3.1.

Production of Cells Expressing Reporter Proteins

The production of genetically modified mammalian cells involves transfection, which is the general term describing a procedure for introducing foreign nucleic acids, such as those constituting a RG, into eukaryotic cells. Transfection methods can be broadly categorized into those that are transient and others that are stable, illustrated in Figs. 3(a) and 3(b) respectively. Transient transfection (TT) is relatively straightforward, involving the introduction of DNA into cells along with an agent that promotes its uptake. However, since the foreign DNA is not integrated into the host genome, the sequence is gradually lost after repeated cell divisions leading to gene expression that lasts only a few days. Alternatively, stable transfection entails integration of the foreign genetic material into the host genome giving rise to long-term expression of the sequence of interest as it is propagated in subsequent cell generations.

Methods for TT are generally physical or chemical.¹⁷ The most commonly used physical method is electroporation, which can rapidly transfect large numbers of cells, but only after an experimentally demanding determination of the optimum conditions. PAI studies to date have predominantly used chemical methods for TT. These methods include use of a cationic liposome,¹⁸ a cationic lipid,¹⁹ or calcium phosphate,²⁰ and are widely available but may vary in their toxicity and efficiency according to the cell type. The RG can be coexpressed with an antibiotic resistance gene and/or a fluorescent marker such as EGFP or an inert membrane tag such as the truncated nerve growth factor receptor (dNGFR). These enable successfully transfected cells to be selected by their survival in the presence of antibiotic, and also the reporter protein expression level to be identified using flow cytometry of the coexpressed marker.

Stable transfection typically involves virus-mediated gene delivery,²¹ which is also known as transduction. The first step in the transduction process is to transfect three or four plasmids—one envelope vector, one or two packaging vectors, and the transfer vector containing the RG—into packaging cells such as human embryonic kidney 293T cells (HEK293T). The virus is then harvested from these cells and used to infect the target cells, resulting in incorporation of the RG sequence into the host DNA. Photoacoustic RG studies have employed common viral vectors including lentiviruses^20,22 and retroviruses.^23,24 Lentiviruses are more versatile since they transduce nondividing cells, unlike retroviruses, which require replicating cells for transduction. A third viral vector system that has been used is the vaccinia virus,²⁵ but this is limited to transient protein production since the virus replicates in the cytoplasm outside the nucleus and the infected cells only survive for one or two days.

Fig. 3

Schematics illustrating (a) TT and (b) stable transfection of an RG into a mammalian cell. The gene must first be introduced into the target cell (for example, by electroporation), and subsequently into the nucleus: these two steps represent TT. The recombinant protein is expressed but after cell division the DNA is only passed on to one of the daughter cells, resulting in the dilution of the RG with subsequent divisions. Stable expression over successive cell generations is achieved by integration of the RG into the target cell’s chromosomal DNA (b). This process typically involves lentiviral or retroviral vectors, which are produced by packaging cells such as HEK293T. The viral genes are expressed by several plasmids (also called vectors): (i) an envelope vector encoding the vesicular stomatitis virus (VSV-G) envelope gene; (ii) one or two packaging vector(s) encoding the pol, gag, rev, and tat viral genes; and (iii) a viral expression plasmid (transfer vector) containing the psi $\mathrm{\Psi }$ packaging sequence and the RG. Approximately 2 days after transfection of HEK293T cells, the cell supernatant contains recombinant viral vectors, which can be used to transduce the target cells. Once in the target cells, the viral RNA is reverse-transcribed ($\mathbf{A}$), imported into the nucleus and stably integrated into the host genome. The RG is transcribed into mRNA ($\mathbf{B}$) and then translated into the reporter protein ($\mathbf{C}$). Expression of the recombinant protein can be detected 1 or 2 days after integration of the viral RNA.

3.2.

Classes of Reporters

PAI RGs can be classified as indirect (enzymatic, which result in nonfluorescent proteins, Sec. 3.3) or direct (fluorescent and nonfluorescent proteins, Sec. 3.4). These classes are shown in Fig. 4, and their spectra are given in Fig. 5. Enzymatic RGs provide inherent signal amplification, as a single enzyme can create multiple molecules of absorbing product, whereas direct RGs provide $1:1$ stoichiometric mapping between the amount of detected protein and the level of RG expression. Table 1 makes further comparisons between the different classes of reporters. The ideal reporter should be: specific to the biological process of interest; exhibit an absorption maximum in the tissue optical window for deep in vivo imaging; be nontoxic to the cell; and avoid photobleaching. The reviewed RGs, including optical properties and demonstrated applications, are discussed in the following two sections and summarized in Table 2.

Fig. 4

Selected studies of PA RG imaging. (a) PA image of a tyrosinase-expressing tumor (shown in yellow) after subcutaneous injection into the flank of a nude mouse.²⁴ The surrounding blood vessels are shown in gray. (b) Left: fluorescent histology of the brain of a transgenic zebrafish expressing mCherry (shown in color); right: corresponding PA image.²⁶ (c) PA image of a kidney tumor expressing reversibly switchable nonfluorescent bacterial phytochrome BphP1.¹⁹ The tumor is shown in color, and the background blood-rich organs are shown in gray. Hb, hemoglobin.

Fig. 5

Summary of all spectra for genetically encoded reporter proteins demonstrated for PAI. (a) Fluorescent proteins: EGFP,^26,27 DsRed,²⁸ mCherry,²⁹ mKate2,³⁰ AQ143,³¹ mRaspberry,³² mNeptune,³³ eqFP670,³⁴ E2-Crimson,³⁵ TagRFP657,³⁶ iRFP670,³⁷ iRFP713 (iRFP),³⁷ iRFP720.³⁷ (b) Chromoproteins: E2-Crimson NF;²³ ultramarine;³⁸ aeCP597;³¹ cjBlue.³⁹ (c) Photoswitchable proteins: rsTagRFP;¹⁹ AGP1;²² BphP1.¹⁹ (d) Molar extinction coefficients and (e) specific extinction coefficients are compared to oxyhemoglobin (${\mathrm{HbO}}_2$, $150{\mathrm{gl}}^{-1}$, red dashed line⁴⁰) and deoxyhemoglobin (Hb, $150{\mathrm{gl}}^{-1}$, blue dashed line,⁴⁰) as well as enzymatic reporters: eumelanin [also endogenous, black line;⁴¹ blue precipitate product of lacZ (blue line,⁴²)]; violacein (yellow line⁴³). Reporter spectra were generated from published excitation spectra, normalized to the excitation maximum and multiplied by the relevant peak molar extinction coefficient. Wavelengths indicate absorption maxima.

Table 1

Summary of the key features of different groups of RGs investigated in PA imaging.

Table 2

Properties of genetic reporters used in PAI.

3.3.

Enzymatic Reporters

Three enzymatically generated reporters have been introduced for PAI: (1) 5,5′-dibromo-4,4′-dichloroindigo, a blue precipitate produced via the enzyme $\beta$-galactosidase (encoded by the lacZ gene), which metabolizes X-gal; (2) eumelanin, a brown pigment produced through action of the tyrosinase enzyme on tyrosine; and (3) violacein, a violet pigment produced from L-Tryptophan via the action of a five-enzyme operon.

LacZ. LacZ is the earliest example of a PAI reporter.⁴⁴ It expresses $\beta$-galactosidase, an Escherichia coli enzyme involved in the metabolism of lactose. Injection of X-gal, a colorless analogue of lactose, leads to formation of a blue product (5,5′-dibromo-4,4′-dichloroindigo) after cleavage by $\beta$-galactosidase [Fig. 6(a)]. The blue product is strongly absorbing between 600 and 700 nm. Dual-wavelength PAM^42,45 and PACT⁴⁵ have been used to detect gliosarcoma cells expressing the lacZ gene under the rat scalp; however, X-gal requires local administration as systemic delivery is inefficient,⁴⁴ which has discouraged further PAI studies with the lacZ gene.

Fig. 6

Schematics showing the reaction pathways of three enzymatic reporters. (a) The hydrolysis of X-gal catalyzed by $\beta$-galactosidase, an enzyme produced from the lacZ RG. (b) Reactions leading from tyrosine to the pigment eumelanin. The enzyme tyrosinase, produced from the tyrosinase RG, oxidizes tyrosine to dopaquinone, which then undergoes various enzymatic and nonenzymatic reactions (represented by the dashed line) leading to eumelanin. Dopa, 3,4-dihydroxyphenylalanine. (c) The violacein biosynthesis pathway.⁴⁹ IPA, indole-3-pyruvic acid.

Tyrosinase. Tyrosinase oxidizes tyrosine to dopaquinone, which then undergoes several reactions that lead to melanin pigments [Fig. 6(b)]⁴⁶ including eumelanin and pheomelanin. Eumelanin exhibits a broad extinction coefficient spectrum (similar to that of pheomelanin), with absorption extending into the near-infrared, enabling it to be detected above the background hemoglobin signal at wavelengths between 600 and 700 nm.^18,24,47 Paproski et al.¹⁸ were the first to develop tyrosinase as an RG for PAI. They subsequently developed an inducible system, in which tyrosinase expression was triggered by doxycycline administered in the animals’ drinking water.⁴⁷ The tyrosinase reporter has also shown multimodal imaging potential⁴⁸ as eumelanin can bind heavy metal ions (${\mathrm{Fe}}^{3+}$) detectable with MRI, and a melanin-targeted ${}^{18}\mathrm{F}$-P3BZA probe for positron emission tomography. Stritzker et al.²⁵ also explored multimodal tyrosinase imaging by TT of tumor bearing mice with a vaccinia virus encoding tyrosinase. This approach offers clinical relevance but prevents longitudinal studies. To maintain tyrosinase expression in subsequent generations of cells requires modification of the cell genome, which, although ethically unapproved for clinical studies, does allow longitudinal animal imaging. Jathoul et al.²⁴ achieved this using stable retroviral transduction, which enabled tumor growth to be monitored over periods up to 52 days, with apparently little effect on cell viability [Fig. 4(a)].

Violacein. The deep violet chromophore violacein is enzymatically generated [Fig. 6(c)] from the sole precursor tryptophan by five enzymes (VioA-E) that were originally cloned from Chromobacterium violaceum. Violacein has a strong absorption peak around 575 nm, enabling differential detection through dual-wavelength excitation at 490 and 650 nm. However, while tyrosinase and lacZ have been expressed in mammalian cells, the violacein operon has so far only been expressed in bacteria.⁵⁰ Violacein may, therefore, find application for in vivo imaging of bacterial infections or theranostic applications, but concerns remain about the toxicity to mammalian cells.

3.4.

Nonenzymatic Reporters: Fluorescent, Nonfluorescent, and Photoswitchable Proteins

Nonenzymatic reporters can be classified by their optical properties and according to their natural origins, such as marine creatures and bacteria [Figs. 5(a)–5(c), Fig. 7]. In general, nonfluorescent proteins are preferable since they exhibit higher photoacoustic generation efficiency and significantly higher photostability (Fig. 8).

Fig. 7

Classification and origins of nonenzymatic fluorescent (wavelengths of absorption and emission maxima given), nonfluorescent (only absorption maxima), and photoswitchable (↔) reporter proteins used in PAI. The colors indicate the corresponding wavelengths. Full molar extinction spectra are given in Fig. 5.

Fig. 8

Photostability of eumelanin-producing cells (the cells express the enzyme tyrosinase), compared with photobleaching of fluorescent and nonfluorescent proteins under prolonged exposure to nanosecond laser pulses. The incident fluence was in the range 1.5 to $1.7\mathrm{mJ}{\mathrm{cm}}^{-2}$ and the pulse repetition frequency was 50 Hz. Reproduced with permission from Refs. 23 and 24.

3.4.1.

Fluorescent proteins

Since the discovery of green fluorescent protein (GFP), fluorescent proteins have been widely applied in biomedical studies, mostly imaged by fluorescence microscopy.^27,51,52 GFP-like proteins are derived from anthozoa (such as sea anemones and corals) or scyphozoa (jellyfish) and have a common structure consisting of a $\beta$-barrel scaffold containing the chromophore, which is formed by folding of the polypeptide chain.⁵³ Within the barrel the chromophore is protected from the outside environment, making it possible to genetically engineer the chromophore microenvironment, for example, to produce enhanced brightness (EGFP)⁵⁴ or spectral modifications. Razansky et al.²⁶ reported the first demonstration of PAI using EGFP and mCherry in fruit fly and adult zebrafish [Fig. 4(b)]. The absorption peaks of even the furthest red-shifted GF homologues^35,55 are generally restricted to the visible wavelength region, making them well suited to PAM but less applicable for PACT deep in vivo imaging [Fig. 5(a)].

In recent years, several near-infrared fluorescent proteins (IFPs) have been derived from bacterial phytochrome photoreceptors (BphPs).⁵⁶ These are targeted toward deep tissue in vivo imaging, exploiting the optical window where endogenous chromophores have relatively low absorption between 620 and 950 nm [Fig. 5(d) and 5(e)]. BphPs incorporate a far-red absorbing bilin, biliverdin (BV), as their chromophore, which is ubiquitous in most eukaryotic organisms. The overall structure of the BphP photosensory module consists of two or three protein domains linked by $\alpha$-helices (Fig. 9), which may exist as a monomer, dimer, or oligomer. As with GFP-like proteins, monomers are preferable and several monomeric IFPs have been developed.^57–59 Compared to GFP-like proteins, BphPs exhibit several advantages for PAI,⁶⁰ such as a relatively low fluorescence quantum yield ($\sim 6\%$ for iRFP compared to 60% for EGFP) and a high intrinsic extinction coefficient ($\sim \mathrm{100,000}{\mathrm{M}}^{-1}{\mathrm{cm}}^{-1}$ for iRFP, compared to $\sim \mathrm{56,000}{\mathrm{M}}^{-1}{\mathrm{cm}}^{-1}$ for EGFP).

Fig. 9

Domain structure of BphPs. The photosensory module consists of PAS, GAF, and PHY domains. The chromophore BV is located in a pocket of the GAF domain and is covalently bound to the cysteine (Cys) residue of the PAS domain. Upon light absorption BV undergoes a conformational change that is detected by the photosensory module, which then transmits the signal to the effector domain, initiating the light driven molecular signaling pathway. Adapted from Ref. 56.

The BphP iRFP713 (also known as simply iRFP⁶¹) was the first to be demonstrated in PAI. iRFP expressed in tumor cells inoculated in the mammary fat pad efficiently incorporated endogenous BV, and an imaging depth of 4 mm was achieved exploiting the peak iRFP absorption of 690 nm.⁶² Two subsequent studies at similar penetration depths imaged mouse brain tumors expressing iRFP713^63,64 and illustrated spectral resolution of two variants of iRFP (iRFP670, 645 nm peak and iRFP720, 702 nm peak³⁷) as well as background hemoglobin signals.⁶⁵ Despite this improved penetration depth of imaging compared to GPF-like fluorescent proteins, the photostability of BphPs remains low, due not only to photobleaching but also to transient absorption effects.^66,67 Furthermore, while their low fluorescence quantum yield is attractive for PAI, fluorescent emission and ground state depopulation can lead to low PA signal generation efficiency. This may partly explain the observed discrepancies between the measured absorption and photoacoustic spectra of these proteins.²³

3.4.3.

Photoswitchable proteins

Photoswitchable proteins undergo a change in their absorption spectra under illumination with a specific wavelength [Fig. 5(c)]. The mechanism behind this change is typically a cis-trans isomerization of the chromophore. For the BV chromophore used in BphPs, the two conformations are referred to as Pr (“pigment red absorbing,” also termed the OFF state) and Pfr (“pigment far-red absorbing,” or the ON state)⁵⁶ (Fig. 10). In the unbound state, switching does not affect the absorption properties of BV, but when covalently bound to the protein barrel the two isomeric states exhibit distinct absorption spectra. This property has been exploited in a comprehensive study of a photoswitchable protein BphP1 derived from the bacterial phytochrome RpBphP1.¹⁹ By taking the pixelwise subtraction between the images of the protein in the OFF and ON states, a $\sim 21$ fold enhancement in the contrast-to-noise ratio (difference between proteins and blood) was obtained relative to the ON-state image. This differential imaging enabled visualization of BphP1-expressing U87 cells in the left kidney in PACT at depths up to $\sim 8\mathrm{mm}$, with an average CNR of $\sim 20$. The photoswitching property was exploited to perform subdiffraction imaging of individual cells in PAM, as BphP1 molecules are switched off at a rate proportional to the local excitation intensity, which is greater in the centre of the Gaussian-shaped laser pulse than at the periphery. In addition to this study, which illustrates the versatility of photoswitchable proteins for high contrast in vivo imaging, a preliminary demonstration has recently emerged of a second BphP derived photoswitchable protein, AGP1.²² Various photoswitchable proteins have also been derived from corals (GFP-like),⁶⁹ two of which have been demonstrated for PAI: rsTagRFP¹⁹ and Dronpa.⁷⁰ With absorption peaks at wavelengths $<570\mathrm{nm}$, these proteins are most suitable for studies in PAM.

Fig. 10

Photoswitching of the BV chromophore between the Pfr (ON) and Pr (OFF) states. Switching from ON to OFF and vice versa is induced by illumination with NIR light ($\sim 730$ to 790 nm) and far-red light ($\sim 630$ to 690 nm), respectively, and involves an out-of-plane rotation of the D pyrrole ring around the C 15/16 bond between the C and D rings. Adapted from Ref. 19.

4.1.

Spectral Unmixing

Spectral unmixing takes advantage of the multiwavelength data acquisition capability of PAI, applying multivariate analysis techniques to retrieve abundance maps of the individual absorbers present in the tissue (Fig. 11). First, the aforementioned challenge of unknown light fluence $\mathrm{\varphi }$ must be accounted for, which is usually achieved with the addition of a preprocessing step. Multiwavelength PA data are typically divided by an estimated value of the fluence ${\mathrm{\varphi }}_{\mathrm{est}}(\mathbf{r},\lambda )$ to give images proportional to ${\mu }_a(\mathbf{r},\lambda )$. For imaging of the superficial tissue layers in PAM, the light fluence $\mathrm{\varphi }$ may be assumed to be spectrally constant⁷¹ and can be estimated experimentally, for example, by invasively inserting a film with uniform absorption to correct for absorption by the dermis.^80,81 For applications in PACT, a simple finite-element method solution to the light diffusion equation assuming homogeneous background optical properties can be applied to derive ${\mathrm{\varphi }}_{\mathrm{est}}(\mathbf{r},\lambda )$.^26,55,62,65 More rigorous approaches to deal with this “spectral coloring” in PACT include Monte Carlo simulations and model-based iterative minimization;^84–90 however, these methods are computationally intensive and may give rise to nonunique solutions. As a result, they are not routinely applied during in vivo imaging.

Having dealt with the data preprocessing, most RG studies perform linear inversion using a form of least-squares regression,^26,55,62,65 in some cases with constraints for positive concentration values exceeding some threshold.^47,68 These least-squares approaches rely on a priori knowledge of the absorption spectra present within a given voxel. Where these are not well known, for example, when the spectra change as a function of concentration, semiquantitative “blind” methods that use the inherent statistical properties of the spectral data can be employed. Some examples relevant to RG imaging include: principle component analysis (PCA);⁹⁷ independent component analysis (ICA);^63,98 vertex component analysis;⁹⁴ and adaptive matched filter (AMF).^64,95 Only AMF⁶⁴ and ICA⁶³ have been applied in PAI of RGs; both methods were able to resolve brain tumors in mice expressing iRFP713. Tzoumas et al.⁶⁴ assessed the impact of the number of excitation wavelengths and different unmixing algorithms on the detection sensitivity for the RG, although their findings contradict prior studies suggesting that the optimum number of wavelengths and unmixing algorithm is likely to be case sensitive.⁹⁵

Fig. 11

Multispectral PA imaging. Images at multiple wavelengths ($\lambda$) are combined with spectral unmixing algorithms to separate different absorbers. Hb, deoxyhemoglobin ${\mathrm{HbO}}_2$, oxyhemoglobin reporter example, iRFP713. Adapted from Ref. 102.

4.2.

Difference Imaging

Spectral unmixing is hampered by fundamental requirements: the preprocessing must be accurate to compensate for light fluence, and the data must have sufficient signal-to-noise ratio (SNR) to avoid corruption of the spectral profile. The recent emergence of photoswitchable nonfluorescent proteins [Fig. 4(c), Sec. 3.4]^19,22,70 mitigates these limitations by switching the protein ON/OFF state between image acquisitions. Taking the difference of images acquired at the same wavelength but with different protein states effectively eliminates the nonswitchable background signals, minimizing the effect of the local optical fluence.⁷⁶ As a result, photoswitchable reporters achieve reliable background removal and have been shown to dramatically enhance the detection sensitivity ($\sim 34$-fold enhancement in contrast-to-noise ratio compared to two-wavelength least-squares fitting¹⁹). For applications requiring quantification of protein concentration, rather than only qualitative spatial localization, photoswitchable reporters therefore represent an ideal solution.

5.1.

Imaging System Sensitivity

For molecular PAI, spatial resolution and penetration depth have to be traded to optimize the detection sensitivity. One parameter used to quantify detection sensitivity is the noise-equivalent detectable concentration (NEC).⁷⁹ NEC is calculated as the concentration of the imaged molecules normalized by the SNR. To achieve a low NEC, one can increase the excitation light fluence (within the maximum permissible exposure, MPE) to increase the SNR. Since the SNR of PAI is proportional to the total number of molecules in a resolution voxel, relaxing the spatial resolution can also improve the NEC, as long as the local optical fluence is maintained. Counter-intuitively, this means that PACT systems, which generally have worse spatial resolution, typically have a better NEC than PAM systems, provided that the light pulse energy is sufficient.

The detection sensitivity of PAI is generally lower than that of pure optical imaging modalities by at least two orders of magnitude.⁷⁹ This is somewhat surprising considering that, at the source (the imaged target), the overall energy conversion efficiencies of fluorescence and PA imaging are similar: for a fluorescent absorber, the transfer of energy from the absorbed photons to fluorescent emission is typically less than 30% efficient, determined by the fluorescent quantum yield, and the remaining absorbed optical energy is deposited as thermal energy, which is subsequently converted to acoustic energy with an efficiency of $\sim 20\%$ at body temperature, determined by the Grüneisen coefficient.⁸² The difference in sensitivity between the two imaging modalities is largely a consequence of the detectors. While a photomultiplier tube used in fluorescence imaging may have single-photon sensitivity, a typical ultrasonic transducer has a noise-equivalent pressure (NEP) of $\sim 2\mathrm{Pa}$,^83,91 which is equivalent to a temperature rise at the target of $\sim 2\mu \mathrm{K}$. Within the skin MPE ($20\mathrm{mJ}/{\mathrm{cm}}^2$), the noise equivalent absorption coefficient ($\mathrm{NE}{\mu }_a$) of the target is $\sim 0.001{\mathrm{cm}}^{-1}$. For EGFP with a molar extinction coefficient of $\mathrm{56,000}{\mathrm{M}}^{-1}{\mathrm{cm}}^{-1}$ and quantum yield of 60% at 488 nm, this $\mathrm{NE}{\mu }_a$ roughly translates to an NEC of $0.1\mu \mathrm{M}$. When the geometric signal loss from the target to the ultrasonic transducer is taken into account,^91,92 the NEC approaches several micromolars, which is generally consistent with experimental results and much higher than that of native proteins in cells.⁷⁹ By contrast, the detection sensitivity of fluorescence microscopy is on the level of several nanomolars, close to the natural concentrations of native cell components.

Fundamental limitations on recorded PA signals are: PA detection sensitivity; PA signal generation efficiency; and the tissue optical and acoustic attenuation. Fortunately, the acoustic sensitivity of ultrasound detectors can be enhanced by better matching their detection bandwidth with the broadband PA signal spectrum. While the detection sensitivity of a conventional piezoelectric ultrasonic transducer degrades with decreasing element size, the sensitivity of optical sensors of acoustic waves, typically based on interferometry, does not. Therefore, optical detection holds promise for high-sensitivity PACT, especially for the cases where the elements of the ultrasonic transducer array are miniaturized to achieve isotropic resolutions.^93,96,99

5.2.

Reporter Detectability

For PAI at depths more than 1 mm, the reporter protein should have an absorption maximum within the tissue optical window (620 to 950 nm) where mammalian tissues are relatively transparent due to the low optical absorption by water, lipids, hemoglobin, and melanin. Even with such optimization of the reporter protein absorption wavelength, the strong background signal from blood still presents a challenge for PA RG imaging, since hemoglobin concentration in tissue is $\sim 2.3\mathrm{mM}$.¹⁰⁰ The problem is shown in Figs. 5(d) and 5(e), which compares the molar extinction spectra and specific extinction spectra for the hemoglobins and a selection of enzymatic and direct PAI reporters. The molecular mass of enzymatically produced proteins is more than two orders of magnitude smaller than the molecular mass of the hemoglobins and directly produced proteins, such as mCherry or BphP1. Thus, care must be taken when comparing required reporter concentrations in terms of moles, grams, or gene-expressing cell numbers. As an example of the challenges of reporter detectability, while the absorption maximum of BphP1 is in the tissue optical window, if the protein concentration is two orders of magnitude lower than hemoglobin concentrations, it remains a challenge to detect a signal above the hemoglobin background. Nonetheless, proteins derived from bacterial phytochromes (iRFP670 and iRFP720) have been detected at estimated concentrations in vivo of 16 to $30\mu \mathrm{M}$,⁶⁵ although the effect of such high concentrations on cell biology remains unknown.