Luminescence based temperature bio-imaging: Status, challenges, and perspectives

The only way to get thermal images of living organisms without perturbing them is to use luminescent probes with temperature-dependent spectral properties. The acquisition of such thermal images becomes essential to distinguish various states of cells, to monitor thermogenesis, to study cellular activity, and to control hyperthermia therapy. Current efforts are focused on the development and optimization of luminescent reporters such as small molecules, proteins, quantum dots, and lanthanide-doped nanoparticles. However, much less attention is devoted to the methods and technologies that are required to image temperature distribution at both in vitro or in vivo levels. Indeed, rare examples can be found in the scientific literature showing technologies and materials capable of providing reliable 2D thermal images of living organisms. In this review article, examples of 2D luminescence thermometry are presented alongside new possibilities and directions that should be followed to achieve the required level of simplicity and reliability that ensure their future implementation at the clinical level. This review will inspire specialists in chemistry, physics, biology, medicine, and engineering to collaborate with materials scientists to jointly develop novel more accurate temperature probes and enable mapping of temperature with simplified technical means.


Introduction
Recent advances in materials science aim to address numerous challenges set by biologists and medical doctors. "Smart" drug delivery nanoparticles, specialized multiplexed luminescent labels (e.g. quantum dots, lanthanide doped nanoparticles, nanodiamonds etc.) for imaging and flow-cytometry 1,2 , photostable multicolour luminescence labels for super-resolution imaging 3 , improved MRI 4 , CT 5,6 , OCT 7 , USG 8 contrast nanoagents and many other purposefully designed materials, have revolutionized current possibilities in the understanding of biology, feedback controlled therapies and personalized medicine 9 . In particular, optical and remotely read, sub-millimetre spatial resolution temperature mapping is very appealing due to the fact that the temperature is a fundamental quantity that is related to many natural biological processes (e.g. activity of enzymes) and that is a prerequisite tool to feedback control hyperthermia therapy of cancer.
The fundamental question regarding remote temperature sensing and imaging in biology is: why is it important to know the in vivo/in vitro distribution of temperature? As known, increased temperature modulate numerous enzymatic and biochemical processes, through modifying the proteins or enzyme activity 10,11 , gene expression 12,13 or cell signalling 14 . Temperature affects membrane stiffness, and this one modulates the permeability of the cell membrane to exogenous compounds or nanomaterials. It was suggested, that up to 50 °C can be reached locally in mitochondria as compared to the cytoplasm 15 .
Despite these results are striking, it is still not clear, why such a high temperature is important and how this exact value and temperature gradients affect biological systems. Furthermore, theoretical considerations have been questioning the interpretation currently given to temperature heterogeneities measured in single living cells at nanoscale scale 16,17 . As a consequence, the temperature reading or mapping at the tissue or subcellular level could be important not only to diagnose diseases, evaluate therapeutic efficiency of drugs, evaluate the origins of toxicity but also may shed some light on the extent of validity of thermodynamic considerations when evaluating cellular processes in individual cells and in tissues. Temperature imaging is particularly compatible and important in combination with fluorescence microscopy imaging in biology, because a morphology (e.g. tissue structures, organelle location in cell) can be co-localized with cells machinery functioning (e.g. Ca influxed in neurons, protein activity). Despite the great interest and some progress in recent years, there are still not too many examples, where the temperature is mapped in vivo or in vitro. Apart from the biological suitability of temperature mapping and sensing, many other technological processes require evaluation of temperature distribution (e.g. electronic circuits, rotating mechanical elements etc.) in a remote manner 18,19 with submillimeter optical resolution.
While conventional temperature recording (e.g. with thermocouples or thermistors) or imaging (with bolometric cameras) techniques are robust and reliable for most types of applications, the remote temperature detection in biology and medicine is definitely technically much more challenging. The existing techniques are not suitable to satisfy sophisticated requirements imposed by the thermometric in vitro microscopy or by the body in vivo temperature mapping. This is true mostly because they are either invasive or they prohibits temperature detection from below the skin surface. On the other hand, luminescent imaging in vitro and in vivo is a well-established method to enhance the visualization contrast of biological specimens. Numerous fluorescent labels (e.g. organic dyes, lanthanide chelates, quantum dots, lanthanide doped nanoparticles) have been successfully used for luminescent imaging because they offer: (i) unprecedented optical resolution with subcellular resolution or even below the diffraction limit of light; (ii) penetration of both excitation and emission light through the skin, thus enable remote / non-invasive or minimally invasive measurement from the volume of sample, (iii) specific targeting of molecular, cellular or tissue components, (iv) large (anti)Stokes spectral shift enabling to easily distinguish optical stimulus from the photoluminescence, (v) fast response to study dynamic processes, (vi) quantitative evaluation of intermolecular energy transfer and interaction (like FRET detection of biomolecules structure and reorientation, DNA hybridisation, enzyme activity etc.).
The spectral features of the same fluorescent probes, are very often susceptible to variation in the local temperature, which, in turn, makes them very promising for constructing remote temperature sensing and imaging systems, with no additional investments in detection equipment.
Many of these luminescent materials (e.g. polymers, quantum dots, lanthanide and transition metal doped nanoparticles, metal complexes, diamonds, organic dyes, fluorescent proteins, hybrid materials, metal nanoparticles, which have been extensively published and reviewed 20 ) display temperature dependence, which made them candidates as actual temperature reporters. The susceptibility of the spectroscopic properties to the temperature change is however only a prerequisite condition and does not necessarily mean that the material can be suitable for temperature imaging in biology and medicine, because such applications put very strict and interdependent requirements on the materials and methods depending on target applications [21][22][23] . Such luminescent thermometer (LT) reporters should satisfy numerous requirements to become suitable for detection and temperature imaging in vitro or in vivo. First of all they should either passively diffuse into cells and tissues or be actively targeted to biomolecules, biostructures, cells and tissues. While the former requirement put demands on the size of the labels to stay below 50 nm hydrodynamic diameter, the latter one, requires additionally appropriate surface biofunctionalization, which has been reviewed in scientific literature [24][25][26] . The nanometric size, together with the narrow size distribution of optical thermometers, is very important for biomedical applications for one another reason. When the collection of nanothermometers strongly varies in their particle diameters 27 , not only different spectral properties and thus temperature response may be expected, but also the local heat transfer between the object of interest and emissive centres will vary thus also introducing artificial temperature fluctuations. The LT reporters should also be biocompatible (usually SiO 2 , PEG shell is used to secure this) and biologically safe, with non-existent or limited nanotoxicity 28 at reasonable doses. They should not undergo dissolution or degradation in the harsh environment within tissues, cells or cellular compartments (e.g. lysosomes). The LTs should be resistant to local chemical and physical conditions other than temperature [29][30][31][32] , for example, should not behave differently in different pH, viscosity, presence of H 2 O or radicals, as these factors are very difficult to quantify and thus are not easy to include in data analysis or correction. Moreover, the temperature readout should not depend on LT reporters density (e.g. owing to spectral self-absorption), or the spectral signatures of the T probes should not spectrally overlap with spectral properties of tissue/cell componentse.g. with variable oxy/deoxy hemoglobin or with the autofluorescence of the tissue components under short wavelength photoexcitation, as these two factors increase background level and hinder quantitative analysis. The LT reporters dose is critically important from the perspective of toxicity, luminescence brightness and signal quality. Additionally, the temperature reporters must provide fluorescent signal, which assures good metrological conditions, i.e.
(i) high relative sensitivity S R of T determination as well as high readout accuracy, trueness, precision and measurement resolution (better than 0.1 °C), Currently, a great challenge is also to prepare high brightness NIR emitting luminescent labels, which should enable deeper light penetration into solid tissue samples and allow for in-vivo 3D imaging of specific tissue types (e.g.cancer) or a biochemical / physical quantities (pH, temperature etc.) 22,33,34 .
These NIR reporters are especially difficult to make due to efficient luminescence quenching by ligand and medium molecules [35][36][37] . While most of fluorescent species are vulnerable to such high energy vibration, inorganic nanoparticles (QDs or lanthanide doped nanoparticles) show great advantage here surface shell can be easily formed, which can efficiently protect the NPs and enhance their brightness 36,[38][39][40][41][42] . Nevertheless, neither materials themselves nor NIR emissive labels are within the scope of the current review, and readers interested in this topic can use the above mentioned references as a starting point for more in-depth discussion.
The developments of novel luminescent thermometer should be concomitantly supported by proper physical models, that not only are capable to allow for the targeted design of sensitive nanothermometers but also to account for robustness, accuracy and highest possible precision. The models should be capable to (i) replace current try-and-error or incremental approaches to finding new optical thermometers with educated guesses for new materials, (ii) understand the limits of particular physical phenomenon and (iii) assure sample-to-sample variation-free operation. For example, based on Boltzmann thermometers, not only sensitivity values and temperature range limits may be wisely optimized and enhanced, but also enable to construct primary thermometers to counteract known issues of thermometer re-calibration in-situ 43 .
Given the novelty of many thermometric reporters, some adjustments on recording and detection methods need to be proposed to assure reliable, reproducible and accurate temperature determination and mapping. These technical issues and their impact on the metrology, however, are rarely discussed.
This lack of consideration, in turn, has motivated us to review the current knowledge and achievements in the optical 2D thermometry in vitro and in vivo.
Although some examples of temperature mapping exist in technology (e.g. temperature distribution in flame 44 , hot spots in microprocessors 45 or rotating mechanical elements 46 ), we think that the fact that T mapping in vitro is not well developed and widely used, originates from a few, well defined reasons: 1. lack of sufficiently good LT probes, which are not only temperature sensitive, but can be adopted as biological reportersi.e. assure individual nanoparticles with no aggregates formation, whose surface can be bio-functionalized for specific targeting, long circulation time and non-toxicity; 2. difficulties in implementation of LT mapping, which come from the unreliability of T determination, owing to the susceptibility of the existing LT agents to numerous experimental factors, such as: excitation intensity, the penetration depth of excitation and emission light, technical complexity of measurement (e.g. spectral overlap, the brightness of LT etc.); 3. difficulties in the implementation of LT sensing with the technology available at the moment, i.e. most of the T sensitive phosphors are evaluated with high resolution and super-sensitive spectroscopes or photodetectors, which hinder adoption of good T probes (satisfying the conditions 1 and 2) for simple, practical and widely accessible microscopic imaging studies. Every new technological solution (here luminescence based thermometry and thermal imaging) follows typical trend described in Fig.1, which is called technology adoption curve 47 . In our opinion, current status of temperature imaging has entered the realism phase, where material science achievements and various luminescent temperature probes are confronted with the demands of real-life applications and many materials fail to pass the disillusionment chasm. This is because, even though they fulfill fundamental requirements related to temperature responsivity and sufficient sensitivity, they lack numerous other features, which are necessary to safely use these temperature reporters in 2D imaging in vivo / in vitro 21,48 . This requires setting a standardization rules to be satisfied, which should evaluate not only temperature sensitivity factor, but also other ones, such as (i) trueness, resolution, reproducibility of T determination (in terms of metrology) as well as (ii) susceptibility of the temperature reporters to the local environment (in terms of chemical and spectral susceptibility of the readout to surrounding tissue), (iii) brightness and (iv) spectral operation range (in terms of inherent properties of the LTs) as they also should be suitable to reliably map temperature with existing readout mechanisms and methodologies. We are convinced, it is impossible to further progress in this field, without joined work of material scientists (majorly responsible for the progress in nanothermometry field so far) with physicists, biologists and engineers. Only then, and taking into account all aspects briefly enumerated above, we may expect transition from the optimistic expectations towards mature technology ready to be used in medical diagnosis and technology 21 .
After a few years of initial luminescence spectra based thermometry field development (reviewed by Vetrone and Jaque 49 , Carlos et el. 50,51 , Dramićanin 52 and others 23,46,53 ), the catalogue of methods suitable for temperature determination can be further expanded to the more actual and complete one.  (Table 1 and Fig.2). Each of these techniques quantifies the thermometric parameter () in a different way and traditionally, materials scientists who develop new thermometric reporters rely on the scientific grade (highly time and spectrally resolved) spectroscopes and rarely discuss how feasible is to make a transition to detectors, which are more cost-effective and more suited for T imaging.
Therefore, our motivation here is to try to determine how easy and reliably can the temperature be visualized in in vitro or in vivo conditions. Starting from these considerations on the technical feasibility to map temperature, we derive favourable properties and necessary conditions to be met by LTs. We also evaluate these technical aspects from the perspective of quality of results, robustness and cost effectiveness of imagers. Finally, we present examples of existing literature reports on in vitro and in vivo temperature imaging. The success of optical thermometry requires cooperation between many interdisciplinary fields, and try-and-error approach in searching of the most suitable thermometers must be proceeded with defining clear requirements for materials and quantification methods 21,23 , understanding the physical phenomena behind thermal sensitivity 21,43,54 and adjusting all these issues to the available and reliable detection instrumentation (discussed here). Especially, the understanding and modelling of optical thermometers is underrepresented in scientific literature, while they are key to knowingly optimize the performance of luminescent compounds. A comprehensive and complementary study discussing the physical mechanism responsible for temperature-induced change in physical parameters was presented recently by Suta et al. 43 and the necessary information regarding a theoretical background and physical and thermodynamic limitations of the physical models can be found there.
We believe that the current review offers an alternative perspective drawing the attention of material scientists to established problems and help them find more targeted solutions in future. We also emphasize, that further progress in the field of luminescence thermometry requires deeper understanding of physical mechanisms and materials factors known to be temperature dependent, and simultaneously increased awareness of methodologies and technical aspects of T readout.

Methodologies and technical aspects of T readout
There are many scientific papers showing new materials as potential candidates for remote thermometry.
Unfortunately, these examples are mostly focused on the single spot detection and the calibration or readout of such LTs usually requires high resolution excitation/emission spectra and rather sophisticated and costly spectrometer setups. However, as this technology has the ambition to be used in real in vitro/in vivo studies, the readout technology has to be rapid, simple, cost-effective and compatible with current imaging detectors, which put some additional technical requirements. Therefore, the major motivation for this article, was to overview existing examples of temperature imaging, from the perspective of technology required to achieve this goal. Table 1 and Figure 2 compare how different physical observations ( definition) can be quantified, using a simple available camera ( imaging). For example, the shift of the maximum of emission band in response to temperature change, requires a spectrograph with spectra resolution of d < 1 nm. However, such solution is usually too complex, too slow and ineffective for imaging in raster scanning mode. Moreover, the speed of imaging is directly related to the brightness of the T probes. In opposite to the spectrograph, a CCD camera can be used to quantify emission intensity simultaneously for multiple spots on the sample in two spectral windows, which shall assure fast and simultaneous readout. Nevertheless, such ratiometric imaging is highly susceptible to artefacts, spectral overlap between T probes and sample spectral properties. This one and the other possibilities are discussed in the next chapters from the perspective of technical feasibility to adopt a given mechanism for 2D temperature mapping.
In particular, Table 1 defines the thermometric parameter in two different ways, the one ( definition) seen by materials scientists and the other one ( imaging) from engineers perspective. While the former one is equipped with highly (temporally or spectrally) resolved fluorescence spectrometers, the latter possess a sensitive CCD/CMOS camera, which is capable to measure 2D intensity images in continuous wave or time resolved mode, possibly in spectral domain as well. The former offers high resolution, which, however, is typically inconvenient for imaging because of the speed of T mapping. The latter solution has technical issues with spectral resolution, but highly parallel detection offers technically simple solution for 2D mapping of spectral signatures. As presented in Fig.2, CCD/CMOS cameras support all the types of spectral features, which are considered to be the indicators of temperature. Apart from the obvious ones, such as absolute intensity or emission intensity ratio, also spectral shift, luminescence anisotropy, kinetics and emission bandwidth can be easily quantified by quantifying 2D spectral images. However, it requires using additional spectral filtering, low background, no autofluorescence of the sample, specific excitation wavelength or time gated detection. Each of these methods will be discussed below in the following chapters, and whenever possible, will be supported with the existing scientific literature on experimental demonstrations in vivo, ex vivo, in vitro or on phantoms.

Figure 2. Schematic illustration of the temperature dependent mechanisms.
Blue and red colours symbolize the luminescent features at low and high temperature, respectively. The grey diagonal filled rectangles indicate e.g. spectral (a-g, i) or temporal (h) windows seen by the camera. Even though there are no examples of h and i methods, they may potentially be suitable for T imaging in the future.

Influence of tissue / sample spectroscopy on temperature readout
So far, not too many reports discussed the robustness of the temperature readout in vivo or in vitro, while serious experimental issues limit the reliability of such sensors. This originates from the fact, that the luminescence of LT may be affected by the sample spectroscopy itself, which is especially troublesome in vivo or ex vivo. The LTs are typically calibrated in spectrally "sterile" conditions, i.e. there is no autofluorescence of the sample taken into account as well as low background signal and non-existent or limited scattering are assumed, while the measurements carried out in vivo and in vitro, must consider the sample properties as well. This is clearly illustrated in Figure 3, where absorption spectra of the tissue components are shown against the temperature dependent spectral features of different luminescent thermometers at high and low temperatures. The tissue spectroscopy may modify e.g. the ratio of emission bands or the shape of the emission band, which is even more complicated, because tissue spectroscopy undergoes local (e.g. from site-to-site) and transient (over time) variations. Pulses of oxygenated blood, which spectrally overlap with e.g. most well-known Yb 3+ /Er 3+ co-doped nanothermometers are good examples of this. Another issues come from the fact, that the tissues may significantly differ in composition (water/fat content, pigmentation, thickness, heterogeneity, blood vessels density etc.), which complicates calibration and reliable temperature readout. These issues are discussed in more details elsewhere 55,56 , but they definitely require attention and careful consideration in practical implementations of luminescence based thermometry. It is very important to mention, the spectral characteristics of the tissue or sample, may significantly affect the temperature being read from thermometric parameter 21,22,48 . In a real application, the conditions for T readout are much more complex than in laboratory conditions. For this reason, from the metrological perspective the sensitivity of temperature readout must not be the only important factor characterising luminescent thermometers 48 . No less important are temperature readout accuracy (proximity of individual measurement to true temperature value T 0 ), trueness (closeness of mean from a set of temperature measurements to true temperature value T 0 ), precision (reproducibility of temperature readout T among a set of readout temperatures) and measurement resolution (smallest change in temperature that produces measurable change in thermometric parameter ) 21 . The high resolution stems from high brightness, while high sensitivity requires high signal to noise/background ratio, to determine the smallest  reliably. Since the calibration is usually performed in conditions other than the actual in vivo environments, and the biologically acceptable temperature variation range is limited to 0-50 o C, it is of utmost importance to examine these parameters carefully, which has been omitted so far either in materials science research or in the view of possible applications of luminescent thermometers in hyperthermia treatment. Assuming Gaussian distribution of temperature readout T() , the mentioned trueness, precision and resolution of the measurement, may be therefore defined as mean temperature of readout (()), variance (()  ) and standard deviation (2()), respectively. The temperature resolution is sometimes calculated as T , which actually is the theoretically expected statistical temperature uncertainty. It does not take into account any systematic errors that can additionally falsify the temperature measurement as discussed above and in numerous excellent research papers 43 . One should be aware, that there is currently no method to incorporate the material and experimental parameters (e.g. size, composition, dopant type and concentration etc.) into universal thermometer characterisation, other than to estimate (), S R (T) parameters from the experimental dataset. Although deep understanding of physical phenomena and modelling may support intentional designing of NT reporters such studies are rare 43,54,[57][58][59][60] . Additionally, the dependence of () from photoexcitation intensity, shown by some LT reporters (e.g. upconverting materials) practically disqualifies such LT materials from in vivo applications as one cannot actually determine excitation intensity in heterogeneous objects with sufficient accuracy. Fortunately, understanding the mechanism behind thermal responsiveness, shall let define the primary thermometers, which in some, but not all cases, should make the T measurements independent from environmental and some of the measurement factors.
To better understand the role of biological tissues when considering temperature readout, one needs to take into account, that many physical processes that take place during the measurement. Firstly, from a thermodynamic standpoint, a living tissue is considered to be a complex system whose temperature is determined by the interplay between its intrinsic properties (such as its mass, specific heat, and thermal conductivity), the physiological parameters of the organism (such as blood temperature, perfusion rate and metabolic activity), tissue conditions (such as water or fact content) and the environment conditions (mainly ambient temperature). All these variables influence the response of living tissues to thermal stimuli, which can be expressed through the well-known Pennes' bioheat equation and the corresponding boundary conditions. In such formalism, this can be written as 61 : where ρ t , k, and c t are mass density, thermal conductivity, and specific heat of the living tissue and ω b , ρ b , c b , and T b are the perfusion rate, mass density, specific heat, and temperature of the blood. In the expression above, Q met refers to the heat production caused by metabolic activity. Pennes´s bioheat equation not only establish the relation between tissue conditions and activity and tissue temperature. It also determines the spatial spread-out of temperature within tissues. For instance, it establishes that even when dealing with well-localized heat sources the temperature changes caused in tissues can expand significantly in the three dimensional space. From a technological point of view, this could be understood as a good new, because it means that we do not need super-high spatial optical resolutions in our thermal images (for sure, at the in vivo level we do not need to reach nanometric resolution).
Spatial diffusion of heat in tissues constitutes an additional difficulty when interpreting the thermal images. For instance, during thermal therapies, and advanced analysis of the thermal images of tumors will be required to determine the location of heating source and, also, to achieve an accurate estimation of the thermal dose 62 . But from a positive point of view, if heat diffusion within tissues is well-know and modelled, the acquisition of multiple thermal images under different conditions (e.g.spatial scanning of heat source within tissues, acquisition of multiple projections of thermal patterns) can be the first step towards the development of thermal tomography. This, of course, would imply a significant effort in the development of algorithms for the analysis of 2D/3D thermal images.
From the standpoint of light propagation into the tissue, on the other hand, the tissue is typically considered as a simple scattering medium in which the path of a photon is considered to be probabilistic in nature. Under this simple assumption, the main impact of the tissue on the thermal images (obtained from the analysis of fluorescence images) is losing spatial optical resolution. Even in an ideal situation, where all the photons are generated form a well-defined location, tissue-induced light scattering causes a relevant spread out of photons at the imaging plane and thus resolution loss. In principle, if the photon path within the tissue is known, it becomes possible to determine location from which it has been emitted within the tissue from the fluorescence image. This, in turns, implies that if the photon trajectory is known, it would be possible to increase the resolution of thermal image and, even, to localize in three dimensions the luminescence and heat source. To determine the photon trajectory through the tissue, the Monte Carlo (MC) method is generally applied 63 . It deals with experiments on random numbers and is applied to situations where a thorough analytical description is either lacking or too unmanageable to yield a solution. Since the scattering of individual photons is governed by the laws of quantum mechanics, this method naturally gained popularity for the description of light propagation in biological tissues. The underlying assumption in MC is that one is dealing with a sequence of random noncorrelated events. 64,65 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.
A combination of both the MC method and Penne's formalism could, therefore, result in a very powerful tool for the description of the effect played by tissues in temperature readout provided by luminescent probes. Combining MC and Penne´s formalisms is however not an easy task at all. The key point is that MC simulations are based on the optical properties of tissues and these depend on tissue temperature that, in turns is given by Penne´s equation 66 . The situation becomes even more complicated with the appearance into scene of recent works that demonstrate how at certain spectral ranges (typically used by infrared luminescent nanothermometers in the in vivo applications) the tissue absorption can be a dominant factor affecting the photon propagation within tissues 67 . Note that the tissue absorption at specific wavelengths is not considered in MC simulations. In addition, the presence of relevant tissue absorption of emitted photons also complicates the solution of Pennes´s equation as it introduces an additional source of heat that will spread out all along the photon trajectory. So, both MC and Penne's formalism should be modified to include photon absorption and its impact on their trajectories and on the temperature distribution within the tissue. The adaptation of both formalisms to consider the effect of photon absorption becomes also mandatory to reconstruct the emission spectra of nanothermometers caused by tissue absorptions. Tissue induced spectral distortions make very difficult to reach a reliable thermal readout at the in vivo level 68 . The development of algorithms, considering tissue scattering and absorption, capable of removing these tissue-induced distortions will make possible to achieve reliable thermal reading from a simple analysis of the emission spectra collected at the in vivo level. In summary, challenges of luminescence-based thermo imaging do not only reside at the experimental side, but a great effort in modelling and image reconstruction techniques will be required in short time to achieve high resolution and reliable thermal images. Table 2. Materials, experimental and sample factors, which affect the performance and reliability of temperature readout with luminescent thermometers. Q Squenching at the surface by ligands, defects, solvent; Q NRquenching by non-radiative processes; QYemission quantum yield;  -extinction coefficient; S/Vthe ratio of a number of surface to volume activators; M(T)mechanism of thermal sensitivity; Bbrightness; ETenergy transfer processes, such as ETU, CR, Q NR ; h MAXmaximum phonon frequency; Eenergy gap between thermally coupled states; (T) -thermoresponsive parameter (e.g. LIR, , ); S R = S R (T)-relative sensitivity; T MINtemperature resolution; |T-T 0 | -accuracy of temperature determination in respect to the actual temperature of the sample (T 0 ); J Sspectral self overlap integral between LT absorption and emission; J Espectral overlap integral between LT emission and sample absorption;  STOKES -Stokes shift between excitation and emission;  LIR -Stokes shift between wavelengths used for =LIR determination; S/Ba signal to background ratio; S/N -a signal to noise ratio.

Temperature detection readout
In general, the thermometric parameter (T) is measured versus variable temperature T during the calibration procedure, which means the experimental relationship between  (T) and T is recorded.
It is desirable to understand the mechanism behind these phenomena in order to relate these parameters in the form of (often unknown) analytical mathematical equation ( ). For example, assuming the temperature phenomenon as simple Boltzmann relationship between the population of two emitting levels n 1 and n 2 , the thermometric parameter may be described by a simple formula 69 : In the limit of zero pump power in a given temperature, the temperature 0 corresponds to no laserinduced heating and the thermometric parameter  0 ( ) can be calculated as Making a simple ratio of Θ( )/Θ 0 ( ) one gets: By making the natural logarithm of both site of the equation, one gets: Therefore the inverse function can be defined as ( )  70 , allows to analytically find the inverse function  = −1 ( ). This concept relies on the mathematical foundation that every power series uniquely converges to a function it represents. Although mathematically correct or acceptable, from metrology perspective it is much more important to understand the physical phenomenon behind temperature susceptibility, which enables to find the constraints of the proposed materials and methods and knowingly optimize such thermometers for targeted applications. Moreover, the lack of appropriate model hinders actual interpretation of thermometers behaviour, determination of its operating range and its reproducibility may not be guaranteed.

A. Absolute intensity of luminescence
The great advantage of temperature mapping by detecting the absolute intensity of luminescence changes  = I() (Figure 2a) is its technically simple implementation and fast readout (Table 1a).
Moreover, most luminescent species may serve as temperature probes in this working mode.
Unfortunately, this technique is suitable only for a rough estimation of temperature rise or fall. It is not very reliable, since the absolute intensity depends on the concentration of NPs per volume, penetration depth of photoexcitation intensity, scattering and absorption properties of the biological sample and is difficult to implement in 3D tomography volumetric imaging. This approach has already been used for in vitro and in vivo temperature mapping (Figure 4).
The thermometric parameter in this relative intensity technique (AIT) can be generally defined as: If the relative sensitivity change over the studied temperature range is constant, e.g. S R = 0.5 %/°C, knowing the initial temperature T 0 (x,y) distribution and assuming homogenous excitation illumination and non-moving objects, one may determine spatial temperature maps as: where −1 is the reverse function, which is based on given (T) reads corresponding to T. Because the starting temperature map I o (x,y) is actually difficult to know before the experiment (e.g. mitochondria may show natural higher temperature than in cytoplasm), the relative temperature rise in respect to the starting point can only be found in response to some stimulus (e.g. drug, toxicant, local heating etc.). In consequence relative temperature change can only be quantified: One of the first demonstrations on thermal imaging of cells using relative changes in emission intensity was presented by Zohar et. al. 71 in 1998, who visualised Chinese hamster ovary (CHO) cell clusters during chemical stimulation causing an increase in heat production. The measurement was based on the temperature-dependent luminescence of europium (III) thenoyltri-fluoro-acetonate (Eu-TTA), which exhibited a narrow emission line around 614 nm. With the increasing temperature, the coupling of the energy levels with the environment through molecular vibrations was increasing, and the rate of nonradiative processes was increasing as well, which led to the quenching of Eu-TTA emission. This sensor could not penetrate inside the cells, but was integrated into the liposomal membranes because of its hydrophobic nature. Unfortunately, the sensor showed photobleaching, which was approximated by the exponential decay and omitted in the processing of the data for thermal imaging. In addition, Eu-TTA emission was also dependent on pH changes (range 5-8). Thermal imaging (Figure 4) was carried out for acetylcholine chemical stimulation and a heat wave was observed in the cells. After the initial phase of acetylcholine dilution, the emission intensity of Eu-TTA was significantly reduced, which was equivalent to the increase in temperature. After some more time, the temperature returned to its original value. An additional control experiments (using a pH-sensitive dye and a pH electrode) evidenced that the addition of the acetylcholine did not change the pH that the sensors are known to be susceptible to. Another work describing both intracellular temperature and imaged living cells was presented by Gota et. al. 72 in 2009. The authors described a fluorescent nanogel thermometer consisting of polyNIPAM thermosensitive polymer and DBD-AA fluorophore, whose luminescence is easily quenched by water molecules. The gelation was performed by emulsion polymerization using a cross-linking agent (MBAM). Precipitation in the intracellular environment was avoided due to the fact that a highly hydrophilic layer formed by sulphate groups (which was used in excess) was formed on the sensor.
PolyNIPAM was changing its conformation depending on the temperature, i.e. at low temperature, it allowed water molecules to get inside and quench the luminescence of the DBD-AA dye, while at higher temperatures, water was pushed out of its interior recovering the luminescence of the dye. In addition, the thermal dependence of luminescence was neither susceptible to changes in pH (range 4-10) nor by the presence of proteins. Thermal images of cells at 29 and 35 °C demonstrated a significant relative increase in emission intensity, which is synonymous with the temperature increase in the cells. Slight 0.45 °C temperature rise was also observed in response to chemical stimulation using the FCCP reagent.
Another type of the temperature sensor, which relied on the relative change of the intensity of one band as a function of temperature was also described by Ke et. al. 73 in 2012. The authors used the L-DNA molecular beacon (L-MB), whose temperature induced conformational changes were modulating the distance between donor and acceptor (quencher) molecule, which, through distance (and indirectly through temperature) FRET energy transfer was changing the luminescence signal from donor molecules. Importantly, the L-MBs were non-toxic to cells because they consisted of nucleic acids.
Interestingly, the L-DNA is not a naturally occurring biological DNA form (in contrast to the D-DNA), which made it resistant to enzymatic degradation. What is more, the MBs provided a fast, accurate and sensitive temperature reading. The measurements of the L-MB emission as a function of temperature had shown that the intensity increased in the range of 20-55 °C resulting in temperature resolution < 0.7 °C. It had also been shown that very small changes in sensor performance occurred when the ionic strength (150-200 mM) or pH (6.8-7.4) varied. Then, the L-MBs were introduced into HeLa cells by liposome transfection, and the temperature imaging was performed using confocal microscopy ( Figure   5) with excitation with 488 nm diode laser and emission intensity quantified at 505-535 nm. In the years preceding 2014, Arai et. al. 74 created a large database -Diversity Oriented Fluorescent Library (DOFL) containing various fluorescent dyes and found ER thermo yellow dye to be suitable for endoplasmic reticulum (ER) targeting. This dye showed the temperature dependent luminescence at 581 nm under 559 nm excitation. The first practical application was shown using an IR laser 1064 nm for spot heating of the cells (by irradiation of the aluminum powders attached to the tip of a glass microneedle). Using a shutter, a "square wave" was obtained as a response to the cycles of local heating and cooling. Moreover, temperature gradients created in this way were visualized in living cells using different laser power (Figure 6). A linear dependence was obtained for normalized (to its initial value) intensity as a function of temperature change reaching sensitivity of 3.9 %/°C (for temperature changes from the initial temperature to 5 °C higher). In the next step, the validation of the received results for live cells was carried out in the fixed cells. In this way, the results were obtained for the cellular environment (which was significantly different from the buffers used in other cases), but with guaranteed accurate temperature control. In addition, the thermometer was found to be insensitive to the presence of Ca 2+ ions and pH changes. Sensitivity also did not depend on the cell line used (HeLa, Chang liver, 3T3, brown adipose tissue and C2C13 myotube). In the case of a long-term measurement, the photobleaching effect could not be omitted and the results were corrected by a single exponential curve.
Ultimately, the ER thermos-yellow was used to visualize the heat production caused by the addition of ionomycin (it is a Ca 2+ ionophore). After the addition of the reagent, a decrease in emission intensity of ER thermo yellow was recorded which was equivalent to the temperature increase. The change in intensity of ER thermo yellow during this process was about -6.8%, which corresponded to a temperature elevation of about 1.7 °C. After a thorough analysis, the measurement accuracy for the area with a radius of 1.6 µm was estimated as 0.8 °C. perceived again a 'square wave' for the point observed at a certain distance from the heating spot for the cyclic switching on and off of the heating laser. Relative intensity was obtained as a function of temperature increase and a sensitivity of 2.7 %/°C was noted, which is much higher than 1.8 %/°C for the Rhodamine B. Moreover the thermal imaging (using 561 nm laser as excitation source) was also carried out and a decrease in emission intensity was visible when switching the heating on. Then measurements were carried out also on other cell lines: 3T3, HeLa, C2C13, Chang and mESC and a sensitivity of 2.5-2.8 %/°C was obtained. In contrast, a lower sensitivity of 2.0 %/°C was obtained for brown adipocytes. This has been ascribed to differences in pH, viscosity, the presence of oxygen, etc., which actually limits their use. Moreover, calibration curves were not possible to get, because introducing the Mito-thermo-yellow into the mitochondria requires active transportation and in fixed cells the sensor was leaking out of it. Although some qualitative changes in intensity have been observed for cycling heating and cooling in multi-cellular spheroidal HeLa cells, the previously mentioned drawbacks did not provide sufficiently reproducible quantitative results.
A method based on an absolute intensity of fluorescence has also been involved in thermosensitive copper nanoclustes (CuNCs) spontaneously biosynthesized in cancer cells after injecting and 24 h incubating the copper precursors directly into cells 77  Based on the promising results Kriszt et. al. 78 in 2018 continued the research on molecules selected from their DOFL database 74,76 in order to determine material highly sensitive to thermal changes. In this case, the ERthermAC thermosensitive dye (based on BODIPY) was described, which was created by modifying the previous ER thermo yellow sensor (that was able to target the endoplasmic reticulum (ER)) to provide greater photostability. Using the organelle tracking ER Tracker green, it was confirmed that ERthermAC was localizing in the ER region in cells. In WT-1 adipocytes cells (cellular uptake during 30 min incubation at 37 °C), a calibration curve was created for normalized intensity as a function of the measurement temperature in the 18-43 °C range. It has been found that as the temperature was increasing, the intensity of the sensor was decreasing (the same course of the curve was obtained while reducing the temperature). For the 18.1-35.0 °C range a linear decrease of the intensity by -1.07%/°C was obtained, while for the range 35.7-42.8 °C a higher slope of 4.76 %/°C was noted. That work focused on brown adipocytes and demonstrated the heat production inside of them. To that end, isoproterenol (ISO) was added and changes in the intensity of the sensor emission in the cells were imaged. Following the ISO stimulation, most cells showed a reduction in the intensity of ERthermAC or FCCP emission, which corresponded to the increase in temperature. It was also shown that physiological concentrations of Ca 2+ ions (10-1000 μM) do not affect the ERthermoAC activity, while the pH range of 7.0-8.1 changes its intensity to a small extent (-8.5%) compared to thermally activated intensity changes.
The green synthesis process of nitrogen-doped carbon dots (N-CQDs) for applications as a sensor of Fe 3+ ions presence, cysteine presence and temperature changes in cells was described by Lu et. al. 79 in 2018. The emission of the sensor was tuneable and by varying the excitation wavelength from 315 to 455 nm, the emission peak shifted from 424 to 512 nm. The maximum emission corresponding to the wavelength of 439 nm was excited at 355 nm. In an aqueous solution, it was shown that after the addition of Fe 3+ ions that act as an effective quencher, the emission intensity of the N-CQDs decreased gradually. Temperature changes in HepG2 cells were visualized with confocal laser-scanning microscopy, and at 20 °C the emission intensity was higher than at 40 °C, which proved the potential of N-CQDs as intracellular temperature sensors.
The nitrogen doped carbon dots were studied also by Y. Yang et al. who presented first implementation of absolute emission intensity based temperature-sensitive probes not only for in vitro but also in vivo imaging. The preliminary studies enabled temperature mapping in Hela cells cultures 80 . Owing to the excitation with a 400 nm beam, a strong blue emission was observed, the maximum of which was located at 475 nm. After heating the cells, the intensity of the luminescence of this single band was analysed, and the intensity at 20 °C was assumed as the reference intensity I o . As can be seen in the Figure 7, the strong blue emission observed at 25 °C decreased when the cells were heated to 37 °C. After the following cooling, the intensity again reached the initial value, thus confirming the reversibility and repeatability of the process. The dependence of the luminescence intensity on the temperature was assigned to the presence of surface functional groups (-C = 0, -NH 2 , -OH etc.) and the breaking of the hydrogen bonds at higher temperatures that was affecting the hydrodynamic radius of the particles. There were numerous advantage of this material found, such as independence of I/I o from the concentration of N-CDs, strong stable emission at various pH (range from 1 to 9), for various ionic forces (unchanged even in NaCl solution with 2mol/L concentration) and the emissive properties stable in time (negligible changes after 8 months). Notwithstanding, carbon is known to be a highly absorbing material for which some of the absorbed energy can be converted into heat, which could affect the obtained results. Temperatures were assigned to a given luminescence intensity based on the heater parameters, however, it could be raised as a result of absorption and conversion of light into heat by carbon dots. The further step was to imprement those N-CDs also for in vivo imaging 80 . The nanoparticles were inserted into the back of the mice and the luminescence intensity of single band as a function of temperature was studied. The results at 3 different temperatures are presented in Figure 8. There was no unambiguous information on how the given temperatures were achieved in the work. Probably, the carbon dots have been heated to a given temperature outside the living body and then injected. The signal in the spectral range of 500-700 nm with a step of 50 nm was collected immediately after the injection, however, this procedure did not allow reliable and quantitative indication of the temperature.
Furthermore, the sample autofluorescence influence was not considered at all, while the photoexcitation wavelength of 400 nm is known to be strongly absorbed or scattered in the cells, reducing the penetration depth. The effects after treatment remained unknown, as the effects of the study, such as the death of mice, damage to organs or scars, have not been described. As indicated by the authors, the main advantage of the measurements based on the intensity of a single band originates from a simple measurement setup with no need to switch between emission filters. with a single 808 nm wavelength, was followed by a non-radiative depopulation resulting in efficient light to heat conversion (~ 43%) and a radiative transition evidenced by the emission band appearing in the area of the second optical window, which allowed luminescent imaging at a greater penetration depth without the autofluorescence from surrounding tissues. For this purpose, the emission intensity at 1270 nm was examined as a function of temperature (Figure 9). The intensity of the temperature dependent emission band was strong enough to be detected with the NIR InGaAs camera even at a low concentration of QDs, and at relatively low irradiation power density. With increasing temperature, the intensity decreased linearly and in a reversible manner. It was checked whether the photothermal heating was not affecting the ability to the temperature reading. The IR emission was monitored twice in 4 minutes cycles after excitation with an 808 nm beam at 2 W/cm 2 . Before and after each cycle, the laser power was reduced by 2 orders of magnitude to observe dots emission without the presence of significant heat. After each cycle, the intensity returned to the initial value which confirmed that the decrease in intensity during the photothermal process was associated only with temperature changes, and not with the physical or chemical degradation of the QDs. The difference between intratumoral and surface temperature as a function of the excitation power density was also examined aiming for hyperthermia of the tumor. As shown in Figure 9c, regardless of the power density, the internal temperature was always lower than the surface temperature, but together with the power density increase, the difference between these temperatures also increased. Such tests were carried out for power density from 0.7 to 2.3 W/cm 2 and 3 ranges were found: (i) below 1.5 W/cm 2 , which allowed to raise the temperature to only 35 °C and merely caused the destruction of the tumor; (ii) between 1.5 W/cm 2 a 2.3 W/cm 2 , the area of effective photothermal therapy was determined. In this compartment, the intratumoral temperature might increase from 35 °C to 50 °C, which overheated the tumor without significant damage to the surrounding cells. The observed skin temperature at the end of

B. Spectral shift of emission band or absorption edge
The spectral shift of emission bands (Figure 2b) or absorption edge (Figure 2i) is relatively complex to the image, since usually high resolution hyper-spectral imaging is required to quantify sub 10 nm shifts per 1 o C. Acousto-optical tuneable filters or 32 channel PMT, which can be used for wide field or confocal raster scanned microscopy respectively, do not offer sufficient spectral resolution. Because the recorded emission wavelengths are of the order of 500 nm to around 1500 nm (the denominator in S R equation), while both spectral shifts and vibrationally induced broadening usually cover only a few nm (the nominator in S R ), the relative sensitivity using spectral shift is usually limited below 1%K -1 .
Moreover, from practical perspective, the nanomaterials showing this effect are almost exclusively semiconductor quantum dots 84 whose emission spectra are relatively broad, thus hinder the determination of exact wavelength of the emission band maximum. While some indirect methods may be proposed to quantify d/dT, such as the one presented in Fig.2b, the emission integrals at two closely located spectral bands must be quantified. This non-trivial task is affected by the background signal or emission band shape. The former occurs when LT photoexcitation directly initiates autofluorescence from other endogenous and exogenous fluorophores. The latter may happen, when the LT luminescence is absorbed by these chromophores, water or other non-fluorescent molecules such as haemoglobin (inner filter effect -IFE). In consequence, not only the readout will depend on the depths the temperature dependent luminescence comes from, but also may vary proportionally to the skin pigmentation and thickness or tissue composition (e.g. in response to periodic variability of oxy-to deoxy-genated blood).
Similar issues are valid for either temperature shifting the emission band or shifting the absorption edge.
The thermometric parameter using relative spectra shift technique (SST) can be generally defined as: where  max (T) is the wavelength at which either emission or absorption band is the highest. One needs relatively narrow bands to reliably determine the  directly. In imaging mode the  (T) can be quantified as the ratio between emission intensities in two fixed, neighbour spectral bands (B 1 and B 2 ), by using a sharp edge dichroic or bandpass filters and a camera suitable to record 2D intensity images at the two spectral channels I B1 (T) and I B2 (T) under single excitation line (Eq.B1). The artefacts can originate from the presence of autofluorescence in spectral windows defined by the spectral filter used for imaging, i.e.
A B1 and A B2 . Even if the autofluorescence will be assumed to not vary with temperature or time, its intensity affects the outcome temperature maps (Eq.B2). If the relative change over the studied temperature range is constant, e.g.S R = 0.5 %/°C, knowing initial temperature T 0 and assuming homogenous excitation illumination and non-moving objects, one may determine spatial temperature maps as:

I x y T A x y T x y T x y C I x y T A x y S
Similarly to intensity based temperature mapping, the initial temperature map is actually difficult to know before the experiment, thus the temperature rise in respect to the starting conditions can be only found, in response to some stimulus (e.g. drugs, toxicants, local heating etc.). Therefore, the relative temperature change can only be quantified: Assuming of non-existent autofluorescence (e.g. by using anti-Stokes luminescence reporters) cancels the A Bn terms. Using very bright reporters (I Bn >> A Bn , n=1,2) diminishes A Bn terms (A Bn →0) as well.In consequence, these two methods simplify the equation Eq.B3. The spectral shift method is additionally susceptible to the spectral width of the emission bands (i.e. the broader the emission band, the smaller sensitivity), a reflection of the filters (i.e. to avoid spectral bleed-through) and steepness of the transmission band of the chosen filters.
The temperature measurement method based on the spectral shift of the emission wavelength as a function of temperature is best known in the case of the quantum dots. This phenomenon occurs due to a number of different effects, for example due to temperature dependent changes in the band gap energy, quantum effects, electron phonon coupling (DOS(T)), a variation of the quantum yield, thermal expansion of crystalline lattice (r(T)), and change of the solvent refractive index. The spectral shift in quantum dots used for thermal measurements are described by Varshni empirical law 85 : The use of previously widely described materials such as CdSe QDs allows the use of their well-known emission redshift with increasing temperature. Moreover, the use of QDs provides known synthesis methods, uniform shapes and sizes, and the possibility of an easy bio-modification of their surface.
One of the first demonstrations of suitability of this method for temperature estimation in the in vitro environment was demonstrated in 2010 by Maestro et al. 86 , who used two-photon fluorescence microscopy of CdSe quantum dots (CdSe spherical nanoparticles of 4 nm in diameter from Invitrogen Inc. (Qtracker 655)) to measure temperature in the HeLa cervical cancer cells. This excitation allowed for a much higher sensitivity than for one-photon excitation. CdSe QDs allowed to measure the temperature in two different ways, because both the emission intensity and the position of their emission peak were strongly temperature dependent. The latter fact enabled the temperature measurement regardless of whether QDs were distributed evenly or not in the imaged region. The variation of 0.16 nm/°C was determined for the spectral shift (based on the calibration curves obtained in the PBS). To incorporate QDs into the cells, a 2-hour incubation was carried out, however the distribution inside the cells turned out to be inhomogeneous. Fortunately, the temperature dependence of the spectral shift of the emission peak was independent of the QDs distribution. The temperature assessed on the basis of the spectral shift proved to be constant for the cell area. Therefore, the cell temperature was changed externally via a microair-heater and an increase in temperature was observed from 25 ° C to 50 ° C after 3 minutes of heating. Despite the uneven distribution of these nanothermometers in the cell, temperature mapping in this case was inaccurate, but showed the potential of quantum dot labels for temperature measurement in the cellular environment.
Another report on the use of commercially available quantum dots (Qtracker 655 Cell Labeling Kit (Q25021MP, Invitrogen)) for measuring cell temperature was presented by Tanimoto et al. 87 .
Although the described measurement was claimed to be ratiometric, actually it was based on temperature dependent emission band red shift (0.105 nm/°C). Therefore, achieving a detection resolution of around 1 °C would require a very high spectral resolution. Instead, the authors proposed to measure the integrated intensity above and below a certain defined wavelength and the change in their ratio as a function of temperature was studied in a confocal laser scanning microscope. The authors emphasized that the use of such methodology allowed to eliminate errors, resulting for example from photobleaching (a special fitting curve was used). The temperature was additionally monitored with a thermocouple and Then, a similar measurement was made for a single QD introduced by incubation into SH-SY5Y living cells. In this case, the relationship between the ratio of the intensity and the temperature was also obtained and the slope of the curve was determined as 0.067/°C (i.e. no effect of the cytoplasmic environment was observed). Additionally, the authors showed the heat production by mitochondria after the addition of 10 µM CCCP (carbonyl cyanide 3-chlorophenylhydrazone) reagent. To locate the mitochondria, cells were labelled with MitoTracker Green FM (M7514, Molecular Probes) and Hoechst 33342. The intensity of a single QD emission near the mitochondrion was measured every 30 seconds after the addition of the reagent and an increase in temperature was recorded. The authors also described a method of imaging temperature differences in the cell body and in the nucleus. The QDs were excited by a 405 nm or a 488 nm lasers, and the images were captured by PMTs through appropriate dichroic mirrors and emission filters. A 0.1 difference in intensity ratio for the nucleus and cell body was found, which corresponded to the temperature difference of 1.6 °C. The maximum relative sensitivity obtained for the described method was 6.3%/ °C .
A new approach allowing for temperature assessment via testing band shift without the need for nanothermometer calibration was proposed for the first time in 2019 by Savchuk et al. 88 who involved commercially available green fluorescent proteins (GFPs). This approach makes the measurement independent of the total intensity of the sample emission. The GFP shows emission in the green spectral range, and by defining two ranges: below the maximum (I 1 being defined as the integral intensity between 495 and 509 nm imaged by the green channel) and above the maximum (I 2 being defined as the integral intensity between 510 and 600 nm imaged by the red channel), they can be observed simultaneously and a reliable parameter depending on the temperature may be defined as PF = (I 2 - Normalizing this difference to the total intensity of the band makes temperature measurement more accurate and independent from photobleaching and other undesirable effects. Measurements were also made on HeLa cells using GFP located at the mitochondria (CellLight BacMam 2.0 Mitochondria-emGFP). Importantly, the impact of pH and ionic strength (by the addition of KCl) was verified in fixated cells, which showed that the GFP used was also independent of changes in these parameters in living cells that may occur due to some biological processes. It was demonstrated that for GFP inside HeLa cells, the shift of emission maximum from 509 nm for 20 °C to 512 nm for 35 °C was observed.
Thermal uncertainty δT=0.25 °C and a relative sensitivity of S r =2.2%/°C was reached. Furthermore, imaging of cells in which heat production occurred under the influence of FCCP uncoupler reagent was also provided, proving that the described method is capable to trace the evolution of temperature growth by around ∆T=3 °C after the chemical stimulus. At the same time, the method was able to evidence gradients of temperature, pointing out temperature heterogeneity. It was noted that the mitochondrial regions closer to the cell nucleus were warmer. The described nanothermometer shows a significant sensitivity of temperature readout in the biological range and enabled temperature imaging with subcellular spatial resolution.

C. Relative emission bands intensities under single excitation
Ratiometric emission LTs are considered to be the most reliable temperature probes. This is because the normalization of thermometric parameter i.e. the intensity at one emission band is related to the intensity of the other emission band, and in principle this ratio shall be independent of numerous experimental conditions, such as excitation intensity or variations in transmission at the path of excitation and luminescence. This is typically true for most in vitro studies, because up-right microscopes make the photoexcitation and luminescence penetrate in a predictable short paths, thus photoexcitation intensity can be controlled. Unfortunately, the in vivo experiments are not always reliable enough, due to the inhomogeneous composition and more complex tissue spectroscopy. Not only heterogeneous tissues scatter light in an unpredictable manner, but also numerous chromophores absorb or emit their own light under direct photoexcitation or indirectlyi.e. being excited by LTs luminescence. Additionally, spectral properties of tissues may dynamically change as a result of oxyand deoxygenated blood pulses or even with temperature itself 66,89,90 . Other issues, such as skin pigmentation, fat content, vasculature, composition, homogeneity may additionally impair reliable temperature mapping, especially because the calibration of LTs in vivo cannot, in principle, be performed.
From the perspective of temperature detection and mapping, the ratiometric technique is relatively simple to implement either in vitro or in vivo. For in vitro T mapping, wide field fluorescence microscope with appropriate switchable filter can be used, followed by image processing, i.e. colocalizing pixels and dividing one image by the other to get. Using dichroic, sharp edge optical filter enables to employ even small (anti)Stokes shifts to determine  More advanced (but more costly as well) solution is using hyperspectral tuneable filters or a 3CCD cameras. On one hand, such alternatives neglect any mechanical and time-ineffective switching between the filter cubes in the microscope, on the other hand, the spectral filtering cannot prevent some spectral bleeding of signal between the channels and is affecting the sensitivity of ratiometric sensing. Confocal raster scanning with 32 Channels PMT is a rather expensive solution with limited spectral resolution sufficient for broadband emitting LTs (e.g. organic dyes), but it may not be satisfactory for narrowband LTs (e.g. QDs, Ln 3+ , TM). Nipkow disks may improve the imaging speed in confocal highly resolvable imaging, but then at least two spectral images must be acquired for ratiometric detection. The in vivo ratiometric temperature mapping exploits either tuneable filters or switchable static filters. The most severe drawback of this approach is the fact, that one is usually not aware of how the spectral properties of tissues affect the , and how reliably the temperature is determined.
To define thermometric parameter using the relative intensity of two emission bands I Bn (T,  exc ) (n=1,2) under a single excitation line ( exc ) is relatively simple, when sample autofluorescence A is absent (i.e. due to anti-Stokes emission or due to negligible intensity):  I T  I  T  A  I  T  T C  C  I T  I  T One may determine spatial temperature maps as: By using simple mathematical recasting of  parameter, Brites et.al demonstrated the so-called primary thermometer 69 , which eliminates the need for recalibration of the temperature sensitive system, due to different absolute behavior of the temperature probe (e.g. owing to different medium or intensity of excitation). Such an approach, presented in chapter 5, which is critical for the reliability of T readout and mapping, is rarely discussed in scientific literature, but knowing the mechanism standing behind the temperature dependent relationship, enables to define other primary thermometers.
In 2011 Ye et. al. 91 used a ratiometric temperature sensor based on semiconductor polymer dots (Pdots) and an attached thermo-sensitive dye, rhodamine B (RhB). To increase the sensitivity of the temperature readout, the Pdot-RhB nanoparticles could be excited with a single excitation line, using an efficient FRET mechanism from the Pdot to the RhB (Figure 11. Left a, b). The use of the Pdots enabled an easy manipulation of the sensor size and ensured very high emission brightness. The rhodamine B is a dye whose emission intensity is strongly quenched by temperature, which provides the possibility of   Figure 12a). The sensitivity obtained from the curve for a ratiometric measurement (Figure 12b) was lower than 1 °C for the range 34.2-40.5 °C (Fig. 12c) and it reached 0.40 °C at 37 °C. What is more, the authors noted the expected increase in cell temperature after the addition of 100 mM glucose (Fig. 12a, last column).  Because the emission of CDs changed insignificantly as a function of temperature, the intensity ratio of 430/605 nm increased with increasing temperature and the ultimate emission color of the nanohybrids changed from red to violet. No spectral shifts of the emission peaks were observed in the studied temperature range. The usefulness of nanohybrids for temperature measurement was confirmed additionally by measurements of repeatability (several cycles between 20 and 60 °C), and insensitivity to pH (range 4-9), ionic strength, presence of biomolecules as well as photostability was confirmed under exposure to the Xe lamp (500W, 150 min). Cytotoxicity was also tested for 293T cells after 10 minutes lasting incubation and low toxicity of nanohybrids was observed. Cells were imaged at different temperatures using a confocal fluorescence microscope and two emission channels were observed: 415-544 nm and 580-620 nm. With the temperature increase, a decrease in the intensity of the red channel was observed, while the blue channel remained unchanged. Therefore a change in the intensity ratio corresponding to the changes in temperature was observed, which proved that the nanohybrids described have the potential to image changes in the temperature of living cells.
Gold nanoclusters were utilized also by Wu et al. 99   In 2017, a DNA nanomachine acting as a ratiometric LT for temperature measurements in living cells was described by Xie et. al. 103 . This LT was of tetrahedron shape and one of its edges was made of a temperature-sensitive molecular beacon (MB) that could shrink or stretch under temperature changes. Below a certain characteristic temperature, the MB was forming a closed structure folded into a loop, so the two faces of the tetrahedron joined by the MB were close to each other. Above this temperature, the DNA tetrahedron became expansive because the MB was unfolding and straightening.
The prepared DNA machine was resistant to the nuclease degradation and could be introduced into the cells without additional reagents. To obtain a sensor with ratiometric luminescent properties, two dyes, FAM and TAMRA, which acted as donor and acceptor, respectively in the FRET process, were attached to the two moving tips of the tetrahedron. Therefore, by changing the temperature, the distance between the dyes could be modified, which enabled the FRET (red emission) at low temperature and prevented  and IR luminescence could be obtained. By selecting a sufficiently high concentration and excitation with a beam of 808 nm (4W/cm 2 ), the simultaneous heating and luminescence in the spectral range of the optical transparency window was achieved. As shown in Figure 16a) the luminescence intensity ratio of two spectrally close localized emission bands at 865 nm and 885 nm of Nd 3+ :LaF 3 NPs was analyzed as a temperature dependent parameter. Unluckily, the close spectral vicinity of these bands strongly hindered the temperature readout. However, the authors showed that in the biological temperature range, the I R1 /I R2 (865 nm/885 nm) ratio revealed a linear dependence on the temperature (Figure 16b). It was shown (Figure 16c) that the value of the relative sensitivity, which reached 0.25 %/°C was only slightly dependent on the type of media used (such as aqueous media, powders and different types of tissues). In order to verify the usefulness of the nanoparticles for in vivo applications, the tests were carried out, confirming the lack of their toxicity and stability (Z-potential of LaF 3 nanoparticles: 5.6% Nd 3+ was -23.7mV). Human breast cancer cells were injected into a mouse (one tumour on both sides) and about two weeks after the injection, when it reached 9mm 3 in size, photothermal therapy was initiated. The LaF 3 :5.6%Nd 3+ nanoparticles dispersed in PBS at a concentration of 10% tumor weight (which is about 7·10 13 nanoparticles per tumour) were introduced into the area of one tumor, while the other tumor served as a reference (Figure 17a). The infrared luminescence images of the mouse, which confirm the ability of these NPs to subcutaneous luminescence imaging are presented in Figure 17b.
As can be seen in Figure 17c, the thermal images verify the capability of LaF 3 :5.6%Nd 3+ NPs to generate heat. Simultaneously with the intratumoral temperature measurements, the surface temperature of the skin was estimated as a function of the irradiation time during 4 minutes-long therapy (Figure 17d).
During the whole cycle, the temperature inside the tumor was c.a. 20% higher than the surface temperature. This was explained by a decrease in temperature as a result of heat exchange between the tissue and the air, which did not occur in the case of heat diffusion within tissues volume. As can be seen in Figure 17d). a similar trend as a function of time was also manifested by the changes in the temperature of a control tumor that has undergone the same exposure procedure. The temperature increase as a result of the residual tissue absorption still existed, but it was less than 50% of the heating induced by Nd 3+ :LaF 3 NP in the tumour subjected to photothermal therapy. Other multifunctional nanoparticles designed for thermal imaging based on the emission intensity ratio were also analysed by X. Zhu et al 106  of this thermometer reached 1%/°C. Therefore, the thermal imaging with 0.5 °C thermal resolution and 0.9 μm spatial resolution was achieved. The goal of the authors was to optimize the radiation dose needed for the photothermal in vivo therapy using a microscopic temperature measurement system that differs from the macroscopic approach. The external temperature was analysed using a thermal imaging camera, whereas the actual temperature in the area of csUCNP@C nanoparticles after 730 nm excitation was determined by means of their emission spectra. The measurements were made for two power densities: 0.8 W/cm 2 , which led to overheating, and 0.3 W/cm 2 causing a subtle, but sufficient to cause hyperthermia, temperature increase of nanoparticles. HeLa cells (107 cells Such mice after 40 days of treatment were still alive, whereas in control animals (either no treatment, or mice exposed to 730nm beam but without nanoparticles or mice injected with nanoparticles, but with no laser irradiation) a continuous increase in tumor size and ultimately the death of the mice after about 22 days was noted.
Unfortunately, this promising approach, had also several disadvantages. The system used two different wavelengths to simultaneously excite luminescence and generate heat, which not only complicated the experimental setup, but also caused some thermal reading uncertainty, because the penetration depths of both excitation wavelengths in the irradiated tissue were not equal. Moreover, the carbon layer could also absorb the 980 nm excitation line and radiation emitted by the core of the nanoparticle. In addition, the bands used to determine the temperature were spectrally close to each other, which could generate a problem related to filters and spectral resolution, which also limited the usefulness of this method.
The aim of the study of Miyagawa was to visualize the thermal changes of the muscle of Dicronorrhina Derbyana beetle during its work 107 . For this purpose luminescent nanoparticles were embedded in the ultrathin, elastic and simple polymer nanosheets with a large contact surface, high adhesion and a small thermal capacity. A nanopolymer on which two separated layers were deposited, was created. The first layer was formed by highly temperature-dependent EuDT and the other one by less temperature sensitive Rhodamine 800 dye. Owing to the confinement of EuDT and Rhodamine in two separated nanosheets, the energy transfer between them was negligible. The system was equipped with two excitation wavelengths: 405 nm and 640 nm, and the temperature was determined based on the EuDT emission intensity ratio at 620 nm compared to 700 nm emission from Rodamine. The emission intensity of EuDT-NS decreased practically linearly with the increasing temperature, whereas the intensity of the Rho-NS was almost temperature independent. The sensitivity to temperature changes in EuDT originated from the fact that the energy transfer rate between Eu (III) and the ligand molecule (βdiketonate) was varying depending on the temperature. It is worth pointing out, that due to the absorption (640 nm) and emission (700 nm) by Rodamine 800 in the red visible range and strong emission of EuDT, the contribution of autofluorescence was relatively small. In the studies of the temperature changes caused by the muscle work, the heat production was observed in the so-called "preflight preparation" or "escape mode", when the flight muscles were activated and warmed up. Such a mode was induced by a mechanical stimulation, e.g. by gently touching the legs of the beetle with a stick. When the beetle was not stimulated, the muscles underwent a natural cooling process. Several beetle stimulation cycles were performed and the result of one of them is visible in Figure 18. In the second image, the temperature distribution due to the activation and warming up of the muscles after the stimulation of the beetle for the preflight preparation was noted. The last view shows the temperature of the muscle during rest, and thus the natural cooling of the muscle. The change of the muscle temperature during such a cycle was also filmed. At 37 °C, a very high relative temperature sensitivity S = 5.26 %/°C was obtained, and as the size of the analyzed area was reduced, the temperature resolution was increased but the maximal value was around 0.75 °C.
Unfortunately, this method only allowed surface imaging.
In order to eliminate numerous experimental obstacles, Wu et al. optimized the phosphors used in the fluorescent short single-stranded DNA (ssDNA) based nanothermometer for in vivo imaging applications 104 . To eliminate the problem of the absorption and scattering of FAM and TAMRA phosphors on cells, an ROX donor and ATTO 647N acceptor were used to shift the emission bands toward infrared direction (donor 610 nm, acceptor 665 nm). The ability to modulate the emission characteristics by choosing the appropriate phosphors between which FRET occurs, so that their emission maxima appear at a desirable spectral range is an undeniable advantage of such a thermometer.
The usefulness of L-ROX-12R1-ATTO for an in vivo application has been demonstrated in mice. The 5·10 6 PC-3 cells (prostate cancer) were injected subcutaneously (3.5 mm in depth) into both legs of the mouse and an imaging experiment was performed 24 days after the injection. Additionally, the DNA L-ROX-12R1-ATTO structure was injected only into the right leg and after 5 minutes from the injection, the mouse was exposed to 40 W microwave radiation to heat the tumor. The Figure 19 showed the distribution of the emission changes of each of the phosphors before (situation a and b) and after (c and d) irradiation. As the temperature increased, the emission intensity of the acceptor decreased significantly, while the intensity of the donor emission increased only slightly. Therefore, an increase in the D/A ratio with the temperature was observed in the area of the biological temperatures (20-50 °C).  To demonstrate the effects of thermal imaging in a living organism, one of the requirements was the use of a water-soluble nanothermometer. Chen et al. accomplished this task using the iridium (III) complexes attached to a thermally sensitive, water-soluble polymer backbonePNNPAM, was applied as the thermometer 110 . The dispersibility of this material in water and the solubility of such a polymer solution changed, what was related to the existence or breaking of hydrogen bonds between polymeric amide bonds and water molecules. With the temperature increase, the hydrogen bonds were broken, the water molecules were released. Therefore, the hydrophobic interactions started to play a bigger role, which led to partial aggregation (formation of micelles), decreased polarity of the iridium complexes and increased stiffness. Such a change in the polymer conformation from longitudinal to globular strongly affected the emission intensity and lifetime of the excited state of the iridium complex in this material, which allowed the creation of two types of thermometers for in vivo applications operating in the 16-40 °C temperature range: the thermometer based on the emission intensity ratio of red (570-620 nm) to green (460-510 nm) channels of iridium complexes and luminescence imaging with time resolution, which will be described later in this article (section G). The studies were carried out in the zebrafish larva, which is widely used in research related to cancer, diabetes and genetic cardiovascular diseases, as it is 87% identical at the genetic level to humans. As can be seen in the confocal laser scanning microscopy images shown in Figure 21, after excitation with a 405 nm beam of 100 μW power the luminescence was distributed in the whole of a zebrafish larva, also in the yolk sac and belly. Both the luminescence intensities of green and red channels exhibited only trifling increase with the elevation of temperatures from 22 °C to 28 °C and due to this fact, the luminescence intensity ratio of the green to red channel did not reveal any obvious increase when the temperature increased, which may be related to the effect of the intense autofluorescence. With the rise of the temperature, as a result of the thermal motions of the molecules, these bonds weakened, the solubility in water decreased, and the hydrodynamic radius increased (up to 107 nm at 30 °C). The hydrophobic interactions within P-Ir 3+ -Eu 3+ began to play a more important role and led to such a chain arrangement that the water molecules were pushed out, whereby the polarity of the microenvironment around the Ir 3+ complexes decreased causing the enhancement of the Ir 3+ luminescence intensity. Thus, after increasing the temperature from 20 to 42 °C, the hydrophilic polymer backbone transformed into a hydrophobic type, which resulted in a decrease in the microenvironment polarity. Then, the change in polarity led to a change in the complexity of the Ir 3+ , as well as the intensity of the phosphorescence and the lifetime of the emission. It is worth noting that these processes were fully reversible. A water P-Ir 3+ -Eu 3+ was injected into the heart of a zebrafish. After 2 hours from the injection, the luminescent signal was collected from the heart area, and after 12 hours the polymer was found throughout the whole body confirming good biocompatibility. The temperature was determined based on the intensity ratio of Ir 3+ (470 nm) to Eu 3+ (615 nm). The influence of other factors on LIR was examined and this parameter was shown to be practically independent of the ionic strength (aqueous KCl, CaCl 2 and MgCl 2 ), pH (6-8) and polymer concentration (from 0.06 to 1.2 mg/mL). What is more, it revealed low cytotoxicity even for high polymer concentrations (0.8 mg / L). However the autofluorescence had a strong impact and the differences in the ratio of the green channel (460-540 nm) to the red one (580-630 nm) as a function of temperature were not clear and obvious, as shown in Figure   22. Moreover, the Eu@SiNPs were found to have lysosome-targeting property. The heating and cooling process between 30 and 45 °C was also investigated and simultaneously cells in both channels were imaged. The same calibration curve for both cooling and heating was recorded which was consistent with the set temperatures for the cells.

D. Relative emission band(s) intensities under double excitation
The most reliable methods to image temperature in 2D are believed to be based on the relative measurements, which are expected to take into account and correct for non-homogenous illumination (if only the fluorescent reporters respond in a linear way) or non-homogenous distribution of the temperature reporters (unless aggregates are formed). However, these methods are less reliable, when two spectrally different excitation lines and emission bands are used. The issues originate from the fact that the spectral properties of the tissues or samples (such as light scattering  S and direction g, absorption  A coefficients) are strongly dependent on the wavelength, spectral range and spectral shift between the wavelengths used. Moreover, in the most complicated case, the fact that two images have to be collected, makes the temperature imaging system either fast but costly (when 2 cameras are used for simultaneous recording of images in two spectral windows) or cost-efficient but slow (when a single camera is used with switchable filters).
The relative emission band intensities under double excitation are solving some of the issues mentioned above. Using single emission band intensity is as cost-effective and technically as simple as intensity based temperature determination. However, the reliability is improved by collecting two images which form a ratio. In contrast to another popular technique, which exploits two emission bands under single excitation line (Figure 2, section C), the method presented here sustains ratiometric imaging, but may offer faster frame rate, because it is technically simple and affordable to allelectronically switch between two LED or semiconductor laser sources, in order to record two images by the same optical setupi.e. the same filer set with single camera.
To define thermometric parameter using the relative intensity of single emission band intensity I B (T,  exc ) under two excitation lines ( n exc , n=1,2) is relatively simple, when sample autofluorescence A is absent (i.e. due to anti-Stokes emission or negligible intensity):  I T  I T  A  I T  T C  C  I T  I T  A  I If the relative change over the studied temperature range is constant, e.g. S R = 0.5 %/°C, knowing initial temperature T 0 and assuming homogenous excitation illumination and non-moving objects, one may determine spatial temperature maps as: Again, because the initial temperature map is actually difficult to know before the experiment, the temperature rise in respect to the starting point can be only found, in response to some stimulus (e.g. drug, toxicant, local heating etc.), that is the relative temperature change that can be quantified: Then, the calibration of such function over temperature shall enable to derive actual absolute temperature. An examples of the thermal imaging taking advantage from this approach was demonstrated on a non-invasive genetically encoded tsGFP cell temperature sensor, consisting of a green fluorescent protein (GFP) and a TlpA thermosensitive protein 114 . The coiled-coil TlpA protein was undergoing conformational changes under the influence of temperature at around 37 °C. It was a reversible structural transformation from a parallel coiled-coil dimer to unfolded monomers. These temperature-dependent conformational changes of TlpA were passed to the attached GFPs, under the influence of two excitation peaks of GFP at 400 nm (neutral, phenolic form) and at 480 nm (anionic, phenolate form) which changed their mutual intensity. An excitation ratio of 400 nm to 480 nm for emission at 510 nm was a measurable parameter informing about the temperature of the medium. It is important to note that unfused GFPs showed a smaller change in the 400/480 nm excitation ratio than tsGFP. Moreover, the authors created tsGFPs containing special sequences that allowed targeting selected organelles (plasma membrane, ER and mitochondria). They also showed the dependence of luminescent properties on the temperature, while their measuring ranges (and the temperature corresponding to the conformation change) differed slightly between the organelles, which the authors attributed to differences in temperature at different points in the cell (and not the presence of targeting sequences). What is more, the authors visualized the mitochondrial thermogenesis in HeLa cells after adding the CCCP reagent (Figure 23). Confocal laser-scanning microscope imaging using the 405 nm line of a diode laser and the 488 nm line of an argon laser for excitation and a 505-525 nm band-pass filter for the emission of HeLa cells showed heterogeneity in the excitation ratio of 405/488 nm (although the sensor was evenly distributed in the mitochondria). However, a decrease in the ratio after the addition of CCCP was noted, which is equivalent to depicting the temperature increase in the mitochondrial areas. With the help of an additional dye JC-1 which reacts to the high potential of the mitochondrial membrane, a negative correlation with the excitation ratio 405/488 nm was shown, which suggested that the temperature was higher there, where the potential of the membrane was high. The authors also described the visualization of thermogenesis in brown adipocytes and in skeletal muscle myotubes.  The most recent optical thermometry in vitro imaging, where 2 different excitation lines (at 373 and 436 nm, for btfa-and DNPF-based micelles, respectively) and two different emission bands I 1 (600-620 for Sm 3+ 4 G 5/2 → 6 H 9/2 ) and I 2 (620-660nm for Eu 3+ 5 D 0 → 7 F 2 ), demonstrated maximal relative sensitivities of both samarium and europium temperature reporters to reach 1.5 %K -1 and 1.7%K -1 for the btfa-and DNPD-containing micelles, respectively, both at 328 K. The corresponding minimum temperature uncertainty reached low value 6 mK (at 328 K) and 0.20 K (at 305 K). Although the demonstrated relative thermal sensitivity values were lower than that reported for similar micelles containing Tb 3+ ions 115 (5.8% K -1 at 296 K), the decrease of the relative uncertainty on the integrated intensity enabled to improve and get δ I/I < 0.4% , which is below 10 mK. The temperature maps wer obtained by reconding the wide field fluorescence images and simple division of the two color channels.
These studies evidenced spots, where c.a. 5 K higher temperature was observed as compared to average temperature in the cells. This was explained by the intense RNA transcription activity of the nucleolus ( Figure 24).

E. Relative emission intensity under two excitation bands
The concept of using relative emission intensity from two bands under two different excitation lines is a combination of the two methods described above. Actually, the advantages of those individual techniques are lost, because although switching excitation sources is simple and reliable (as in relative  Another paper describing the ratiometric sensor with 2 excitation lines was presented by Xie et. al. 117 in 2017. The authors desired to thermally visualize the cells of brown adipocytes (BA) and they used for this purpose two components, which were very similar due to their chemical structure, Rhodamine B methyl ester (RhB-ME) and Rhodamine 800 (Rh800). RhB-ME emission was quenched at higher temperatures, while Rh800 emission was insensitive to temperature changes. Both dyes had mitochondrial targeting properties because they were cationic dyes that could be distributed in the cell in response to a negative electrical potential, especially in the mitochondria. The imaging of the dyes in aqueous solution was performed using a confocal microscope with simultaneous excitation with 559 nm and 635 nm and collecting the fluorescence at 575-620 nm and 655-755 nm channels. The visualization of the HeLa cells showed that both dyes were located exactly in the same places (RhB-ME emission was observed in the red channel, Rh800 in the green channel). The ratio of the intensities allowed to observe the temperature distribution in the mitochondria of HeLa cell, with higher temperature inside (which is explained by the geometry of this cell). In order to further evaluate the mito-thermometry, the described method was used to visualize thermogenesis of BA (cells were co-stained with the dyes for 1 h at 33 °C). Imaging of the temperature changes was carried out by studying the ratio of the intensity of Rh800 (green channel, excited at 635 nm, collected at 655-755 nm) to RhB-ME (red channel, excited at 559 nm, collected at 575-620 nm) emissions. Thermal changes were caused by the addition of 0.1 μmol/l NE or 10 μmol/l carbonyl cyanide m-chlorophenylhydrazone (CCCP), however, in the second case the temperature increase was much higher.    focused on gold nanocrystals inside the nematode was turned on after 20 seconds, which can be seen as the rise in the temperature depicted by a blue curve in the Figure 27, while the grey curve shows the result of a control experiment that took place in identical conditions, but for the reference C. elegans without gold nanoparticles. The use of GFP enabled imaging of small changes in temperature distribution and observed laser-induced temperature changes were almost 3 °C. The reversibility of this process was confirmed by several heating and cooling cycles. However, this approach encounters several issues that exclude this method from a real application. Firstly, the sensitivity depended on the medium to which GFP has been introduced (0.4 %/°C for those attached to GAD proteins located in cell bodies, axon branches and synapse regions in nematodes and 0.1 %/°C in water) and proteins to which they were bounded, due to the different rotational time of different compounds. For another complex, new calibration curves have to be determined. Secondly, although the FPA did not depend on the GFP concentration, the photoheating effect depended on the concentration and distribution of the gold nanoparticles. Thirdly, it was a green protein for which the depth of penetration can be disturbed by absorption and scattering on cells or by tissue components, thus a more appropriate solution could be red proteins or their derivatives so that the spectrum of luminescence coincided with the biological window.

G. Luminescence decays
The usage of the luminescence lifetimes for temperature mapping is conceptually straightforward and reliable, and stems from the fact most of the fluorophores demonstrate temperature dependent spectral and kinetic properties. The technique is relatively simple to implement, because the kinetics of a single emission λ emi band is monitored in time after the single excitation wavelength pulse λ exc and Thus, the imaging shall be directly related to the luminescence lifetime image: Despite the conceptual simplicity of this method, the implementation is however hindered by the nature of the luminescence decay and some technical constrains. The simple equation G2 does not take into account the fact, that the luminescence lifetimes are often non-single exponents and that not only the temperature affects the spectral properties of the fluorescent sensors, but also the chemical environment such as viscosity or pH. Moreover, in many cases, especially given e.g. energy transfer interactions, the decay becomes non-exponential, often accompanied with slow risetime and reliable determination of the lifetime becomes difficult. In that case, an agreement on the lifetime to be used has to be met, e.g.
by calculating the average decay time that does not require explicit fitting, i.e. Eq.G3 Further, more simple, rapid lifetime determination algorithms exist, which integrate luminescence intensity in 2 or more time gated windows with variable delay after the pulse 121 . Frequency domain lifetime determination are also known 122 , but they are also not perfectly suited for multi-exponential decays. It is then possible to assess the statistical uncertainty of this luminescence decay value (by calculation of the second moment) and thus, indirectly assess the theoretically expected temperature uncertainty.
The other serious issue related to the fluorescence lifetime based temperature sensing with organic probes is the fact, their fluorescence lifetimes are similar to the lifetimes of other endogenous organic molecules (i.e. collagen, elastin, NADH, FAD etc.), which requires careful interpretation of experimental results. In the case of substantial autofluorescence, the resulting low signal to background signal ratio is a serious issue, unless fluorescent probes are used with significantly longer (e.g. ~150 ns for InCuS 2 QDs, 0.3-0.6 s for iridium complexes, 1-1000 s for lanthanide chelates or Ln 3+ doped nanoparticles) luminescence lifetime. Secondly, the very short (orders of nanoseconds) luminescence lifetimes require very short excitation pulses. Sub-nanosecond pulses from laser diodes may be used, but such light sources, due to limited intensity, are suitable for raster scanned imaging, which, by its nature, makes the imaging rather slow. Whole field lifetime imaging requires much larger energies to illuminate the whole field of view. In consequence, mode-locked femtosecond lasers are typically used, which significantly increase the costs of such experiments. Thirdly, since the nanoseconds time scale kinetics has to be quantified, ultrasensitive imagers have to be used. Electron-multiplied CMOS FLIM and the appropriate organelle markers, the temperature differences between the cytoplasm and some cell elements were noted (Figure 28a, b). Analysis of the data from many cells allowed to estimate the average temperature difference between the cytoplasm and the nucleus as 0.96 °C (Figure 28c) and it was shown that this difference was dependent on the cell cycle. In more than half of the examined cells, an additional warmer point (of about 0.75 °C) identified as centrosome (marked with arrows in Figure 28a) was also noticed. Moreover, the production of heat by the mitochondria was shown. The red emission of MitoTracker Deep Red FM used in the experiment allowed to locate the mitochondria in the cell by luminescence measurements (Figure 28). The FLIM image enabled to state that in these places the temperature is higher (arrows on the lower right Fig. G2a). In addition, the increase in heat production by mitochondria after the addition of FCCP reagent (30 min after addition) has been shown and an increase in temperature of about 1.02 °C has been noted (Figure 28e). Importantly, the additional measurement for the control polymer showed no change after the addition of FCCP (Figure 28f).
Analogous measurements were conducted for HeLa cell line.  To obtain a photon count rate lower than 2% of pulse countirate (2·10 7 ) the laser power, pinhole size and detector sensitivity were carefully controlled. Fluorescence decay curves were fitted with double-exponent in each pixel using SPCImage software (after the binning procedure). FLIM performed at living HeLa cells showed that at higher temperatures of the medium, the lifetimes of the fluorescence were longer. In the next step, the temperature of the medium was maintained at 30 °C and the higher temperature (extended lifetimes of FTP luminescence) of the nucleus was observed than of the cytoplasm for single cells. After the examination of many cells, the average difference of 0.98 °C was found. The heat production near mitochondria, whose positions were found using a mitochondrial indicator MitoTracker Deep Red FM was also detected (arrows in Figure 30a). It was also confirmed that the heat production within mitochondria was increased by the addition of the CCCP reagent. Indeed, longer fluorescent lifetimes were obtained after the addition of the reagent (Figure 30b)   ionomycin. Subsequently, C2C13 myotubes were tested using this thermometry (Figure 31). The authors also managed to monitor changes in heat production by myotubes after the addition of 1 mM caffeine -the luminescence lifetime of ER thermo yellow shortened after adding caffeine and after 10 minutes it returned to its previous state. The temperature sensitivity of the fluorescence lifetime of ER thermo yellow was between -24 and -26 ps per 1 °C in living cells regardless of cell types. Moreover, the thermo yellow insensitivity to changes of pH and ionic strength was found. In addition, calibration curves in live and fixed cells were similar, which suggested the insensitivity to the viscosity of the cellular environment. As mentioned in the previous section, the extensive research on the zebrafish carried out by Chen et al.
included also an analysis of decay times to determine the temperature 110 . Using the fluorescence lifetime approach, the unwanted autofluorescence occurring for LIR based thermometers was eliminated. Figure   32 shows ns (gate 0-150nm), the autofluorescence from the yolk sac and belly of the fish was eliminated. Time resolved luminescence imaging using a polymer with attached iridium complexes, due to a good reversibility during heating-cooling processes, biocompatibility, independence on particle concentration, pH values (4-10) and ionic strength and no autofluorescence effect made them a very promising tool for in vivo research, however it also revealed a disadvantage, namely, due to the lifetime of the order of ns, the pulses on the pico or femtosecond scale are needed for the excitation, therefore, the detection costs increase. The lifetime approach using an acryl-based polymer thermometer with both Ir 3+ and Eu 3+ compounds was also studied 111 . As already mentioned ( chapter E), both the emission intensity and fluorescence lifetime of iridium compounds increased with increasing temperature, while Eu 3+ complex acted as a reference signal, due to the constant emission intensity and lifetime at temperature changes.  To study biological phenomena important for the field of neuroscience in mice brain tissues, Hoshi, Okabe et al. 131 used the fluorescent polymer thermometer (FPT) previously described by the same authors 127 , which is now commercially available as a nanothermometer for measuring intracellular temperature using FLIM from Funakoshi (Ultrasensitive Fluorescent Thermometer for cellular temperature monitoring -Diffusive Thermoprobe). The phenomenon of brain edema, during which the volume of the brain increases because the water content increases, has been studied. Ischemic edema was studied in model mice brain samples treated with oxygen-glucose deprivation (OSD). Edema pathology has been shown to be associated with thermosensitive cation channel TRPV4 (transient receptor potential vanilloid 4) activation, which is enhanced after elevated brain temperature, as demonstrated by tissue temperature imaging using FPT (Figure 34A and B). OGD is associated with the release of glutamate also by the activation of glutamate receptors and TRPV4 receptors. Therefore, it has been checked that pharmacological blocking of glutamate receptors suppresses OGD-related temperature rise( Figure 34C). It was also confirmed that TRPV4 was involved in edema by testing in mice that lacked it. Thus, hyperthermia has been shown to cause TRPV4 activation and induces brain edema after ischemia. These studies can contribute to the development of ischemia intervention techniques that can help fight brain edema. Figure 34. The oxygen-glucose deprivation (OGD) treatment raises the intracellular temperature of brain slices via glutamatergic signaling. A: Representative images of control (Aa-Ac) and OGD-treated (Ad-Af) brain slices when the temperature of the perfusate was set at 30°C. Scale bars, 50 μm; B: The temperature estimated by fluorescence lifetime of FPT was shown as difference between the temperature of each sample and the average of temperatures of control samples (estimated Δ temperature, i.e., T-<T Control >). The estimated Δ temperature was significantly higher in the OGD group compared with control. The average difference of calculated intracellular temperature was 2.12°C; C: The OGD treatment significantly increased the estimated temperature of brain slices, a phenomenon blocked by co-applying the glutamate receptor antagonists CNQX (50 μm) and AP-5 (50 μm) 131

I. Absorption edge shift
Temperature affects the absorption and emission properties of many luminescent nanomaterials. The temperature dependent changes in the absorption coefficient or the absorption edge shift will definitely translate to changes in luminescent propertieseither luminescence intensity or spectra. These changes differ between particular materials and various mechanisms can be ascribed.
In lanthanides the narrowband absorption bands become thermally broadened due to electronphonon coupling and temperature induced Stark components population. However the strongest various are observed in cryogenic temperatures. Temperature dependent charge-transfer CTB edge shifts to longer wavelength, leading to enhanced excitation efficiency at a certain wavelength in the tail of the CTB. Under appropriate excitation wavelength, the luminescence intensity of the emitting ions (e.g. lanthanide or transition metals) will increase rapidly with temperature increasing. Temperature dependent charge-transfer between vanadate (VO 4 3− ) group and Sm 3+ or Er 3+ lanthanide has been shown to be useful for temperature sensing with reasonable >2%/ o C relative temperature sensitivity 133 134 .
Because the CT bands are stronger than absorption bands of lanthanides themselves, this is a promising direction for achieving bright and sensitive thermometers. Alternatively, the temperature dependence of two intervalence charge transfer states (CTS) between Tb 3+ , Pr 3+ in NaGd(MoO 4 ) 2 were proposed and a maximal S R value of 2.05%/ o C was obtained at 403 K. 135 . A serious disadvantage is the fact, the CT bands are located in UV spectral region, which is not compatible with biomedical requirements, neither in terms of light absorption/scattering and safety reasons.
In transition metal complexes a change in electron distribution between the metal and a ligand gives rise to charge transfer (CT) bands, which is known to strongly depend not only on temperature, Although there are experimental studies on using the above mentioned effects for luminescence temperature determination, we are not aware of biomedical imaging using the spectral shifts of absorption edge of luminescent reporters. The fundamental problem with such methods originates from the fact, that usually short wavelength absorption of these fluorophores overlap strongly with the absorption of various endogenous dyes and molecules. Moreover, short wavelength light is particularly susceptible to light scattering in highly heterogeneous tissues and cells, which hinders proper calibration and forces to perform this unavoidable step specifically for every sample to be studied.

Summary
Below, a summary of existing temperature imaging studies is presented for in vitro (Table 3) and in vivo (Table 4) experiments. Reading method, sensitivities, artefacts and issues, heating methods and additional information are provided in a comparative way.  GFP can be bounded with selected enzymes found in only one cell type; the concentration of GFP used for imaging doesn't affect the measurement and the FPA is insensitive to fluctuations in the intensity of the excitation laser, photobleaching and sample migration; allows the imaging of thermal changes of individual neutrons

Outlook and perspectives
Although temperature imaging is a well-established technology and advanced T imagers (i.e. bolometric cameras) are commercially available, this technology is only suitable to image temperature of the surface of objects. Such technology is sufficient for industrial (e.g. thermal hot spots detection in electronic circuits or mechanical devices, thermomodernisation of buildings etc.), military or security (e.g. noctovision, border inspection) and limited medical applications (e.g. diabetes foot monitoring, breast cancer screening, remote fever detection on airports). The principle behind bolometric camera operation relay on using 7-15 μm emission and black-body radiation law. Due to spectral detection range, neither below skin surface nor sub-cellular optical resolution (using optical microscopes) temperature mapping is nevertheless possible with present technologies. In the view of existing needs created by novel therapies (i.e. hyperthermia of cancer, targeted therapies, temperature induced drug release) and to understand fundamental processes in biology (like thermogenesis, enzyme activity), in this review, we focused on summarising current knowledge about remote imaging (i.e. 2D mapping) of temperature in biology and medicine using luminescent thermometry. Despite some examples of experimental biomedical T imaging can be found, most of the current research activities are dedicated to develop reliable T sensors using fluorescent molecules, proteins, polymers and numerous inorganic nanoparticles. However, understanding how they work, how their spectroscopic properties depend on microenvironment (pH, viscosity, pressure etc.) and how sensitive they are, teaches, that performing reliable absolute temperature determination in 2D space is not a trivial task. The influence of the environment on T accuracy determination in not negligible, as not only chemical but also spectral properties of biological samples may affect spectral properties of the LNTs owing to the highly heterogeneous composition of the biological samples, sample autofluorescence and light scattering, re-absorption by the sample and by the LT probes themselves. Such studies definitely require further attention and work.
As more and more reliable luminescence based temperature (nano)reporters are demonstrated and the luminescent (nano)thermometry matures, a transition from purely materials science and chemistry research, to application driven technology requires comprehensive understanding and ranking of various materials but also, numerous reading methods have to be evaluated from various perspectives. Obviously, not only thermal relative sensitivity is a single figure of merit parameter, but other factors, such as the simplicity and cost-effectiveness, spatial and temporal resolution, 2D and 3D T determination capability in-vivo/in vitro, background and artifact free detection, specificity of T detection, simple data correction and analysis must be taken into account, when new versatile technology is to be developed.
The comparative analysis presented in this review, which considers the advantages and drawbacks of each of the particular nine temperature readout approaches and thermographic material used, enables to enumerate the features of the model-based luminescent thermometers.
Temperature variable luminescence intensity is the most simple approach to temperature determination, but relative temperature changes may only the detected, and numerous artifacts, such as excitation intensity at the LT probes location, sample spectral properties on the way for excitation source to the LT probes and from LT probes to the detector may change to the absolute temperature being derived. Spectral band shift or band-width are relatively simple to implement for visualisation, both with spectral detectors or 2D cameras, but this detection method is susceptible to some artifacts which disturb the pure LT probes luminescence spectra shape, such as sample autofluorescence or LT probes emission reabsorption, thus extensive data correction may be necessary to get reliable absolute temperatures. Ratiometric luminescence approach was originally considered as most reliable, but in the course of the studies it is now obvious the self-reabsorption and interaction with the sample absorption or scattering may significantly change the outcome of the readings. Primary thermometry in this case is of some help, and selection of emission bands which are spectrally close to each other but all these new rules, change the paradigms and require more detailed studies. Single emission ratiometric excitation gives some hope as a new temperature determination mechanism, in which only one emission band is studied under two excitation wavelengths. Such an approach allows to speed up and simplify optical setups to acquire temperature maps, offers also Stokes emission (unlike many anti-Stokes upconverting nanothermomters) and positive temperature sensing coefficients (i.e. rising temperature increases the brightness). The artifacts, that however must be considered originate from the spectral transmittance of the sample at two distinct excitation wavelengths. From the presented analysis it can be concluded that the most reliable technique of remote temperature sensing, which shows the smallest impact of the object itself on the detected temperature, is based on the kinetics of the excited statei.e. luminescence lifetimes of the temperature probes. On the other hand, depending on the optically active luminescent material used, the considered lifetime may vary from sub-nanoseconds to tens of miliseconds.
The shorter the luminescence lifetime, the more complicated and costly the equipment and brighter the probes must be to acquire sufficient number of photons to derive the temperature.
Although commercial FLIM systems exists, the slow frame-rate and/or constricted field-ofview may limit the wider adoption of photo-instable fluorescent molecular probes for this purpose. Therefore, lanthanides and transition metals, with their long sub-ms to ms luminescence lifetimes seem to be the optimal choice, but such reliable LT probes are still under construction.
The balance between these materials science and imaging capability technical aspects needs to be found depending on the application. Future work should be oriented to develop nanosized temperature sensor that exhibits bright emission, high chemical and mechanical stability without photobleaching effect. Additionally, as shown in one of the previous sections, the interactions between nanoparticles themselves (like energy diffusion or reabsorption) may lead to the erroneous temperature readout. Therefore phosphors of large Stokes shift are especially desired from this perspective. The model-based LT should possess absorption and emission bands which overlaps with the optical biological windows (OBWs) spectral range in order to minimize the effect of object-thermometer interaction and to maximize the light penetration depth in the biological system. Last but not least the spectroscopic properties of such an ideal photoluminescent sensor should be strongly susceptible to temperature changes providing high accuracy and precision of temperature determination. If all of these requirements are met by one thermometric phosphor, the dissemination of luminescence thermometry technology, as one of the most reliable and accurate methods of temperature determination in vitro and in vivo imaging will become possible. In consequence, not only new information about the dynamics of the biological processes will be revealed but, above all, the effectiveness of a new targeted therapies will be enhanced above the extent, which enables their wide and commercial use.