Upconverting nanoparticles for pre‐clinical diffuse optical imaging, microscopy and sensing: Current trends and future challenges

Upconverting nanoparticles (UCNPs) are a class of recently developed luminescent biomarkers that – in several aspects – are superior to organic dyes and quantum dots. UCNPs can emit spectrally narrow anti‐Stokes shifted light with quantum yields which greatly exceed those of two‐photon dyes for fluence rates relevant for deep tissue imaging. Compared with conventionally used Stokes‐shifting fluorophores, UCNP‐based imaging systems can acquire completely autofluorescence‐free data with superb contrast. For diffuse optical imaging, the multi‐photon process involved in the upconversion process can be used to obtain images with unprecedented resolution. These unique properties make UCNPs extremely attractive in the field of biophotonics. UCNPs have already been applied in microscopy, small‐animal imaging, multi‐modal imaging, highly sensitive bioassays, temperature sensing and photodynamic therapy. In this review, the current state‐of‐the‐art UCNPs and their applications for diffuse imaging, microscopy and sensing targeted towards solving essential biological issues are discussed.


Introduction
Upconverting nanoparticles (UCNPs) constitute a novel type of contrast agent with highly interesting and unique properties for luminescence bioimaging. The aim of this review paper is to illustrate the great potential of this emerging field. The phenomenon of upconverting luminescence has been studied for decades, with the earliest experimental works presented in, e.g., Refs. [1,2]. The idea originated from Bloembergen in 1959 [3], who proposed that infrared (IR) light could be detected by sequential stepwise absorption of an ion in a solid material. An extensive review of the early work was given by Auzel in 2004 [4]. Most efforts were focused on rare-earth (RE) ions doped into a lattice host. Such ions are ideal because they exhibit very long lifetimes for the intermediate states in the upconver-10 improvement) than conventional luminescent reporters, which is most likely caused by the absence of sample autofluorescence background. It was known from the early beginning that host materials doped with two types of ionssensitizers and activators -were the most efficient [10][11][12]. With the upconverting materials attracting an increasing amount of interest, further efforts were invested into the understanding of the UC process, leading to particles that were acceptable both in terms of size and brightness for in-vivo biomedical studies [13][14][15][16]. The power dependence of the UC signal (I ) was known to be I ∝ I n ex , where n is the number of photons absorbed in the process, and I ex is the power density of the excitation light. Later, it was realized that this relationship is only valid in a limited range of power densities, as saturation alters the power dependence and the efficiency was shown to have a more complex power-density-dependent behavior for high power densities [17,18].
During the last few years, the great potential of UCNPs for in vitro and in vivo use in, e.g., tissue microscopy applications and also deep tissue imaging and tomography, is becoming obvious [19]. Today, the research topic of upconverting nanoparticles for biomedical applications is extremely popular, in particular related to optical bioimaging. Much progress has been made in developing nanoparticles with outstanding properties, motivated by the huge interest for biomedical applications. In this review, recent and present trends in optical imaging, microscopy and sensing using upconverting nanoparticles are discussed, with emphasis placed on the biological applicability rather than the nanoparticle materials. Recent reviews which focus more on the properties of the UCNPs themselves can be found in, e.g., Refs. [20][21][22][23].
This review paper will first introduce the UCNPs and how they are synthesized. This section will be followed by a discussion on means to characterize these particles and determine their most interesting properties. The importance of careful characterization of their luminescence efficiency in a reproducible manner will be stressed, as this appears to be neglected in the literature, making it difficult to compare results from different studies. Furthermore, the necessary surface modifications for in vitro and in vivo applications, and health issues related to UCNPs are discussed. Finally, the central theme of this review is developed, i.e., biomedical applications of UCNPs which is followed by a discussion of the outlook as well as the present and future challenges.

Composition and synthesis
Upconverting materials have been known and studied over a long period of time, however, only recently have they become interesting for biomedical applications. For this reason, there are still plenty of unexplored aspects of the UCNPs which can lead to improved and optimized properties for biomedical applications. In this section, the composition and synthesis methods for UCNPs used in biomedical imaging will be briefly reviewed.
UCNPs are generally comprised of an inorganic host doped with a sensitizer and an activator. The dopants, especially the activator, are usually incorporated into the host lattice at a low doping concentration in order to avoid quenching caused by undesired cross relaxations [24,25]. Efficient UC emissions can be obtained by manipulating the energy transfer between the sensitizer and the activator with the assistance of the host lattice [4]. The sensitizer, which displays a considerable absorption cross-section, absorbs the energy from the excitation light, and transfers it to the activator, mainly through non-radiative and phononassisted processes. During the last decade, many kinds of UCNPs which incorporate RE ions into various host materials have been developed [26][27][28][29][30][31][32][33][34][35][36]. However, up to date, efficient UC emissions with good potential in bioapplications have only been observed in very few dopant-host combinations, such as NaYF 4 :Yb 3+ /Er 3+ , NaYF 4 :Yb 3+ /Ho 3+ and NaYF 4 :Yb 3+ /Tm 3+ [37,38].
Well-crystallized nanoparticles are highly desirable in biological applications as luminescence markers, since they can exert a strong field on the doped ions and energy losses caused by crystal defects can be minimized. Uneven components of the field increase the f − f transition probabilities of the dopant ions [39,40], resulting in efficient UC emissions [41,42]. The crystal structure of the host thus plays an important role in the process of UC emission, as it determines the crystal field and the doping concentration [43][44][45][46][47]. NaYF 4 UCNPs constitute a good example of this, since NaYF 4 exists in two polymorphs at ambient pressure: cubic (α) phase (metastable high-temperature phase) and hexagonal (β) phase (thermodynamically stable low-temperature phase) [48,49], which are closely related to the quantum yield (QY). The β-NaYF 4 UCNPs have approximately one order of magnitude higher QY than their α-phase counterparts [43,44,50,51]. Due to the stringent requirements on the crystallinity and phase purity of the host materials, during the last decade, considerable efforts have been invested into developing synthesis methods which yield highly crystalline structures for efficient UC emissions.
In the following, the three most important aspects that determine the quality of UCNPs will be discussed, i.e., the commonly used host materials, activators and sensitizers; the typical synthesis methods; as well as the phase-and-size control approaches.

Host materials, activators and sensitizers 2.1.1. Host materials
Host materials play a key role in UC emissions. Ideal host materials should be transparent in the spectral range of interest, have high optical damage threshold and chemical stability. They are also required to have low lattice phonon energies in order to minimize non-radiative energy losses. Fluorides satisfy these conditions and are commonly used as host materials for UCNPs due to their relatively low phonon energies (∼ 350 cm −1 ). Oxides have also been

Activators
Upconversion emissions are theoretically expected from most RE ions. However, under low excitation power densities, efficient UC emissions can only be generated by very few RE ions, such as Er 3+ , Ho 3+ and Tm 3+ . This is due to the ladder-like arrangement of their energy states and good match with commercially available high-power diode lasers (Fig. 1). For example, Er 3+ ions have mainly three UC emissions bands, including two green emission bands at around 525 nm and 545 nm originating from the transitions 2 H 11/2 → 4 I 15/2 and 4 S 3/2 → 4 I 15/2 , respectively, and a red emission band at around 652 nm originating from the transition 4 F 9/2 → 4 I 15/2 [80]. The Ho 3+ ions have two main UC bands of green and red emission at 541 and 647 nm, corresponding to the transitions 5 S 2 / 5 F 4 → 5 I 8 and 5 F 5 → 5 I 8 , respectively [81]. The main UC band of the Tm 3+ ions is in the near infrared (NIR) range at around 800 nm, originating from the transition 3 H 4 → 3 H 6 [82]. This NIR UC band is located within the "window of optical transparency"for biological tissues, in which both light absorption and scattering are significantly reduced. This feature makes Tm 3+ -doped UCNPs particularly interesting for imaging of deeply located bio-targets. The Tm 3+ ions have another two generally less efficient UC emission bands at around 479 and 648 nm, generated by the transitions 1 G 4 → 3 H 6 and 1 G 4 → 3 F 4 , respectively [82]. The blue emission band is less suitable for bioimaging due to much higher light absorption and scattering in biological tissues at this wavelength. The intensity ratio between different UC bands is host and doping concentration dependent, even for the same activator.
For other RE ions, Wang et al. [83] has made pioneering work to exploit their UC emission ability. They suggested a general approach for realizing efficient UC emissions through gadolinium sublattice-mediated energy migration, by incorporating a set of RE ions into separated layers at precisely defined concentrations. In this way, they have demonstrated efficient UC emissions from a wide range of activators, such as Tb 3+ , Eu 3+ , Dy 3+ and Sm 3+ . Generally, the doping concentration of activators is relatively low (usually < 2 mol%) in order to minimize cross-relaxation energy losses.

Sensitizers
In single-doped nanoparticles, the efficiencies of UC emissions are relatively low. This is due to the difficulty in finding an equilibrium point for minimizing the quenching effect by reducing the RE-ion concentration, and maximizing the absorption of pump energy by increasing its concentration. To enhance the UC luminescence efficiency, a sensitizer with a sufficiently large absorption cross-section in the NIR region is usually incorporated together with the activator. Certainly, it is required that efficient energy transfer between the sensitizer and activator can occur. These types of sensitizers can be called direct sensitizers. For example, Yb 3+ fulfills the condition and is commonly used as UC direct sensitizer for Er 3+ , Ho 3+ and Tm 3+ under excitation by 975 nm light. The optimal concentration of Yb 3+ is dependent on the host and the activator [84], but is usually kept high (∼ 20% for fluoride nanoparticles). Another type of sensitizers, called indirect sensitizers, are used to quench and enhance certain emission bands. For instance, Nd 3+ , Ce 3+ and Ho 3+ have been used as co-sensitizers to enhance the blue emission band of Tm 3+ , red emission band of Ho 3+ and NIR emission band of Tm 3+ , respectively [81,85,86]. It should be noted that the cations of the host materials themselves can be used as indirect sensitizers, as indicated in recent reports [83,87].
Up to date, most efforts have been devoted to developing Yb 3+ -sensitized UCNPs pumped at around 975 nm. Although biological tissues have relatively small scattering at this wavelength, the applications of this type of UCNPs are still limited in biomedical imaging, due to the non-negligible absorption of water, which comprises the major part of biological tissues. In view of this and benefiting from the development of commercially available high-power tunable laser sources, the research on the sensitization capability of different RE-ions excited by different pump wavelengths constitutes an interesting topic. Such research could eventually lead to the emergence of a new group of UC materials.

Thermal decomposition method
In a typical thermal decomposition procedure, metal trifluoroacetates are thermally decomposed to corresponding metal fluorides. Zhang et al. [55] first reported the synthesis of single-crystalline and monodisperse LaF 3 triangular nanoplates via the thermal decomposition of lanthanum trifluoroacetates (La(CF 3 COO) 3 ) in a mixture of oleic acid (OA) and octadecene (ODE). The approach was later developed as a general route to synthesize high-quality REF 3 and NaREF 4 nanoparticles [61,62,67,80,98] (Fig. 2(a-b)).
Although the thermal decomposition method is an efficient approach in synthesizing high-quality and monodisperse UCNPs, it has some disadvantages, including the harsh conditions needed (∼300 • C, anhydrous, oxygen-free and inert gas protection), expensive and toxic metal precursors, and hazardous by-products-as mentioned above. Additionally, post-synthesis processes are required to introduce hydrophilic and biocompatible coatings on these nanoparticles due to the presence of hydrophobic capping ligands on the surface of the UCNPs.

Hydro(solvo)thermal method
The hydro(solvo)thermal method is a typical solution-based chemical synthesis approach in which reactions occur in a sealed environment under high pressure and temperature, usually above the critical point of the solvent in order to increase the solubility and reactivity of the inorganic substances. In a typical synthesis procedure, RE and fluoride precursors, solvents and certain surfactants are mixed  Fig. 2(c)). Wang et al. [13] developed a one-step synthesis of polyethylenimine (PEI)-coated NaYF 4 :Yb,Er/Tm UC-NPs via a solvothermal approach. The prepared nanoparticles were hydrophilic and biocompatible directly after production due to the free amine groups capped on the surface of the UCNPs, thus, no further surface modification and functionalization were needed. Zhang et al. [66,109] reported the synthesis of uniform nanostructured β-NaREF 4 arrays by an OA-assisted hydrothermal route without the assistance of templates, applied fields, and undercoating on substrates ( Fig. 2(d)). Li et al. [68,110] reported a novel user-friendly solvothermal-like method in a glass flask rather than in a sealed autoclave for fabricating high quality hexagonal-phase NaYF 4 :Yb,Er/Tm nanoparticles without the use of excess fluoride reactants ( Fig. 2(e)). This synthesis strategy has been widely used in the synthesis of fluoride UCNPs [42,83,[111][112][113][114][115]. Possible advantages of the hydro(solvo)thermal method include cost-effective raw materials, excellent control over the crystalline phase, particle size and shape under much lower reaction temperatures (generally below 200 • C). Disadvantages of this method include the need for long reaction time (hours-days) and specialized reaction vessels. In addition, surface modifications are usually likewise required because of the insufficient hydrophilicity of the prepared UCNPs in most cases.

REVIEW ARTICLE
The optimization of the synthesis parameters is generally very time-consuming, and the relatively long reaction time makes it even more challenging. Recent work conducted by Wang et al. [116] and Chan et al. [51] showed great progress along these lines. Wang et al. greatly shortened the reaction time to 5 min by employing microwave heating. This microwave-assisted approach has been further developed and is currently another widely used approach for producing high quality fluoride UCNPs [117][118][119]. Chan et al. used a different approach and developed a parallel reaction workstation called WANDA (workstation for automated nanomaterials discovery and analysis) for reproducible and high-throughput synthesis of colloidal nanocrystals including NaYF 4 UCNPs [51].
Although the above discussed batch-control methods are proved to be efficient in the synthesis of UCNPs and enable control over the phase, size and shape, they are not of complete satisfaction. Inherent to batch-control methods, it is difficult to monitor the UCNPs in real time and accordingly control their growth by adjusting experimental parameters. Microfluidic methods have been shown to be excellent in the synthesis of nanostructures including QDs, gold and silver nanoparticles [120][121][122][123][124]. Despite this, currently very few attempts have been made to synthesize UCNPs in microfluidic systems [125]. In the reported microfluidic synthesis, the reaction time, limited by the final length of the microchannel or capillary, seems to retard the formation of uniform UCNPs with desired hexagonal phase, and the prepared UCNPs suffer from aggregation problems probably due to the absence of proper surfactants during the synthesis. Further studies are therefore required.

Phase and size control
Control of the crystal structure of the fluorides is critical to achieve a high QY. It has been shown that in the thermal decomposition and hydro(solvo)thermal synthesis approaches, α-phase NaYF 4 nanoparticles are generally first formed, and then β-phase can be formed through α → β phase transition by overcoming the free-energy barrier [61,126]. Hence, any method which could facilitate such a phase transition is helpful in obtaining β-NaYF 4 nanoparticles. Here, several commonly used methods for the phase control of NaYF 4 UCNPs will be summarized.
(a) Controlling the reaction temperature and time Generally, prolonged reaction time and high temperature are needed to overcome the free-energy barrier for the α → β phase transition [59,61,127]. Usually, a reaction time of at least 30 minutes and temperatures of 290-310 • C are needed in order to obtain hexagonal NaYF 4 UCNPs [110,126].
(b) Adjusting the molar ratio of Na + /RE 3+ and F − /RE 3+ β-NaYF 4 is disfavored under most conditions, except within a narrow window in which the 1:1:4 stoichiometry of Na + , RE 3+ , and F − is strictly maintained [29,126], indicated by the phase diagrams of bulk sodium yttrium fluoride [48]. However, influenced by the mechanisms of Na + , RE 3+ , and F − liberations, previous reports have suggested that higher Na + / RE 3+ and F − / RE 3+ reactant ratios favor the formation of β-NaYF 4 [61,[128][129][130]. Thus, when using different Na + and F − sources, different reactant ratios may be required to promote the α → β phase transition.
(c) Ligand-mediated phase transition Wei et al. [67] found that EDTA molecules capping on the surface of the UCNPs could suppress the cubicto-hexagonal phase transition, prohibiting transformation from the α-phase to β-phase, even when annealed at 600 • C for 5 hours. OA was also found to be in favor of the formation of cubic-phase UCNPs, because negatively charged oleate ligands strongly binds electrostatically to the positively charged (100) surfaces of small α-NaYF 4 particles and stabilizes the α-phase relative to βphase [ from cubic to hexagonal [63,126], while TOP can be used as combined ligands together with OA and ODE to change the surface energy and control particle phase and shape. A ligand formed between oleate and TOP at high temperature promotes the phase transition from cubic to hexagonal [64,128]. TOPO was found to be able to reduce the energy barrier of the α → β phase transition, thus, it can be used as a single solvent, both a boiling solvent and a capping reagent, to control crystalline growth by providing a broad temperature window for the β-NaYF 4 UCNPs [14].

(d) Lanthanide and transition metal ions doping
Yu et al. [131] and Wang et al. [114] reported that lanthanide dopants with larger ionic radius, such as La 3+ , Ce 3+ , and Gd 3+ , can decrease the energy barrier and tip the balance in favor of the formation of β-NaYF 4 . Chen et al. [47] reported that Ti 4+ doping induced α → β phase transition in a liquid-solid-solution reaction system at as low temperature as 130 • C. Tian et al. [132] reported a facile strategy for controlling the phase (β → α) and UC emission behavior (green → red) of NaYF 4 :Yb/Er UCNPs through Mn 2+ ion doping.
Varying the parameters mentioned in the above procedures will also affect the size of UCNPs. Thus, the control over the crystalline phase is always accompanied with the control over the size. The size is an aspect of significance which influences the uptake, biodistribution and clearance of nanoparticles in living organisms. Although research on how the size of UCNPs influences their uptake in cell or animal models is limited, a large amount of effort has been devoted towards the control of the size in order to produce UCNPs with various diameters which are needed for different biomedical applications [63,77,109,114]. For, in particular, in vivo imaging applications, nanoparticles are usually required to have comparable sizes to the targeted molecules, in the range of 4-10 nm for most membrane and globular proteins [133]. Significantly larger nanoparticles may have limited accessibility to smaller subcellular structures, perturb trafficking patterns, retard diffusion, interfere with protein functions or binding events, or alter pharmacokinetics [134][135][136]. Smaller nanoparticles means lower brightness and thus the challenge is to synthesize UCNPs small enough while maintaining their brightness. Much progress has been made in producing sub-10 nm NaYF 4 nanoparticles [77,126,137]. Most notable of these studies has been the determination of the window for synthesis of sub-10 nm β-NaYF 4 by Ostrowski et al. [126]. They described the conditions for controlled synthesis of protein-size β-NaYF 4 from 4.5 to 15 nm in diameter with efficient UC emissions, by varying the concentration of basic surfactants (OA and OM), Y 3+ /F − ratio, and reaction temperature ( Fig. 2(f)).
It is noteworthy to point out that the phase and size control of fluoride UCNPs often requires a precise control over many experimental parameters, since the phases are affected by multiple factors instead of a single one. In addition, the above approaches can be employed synthetically in order to obtain desired products with reasonable con-ditions [77]. To fully meet the demands of different sized UCNPs for various applications, approaches for phase and size control need to be further explored.

Optical properties and characterization
In contrast to traditional fluorescent biomarkers, UCNPs can be excited by NIR rather than ultraviolet (UV) radiation, thereby significantly minimizing photo damage of biological specimens and maximizing the penetration depth of the excitation light. The anti-Stokes nature of the UC emissions enables autofluorescence-free detection which results in excellent signal-to-noise ratio (SNR) and improved detection sensitivity. Distinguished from other anti-Stokes processes -including second harmonic generation, multi-photon absorption and anti-Stokes Raman scattering -UC emissions are based on real intermediate states, thus allowing for more efficient frequency conversion. This means that UC emissions can occur under moderate light intensities, which is often a basic requirement in biological studies. UCNPs can be excited by compact, inexpensive and low-power (1-1000 W/cm 2 ) NIR lasers. UCNPs show non-blinking characteristics under continuous irradiation and are resistant to photobleaching as well as photochemical degradation. Additional advantages of UCNPs include narrow and well-defined emission peaks, a large anti-Stokes shift, and convenient emission color tuning [81,84,138].
The UC photoluminescence spectra of the UCNPs constitute one of their most important characteristics and typical luminescence spectra are presented in Fig. 3. Since the discovery of the UC phenomenon in the 1960's, extensive efforts have been invested into the research of the UC emission mechanisms. To date, several basic processes have been identified, including ground state absorption (GSA), excited state absorption (ESA), energy transfer upconversion (ETU), cross relaxation (CR) and cooperative sensitization (CS) [4]. Through the combined action of these processes, complex multi-photon UC phenomenaincluding photon avalanche (PA) -and anti-Stokes spectra can be achieved. The details of UC mechanisms have been summarized in previous reviews [4,140].

Rate equation analysis
In principle, the UC process can be quantitatively expressed by a set of coupled differential equations describing the population density, N i , of each lanthanide 4 f N manifold, taking into account all population and depopulation rates involved [141][142][143]: where A ED i j and A MD i j are Einstein coefficients for electric dipole (ED) and magnetic dipole (MD) radiative transitions from manifold i to j; W NR i,i−1 is the nonradiative multiphonon relaxation (MPR) rate constant from manifold i to i − 1; C ET i j,kl is the microscopic energy transfer parameter for the transfer of energy via the donor transition i→ j and the acceptor k→l transition. In this model, the interactions among more than two ions (such as the CS process) are not considered. The intensity of any given UC emission peak is proportional to the product of the population density of the emitting state and the microscopic rate constants for the radiative transition. Obviously, these rate constants play critical roles in depicting the UC emissions. Once they are known, the characteristics of UC emissions, including the QYs and spectral purities, can be obtained by calculation directly, as shown in a recent work of Chan et al. [142]. This model can also be used to determine critical energy transfer transitions involved in UC process [142]. However, the use of this model is demanding mainly due to the difficulty in determining the rate constants, although in theory the ED, MD transition rates and energy transfer rate C ET i j,kl can be calculated using the Judd-Ofelt theory [39,40], while the quantum mechanical magnetic dipole operator [144] and the nonradiative MPR can be treated with a modified energy gap law [145]. Thus, this model is usually simplified to only include the major transitions and MPR processes identified by previous studies.
The investigation on the time dependent behavior of UC emissions is a good way to verify the validity of the proposed UC pathways, and the rate constants could be extracted by fitting the measured time-dependent emission intensity with time-dependent rate equations [146][147][148].
There are very few reports on the use of this approach on UCNPs in the literature and an interesting topic could be to correlate the microscopic rate constants and the reaction conditions in order to guide the synthesis of UCNPs.

Power dependence
In addition to studying the time-resolved behavior of UC emissions, power dependence analysis of UC emission intensity under CW excitation, combined with steady-state rate equation analysis, also provides useful information for the UC mechanism. A theoretical model for this was systematized by Pollnau et al. and Suyver et al. [17,18], and is now one of the most important aspects of the optical characterization of UCNPs. Briefly, the UC emission intensity, I, is related to the absorbed excitation power density, I ex , by the following equation By plotting the emission intensity versus the excitation power density in a double-logarithmic diagram, the order n of the UC process, i.e., the number of pump photons required to excite the emitting state, can be obtained by the slope of the power dependence curve. This slope indicates the multi-photon nature of the UC emission. An example of the power dependence of the NIR UC emission for Tm 3+ is shown in Fig. 4 together with the corresponding UC spectrum, indicating that this UC emission line originates from a two-photon process.
It should be noted that the power density dependence of the UC emission described by Eq. ( (2)) is only valid under weak excitation power densities and will become more complicated for higher excitation power densities due to the competition between the ETU rate (excitation power density dependent) and the linear decay rate in the individual excitation steps. The slope of the power dependence curve is known to decrease with increasing excitation power density. When the excitation intensity is high enough such that saturation of the intermediate energy state involved in the UC process occurs, the two-photon UC luminescence will appear with a slope of 1 [17]. Smaller slopes and saturation power densities are often used to indirectly indicate better performance of UCNPs in energy upconversion [103,149].

Quantum yield
The QY measurement constitutes another important aspect in the optical characterization of UCNPs, playing a crucial role for their practical applications. The QY is generally defined as the ratio between the number of the emitted UC photons and the number of absorbed excitation photons. Currently, QY measurements on UCNPs are directly adapted from the QY characterization of conventional fluorescent materials, such as fluorescent dyes and QDs. Two different experimental setups can be employed: a spectrofluorometer-based setup and an integrating-spherebased setup. The former needs a fluorophore with known QY as a reference, while the latter is self-calibrated. Different from linear fluorescent dyes and QDs, the QYs of UCNPs are power density dependent rather than constant [150,151]. Thus, the determination of the excitation power density is critical for QY measurements on UCNPs. In an integrating-sphere-based setup, it is typically challenging to determine the true excitation power-density, since the excitation light repeatedly passes through the sample due to reflections from the wall of the sphere, which could lead to errors in the measurements. Compared with an integrating-sphere-based setup, determining the excitation power density of a spectrofluorometer-based setup is much more straight forward. Reports in the literature on the absolute QY of UCNPs are generally very scarce [50,[150][151][152]. Page et al. [150] measured the QYs of several UC phosphors using an integrating-sphere-based setup. For bulk NaYF 4 :Yb 3+ /Tm 3+ material, they determined the power conversion factor of the blue emission band to be 2 × 10 −4 at an excitation intensity of 1 W/cm 2 . Size-dependent effects have also been considered by Boyer et al. [50] for Yb 3+ /Er 3+ co-doped NaYF 4 nanoparticles, where 10 times lower QY of core-shell nanoparticles (30 nm) as compared with bulk material was found under an excitation intensity of 150 W/cm 2 . Recently, Xu et al. [151] measured the QY of NaYF 4 :Yb 3+ /Tm 3+ @NaYF 4 core-shell nanoparticles using a spectrofluorometer-based setup and reached a value of 3.5% under an excitation intensity of 78 W/cm 2 , as illustrated in Fig. 5. For comparison, the QYs of the most efficient two-photon dyes were simulated under identical experimental conditions and are also shown in the same figure. It can be seen that the required excitation intensity is Figure 5 (online color at: www.lpr-journal.org) Quantum yield of the 800 nm emission band in core-shell NaYF 4 :Yb 3+ /Tm 3+ @NaYF 4 nanoparticles and core nanoparticles NaYF 4 :Yb 3+ /Tm 3+ . The quantum yield increases linearly with the excitation intensity until a saturation point, from which the quantum yield approaches a constant value. Solid lines show simulated corresponding QYs for highly efficient two-photon dyes under identical experimental conditions. (Reprinted with permission from Ref. [151]. Copyright 2012, American Chemical Society.) clearly too high to be used in scattering tissues. The reason is that these dyes require simultaneous absorption of two photons via a virtual state, in contrast to UCNPs, which display long-lived real intermediate states.
An important aspect of proper QY measurements is to make it possible to compare the results between different studies. However, up to date, the QY characterization on UCNPs still does not follow a harmonized protocol, instead the QYs are usually provided at one specific power density, ignoring their power density dependencies [50,77,126]. As mentioned above, UCNPs are different from conventional Stokes-shifting fluorophores, since saturation can occur due to competition between ETU and linear decay. This leads to QYs that in general have complex power density dependent behaviors. Thus, the reported QYs from different studies are not directly comparable. Here we propose a protocol for a standardized QY measurement for UCNPs, where the whole power density dependence is measured in order to provide complete information on the UC energy conversion system. In addition, the intensity of the excitation light generally has an inhomogeneous rather than a "top hat"distribution. This inhomogeneity should also be compensated in order to provide accurate QY information. If possible, finding a few parameters which are able to characterize both the absolute QY and the change on excitation power density are also highly desirable. Such a careful characterization of the luminescence efficiency in a reproducible way -which appears to be neglected in the literature -would make direct comparisons between results from different studies possible. This would stimulate the further development of the UCNPs, improving their performance and lead to more optimized synthesis methods.

Enhancement of upconversion efficiency
In most cases, the emission efficiency of UCNPs is relatively low because of two reasons: (1) non-radiative decay due to surface defects and (2) two-photon nonlinear processes. Hexagonal phase NaYF 4 UCNPs usually have one order of magnitude higher QY than their cubic counterparts. Thus, by phase control approaches UCNPs with enhanced UC efficiency can be obtained. Another trivial method to increase the QY is to synthesize larger UCNPs. Smaller particles have larger surface-to-volume-ratio, which naturally leads to more defects. Ions on or near the surface of the UC-NPs are sensitive to the local environments, thereby causing higher non-radiative energy losses. However, the maximum size of UCNPs for specific applications is normally limited. Therefore, this approach is not generally valid. Other versatile approaches will be presented as follow.

Ion doping and composition tuning
Ion doping can modify the crystal field surrounding doped RE ions, thus changing the UC emission intensity by altering the transition probabilities of the RE ions. Introduced by Chen et al. [16,153] and Bai et al. [154][155][156], Li + -doping was found to be able to enhance the UC emission in oxide UCNPs. In some recent reports [116,157] it was shown that Li + -doping gives similar results in fluoride UCNPs. Ho 3+ -doping is found to enhance the NIR UC emission of Tm 3+ [86]. Composition tuning is another commonly used approach to enhance the QY, where the recent work of Chen et al. [137] is a good example.

Coating with a shielding layer
A shield layer on the surface of the UCNPs can reduce crystal defects and protect the active optical ions from the coupling with vibrational modes in the solvent, thereby reducing the non-radiative energy losses [42]. The shell could be either inert (i.e., with no optical active ions) or active (i.e., containing active ions, mainly Yb 3+ ). For example, increased QYs were observed after growing of an undoped NaYF 4 shell over Er 3+ -or Tm 3+ -doped core UCNPs [50,151]. In another report, a 50-fold higher QY was observed for sub-10 nm UC-NPs, and the emission from 9-nm core-shell UCNPs was larger than that of comparable 37-nm cores when normalized to the absorbance at 980 nm [126]. For the use of active shells, examples include comparison of the pure BaGdF 5 :Yb 3+ /Er 3+ core with the active-shell coated counterparts, where it was shown that the luminescence intensity could be enhanced by several hundreds of times [158].

Surface plasmonics coupling
It is well known that the unique surface plasmon properties of metallic structures can be exploited to enhance the fluorescence from adjacent fluorophores (organic dyes and inorganic QDs) [161,162]. Similarly, surface plasmons with strong local field can also be used to enhance the efficiency of UC emission [163,164]. Zhang et al. and Sudheendra et al. have successfully attached gold nanoparticles onto UCNPs to modulate the UC emission [165,166]. A specifically designed plasmonic gold surface coupling with 980-nm radiation was shown to clearly enhance NIRto-visible upconversion luminescence from the nanocrystalline layer [167]. It is very important to study the surface plasmon enhancement mechannism at single nanoparticle level in order to provide an optimal design of hybrid UCNPs and metallic nanoparticles. Schietinger et al. coupled single NaYF 4 :Yb 3+ /Er 3+ nanocrystals with gold nanospheres (30 and 60 nm in diameter) to enhance UC emission in a combined optical and atomic force microscopy (AFM) system, gaining an overall enhancement factor of 3.8 [168]. Time-resolved measurements revealed that both the excitation and the emission processes are influenced by the coupling to plasmon resonance in the gold nanospheres.
Several studies have shown that the separation critically determines whether enhancement or quenching eventually dominates [161,169]. The enhancement of UC emission is highly spectral dependent. A more-photon-involved UC process resulted in a larger enhancement factor under the same excitation power density [170]. The enhancement is known to result from the modification of the radiative and nonradiative decay rates and the enhancement of the excitation intensity by the localized surface plasmonic resonance of metallic nanostructures [163,168,171,172]. However, some details remain unknown, e.g., is the change of decay rates or the increase of excitation intensity contributing the most to the enhancement of the intrinsic QY of the nanocrystals, and by how much the ultimate (maximum achievable) intrinsic QY will be increased due to the change of the decay rates (the enhancement of the excitation intensity will not change the ultimate intrinsic QY). Further experimental studies on the QY enhancement are needed. Regarding the QY of the system, although it always tends to increase due to the enhanced local field, the enhancement is, however, related with the intrinsic power-density dependent QY of the crystal. Thus the understanding of the influence of the surface plasmonic coupling on the intrinsic QY could address questions related to the amount of gain that can be expected.
Although the UC emission enhancement depends on the fourth or higher power of the local field, solid theoretical analysis revealed much weaker gains than that of Raman enhancement [163]. Furthermore, chemical sample www.lpr-journal.org preparation and precise physical control are essential to ensure a satisfactory enhancement, making the process complicated and demanding. Thus, considerable challenges involved in using metallic nanoparticles to enhance UC emission for biodetection remains and there is a strong need of simplifed and reproducible synthesis strategies.

Surface modification
Similar to other nanosized biomarkers, UCNPs also need to be water dispersible, highly stable, biocompatible, sensitive and biotargeting for successful bio-applicability. However, most of the commonly used UCNPs were synthesized using oil-phase methods which render UCNPs with no intrinsic aqueous dispersibility and functional organic groups on the surface. Recently, several single-step synthesis methods, mainly including polyol process [173,174], one-pot synthesis assisted by hydrophilic ligands [175,176], hydrothermal microemulsion synthesis [177] and ionic liquid-based synthesis [94,178], have been developed to directly synthesize hydrophilic UCNPs. However, it is still very demanding to obtain monodisperse and hydrophilic UCNPs with small size via single-step methods, although they simplify the reaction procedure and reduce the post-processing. For hydrophobic UCNP synthesis strategies, it is easier to control the size, morphology, phase and crystallinity of the UCNPs, and the procedure of post-processing is straight forward and controllable. Furthermore, the directly synthesized hydrophilic UCNPs without any treatment also show toxicity to cells, tissues or whole organisms. Therefore, UCNPs synthesized from hydrophobic methods are still the most commonly used in biological applications. In the following, more emphasis will be placed on how to re-process the as-synthesized hydrophobic UCNPs into a hydrophilic state with high stability, high biocompatibility and bioconjugation platform.

Hydrophilic processing of UCNPs
In general, a multitude of methods exist for converting hydrophobic UCNPs into a hydrophilic state [38]. Here, four main methods commonly found in the literature are summarized: ligand exchange, ligand oxidation, silanization and ligand free. These methods are versatile and do not have any obvious side effects upon the morphology, size, composition, crystallization and optical properties of the resulting UCNPs.

Ligand exchange
Ligand exchange is an effective method for modificating the surface of nanoparticles. When the hydrophobic ligands on the as-prepared nanoparticles are replaced by some hydrophilic ligands, the nanoparticles will become water dispersible. Due to the surfactant OA ad- sorbed on the surface, UCNPs contain a large number of carboxyl groups (-COOH), which strongly interact with RE ions. Therefore, for the OA-coated UCNPs, multichelated ligands or an excess of single-chelated ligands are required to exchange the OA ligands. In recently reported works, poly(ethyleneglycol) (PEG) [139,179], 3mercaptopropionic acid (3-MSA) [82], 5-mercaptosuccinic acid (MSA) [139], polyacrylic acid (PAA) [71,180,181] and citrate have been used to replace the OA ligands via ligand exchange. These replacement processes are simple, highly repeatable and do not change the unique optical properties of UCNPs in any significant way. These commonly used organic molecules all carry functional groups and facilitate further biofunctionalization and bioconjugation. For UCNPs which are coated by charged organic molecules, it is possible to add an additional polymer coating of opposite charge. Once the UCNPs become water dispersible, further process could be performed. Recently Zhan et al. [139] proposed to encapsulate negatively charged MSA-UCNPs using positively charged polyallylamine hydrochloride (PAH). The significant decrease of OA, the MSA encapsulation and PAH coating were confirmed by Zeta potential measurements and Fourier transform infrared (FTIR) spectroscopy, as shown in Fig. 6. This attraction is based on electrostatic interaction. Repeated coating via this attraction can be useful and this approach is called layerby-layer method [182], which, however, normally requires repeated coating and complicated washing procedure.

Ligand oxidation
Apart from ligand exchange, OA ligands on the surface of the UCNPs can also be oxidized into azelaic acids (HOOC(CH 2 ) 7 COOH), which results in the generation of free carboxylic acid groups on the surface. After ligand oxidation, the OA-coated UCNPs will become water dispersible. A strong oxidizing agent is very important for the effective ligand oxidation process. In 2008, the group of F.Y. Li proposed a simple and versatile strategy for direct oxidization of OA ligands without any intermediate procedures using a strong oxidizing reagent named Lemieux-von Rudloff reagent [16,183]. Naccache et al. [71] proposed the utilization of permanganate/periodate to oxidize and break the double bond of the long OA C18 chain with the -COOH moiety maintained, facilitating better dispersibility in water. The OA ligands could also be directly oxidized by ozone -a clean and readily available strong oxidant -under specific conditions, which enabled the presence of -COOH or -CHO groups on the surface [184]. These oxidization reactions were reported to have no significant negative effect on the chemical and optical properties of UCNP, and the introduction of carboxylic or aldehyde groups not only rendered high water dispersibility, but also facilitated further bioconjugation with biomolecules through covalent methods. It is worth to point out that this method involves oxidation of the double C=C bond of the ligand, and thus the types of available ligands are limited and dependent on the used surfactants. Other disadvantages of the oxidation strategy include the long reaction times and the low yields.

Silanization
Silicate systems have been used to synthesize bulk, film, and particle silicate mesoporous structures for a wide range of applications. As one of the most frequently used methods of surface modification for nanoparticles, silica coating is highly stable (chemically inert), biocompatible, optically transparent and offers nanometer-precision thicknesses. It is also suitable for use as a coating material for UCNPs [185]. In recent years, the group of Y. Zhang has employed silica to encapsulate hydrophobic OA-coated UC-NPs [90,111,[186][187][188][189][190][191]. Surface silanization methods can flexibly offer abundant functional groups (e.g., -COOH, -NH 2 , -SH, etc.) and thus satisfy various needs of conjugation with biological molecules. The most commonly used approach of silica coating is the reversed microemulsion system, which is based on a homogeneous mixture of water, oil, surfactant and tetraethyl orthosilicate (TEOS) [115,192]. This method can precisely control the silica shell thickness via altering the reagent amount or the reaction time, which is very useful to control the distance between the nanoparticles and the molecules of interest [115,193]. Furthermore, another advantage of silica coating is that mesoporous silica shell can be easily obtained and greatly facilitates the loading of drugs and biomolecules onto the surface of UCNPs. Thus, such surface modification of the UCNPs can allow drug delivery to specific cells or receptors. An representative application is to employ mesoporous-silica-shell-coated UCNPs as nanocarriers for PDT drugs [194].

Ligand free
The surface OA ligands of NaYF 4 :Yb 3+ ,Er 3+ nanoparticles could be released by thorough washing using excess ethanol [90] and HCl [195] under ultrasonic treatment. This ligandfree method is very simple and was proposed very recently. UCNPs without any ligand on the surface are highly water dispersible. However, the problem would be that biofunctionalization and bioconjugation is nontrivial due to the lack of functional groups on the surface. A solution to efficiently biofunctionalize these ligand free UCNPs would be the key.

The impact of surface modification on optical properties
The described surface modification methods have different applicabilities due to their distinctive features, and thus the method selection should rely on the specific application of interest. As previously discussed, a large number of successful bioapplications of UCNPs were substantially demonstrated with the assist of these surface modification methods. It is reasonable to believe that these methods are essential and effective, and they hardly have considerable side effect on the optical properties of UCNPs. However, it is still of significance to investigate the impact of these processes on the luminescence intensity in order to improve or maintain their brightness. In very recent years, there have been an increasing number of studies revealing the impact of surface modification on the optical proterties of UCNPs, although they are not covered in most review papers. According to some reports, the luminescence intensity, to some extent, was actually decreased after treatment with certain modification processes [113,115,196,197]. As a case of point, decreased luminescence intensity can be observed for PAA-coated UCNPs due to the interaction between the surface of UCNPs and PAA [196,197]. Besides, in the silanization method the silica shell can slightly scatter both excitation and emission light and thus weaken the luminescence to some extent [115]. The type of solvent is also an important factor affecting the brightness. UCNPs dispersed in aqueous solvent were reported to have a decreased brightness when compared to the same UCNPs dispersed in organic solvent [113]. The reason is that water molecules have high energy vibrational modes, which probably results in an increased nonradiative relaxation of the excited states and thus an overall quenching of the luminescence [113]. These problems could be overcome partially by a protective shell of NaYF 4 , which has low phonon energy and can act as an isolation layer in order to greatly weaken the negative interaction between UCNPs and surface ligands/local environment [113,196]. The other strategies of enhancing QY previously discussed in section 3 can also be exerted to compensate the possible brightness decrease caused by surface modification and aqueous solvent. The key is to keep or even improve the unique optical properties of UCNPs while performing surface modification.

Surface functionalization of UCNPs
The previously mentioned surface modifications of UCNPs enable their dispersion and stability in aqueous solutions. This post-treatment is quite indispensable for hydrophobic UCNPs, but not sufficient for further bioapplications.
In most applications, including bioimaging, biosensing and biotherapy, UCNPs also need to be capable of targeting specific cell lines, tumors or biomolecules. Living cells do not interact directly with the assistant biomaterials, but with the proteins or molecules adsorbed on their surface. Thus, it is needed to develop a bioconjugation of UCNPs interacting with such proteins and molecules. After the surface modification, UCNPs either gain functional groups (e.g., carboxyl, amino, thiol, etc.) or are strongly charged (positive/negative) on the surface, which both enable them to conjugate with various biological or polymeric molecules.
In the field of bionanotechnology, there exists several wellknown approaches for bioconjugation of nanoparticles with biomolecules. Generally, these biomolecules can be bioconjugated to the surface of UCNPs via physical interaction (e.g., electrostatic adsorption) or chemical interaction (e.g., covalent link).
As one of the most commonly used procedures, the electrostatic adsorption method is straightforward and effective via non-covalent forces. Zhan et al. [139] successfully conjugated positively charged PAH-UCNPs with negatively charged antibodies (anti-CEA8 with isoelectric point (pI) in the range pH 5.8-6.5) in phosphate buffer solution (PBS) (pH 7.4). Without chemical bonding, the proteins or molecules attached on the surface of UCNPs will not change or lose their activity. Non-specific binding cannot be eliminated in the case of physical methods. In comparison, chemical binding is more specific and also more complicated to perform. Chemical binding needs to make use of covalent interaction between some specific functional group pairs, such as, a carboxylic acid and a primary amine to form an amide bond, two thiols to form a disulfide bond, a thiol and a maleimide to form a thioether bond, and an aldehyde group and a hydrazide group to form a hydrazide bond [23]. Covalent binding of biomolecules often requires some intermediate process and are assisted with some linker agents. As a case of point, EDC (ethyl(dimethylaminopropyl) carbodiimide) and Sulfo-NHS (N-hydroxysulfosuccinimide) are always employed to couple amino and carboxyl, conjugating surface modified UCNPs and biomolecules [182,198,199]. The effectiveness of this strategy relies on the precise control of the molar ratio of the reagent molecules. In addition, there are many other useful bioconjugation strategies. As a very popular protein binding mechanism, biotin binding to streptavidin were also introduced to the bioapplications of UCNPs [200,201]. Some peptides bearing -COOH, including RGD (arginine-glycine-asparatic acid) and CTX (chlorotoxin)), were directly conjugated to the surface of -NH 2 modified UCNPs for desired function activity [199,202]. DNA and folic acid (FA) were also successfully used to directly bioconjugate UCNPs to specifically target cancer cells [187,201].

Toxicity and health issues
Despite all of the advantages of UCNPs as compared to conventional fluorophores, in order for them to be an attractive choice for biological imaging studies, it is obviously of utmost importance that the toxicity and potential health issues are thoroughly investigated. Even though UCNPs have not been available for very long, the importance of this topic has already attracted a great amount of interest. Most of the studies have focused on in vitro cytotoxicity, however, results concerning long-term effects within small animals have recently started to appear. In this section, a few of the most important results concerning the cytotoxicity of UCNPs will be highlighted.
Following the increasing interest of UCNPs, several studies of in-vitro cytotoxicity of UCNPs have been conducted. Already in 2008, Chatterjee et al. [203] used murine stem cells to assess the cell biocompatibility of PEI-coated UCNPs, while Shan et al. [204] studied the toxicity of silica coated UCNPs in human osteosarcoma (HOS) cells, using the methylthiazol tetrazolium (MTT) assay to evaluate the cytotoxicity. Both of these early studies found the cytotoxicity to be very low. To date, the cytotoxicity of a large number of both human and animal cell lines have been studied, including HeLa cells [177], human glioblastoma (U87MG) cells [199], human nasopharyngeal epidermal carcinoma (KB) cells [205], and human hepatic (L02) cells [206]. An example of the cell viability of HeLa, KB and L02 cells are shown in Fig. 7. Currently, no severe adverse effects have been found that can be directly related to the UCNPs, indicating them to be of high biocompatibility.
An important aspect that in-vitro cell studies cannot provide an immediate answer to is the biodistribution and clearance of UCNPs following an injection into an animal. The long-term biodistribution of intravenously injected UCNPs has been reported by Xiong et al. [205]. In this study, the animals were followed for 115 days and the results showed that the UCNPs mainly accumulate within the spleen and liver. Furthermore, the UCNPs were found to have a clearing time longer than 7 days, in contrast to a previous study which showed a more rapid clearance speed [186]. This shows that sample preparation is of high importance in order to determine the pharmacokinetics of the UCNPs within an organism. However, perhaps more importantly, for these studies, no significant toxicity effects could be seen in the animals under the moderate doses (∼ 15 mg/kg) used.
In addition to the studies of small animals, systematic investigation of the effects upon Caenorhabditis elegans (C. elegans) has also been recently conducted [207]. The worms were fed with a mixture of B-growth media and UCNPs, and the toxicity assessments were based on green fluorescent protein (GFP) expression, life span, egg production, egg viability, and growth rate. The results showed no significant differences between the worms fed with the mixture of Bgrowth media and UCNPs compared with those fed with only B-growth media.
Predicting the toxicity and health issues associated with nanoparticles in a general manner is very difficult. Since the toxicity is not only related to the composition of the nanoparticles (including surface modification), it is clear that the properties of the bulk material very seldom can be directly translated to the nanometer scale. For example, size-dependent factors and the local environment within an organism will play a significant role in the pharmacokinetics of the nanoparticles. For the case of UCNPs, the material itself is still relatively new and despite the current studies showing no significant toxicity effects, further studies on an even longer time scale are certainly required before they can be applied in clinical settings. For example, it is known that RE elements could induce autophagy in HeLa cells, which is a common biological effect for the lanthanide elements [208]. It is also worth to point out that for the purpose of small-animal studies, the effects on a very long time scale may not be of extreme importance. Similarly, the use of experimental drugs on patients carrying terminal diseases is approved in certain regions of the world, e.g., Europe. Thus, even though the effects on a very long time scale may not be clear at this time, there are still numerous of compelling in-vivo applications of UCNPs.

Applications of UCNPs
In recent years, UCNPs have attracted remarkable attention in the biophotonics area due to their merits of autofluorescence free, large anti-Stokes shifts, sharp emission bandwidths, high resistance to photobleaching, nonblinking emission behavior, deep detection ability and high spatial resolution. As shown in Fig. 8, UCNPs have widely been employed in in vitro cell microscopy, in vivo animal diffuse imaging, luminescence diffuse optical tomograpy, in vivo multimodal animal imaging (MRI/PET), highly sensitive bioassays (luminescence resonance energy transfer (LRET)), in vitro temperature sensing and photodynamic therapy (PDT). In the following, all the above biological applications will be covered and discusssed in detail.

Bioassays
Bioassay techniques are of fundamental importance in bioanalytical chemistry and biological sciences. They can of- fer qualitative assessment or measurement of the presence, amount and the functional activity of the analyte, which can be a drug, a biochemical substance or a cell in an organism under study. Concerning single biomolecule detection or nano-scale bioprocess monitoring, the SNR performance of the employed bioassays is very critical and needs further improvements. UCNPs can facilitate the weak signal detection due to the improved SNR as compared to conventional fluorophores. Therefore, in recent years, UCNPs have gained much popularity in applications towards bioassays. In the year of 2001, Corstjens et al. [209] developed UCNP-based lateral-flow bioassays to identify Human Papillomavirus type 16 infection via detection of specific nucleic acid sequences. Later, the same group further applied UCNPs in immunohistochemistry in lateral flow assay formats, and in immunochromatographic assays for human chorionic gonadotropin [210]. A host of UCNP-based nucleic acid assay has also been exploited by Tanke et al. and coworkers to achieve a detection limit of 1 ng/ml oligonucleotides [54]. A sensitive luminescent bioassay for the simultaneous detection of Salmonella Typhimurium and Staphylococcus Aureus was developed for both recognition and element concentration evaluation [211]. UCNP-based assays are highly sensitive, inexpensive, allow for multiplexing, and are suitable for quantitative detection. Successful on-site detection with UCNP-based assays and portable readers has been performed in Europe for the detection of drugs abuse via oral fluids. These reports show the very large potential of the UCNP-based bioassay in biochemical testing.
As one of the most powerful bioassay tools, FRET (Förster resonance energy transfer)-based assays are advantageous in detecting bioaffinity interactions and conformational changes of biomolecules on nanometer scale www.lpr-journal.org (< 10 nm) [211]. Here, UCNP-based probes could assist in circumventing some of the challenges with traditional FRET system caused by the commonly used downconversion organic dyes or QDs. The cross-talk between the donor and acceptor absorption and/or emission spectra probably disables the detection of the weak FRET signals. A significant overlap between these spectra could lead either to direct excitation of the acceptor by the excitation light of the donor or to incomplete discrimination between acceptor and donor emissions. Another limiting factor in traditional FRET is that the background autofluorescence of biological materials caused by the excitation light could disable the detection of weak FRET signals. Such drawbacks limit the efficiency and feasibility of FRET and could be overcome by using UCNPs as donors because of their large anti-Stokes shifted and narrow-band emission. As a derivative of FRET, the working principle of upconversion luminescence RET (LRET) is schematically introduced in Fig. 9. The group of Soukka has performed several studies on UCNP-based LRET bioassays [200,[212][213][214][215]. In 2005, they developed a novel homogeneous upconversion LRET assay technology and its potential was well demonstrated [200]. Since then they applied this technology in many applications, such as immunoassay for E2 (17β-estradiol) in serum [212], enzyme activity assay [214], dual-parameter DNA hybridization assay [215]. A highly sensitive nucleotide sensor has also been exploited with a detection limit of 1.3 nm [201,216]. Very recently, UCNP-based LRET was used to perform detection of Matrix Metalloproteinase, which is a very important biomarker in blood, while, also challenging for sensitive and selective detection [211]. These examples clearly demonstrate that upconversion LRET can extend the applications of FRET technique and enable the realization of effective and highly sensitive assays to be utilized in diagnostic applications and also in high-throughput screening approaches. As a donor, UCNPs with two-photon emission processes have a relatively low quantum yield compared to the downconversion donors of traditional FRET. Further enhancement of the UCNP luminescence will improve the detection limit of the UCNPbased LRET.

Optical thermometry
As a property of the Boltzmann distribution, the relative intensity of the different emission bands of UCNPs will be dependent on the surrounding temperature. For this reason, UCNPs have been proposed as sensitive nanothermometers. For the UC emission bands originating from two states in close proximity, separated by an energy gap of E (usually on the order of several hundred wavenumbers), a thermal equilibrium exists governed by the Boltzmann factor [217]: where I 1 and I 2 are the integrated intensities of the emissions from the higher state and the lower state, respectively; C is a constant which depends on the degeneracy, spontaneous emission rate, and photon energies of the emitting states in the host materials; k is the Boltzmann constant, and T is the absolute temperature. Up to date, several emission bands of different activators have been used in temperature sensing, including the green emission bands of Er 3+ [219][220][221][222][223][224][225] and Ho 3+ [226], red emission bands of Er 3+ [227], and blue emission bands of Tm 3+ [228]. In addition, the ratio between two well-separated emission bands has also been used as temperature-sensitive measurables [218,229]. A theoretical model is, however, needed to evaluate such information.
Recently, optical thermometry based on UCNPs has been used in cell models. Vetrone et al. [217] reported the use of green UC emissions from NaYF 4 :Yb 3+ , Er 3+ nanoparticles for temperature sensing in HeLa cervical cancer cells (Fig. 10(a)). In this study, the excitation light at 920 nm with excitation intensity well below 0.5 kW/cm 2 was used in order to avoid any pump-induced heating. Fischer et al. [218] reported the use of the same kind of nanoparticles for temperature sensing in human embryo kidney cells (Fig. 10(b)).
It is worth pointing out that the ratiometric optical thermometry using UCNPs, although reliable, is primarily applicable for superficial imaging. When the UCNPs are embedded in biological tissue, the luminescence light will be spectrally distorted by absorption and scattering of the tissue. The intensity ratio of two emission bands can thus not directly be related to a temperature. In addition, considering different power-density dependencies of various UC emission bands, even in superficial imaging, optical thermometry based on the use of two well-separated UC bands is difficult because of a general lack of adequate control of the excitation intensity.

Optical Microscopy
Photoluminescence microscopy using either organic dyes or fluorescent nanoparticles can offer high resolution and sensitivity, thus constituting a powerful tool for biological studies as well as clinical medical applications. Microscopy techniques vary a lot depending on the involved emissive features. Due to their unique properties, UCNP-based microscopy exhibits a lot of advantages over other traditional fluorophores.

Ideal properties for single molecule imaging
Due to the multiphoton excitation process involved, UCNPbased microscopy can yield high resolution images under CW excitation. In most cases, a Gaussian beam of light is used to excite the emissive samples and this Gaussian intensity profile (unsaturated level) of the excitation beam can be expressed as where r is the radial distance from the center axis of the beam, ω 0 the beam waist size, and I 0 the center intensity. The power dependence of multiphoton process is nonlinear, which can be seen in Eq. ( (2)). Thus the corresponding emission intensity profiles via one-photon (linear/conventional fluorescent dyes), two-photon and three-photon processes Figure 11 (online color at: www.lpr-journal.org) Calculated photoluminescence intensity profiles upon excitation by a Gaussian beam for three cases: one-photon (conventional fluorophores), two-photon (red and green emission of UCNP) and three-photon luminescence (blue emission of UCNP).
(nonlinear upconverting nanoparticles) are shown in Fig. 11. Apparently such nonlinear power dependence can be exploited to improve the spatial resolution in microscopy applications. Apart from their anti-Stokes emission characteristics, UCNPs are highly photostable and display nonblinking emission in contrast to quantum dots. Compared with the traditional organic dyes (red and blue emission in Fig. 12c), UCNPs (green emission in Fig. 12c) exhibit exceptional photostability after the same long-time exposure to the excitation [230]. As shown in Fig. 12b the time-resolved emission intensity did not show on/off behavior for single UCNPs [231]. The absence of photobleaching and photoblinking enables rapid and precise tracking of single UCNPs. Thus, individual UCNPs possess such ideal properties suitable for single nanoparticle imaging [72,126,231], as shown in Fig. 12d. To use sinwww.lpr-journal.org gle UCNPs for probing single proteins, the particles should be small enough in order not to affect the protein itself. However, smaller nanoparticles also means less emission light. Very recently, Ostrowskigrow et al. [126] reported on the successful synthesis of light-emitting nanocrystals small enough not to disrupt cell activity but bright enough for single detection, enabling single protein imaging. This breakthrough will broaden the applications of UCNPs, such as mapping single proteins moving through a cell, neurons cell interaction, and the process in brain cells connecting together to form a synapse.

LASER & PHOTONICS REVIEWS
With the merit of single particles imaging, UCNP can be envisaged as a superresolution probe. Generally, superresolution can be realized using centroid localization and computer rendering as long as the detected light spots in the microscopy field of view are confirmed from single or separate nanoparticles. It is well known that photoswitchable emission or blinking behavior are the paramount for successful excitation of a single fluorophore molecule in organic dyes or a single QD in superresolution imaging systems (e.g., PALM (photoactivated localization microscopy), STORM (stochastic optical reconstruction microscopy)) [232,233]. For the case of multiple UCNPs in close proximity of each other, differentiation is not possible due to the fact that they have no dark/activate states or stochastic blinking. To develop other time-division mechanisms for exciting UCNPs is challenging but required to address this problem. Different from single-moleculeimaging based superresolution microscopy, STED (stimulated emission depletion) and SIM (structure illumination microscopy) superresolution technologies break through the diffraction limit optically instead of using photobleaching or blinking properties [234,235]. It is reasonable to envisage that UCNP with many unique optical properties can be an ideal probe for these modalities.

In vitro non-specific imaging for cells
In 1999, Zijlmans et al. [8] demonstrated for the first time UCNP-based high-contrast bioimaging. In their work Y 2 O 2 S:Yb 3+ /Tm 3+ particles were used to study the distribution of prostate specific antigen in paraffin-embedded human prostate tissue. At that time, the size of upconversion particles was in the range of hundreds of nanometers, and surface modification engineering was in its infancy. In recent years, with the improvement of the UCNP performance, UCNP-based microscopy techniques have been widely exploited for high-resolution and high-contrast imaging of cellular specimens. Non-targeting UCNPs are found to be attached to the membrane or to be endocytosed by various cell lines incubated with those UCNPs. In 2008, Nyk et al. [82] successfully employed MSA-UCNPs to label Panc 1 cells to produce high-contrast images. PEG and some polymers modified UCNPs can also be utilized for in vitro nonspecific cell imaging [236]. For non-specific binding, two main interaction mechanisms are electrostatic interactions and ligand interactions with cell membranes. Non-specific binding depends on the charge and hydrophobicity of a ligand, but the dependence has not yet been clarified. Very recently, Jin et al. [181] prepared three types of polymer-coated UCNPs, which resulted in the discovery that positively charged PEI-UCNP have enhanced cellular uptake, more than its neutral and negative counterparts as shown in Fig. 13(b)-(d). UCNPs doped with RE elements could induce autophagy in HeLa cells and it is a common biological effect for RE elements with this process being dose and time dependent [208]. Nonspecific binding can, however, also occur for molecules or cell lines of no interest for the study, and thus this process is not suitable for targeted diagnosis and therapy.

In vitro specific imaging for cells
Having obvious advantages over non-specific binding imaging, specific imaging of tumor cells has been widely studied using surface-functionalized UCNPs via biomolecular recognition. In 2009, Wang et al. [189,198] proposed anti-CEA8 conjugated UCNPs to perform cell imaging. Zhan et al. [139] have recently performed a set of solid control experiments and demonstrated that NaYF 4 :Yb 3+ /Er(Ho) 3+ nanoparticles conjugated with anti-CEA8 antibody can be utilized for highly specific binding and imaging of HeLa cells with antigen expressed on the cell membrane, as shown in Fig. 14. In their report, the assynthesized OA-capped UCNPs were first rendered aqueous dispersible through MSA encapsulation. Then negatively charged MSA-UCNPs were further polymer-coated through physical adsorption by positively charged PAH, which allows for antibody protein bioconjugation. Other representative work have been reported by Zhang et al. [177,187], where FA-modified UCNPs were introduced to specifically bind to the folate receptor overexpressed by cancer cells for targeted imaging. High-affinity polypeptide have in recent years been shown to be effective agents for probing biological systems with high specificity. Utilized as peptides, RGD-conjugated UCNPs have been successfully used for cell targeted imaging [199,237]. Yu et al. [202] conjugated PEI-coated NaYF 4 : Yb 3+ , Er 3+ /Ce 3+ with recombinant chlorotoxin -a typical peptide neurotoxin that could bind with high specificity to many types of cancer cellsand incubated the modified UCNPs with C6 glioma cells for targeted imaging. Utilized as target agents, UCNPs could also be nanocarriers and indicators for some molecules and drugs. In a recent report, Zhang et al. [238,239] used UCNPs for the intracellular investigation of smallinterference (siRNA) in living cells. Their results have shown that UCNPs are capable of delivering and tracking siRNA. In principle, the long lifetime in emissive organic dyes and nanoparticles would result in a relatively low photoluminescence intensity under the unsaturated power density level. Due to very long phosphorescence lifetime (millisecond to microsecond) and relatively low QY, UCNP-based microscopy requires relatively low scanning speed to obtain high lateral resolution [240]. Saturated excitation could speed up the scanning while preserving the same lateral resolution. However, in this case it is difficult to obtain a satisfactory axial resolution as the 3D sectioning ability of multiphoton processes will be significantly affected. Another drawback of UCNP-based laser scanning microscopy is that in the cases of 3D spatial scanning, emission wavelength scanning or real-time monitoring, the cell sample has to be exposed to excitation for several minutes or even hours of time [139]. Thus, it is not possible to record very quick bioprocesses. Wide field microscopy could be an alternative method for time-consuming laser scanning upconversion microscopy [82,240]. Up to date, almost all excitation wavelengths for UCNP applications were selected around 980 nm. It is worth pointing out that light around 980 nm suffers from an intrinsic disadvantage: water -being the most significant ingredient of animal and human body -heavily absorbs light around this wavelength. In the cases of long-time scanning, the excitation light of 980 nm can significantly heat up the cell growth medium, which probably results in damage to the cells, as shown in  excitation light around 920 nm (where the optical absorption coefficient of water is much lower and the excitation can still occur effectively) to replace 980-nm light in order to avoid cell damage [139].

Diffuse optical imaging
As is the case for novel microscopy techniques, deep-tissue diffuse optical imaging (DOI) has over the last decades also attracted an increasing amount of attention [241][242][243][244]. Compared with conventional non-invasive imaging systems, such as X-ray computed tomography (CT), magnetic resonance imaging (MRI), and Positron emission tomography (PET), optical imaging systems are fast, compact and very sensitive to luminescent contrast agents. However, due to the relatively high scattering and absorption of light in tissue, the penetration depth is typically limited to ≤ 10 cm [243,245]. Although the penetration depth may seem to be a severe limitation, DOI has already been applied to monitor a wide range of biological processes and systems in both small animals and humans. In small animals, DOI has, for example, been used to follow the development in time of Alzheimer's disease [246], brain metabolism [247], proteases [248] and various drug effects [249]. In humans, DOI has, for example, been employed to monitor and detect breast cancer tumors [250][251][252][253][254], brain activity and brain metabolism [252,255]. However, the images obtained from DOI techniques can be of relatively poor quality due to a number of reasons, including endogenous background tissue autofluorescence, the diffusive nature of light propagation in tissue, and the ill-posedness of the mathematical problem formulation. In this section, we will focus on luminescence/fluorescence diffuse imaging, and in particular luminescence/fluorescence diffuse optical tomography (LDOT/FDOT) and discuss how the upconversion luminescence (UCL) of UCNPs can be applied to obtain images of superior qualities as compared to conventional Stokesshifting contrast agents.

Autofluorescence of tissue
Biological tissue contains a large number of endogenous fluorophores. The fluorophores can either be components of the tissue structural matrix, for example, collagen and elastin, or be formed during metabolism processes, for example, nicotinamide adenine dinucleotide (NADH) and flavins [256,257]. Although the fluorophores are individually well known, any given tissue, however, typically consists of a mixture of a large number of endogenous fluorophores. The distribution and optical characteristics of the fluorophores within the tissues are not only dependent on the tissue type, but also on the chemical environment as well as the metabolic state at the cellular level. By measuring the changes in the autofluorescence spectra, it is possible, for example, to perform cancer diagnostics [258][259][260]. However, for diffuse optical imaging, and in particular

REVIEW ARTICLE
Laser Photonics Rev. 7, No. 5 (2013) 681 luminescence diffuse optical imaging employing an exogenous contrast agent, the ever-present tissue autofluorescence will certainly deteriorate the signal-to-background ratio and limit the sensitivity of the system, and can at the same time cause severe artifacts in the reconstructed tomographs [261]. In order to obtain accurate representations of the fluorophore distribution as well as to increase the sensitivities, significant efforts have been invested to develop methods to overcome the tissue autofluorescence. Suggested approaches include subtraction-methods which model the tissue autofluorescence [262], time-domain separation using fluorophores with suitable lifetimes [263], spectral unmixing and multispectral methods [264,265], transillumination and normalized approaches [266], and large Stokes-shifting markers, such as QDs [265,267]. Although these methods can reduce the effects of unwanted tissue autofluorescence, they are usually associated with complex instrumentation and heavy computational needs. The use of UCNPs can, however, completely eliminate the tissue autofluorescence since the emission from endogenous fluorophores is Stokes shifted, as shown in Fig. 16 [19,38,199,236]. As illustrated in Fig. 17,  this will significantly reduce the numbers of artifacts in the particularly sensitive LDOT problem [268,269]. Furthermore, in contrast to the approaches discussed above, a system using UCNPs is very straightforward to implement and does not require any complex instrumentation in terms of, for example, excitation-light rejection due to the large anti-Stokes shift and the narrow bandwidth of the emission.

Improving image quality by exploiting the nonlinear power dependence
In addition to the anti-Stokes shifted emission, it has been shown that the nonlinear power dependence of the UC emission can be further exploited to enhance both the quality and the contrast of the resulting images. Since the emissions utilized in diffuse imaging correspond typically to two-photon processes, this can be used to confine the excitation volume, leading to an increase in the spatial resolution. The enhancement of the spatial resolution has been demonstrated for the case of diffuse planar imaging [270] as well as LDOT [151] (Fig. 18). Related to the case of microscopy, the improved spatial resolution can be explained by considering the sensitivity profiles for a given source-detector pair, shown in Fig. 19. It is clear that a multi-photon process will cause a confinement of the sensitivity profiles, thus effectively resulting in selective excitation of each individual luminescent target within the medium. From Fig. 19, it can also be realized that not only will the lateral resolution be improved, but also the axial resolution. This is due to the fact that the resolution strongly depends on the gradient in the sensitivity profiles, and such sharp gradients can be found along all spatial dimensions when nonlinear fluorophores are employed.
An interesting aspect of the nonlinear power dependence of the UCNPs is their potential to increase the information density from a given volume by carefully designing the excitation scheme to include multi-beam excitation. This was demonstrated by Liu et al. [271], where it was shown that the use of two excitation beams www.lpr-journal.org results in a luminescence intensity which can be expressed as

LASER & PHOTONICS REVIEWS
where is the detected luminescence intensity and I 1,2 denote the excitation intensities of the two beams. It can be seen that in contrast to conventional linear fluorophores, the use of two excitation beams for a quadratic luminescent marker results in a cross term which contains additional information of the distribution. Figure 20 shows the effect of including the cross term in the reconstructions, which makes clear that the inclusion of the cross term results in a much more correct representation of the true UCNP distribution. Furthermore, by using even more excitation beams, it should be possible to extract additional orthogonal information. Although the benefits may diminish and the system complexity will increase for more excitation beams, this approach should still be very relevant for measurements on small surfaces, such as those found on mice. Using the UCNPs, even a small surface can provide sufficient data for an accurate reconstruction, whereas the amount of orthogonal data points while employing a conventional fluorophore may be very limited.

Improving the penetration depth and avoiding overheating
As has been previously discussed, few reports present the efficiency of UCNPs in absolute terms [50,151]. However, it is relatively safe to conclude that compared with conventional dye molecules, the QY of UCNPs is usually a few orders of magnitude lower. In most cases, the absence of background tissue autofluorescence will still produce images that are superior to those produced with autofluorescencesensitive fluorophores. However, the maximum depth that can be imaged while employing UCNPs is of course ultimately limited by the QY as well as the attenuation of the excitation and the emission light in biological tissue. As previously mentioned, the most commonly excitation light (975 nm) can be strongly absorbed by the water in tissue, which may result in limited penetration depth as well as overheating. It is of importance to systematically investigate these issues in order to optimize UCNP-based optical imaging. Recently, it has been demonstrated that by shifting the excitation wavelength to the absorption band of Yb 3+ at 915 nm, it is possible to increase the penetration depth as the absorption of water is lower than the typically employed absorption band at 975 nm [139]. The experimental and calculated results shown in Fig. 21 both confirm that 915 nm laser excitation is advantageous for deep tissue imaging as compared to 980-nm laser excitation. This is highly relevant when imaging tissue types which contain a high water content, not only to increase the penetration depth but also to avoid the need of excessive laser intensities that may cause tissue heating and damage [159]. The possible overheating effect induced by 980-nm laser irradiation during UCNP-based diffuse imaging was computationally and experimentally studied by Zhan et al. [139]. As shown in Fig. 22, the in vivo experimental results of a mouse show that overheating of the mouse can clearly be induced by 980-nm laser irradiation, which can be effectively overcome by using a 915 nm laser. However, it is worth to point out that the absorption of Yb 3+ at 915 nm is significantly lower and thus the shift of the excitation wavelength is most relevant for water-dense tissues at depths greater than ap-

Small-animal imaging
The rapid development of UCNPs has made them viable for employment in in-vivo small-animal imaging. To date, several aspects within small-animal imaging have been studied with the use of UCNPs. Early studies involved imaging of subcutaneous injections [139,272], and biodistribution of UCNPs. Salthouse et al. [273] studied the accumulation of UCNPs in mice following a tail vein injection and showed that PEG-coated UCNPs with no specific targeting accumulate primarily within the liver and spleen of the animals. This was later confirmed by Xiong et al. [205] using polyacrylic acid (PAA) modified UCNPs, see Fig. 23. In addition, long clearing times were found and the UCNPs could still be observed 7 days post injection in an in-vivo setting [205].
Mapping of the lymphatic systems within small animals is another research topic that has been of great interest for UCNPs. Knowledge of the lymphatic system is important to predict the spread of certain kinds of cancer metastasis. Due to the small sizes of the sentinel nodes, it can be difficult to accurately detect and quantify the distribution of fluorophores if autofluorescence is present. Thus, UC-NPs are very promising for this application where imaging of the lymphatic vessels and lymphatic nodes has been The lymphatic drainage UCL imaging after removal of skin and fatty tissues was also measured. All images were acquired under the same instrumental conditions (power density of 120 mW/cm 2 and temperature at 25 • C on the surface of the mouse). The mean luminescence intensity of the regions of interest (blue areas) of ROI 1 (specific uptake), ROI 2 (nonspecific uptake) and ROI 3 (background) were selected for the in vivo SNR calculation. (Reprinted with permission from Ref. [274]. Copyright 2011, Elsevier Ltd.) demonstrated [236,274,275]. As demonstrated in Fig. 24, by using UCNPs codoped with different RE ions, it is possible to optically separate injections at different times and sites with virtually no cross talks, enabling means to probe and separate movement speeds.
For targeting of specific tissue types, such as cancer tumors, several approaches exist. The use of FA to label UCNPs for the targeting of folate receptor overexpressing cell lines, for example, HeLa and KB, is among the first to be employed in an in vivo setting [177]. Using nude mice bearing a HeLa tumor on the hind leg, it was demonstrated that specific targeting could be achieved and imaged in vivo after intravenous injection of the UCNPs. Other reports on targeting methods include peptide labeling, which could provide a more efficient uptake in vivo [199,202]. The specificity was found to be very good and the lack of background tissue autofluorescence resulted in SNRs which were comparable with those found in bioluminescence imaging.
The long-time stability of UCNPs in biological environments has also enabled the possibility to perform live cell tracking of transplanted cells in an animal [112]. Traditionally, cell monitoring of transplanted cells is performed using histology since the staining agents used typically operate within the UV-blue range of the spectrum. As known, light at these wavelengths has extremely high attenuation and thus in-vivo imaging is difficult except for very superficial targets. In addition, these dyes are highly sensitive to photobleaching, which sets a limit on the time span for longitudinal studies in contrast to the stable UCNPs. Idris et al. [112] first demonstrated the use of UCNPs for both in-vitro and in-vivo tracking of stem cells over a time span of more than one week using confocal microscopy. On the macroscopic scale, Cheng et al. [236] have shown that human cancer cells can be tracked after injection into an animal and the further development of the tumor itself can also be monitored (Fig. 25).

Photodynamic therapy
Photodynamic therapy (PDT) is a treatment modality that has gained clinical acceptance and is now in many places a first line treatment for selected indications, including nonmelanoma skin cancer [276]. For other indications, PDT is still being developed and evaluated. In particular, efforts are made towards developing PDT for solid and deeply lo-cated malignancies [277,278], antibacterial and anti-fungal treatment [279,280]. Another area of great interest in connection to photodynamic actions is drug delivery. Berg et al. [281,282] have developed photochemical internalization (PCI) as a technique to assist drug delivery into cells. Photodynamic reactions are used here to destroy the membrane of a vesicle after endocytosis, and consequently to release its content into the intracellular liquid.
Recently, a number of studies have been conducted to evaluate whether UCNPs could be beneficial for PDT, especially due to the long excitation wavelength which provides good light penetration. Commonly used drugs for PDT typically require light in the visible region, however, the penetration depth at these wavelengths is clearly a limiting factor. The strong need to shift the excitation wavelength to the NIR region is therefore motivated. The optimal region for penetrating biological tissue resides around 800 to 1 μm. However, these NIR photons have too low energy to generate 1 O 2 . Efforts have been invested into using two-photon absorption to enable the use of NIR photons for PDT [283]. Unfortunately, the usefulness of this technique for deep-tissue PDT is limited by the required intensities for two-photon absorption. As UC processes in UCNPs are efficient, they should be more suitable for the use of deep-tissue PDT as compared to two-photon dyes. The use of UCNPs for PDT was already reported in 2007 by Zhang et al. [194]. The obtained efficiency for generation of 1 O 2 was, however, very low, although it was envisaged that the rapid development of UCNP materials in general will significantly enhance the efficiency. Several approaches have since been developed to obtain sufficient efficiency to enable deep-tissue PDT [187,188,[284][285][286][287]. These approaches include the encapsulation of the photosensitizer within mesoporous nanoparticles to overcome the hydrophobic properties of many photosensitizers, thus enabling a broader range of photosensitizer drugs. However, most demonstrations have been on cell plates in a microscopic environment and further studies are certainly required to investigate the applicability of deep-tissue PDT using UCNPs.

Dosimetry
For deep-tissue PDT with the currently available photosensitizers, it is of utmost importance to understand the lighttissue interaction in order to provide a useful dosimetry [288][289][290]. Various kinds of photosensitizers are available, and we do not intend to review the current state-of-theart photosensitizers. Instead, using simple calculations, we would like to highlight the importance of understanding light-tissue interaction in order to estimate the outcome of a treatment.
For simplicity, we will assume an infinite homogeneous medium and a conservative light-dose threshold of 1 J/cm 2 to reach a satisfactory treatment. The distribution of light for a source of power P 0 for steady state diffusion is given www.lpr-journal.org

LASER & PHOTONICS REVIEWS
where D = 1/3(μ a + μ s ) is the diffusion coefficient, μ a the absorption coefficient, μ s the reduced scattering coefficient, μ eff = (μ a /D) 1/2 , and r denotes the distance from the source. Using this well-known and simple relationship, we will show that the time needed to reach a threshold dose for various r , i.e., distances or depths from the source, can be estimated. We assume that the photosensitizer drug is excited in the red wavelength region (∼ 660 nm), which coincides well with the emissions of the commonly used activators. By knowing the tissue optical properties at this excitation wavelength, it is thus possible to calculate the light distribution and hence the time needed to reach a threshold dose. In the following, we assume PDT treatment of human prostate tissue with μ a (660 nm) = 0.5 cm −1 and μ s (660 nm) = 9 cm −1 [292]. The treatment time needed when using UCNPs will be determined by the absorbed fraction, the QY of the UCNPs as well as the energy transfer efficiency from the UCNPs to the photosensitizers. These numbers are difficult to assess in a general manner, and we assume in these quick estimates that all excitation light is absorbed, an energy transfer efficiency of 1 and a QY that is similar of that presented in Fig. 5. The UC fluence rate can thus be expressed as where η denotes the QY, and the subscripts ex and em denote the excitation light and the emission light, respectively. Furthermore, we can estimate the optical properties at the excitation wavelength of UCNPs at 975 nm [245,292] to be μ a (975nm) = 0.7 cm −1 and μ s (975nm) = 5 cm −1 . Using these values we can then again calculate the treatment time needed to reach a threshold dose for UCNP-mediated PDT. Figure 26(a) shows the calculated fluence rates of the excitation light at 660 nm and 975 nm, with the inset showing the corresponding QYs for each excitation depth for the UCNPs. The strength of the source was chosen to be 500 mW and the QY of the UCNPs was obtained by a linear fit (in a log-log scale) of the low-intensity values from previously published work [151]. The time needed to reach a threshold dose for direct excitation of the photosensitizer and UCNP-mediated PDT is shown in Fig. 26(b). A few conclusions can be immediately drawn from these results. Firstly, for prostate tissue, the attenuation factor of light at 660 nm and 975 nm is similar. This certainly depends on the tissue type, with prostate tissue being relatively rich of hemoglobin and water. However, even for tissue types with less amount of water and hemoglobin, the difference in the fluence rates between the two excitation wavelengths will still be relatively negligible for depths less than 1 cm. Perhaps the most remarkable result is the treatment time that will be needed if UCNPs are used to excite the photosensi-tizer drugs. From Fig. 26, it can be seen that the treatment time while using UCNPs will be at least 4 orders of magnitude longer. This treatment time can be shortened by either increasing the excitation fluence rates (for instance by employing a pulsed excitation source) or by increasing the QY of the UCNPs. However, it is quite clear that it is unrealistic to increase either of these factors by several orders of magnitudes directly, especially since the excitation power is already chosen to be very high. It may be possible to use a pulsed light source to keep the average power down, while using the peak power to excite the UCNPs. Pulsed light will allow higher fluence rates in the tissue without causing tissue heating. Ideally one would like to work with fluence rates well above the saturation of the UCNPs, meaning that the QY ideally would be in a region where it is not power dependent. For tissue regions with such high fluence rates, the PDT efficiency will not decrease with depth as quickly, and would have a depth profile more similar to the case of direct excitation of a photosensitizer. However, this approach most probably cannot fully close the gap between direct excitation and UCNP-mediated excitation, as the highest reported QY (obtained under very high intensities) for UC-NPs are still on the order of a few percents [50,151]. Thus, the treatment time while using UCNPs will still be longer than direct excitation for most practical cases.
The efficiency with depth depends on two things, attenuation of the excitation light (in favor for long excitation wavelengths) and the fact that the excitation is a two-photon process for UCNPs and a linear process for conventional PDT drugs (in favor for the linear excitation process). Our simulations clearly indicate that the non-linear excitation process hampers the excitation efficiency more than the gain in better light penetration. This is a very similar observation as for molecular two-photon PDT sensitizers. The calculations above are performed under extremely simplified conditions. Clearly, they are not meant to serve as an accurate representation of the reality, but rather to highlight the trends and needs in order to make UCNP-mediated PDT feasible. The quality and efficiency of UCNPs have been increasing in a very rapid fashion over the last years. Even though the results above may indicate that it is difficult to perform PDT with UCNPs directly, with the ever-increasing knowledge of the UCNPs, and the use of properly pulsed light sources, UCNP-mediated PDT may still be feasible within the foreseeable future.

Multimodality imaging
Multimodality imaging is a research field that has been explored extensively in the last few decades. Until recently, efforts have mainly been invested into the modalities such as X-ray computed tomography (CT), magnetic resonance imaging (MRI), single-photon emission computed tomography (SPECT), positron emission tomography (PET), and ultrasonography [293][294][295][296], mainly aiming for molecular imaging. In recent years, the availability of a large numbers of photoluminescent molecules has catalyzed a growing interest of enabling optical molecular imaging for multimodal contrast agents [297,298]. The high sensitivity, compactness and speed of optical imaging systems have made luminescence molecular imaging extremely attractive as a compliment to the more established modalities. However, luminescent contrast agents that are optically bright and stable are difficult to find. Thus the luminescent probes on the multimodal contrast agents tend to degrade and bleach, making it difficult to perform truly longitudinal studies in parallel with other modalities. UCNPs, on the other hand, are extremely stable and do not in general suffer from bleaching and aging effects. In addition, since UCNPs are based upon crystal hosts, multimodality imaging can, in some cases, be enabled by modifying the crystal host. Diffuse optical imaging is often associated with poor spatial resolution, although the use of UCNPs can significantly improve this characteristic [151]. The spatial resolution of CT and MRI are in general higher, however, the information obtained is quite different. While CT and MRI can provide detailed anatomical information, for molecular imaging, luminescent optical contrast agents usually have a much higher sensitivity, comparable with PET. However, while PET can provide detailed pre-operative data, they are not practical for acquiring intra-operative data due to both complexity and the relatively short half lives of the radioactive tracers. In this section, the current state-of-the-art multimodality imaging agents which are based on UCNPs will be discussed.

Magnetic-optical imaging
UCNPs have been demonstrated for both T 1 -and T 2weighted MRI. For T 1 imaging, the UCNP crystal can be modified to provide contrast in an MR system directly. For MRI, gadolinium (Gd) has been an extensively used re-laxation agent. Indeed Kumar et al. [299] co-doped Gd 3+ ions into the NaYF 4 crystal host of UCNPs, while Park et al. [72] used the NaGdF 4 crystal host, with both approaches showing magnetic properties under T 1 -weighted MRI (r 1 = 0.14 s −1 mM −1 for the co-doping approach and r 1 = 1.4 s −1 mM −1 for the NaGdF 4 crystal host approach). The QY of NaGdF 4 UCNPs is expected to be comparable to that of NaYF 4 UCNPs, however, absolute numbers are seldom reported, which makes it difficult to draw final conclusions on the effects upon the optical efficiency.
Another strategy that can be employed is to use a coreshell structure. For example, Li et al. [300] used NaYF 4 cores coated with Si-DTTA-Gd 3+ shells to generate multifunctional UCNPs for multimodal imaging. This design was motivated as it should yield the most UC efficient core, while high Gd 3+ payload is achieved on the surface which is suitable for functionalization. Using this approach they report r 1 = 20.1 s −1 mM −1 and r 2 = 55 s −1 mM −1 . However, the lack of reports upon the QY makes it once again difficult to obtain absolute numbers in terms of the efficiency.
For in-vivo studies, Zhou et al. [183] have shown that NaGdF 4 :Yb 3+ ,Er 3+ ,Tm 3+ UCNPs can be accurately imaged, using both MRI and optical imaging, inside white Kunming mice following an intravenous injection. As expected, the UCNPs were found to accumulate mainly inside the spleen and liver of the animal.

Nuclear-optical imaging
Nuclear imaing is extensively used within the field of biology and medicine. PET imaging, for example, can achieve sensitivities down to picomolar range. Recently, the use of multimodal UCNPs for PET imaging was demonstrated [206,301], thereby the commonly used 18 F isotope was www.lpr-journal.org chosen as the PET agent and very promising in vivo experiments were conducted. In addition, Zhou et al. used a gadolinium-based crystal host, thus simultaneously enabling tri-modal imaging (further discussed below). Such multimodal agents may prove to be very useful in future clinical development. However, due to the limited half-lives of the radioactive tracers, it is important to develop reaction processes with high yield. Another aspect that is worth to consider is information orthogonality. It is obvious that the largest gain from a certain contrast agent can be obtained if each modality provides independent information as compared with the other modalities. For the case of PET and UC emission, although the information space may overlap, the time window for acquiring PET data is much more limited as compared with UC emission, however, PET may be able to provide more detailed data.

CT-and trimodal imaging
The high-atomic number associated with the RE elements in the host crystal of UCNPs can lead to effective attenuation of X-rays. Thus, very recently several groups have proposed the use of UCNPs for X-ray CT imaging [302][303][304][305][306]. Ytterbium (Yb) and lutetium (Lu) have received most attention due to their high atomic numbers that match the operating conditions found in CT systems. For the case of Yb, the host material NaYbF 4 has been demonstrated to provide contrast superior to iobitridol [302,304]. Due to the nature of the electron structure within the RE elements, Lu in a similar fashion provides an enhanced CT contrast [304,306].
Since the attenuation of X-rays is an inherent property of the most efficient crystal hosts, the application of tri-modal imaging (CT/MRI/optical) is often discussed and proposed. Similar to the discussion in section 6.6.1, this can be accomplished either by co-doping the host crystals with Gd or by using a core-shell structure that provides the MRI contrast, such as a Gd shell or an iron shell [303,305,306].
It is well known that optical systems are generally compact, fast, inexpensive and easy to operate. Thus, the possibility to perform imaging in multiple modalities can provide new opportunities in the understanding of biological systems. Perhaps even more importantly, the fast screening process enabled by optical imaging systems can lead to a more rapid development of new contrast agents as well as new drugs.

Challenges and Outlook
As clearly illustrated in this review article, UCNPs have many potential advantages and are extremely interesting for microscopy and in vivo applications of diffuse light imaging. Consequently, the use of UCNPs as improved contrast agents for optical bioimaging has grown enormously in just a few years. A main remaining challenge is obviously to make these particles compatible with currently used labeling and imaging technology and to be employed in important biomedical studies. This will require reproducible and commercially available nanoparticles in user-friendly kits and in large quantities. Also corresponding imaging systems must be developed for such applications. Apart from these obvious technical aspects, a few research challenges remain to be solved in order to fully explore the potential of these very promising particles as contrast agents. These challenges as well as ideas about how these could be addressed will be discussed below.
One challenge -despite the strong and rapid development of the particles -is their still relatively low QYs. We believe that this can be partly overcome by utilizing pulsed excitation with pulse lengths that match the relatively long lifetimes of UCNPs, as indicated in our discussion about QY in section 3. A similar case is the use of femtosecond pulsed excitation in two-photon fluorescence microscopy. Since it is known that the QY is power density dependent -as clearly illustrated in Fig. 5 -consequently, a high QY could be achieved by increasing the fluence rate of the excitation light. In order to limit the heating of the tissue to acceptable levels, strictly speaking, the only possibility is to utilize pulsed excitation with an acceptable average power. A difference from the two-photon fluorescence microscopy is that the excitation of UCNPs relies on long-lived intermediate states, which mean that pulses shorter than a few milliseconds do not improve the QY substantially. By utilizing pulsed excitation we foresee that improved sensitivity of deep tissue targets could be accomplished without much side-effects in terms of heating. This is a simple improvement which could be realized with off-the-shelf diode lasers, thereby drastically increasing the applicability of UCNPs for diffuse imaging.
For PDT applications, the potential of using UCNPs would also be drastically improved by pulsed excitation. The presented UCNP-PDT approaches seem to be severely limited by the relatively low efficiency in generating cytotoxic agents in the PDT process. The low efficiency in the excitation process using UCNP is here a limiting factor. The possibility to produce a sufficient PDT effect could probably be drastically improved by utilizing millisecond pulsed excitation. We suggest to further study this approach and to investigate its potential for improved efficiency. It seems more feasible to reach the threshold dose with improved QY of the production of cytotoxic agents in the PDT process via pulsed excitation.
The commonly used UCNPs employ Yb 3+ ions as sensitizers, which have a strong absorption band at 975 nm. However, this absorption band coincides with that of water which causes light attenuation and heating. To improve the penetration depth and reduce the heating of tissue in imaging applications, it is possible to shift the excitation wavelength to the 920-nm absorption band of Yb 3+ . Although this absorption band is significantly weaker than the 975-nm band, excitation at this wavelength has been demonstrated to be very feasible and motivated for in vitro cell imaging, and imaging of very deeply embedded targets within tissues. The conditions determining which

REVIEW ARTICLE
Laser Photonics Rev. 7, No. 5 (2013) 689 excitation wavelength to use is not immediately trivial and relates to the optical properties of the studied tissue and the imaging depth of interest, and thus consideration must be taken based on the experimental conditions. Development of protocols for improved synthesis of UCNPs with optimized properties for bioimaging would be highly desirable. Several aspects relate to this issue. A main issue is to develop reproducible and scalable production procedures -here the microfluidic approach may be a possibility. The use of other sensitizer ions, with an excitation wavelength better suitable for bioimaging applications would also provide a great advancement. In this respect, it would be more optimal with a slightly shorter or slightly longer excitation wavelength to decrease the attenuation caused by water absorption. This would provide a better penetration depth in tissue and at the same time avoid some of the side-effect of heating the tissue.
Another future challenge for diffuse optical imaging is to characterize the particles well enough to allow for direct comparison of results between different studies. In particular an absolute value of the efficiency of UCNPs needs to be provided, so that data can be directly inter-compared, thereby promoting the development in the field. A recommendation would be to measure the QY for a number of power densities, and also for different pulse lengths, at least if pulsed excitation is considered.
A further aspect of vital importance for the future utilization of UCNPs is to fully understand the health issues connected to these particles, in particular considering possible clinical applications that may develop. For cell or small animal studies, the toxicity may be slightly more relaxed, as the long-term effect may not be of critical importance.
Finally, multimodality approaches could become important in some clinical specialities of UCNP-based diagnostics. This could in particular be of interest as various modalities could provide different diagnostic information as well as employment in different clinical situations. Multimodality approaches have been explored a lot to gain complementary diagnostic information. The various tools could also be used in different conditions -for instance, one technique could be utilized in the initial diagnostic preparation phase and another technique could be used later as a bedside tool during a therapeutic procedure or as a monitoring device. For the latter cases simple optical techniques could present obvious benefits.