertain ageing and neurodegenerative derived damage as described by Seehafer and Pearce in their review on the topic.¹¹ 

A third group of endogenous fluorophores in the NIR was recently identified by Htun et al. when they studied the autofluorescence of intra-arterial hemorrhages (see Fig. 4).⁷⁰ They showed that NIR tissue fluorescence can be employed to track and diagnose unstable atherosclerotic plaques that present a high risk of complications such as embolisms, cerebral, and myocardial infarcts. They proposed that the main molecular agents responsible for the NIR emission are the products of blood degradation, especially bilirubin as key metabolite of the heme groups in hemoglobin. This observation was based on NIR imaging of plaques and isolated protoporphyrin IX (PPIX), hemoglobin, biliverdin, and bilirubin in a commercial imaging system, exciting at two wavelengths and observing in the NIR-I [Fig. 4(a)]. This observation can actually be corroborated with the early observations made in the field of in vivo NIR autofluorescence by the group of Weagle and Pottier,^52,54 when they identified the influence of chlorophylls as described above. Pottier et al. also realized the role of protoporphyrin IX, the common precursor of chlorophylls and hemes, in skin fluorescence and investigated its role through experiments, employing the starting point of its biosynthesis 5-aminolevulinic acid and dosage with protoporphyrin IX in vivo [the spectrum is shown in Fig. 4(b)].⁷¹ 

The biosynthetic pathway of chlorophylls and hemes is highlighted in Fig. 5, showing their molecular structure with extended systems of conjugated double bonds and aromatic groups (shown in red). These functional groups form the chromophores that are responsible for the color of these molecules and also explain their potential as NIR fluorophores. Of the molecules in Fig. 5 chlorophylls have been identified as NIR-I and NIR-II fluorophores (see Fig. 3) and protoporphyrin IX and bilirubin as NIR-I fluorophores (see Fig. 4). This does not mean that these or the other molecules do not show autofluorescence in the NIR-II. However, there are no data published for them which includes measurements or specific NIR-imaging above 1000 nm. The importance of the porphyrin-based molecular structures for autofluorescence is well in line with the liver being the organ that shows the most tissue fluorescence in in vivo and ex vivo NIR imaging, as the principal organ that is responsible for hemoglobin degradation and its metabolism.

The use of native tissue fluorescence (autofluorescence) has been proven a useful non-invasive diagnostic tool, causing minimal tissue disturbance due to the relatively low intensity dose required for the excitation light. Once the community became aware of the fact that the presence of endogenous fluorescence signals could be used as potential sensors or indicators of changes in the state of tissues/organs, a great number of techniques were developed to take advantage of autofluorescence. Some of the methods were sensitive enough to discriminate the alterations suffered by cells or tissues in several contexts.⁷² In the visible domain, this potential was explored in the study of the NADP/H pair in cancer diagnostics and in the analysis of the metabolic activity of cells or tissues.^73–75 In the NIR, protoporphyrin IX stands out in its use as it has been shown to accumulate in liver, spleen, blood, and even in cancer cells.^76–79 

Given the well-reported connection of autofluorescence with many features of important biological tissues, its potential has been thoroughly explored over the last years. Table I separates some of the most relevant works in the field into the following categories: application, organ/tissue studied, and optical imaging technique utilized.

As one can see from Table I, several techniques based on autofluorescence signals obtained from different tissues in healthy and pathological conditions have shown to be effective in order to discriminate and track tissue changes over time. These changes, in turn, could be tied to alterations in the relative proportion of specific endogenous fluorophores and, consequently, the presence of incipient diseases. So far, most of the works typically explored the UV, the visible, and the NIR-I wavelength ranges. Specifically, the last two applications of the table are beginning to be used as in vivo NIR autofluorescence imaging technique that can visualize the studied organs (eye and parathyroid glands) in humans.^104–111 The potential of non-invasive in vivo imaging in the NIR-II or NIR-III remains sparsely explored. For a good overview, see the examples presented by del Rosal et al. in their recent chapter.³³ Here, we will focus specifically on the tissue fluorescence generated by the liver and the brain under NIR-I excitation, which demonstrate great potential for further investigation.

Recent research has demonstrated that the liver possesses significant autofluorescence in the NIR-II when excited with an 800 nm laser,^32,61 while to our knowledge the potential of NIR-II intravital liver spectroscopy has not been explored so far. Figure 6(a) provides an illustrative example of the ex vivo NIR-II autofluorescence presented by selected abdominal organs from a C57BL/6J mouse, including the kidneys, spleen, and liver. As it can be observed, there is an obvious difference in relative intensities, the liver being the brightest organ under 808 nm excitation (50 mW/cm² power density). Furthermore, a subtle but significant peak can be resolved around 1015 nm for the liver, employing a hyperspectral NIR imaging system. This peak was not present in the spleen or the kidneys nor observed in previous works that described for the first time liver NIR-II autofluorescence.³² The identification of the endogenous fluorophores that contribute to the 1015 nm peak in the liver's autofluorescence spectrum could thus potentially be translated into a biomarker through NIR-II in vivo imaging.

The brain possesses a unique autofluorescence spectral fingerprint when excited with an 808 nm light source [Fig. 7(a) bottom and Fig. 7(b) black spectrum]. While it shares the presence of the 1015 nm peak with the liver, one can discriminate a second peak located around 1075 nm.⁶¹ It is important to note that if one uses a different excitation wavelength, such as 690 nm, the peaks are not visible anymore [Fig. 7(a) top and Fig. 7(b) red spectrum]. The selection of the initial excitation wavelength will determine the emission spectra obtained, based on the endogenous fluorophores present in the tissues, and their excitation wavelength peaks. Thus, when working with tissue autofluorescence spectroscopy, it is important to take into account that a rational selection of the excitation wavelength should be performed prior to the beginning of the study.

Figure 8 demonstrates that in vivo autofluorescence imaging of organs is a possibility. Additionally, the endogenous autofluorescence NIR-II spectra of brain parenchyma and liver can be tracked in vivo in mice. Intravital surgery on anesthetized animals was performed to expose the brain [Figs. 8(a) and 8(b)] and liver [Fig. 8(c)] using a cranial and abdominal window, respectively. An 808 nm continuous excitation laser was focused on the area of interest, and the beam spot size was adjusted depending on the studied organ. Through hyperspectral imaging with a power density of 50 mW/cm² and an integration time of 15 s (spectral resolution 5 nm), autofluorescence spectra were obtained [Fig. 8(d)]. Changes in these spectra could potentially be employed to diagnose and monitor pathologies.⁶¹ 

Autofluorescence can be viewed as a double-edged sword when performing in vivo fluorescence imaging. Though in certain conditions, it might be an endogenous marker that provides tissue information, under different circumstances it might introduce errors into the measurements. Therefore, independently of the application, some general procedures need to be followed when performing in vivo fluorescence imaging. Generally speaking, only a camera, an illumination source, and/or a fluorescent agent are necessary for imaging. But equally important is the chosen in vivo model in order to study the scientific problem/medical question at hand, which determines the veterinary equipment necessary to maintain and monitor life functions, anesthesia, and temperature of the animal in the imaging setup. A careful choice of the model and its experimental necessities, on the other hand, also allows avoiding undesired tissue autofluorescence, or conversely taking advantage of it as an endogenous signal by itself. This means, for example, a mouse strain with less absorbance and autofluorescence could be chosen (nude mouse), a transparent animal model might even be a possibility (zebra fish), or a dietary protocol that avoids feedstuff fluorescence (fasting or alternative diet) could help in improving the in vivo imaging studies. The answer to these questions depends on experimental and monetary needs. While the NIR imaging equipment itself has fallen drastically in price, the price tags of some animal models low in fluorescence (nude mice vs agouti) remain high.

First of all, when performing in vivo bioimaging, animal welfare should be taken into account. An experiment in which the physiological parameters are controlled is likely to give better and more consistent results.¹¹² Animal monitoring includes control of the core temperature (e.g., through a rectal thermometer) and regular checks for any sign of hypoxia or cardio-respiratory depression.¹¹³ All three controls are a consequence of the anesthesia mandatory for in vivo imaging in most cases, which will lead to a decrease in temperature of the mouse and should therefore be maintained with the help of a heating pad (see Fig. 9). The optimal selection of the anesthetic agent(s) used during the procedure will also depend on the requirements of the study. Though a large number of anesthetic procedures are available, inhaled isoflurane anesthesia provided through a nasal cone is one of the preferred methods due to its safety, short recovery periods, the possibility of induced repeated and long periods of anesthesia, and regulating the administered dose in real time according to animal physiological requirements. When performing long imaging experiments using mice, it is important to lubricate the eyes with an ophthalmic gel in order to avoid damage caused by dehydration. Furthermore, long surgeries with extended anesthesia periods require fluid replacement. This can typically be achieved through the intravenous administration of no more than 0.2 ml/h of saline for protocols of 2 h or more. Two excellent protocols for imaging procedures of up to 4 h of non-stop inhaled anesthesia and intravital imaging can be found here, and the reader is encouraged to read it carefully prior to start any experiment involving animals.^114,115

While core temperature is of importance for the general welfare of the animal, surface temperature in the illuminated region of the animal also needs to be carefully monitored. Depending on the excitation wavelength, power density/intensity as a consequence of the spot size, and exposure time, significant heating of the animal's skin can occur (compare with Sec. III). Dark strains are also more vulnerable to undesired laser-induced heating due to the increased light absorption of both the hair and skin.³⁷ Therefore, safety standards for laser exposure have been set but can still result in overheating of the animal and damage depending on tissue-specific properties.^116,117 As a consequence, one should also monitor the surface temperature of the animal, for example, by including an external thermal camera in the experimental design (Fig. 9). Temperature affects not only the animal health but also a number of central parameters such as blood flow, which in turn also influences the tissue fluorescence and background autofluorescence in the case of imaging. Therefore, when performing in vivo fluorescence imaging, one should tend to limit the laser spot to the minimal size necessary to explore the area of interest in order to reduce this undesired heating effect. A more precise monitoring of temperature is required in experiments including photothermal therapy studies besides imaging. This not only affects the animal temperature on the surface but also inside the illuminated organ/tissue and therefore calls for additional considerations going beyond this Tutorial.^118–120 

In order to achieve ideal imaging conditions, one also needs to take into account the material and color of the surface on which the animal is placed when performing fluorescence imaging. A large illuminating spot can result in surface illumination and reflections. If this surface is dark, it could get heated and thus damage the animal and negatively influence the experiment. On the other hand, if it is white, it could reflect the autofluorescence coming from the animal (or any other unintended illuminated region) and, as a result, decrease the signal-to-noise ratio. The use of a dark, non-reflecting background is recommended when performing experiments as long as the spot laser is only illuminating the mouse. On the other hand, if a larger spot than the mouse body is necessary, a white, non-reflecting background should be chosen as long as the possible reflections are taken into account in the background correction. Alternatively, prior illumination experiments of the surface can help to ensure avoiding unwanted heating or reflections.

As mentioned above,³⁷ the skin is an organ that presents a high tissue fluorescence signal. Considering its physical properties acting as a natural filtering barrier for a big range of wavelengths, the choice of animal strain and therefore skin color is of utmost relevance for deep tissue in vivo imaging. Dark strains (such as C57BL/6J) present more autofluorescence than white mice (such as CD1) or specific hairless strains (see Fig. 3), but the latter ones are more expensive and can introduce undesired changes in the experiment such as affected immunity. As an easy and inexpensive alternative, shaving the mouse prior to any imaging experiment using a shaver or a shaving cream avoiding skin irritation reduces absorption and autofluorescence.^114,115

The following are some of the key points that we consider relevant for designing an in vivo fluorescence imaging experiment:

In order to begin, it is important to define the objectives of the study, the targeted tissue/structure, and which kind of probe (endogenous autofluorescence vs exogenous fluorophores such as nanoparticles) is going to be employed. Based on this, the minimal spot size that is necessary to correctly visualize the area of interest while reducing the external heating induced by the laser is determined. Furthermore, one should perform preliminary ex vivo studies using phantom tissue in order to determine if the exposure time and intensity could induce relevant tissue heating. The most widely employed laser wavelengths for in vivo biomedical applications are 690 nm, 808 nm, and 1064 nm based on their penetration depth and the intensity that can be reached internally. Lasers centered at 980 nm are generally avoided due to strong absorption by the water present in tissues.¹³ Hence, the rational selection of the laser(s) to be used in a study will depend on the type of tissue, its location within the body, or even the specific excitation of the endogenous autofluorescence markers.

Of importance is also the positioning of the illumination source, typically a laser fiber, in order to obtain the desired spot size. This can either be achieved through the right distance, which requires space, and/or through the use of collimator lenses. Ideally, the illumination and the camera/detection system would have the same orientation toward the imaging field. In practice, they should be as close as possible to avoid shadows and intensity gradients without blocking the field of view. In most approaches, this means that the actual illumination and imaging is performed from the top as illustrated in Fig. 9. This stems also from the fact that the animal welfare equipment (heating, anesthesia, and temperature control) is located at the bottom and sides in direct contact with the animal.

Similarly, the detection system will depend on the targeted type of application. If the focus of a given study is directed toward the autofluorescence in the visible range of the electromagnetic spectrum, then CCD cameras are generally utilized. If, on the other hand, the goal is to avoid the visible component of the autofluorescence, the detection system has to work in the near-infrared. For this, InGaAs cameras are the optimal candidates as they have high quantum efficiency (>85%), a wide spectral range, a fast response, which produces very low dark current, and are operational at room temperature by thermoelectric cooling. The InGaAs cameras are also commonly utilized in hyperspectral imaging for the NIR range alongside with indium-antimonide (InSb), mercury–cadmium–telluride (HgCdTe), and quantum well infrared photodetectors (QWIPs).¹²¹ Finally, the camera needs an objective or optics that are suitable for the NIR and allow to visualize and focus the region of interest in the animal at the distance determined through the construction of the imaging setup in synergy with the illumination.

Section IV illustrates the benefits that autofluorescence measurements could provide in research. There are situations, however, where autofluorescence is seen as a grave obstacle in the study of the phenomena of interest.

Hence, we will demonstrate how to overcome these hindrances and develop optical in vivo imaging into a promising technology. Principally, the advent of novel imaging/contrast agents that work in the NIR-I but even more so in the NIR-II window made this possible, together with improvements in infrared imaging technology. The former come in the form of organic dyes but especially infrared-emitting nanomaterials, while the latter led to a wave of cameras, detectors, and illumination sources (diodes and lasers) due to semiconductors/diodes with emissions in these spectral regions becoming more available.^28,38 The combination of these factors principally covers the excitation, emission, and detection in the NIR regions, allowing to ignore the issues with autofluorescence and opacity in the visible. Thus, the focus shifted to the infrared tissue fluorescence as the remaining stumbling block and how to drastically improve the signal-to-noise ratio regarding this endogenous tissue fluorescence background. In order to do so, the currently investigated techniques can most easily be separated into two approaches based on the method of filtering.^38,122

Managing the various contributions required to attain high-fidelity fluorescence images constitutes a challenging but critical task. After all, almost every component of the experiment can generate what is known as “optical noise.” This knowledge is generally obtained through practical experience with the experimental setup. This, however, does not exclude general rules behind the filtering process. For didactical purposes, we can divide the sources of optical noise into two categories: (1) excitation light noise and (2) background fluorescence that does not originate from the desired targets. Depending on the intended application, the researcher might have to focus on the avoidance of the former, the latter, or, in some cases, both. In our discussion, autofluorescence of the tissues is considered to be contained within the latter.

In many situations, the most common result of background fluorescence is an inconvenient increment in signal that is hard to circumvent. It meddles with the detection of fluorescence emitted by the fluorophore of interest and complicates the detection of weak fluorescence signals. Autofluorescence pollutes single-channel fluorescence microscopy and requires special attention, especially when one is looking for quantitative results (e.g., in ratiometric imaging). In the most generic approach, autofluorescence should be removed by a mathematical technique known as spectral unmixing.¹²³ It is most successfully applied when the spectral signatures of the targeted signal(s) as well as the undesired light are known. The mathematical task is made more complicated and computationally challenging for most in vivo situations, when unwanted background light is not so easily determined. In some cases, however, practical experimental steps can be taken to reduce its influence. The most straightforward is the use of long-pass spectral filters (often placed in a filter wheel, compare Fig. 9). The nature of the filters selected (cut-on wavelength region) is determined by the intended target and the overlap of its emission with the autofluorescence (schematically illustrated in Fig. 10).

Though background fluorescence is targeted for reduction in most experiments, the artifacts induced by the excitation light itself can also present a problem. This is, for instance, the case, when determining the autofluorescence spectra to later subtract them from the main results (as is the case in the spectral un-mixing technique). Though in principle the light sources used in fluorescence imaging emit in a narrow range of wavelengths close to their peak of emission, the spectral tails (on both ends) could not only be exciting other fluorophores but also adding to the signal measured in the spectrum. Thus, similar to the precautions taken against background fluorescence, the blocking ability of thin-film interference filters reveals them as a primary tool to prevent such artifacts. Since filters with limited blocking capability ultimately reduce image fidelity, careful consideration is demanded in their selection. The filters should spectrally select the desired light with high transmission and prevent out-of-band light by providing superior blocking with high optical density (OD) levels.

Another approach takes advantage of the short fluorescence lifetimes of organic fluorophores (∼ns), which includes the endogenous molecules responsible for the autofluorescence described above. In contrast, some inorganic materials can present lifetimes orders of magnitude larger (μs–ms).¹²⁴ In principle, this means that fluorescence from these materials can still be observed “long” after the excitation has stopped, while any tissue fluorescence has disappeared more or less directly with the excitation (see illustration in Fig. 10). Therefore, with the right technical equipment, a differentiation in lifetimes will also allow a separation of tissue fluorescence from the desired fluorescence signal of a lifetime contrast agent. However, this requires a technically more advanced experimental setup than the one described in Sec. V. First up, a stable, pulsed excitation source is required so that the fluorescence can decay in the off-periods. It needs to be connected to the IR camera through an electronic circuit that can trigger the recording of the picture with a short delay after the laser’s pulse but before the following pulse. This could also be achieved with a pulse-synchronized chopper wheel in front of the camera as a more inexpensive solution. That, however, would introduce more insecurities in the alignment of the whole imaging setup.^125,126 Theoretically, a delay of a hundred nanoseconds will suffice in order to filter the typical autofluorescence completely. From a practical viewpoint, the requirement is somewhere around 10–20 μs due to the properties of the laser, the circuit electronics, as well as the characteristics of the typically employed InGaAs cameras.¹²⁷ From these considerations stem the requirement of contrast agents with lifetimes greater than the above mentioned microseconds on top of the usual requirements a biocompatible in vivo imaging agent has to fulfill.

Over the last few years, this time-gating technique has begun to be developed for in vivo imaging, resulting in some exciting applications, especially in combination with rare-earth-doped nanoparticles, which present long lifetimes due to the characteristics of their forbidden f-f transitions, and are readily enough converted into biocompatible imaging agents with sufficient brightness. For a more detailed overview, there are also reviews on the topic available.^38,122,128 Here, we will just discuss a few examples that illuminate the potential of the technique further, while working in the NIR transparency windows. While this approach also removes autofluorescence reliably in the visible, it does not resolve the penetration issues.^129,130 The first in vivo imaging by this time-gated approach in the NIR-I was presented by Zheng et al., employing the 800 nm Tm upconversion emission by pulsed excitation of their upconverting nanoparticles at 980 nm. This not only generated autofluorescence-free in vivo images but also demonstrated that the heating of the animal by the laser was reduced by 35% due to the pulsed mode.¹²⁵ A first proof-of-concept for time-gated in vivo imaging in the NIR-II was presented by del Rosal et al., working with rather large (∼500 nm) Nd-doped NaGdF₄ particles in order to achieve sufficient emission intensity.⁵⁵ A later work showed that a careful dopant strategy and engineering of the nanoparticles improved the lifetimes, while at the same time providing NPs with different lifetimes in controlled manner, and allowed the use of real nanoparticles (13 nm and 9 nm, respectively) for in vivo imaging.¹³¹ 

Currently, this field is expanding, taking advantage of the simultaneous use of NPs with different lifetimes at the same wavelength in the NIR for multiplexed imaging,^{127,132–134} and of the independence of the fluorophore’s lifetime from its concentration in order to employ them not only as autofluorescence free imaging agent but also as nanosensor in vivo.¹³⁵ 

Another technique that is also based on temporal filtering by long luminescence lifetime materials is the use of persistent luminescence nanoparticles. Persistence luminescence consists in the emission of light by a material a long time after the excitation source of the luminescence has been removed. Therefore, this process occurs usually in two steps: first, the material is illuminated, usually with UV light, which promotes electrons from the valence band of the material to traps (defects with associated long-lifetime energy levels), where the electrons can stay for a long time. Then, the illumination source is removed and the electrons are gradually released from the traps, thanks to thermal energy or an external NIR excitation source, so emission of light can be seen for a certain period of time (see Fig. 10). Depending on the material, the persistence time (i.e., the time when the luminescence is still detectable) can be several hours or even days. The interest in these materials for bioimaging stems from the fact that it is possible to develop persistent-luminescence materials with emission wavelength in the biological windows. For this application, the nanoparticles are charged outside the body with UV light and injected afterwards, and the persistence emission allows the tracking of the nanoparticles inside the body. As no excitation light is used during the acquisition of the images, no autofluorescence signal is detected and high contrast images can be obtained, as was shown by le Masne de Chermont et al.¹³⁶ The drawback of this approach is that the persistence time of this kind of materials shortens as the emission wavelength is shifted to lower energies (i.e., to the NIR) so the time available for bioimaging is limited to a few minutes. Maldiney et al. demonstrated that the activation of the nanoparticles once injected into the body is possible, solving therefore the problem.¹³⁷ 

In vivo infrared imaging has advanced rapidly over the last few years, especially in the near transparency windows. First applications are even in the clinic.¹³⁸ This has also brought the spotlight back on tissue fluorescence and the consequence of the presence of endogenous fluorophores. They are well established in the visible range together with ways to avoid them and have even found some applications, e.g., in metabolic studies or diagnostics of skin cancer as described above. The newfound attention for tissue fluorescence came with the realization that there was more to deal with in infrared imaging than just absorbance and scattering of tissues. These are emissions, not as strong as in the visible, but often in key positions (NIR transparency windows) that present challenges for in vivo imaging relying on NIR fluorophores as contrast agents.

Nevertheless, tissue fluorescence in the infrared and its molecular sources are only at the beginning of being investigated or fully understood because the primary focus has been on improving the brightness of the NIR contrast agents employed in bioimaging. Section III summarizes the current knowledge on endogenous fluorophores in the infrared and the identification of tissue fluorescence in that part of the electromagnetic spectrum but also highlights the absence of spectroscopic characterization and clear identification for most sources of autofluorescence. This demonstrates that there is still a need for more information and investigation of the origins of autofluorescence in the NIR. A better understanding of NIR tissue fluorescence would also present more options to researchers interested in a specific phenomenon or pathology of a tissue and would allow them to decide whether an external contrast agent or fluorophore is really necessary or whether the object of the study can be achieved through analysis of the tissue fluorescence, as described in Sec. IV.

The availability of a clear identification of the sources of tissue fluorescence, together with the methods for avoiding autofluorescence and improving the contrast, will greatly advance the possibilities of infrared in vivo imaging. The technological improvements of excitation sources and imaging devices in the infrared together with modern computing power are making approaches like hyperspectral imaging more readily available and are currently resulting in a new wave of investigation into tissue fluorescence and toward the understanding of autofluorescence. The importance reaches beyond just in vivo imaging. Examples are the growing and connected fields of luminescence nanothermometry and photothermal therapy, which both rely heavily on infrared in vivo imaging. A better understanding of tissue fluorescence and hence the autofluorescence background will make these applications more precise and help in vivo realization. The combination of NIR imaging with already clinically established endoscopic/fiber-based NIR techniques like OCT^139–141 together with the identification of autofluorescence markers in the bloodstream (compare Sec. III) will, in our opinion, also help to expand the clinical use of in vivo NIR imaging. We hope that we were able to present the important aspects and underlying principles sufficiently in this Tutorial review so that the interested reader can follow the exciting developments in the field of tissue fluorescence and in vivo imaging that are about to come. In our opinion, these achievements will help to elevate infrared imaging into a fully fledged medical technique.

A hyperspectral imaging video of the organs in Fig. 6 is available in the supplementary material.

This work was supported by the Spanish Ministry of Economy and Competitiveness under Project No. MAT2016-75362-C3-1-R, the Spanish Ministry of Sciences, Innovation and Universities under Project No. PID2019-106211RB-I00 (NANONERV), by the Instituto de Salud Carlos III (Nos. PI16/00812 and PI19/00565), and through the Comunidad Autónoma de Madrid (No. B2017/BMD-3867RENIMCM), and co-financed by the European Structural and investment fund. Additional funding was provided by the European Union's Horizon 2020 FET Open project NanoTBTech (Grant Agreement No. 801305), the Fundación para la Investigación Biomédica del Hospital Universitario Ramón y Cajal under Project No. IMP18_38(2018/0265), and also COST action CA17140. Y.S. acknowledges a scholarship from the China Scholarship Council (No. 201806870023), E.X. is grateful for a Juan de la Cierva Formación scholarship (No. FJC2018-036734-I), and D.H.O. is thankful to the Instituto de Salud Carlos III for a Sara Borrell Fellowship (No. CD17/00210). The authors thank Dr. Blanca del Rosal for the helpful discussion and input on the manuscript.

The data that support the findings of this study are available within the article and its supplementary material.