2.1.

Photoswitching Photoacoustic Overview and Setup

We summarize our approach to multiplexed PA molecular imaging using distinct photoswitchable chromoproteins as follows. We use a far-red pulsed laser to photoswitch one of the molecules (e.g., BphP1) from its far-red natural state to an orange absorption state. A difference image is formed by subtracting PA images before and after photoswitching, which removes the hemoglobin background. A red wavelength laser is used to drive the chromoprotein from its orange state back to its far-red state. Subsequently, a third wavelength is used to photoconvert the second reporter (e.g., sGPC2). Again, a difference image retains the photoswitched reporter while removing signal from background hemoglobin. Note that as more distinct photoswitchable chromoproteins are developed, they may have partially overlapping absorption spectra that prevent photoswitching one without partially photoswitching another in the same image. To mitigate the unwanted cross talk associated with unintended photoconversion of reporters, we propose a spectral demixing approach using photoswitchable difference spectra. Figure 1 shows the principles of difference-spectra demixing using images of sGPC2 and BphP1 that have been adapted for clarity. Figure 1(a) shows a timeline detailing the change in optical absorption states in response to laser stimulation at 760, 608, and 710 nm. Figure 1(b) shows the current optical absorption states when each PA image is taken. Figure 1(c) shows the difference images and difference spectra involved in difference-spectra demixing. Figure 1(d) shows a relative concentration image of sGPC2 and BphP1 obtained via difference-spectra demixing in vivo. In this example, sGPC2 does not change significantly with 760-nm laser pulses and vice versa for BphP1 at 710 nm. However, as more distinct photoswitchable chromoproteins are developed and multiplexed, there will not always exist a wavelength that photoswitches only one of the multiplexed chromoproteins.

Fig. 1

Illustration of difference-spectra demixing. The timeline shows an imaging cycle that consists of a series of laser pulses (wavelengths: 760, 608, and 710 nm) separated by 0.1 s. The laser pulses are used to image and photoconvert sGPC2 and BphP1. (a) Illustrates the change in absorption states for sGPC2 and BphP1 following exposure to laser pulses. (b) Shows the PA images obtained during the imaging cycle and the corresponding optical absorption states of sGPC2 and BphP1. (c) Shows the difference images obtained via image subtraction and the experimentally determined difference spectra used in difference-spectra demixing. (d) Shows a relative concentration image of sGPC2 and BphP1 obtained via difference-spectra demixing in vivo. This illustrates the simple case involving only 710- and 760-nm imaging wavelengths, but in this paper, we will also use 700-, 720-, 730-, 740-, and 750-nm wavelengths. [Animal art was modified from Servier Medical Art³⁴ (CC by 3.0 https://creativecommons.org/licenses/by/3.0/)].

We built a custom three-dimensional (3-D) photoswitching PA imaging system to acquire simultaneous PA and US data with photoconverting and 3-D imaging capabilities, Fig. 2. This imaging system is capable of both traditional PA imaging as well as photoconversion of reversibly photoswitchable chromoproteins. It employs two Nd:YAG lasers: a Surelite OPO Plus (Continuum, 10 Hz repetition) and a ND6000 (Continuum, 10 Hz repetition). A research US imaging platform (Vantage 256, Verasonics) was used to acquire the PA and US signals. A function generator (AFG 3021B, Tektronix) was used to maintain a 10-Hz laser pulse frequency. The delay function generator (DG645, Stanford Research Systems) controls the triggering of the lasers and signal acquisition card (Octopus, GaGe). The signal acquisition card acquires photodiode (DET100A/M, Thorlabs) signals from the imaging laser for laser intensity normalization. A stepper motor (23Y002D-LW8, Anaheim Automation) translates the PA/US transducer (LZ250, VisualSonics, Fujifilm) linearly to obtain 3-D images when desired. The LZ250 transducer has a laser input cable as well as a signal cable, which sends US and PA signals to the Vantage 256. The LZ250 transducer in combination with the Vantage 256 imaging platform is capable of acquiring PA images overlaid on an US background. US and PA images were acquired in an interlaced manner. Due to limited PCI slots, the main computer remotely controls a second computer to use the stepper motor. To allow two laser sources to enter the LZ250 laser input for reversible photoconversion, a dichroic mirror (DMLP650, Thorlabs) reflects wavelengths $<633\mathrm{nm}$ and transmits wavelengths $>685\mathrm{nm}$.

Fig. 2

Diagram shows the custom photoswitching PA imaging system used to obtain 2-D/3-D US and PA difference images with reversibly photoswitchable chromoproteins. The imaging target is imaged by the LZ250 transducer. DFG, delay function generator; FG, function generator; PD, photodiode; PC, personal computer; PC2, personal computer 2; and fire, flashlamp trigger.

In the PA experiments requiring photoconversion in this paper, we used the switching cycle shown in Fig. 3. The laser pulses at a 10-Hz repetition rate. 608-nm wavelength light converts the chromoproteins from the orange state to the far-red state in the first half of the imaging cycle. Image acquisition occurs at a 10-Hz rate in the second half of the imaging cycle using 700 nm+ wavelength light, which simultaneously photoconverts the chromoprotein from the far-red state back to the orange state. Difference images result from subtracting the final image from the first. The intermediate pulses are necessary to ensure full photoconversion between the far-red and orange states. It may be possible to remove the intermediate pulses with a large enough laser fluence. By reducing the number of intermediate pulses, this custom imaging system has acquired difference image data at a rate of 1 frame per second, which is 32-times faster than previously reported photoswitching PA imaging speeds.³⁰

Fig. 3

A single imaging cycle for imaging reversibly switchable chromoproteins.

2.2.

sGPC2

sGPC2 is an engineered protein that undergoes reversible photoswitching between a far-red (absorbance maximum $\sim 692\mathrm{nm}$) and an orange state (absorbance maximum $\sim 620\mathrm{nm}$). sGPC3 is a variant of sGPC2 with slightly different absorption spectra curves and a faster dark recovery (switching between states without light). For the purposes of this paper, they are essentially the same. These proteins were developed by the Campbell group at the University of Alberta. sGPC2 and sGPC3 were engineered from the biliverdin-binding second GAF domain (AM1_1557g2, residues $\sim 220$ to 364) of the cyanobacteriochrome of Acaryochloris marina MBIC11017 cyanobacteria.³⁵

Escherichia coli expressing either sGPC2, sGPC3, BphP1, or mIFP were used in our phantom and in vivo PA experiments. To prepare these samples, E. coli strain ElectroMAX™ DH10B (Invitrogen) was transformed by electroporation with pBAD/His B plasmids containing genes encoding the protein of interest and HO1. HO1 catalyzes the breakdown of heme to produce the pigment biliverdin, which binds to the chromoprotein. A 200-mL culture of lysogeny broth medium containing $0.1\mathrm{mg}/\mathrm{mL}$ ampicillin and 40 ppm L-arabinose was inoculated with a single colony. This culture was grown in a 500-mL baffled shake flask overnight ($225\mathrm{rev.}/\min$, $\sim 24\mathrm{h}$) at 37°C before cells were harvested by centrifugation and resuspended in PBS (pH 7.4).

2.3.

Difference-Spectra Demixing

Difference-spectra demixing can be used to obtain relative concentrations of reversibly switchable chromoproteins and offers improved differentiation and CNR in comparison with traditional spectral demixing. We describe the methodology in this section using sGPC2 and BphP1.

The PA signal at a location $r$ in vivo with sGPC2 and BphP1 present can be described by $P(r,\lambda ,{\eta }_S,{\eta }_B)=\mathrm{\Gamma }{\mu }_a(r,\lambda ,{\eta }_S,{\eta }_B)\mathrm{\varphi }(r,\lambda )$, where $\mathrm{\Gamma }$ is the Gruneisen parameter, ${\mu }_a$ is the optical absorption coefficient, $\mathrm{\varphi }$ is the fluence, $\lambda$ is the laser wavelength, and ${\eta }_{S/B}$ are the states of photoconversion for sGPC2/BphP1. $\eta =0$ represents the protein in its orange state, and $\eta =1$ represents the protein in its far-red state. The absorption coefficient can be written as

With several of these equations and assuming fluence is approximately wavelength independent, we could solve

In the case where imaging wavelengths only photoconvert one chromoprotein at a time as is shown in Fig. 1, Eq. (2) reduces to the following assuming two chromoproteins and two imaging wavelengths

In this case, the difference images may simply be normalized by their difference molar extinction spectra to produce relative concentration estimates. However, Eq. (2) is more general.

3.1.

Characterization

3.1.2.

Optical properties

The molar extinction coefficients of the proteins were determined as described by Filonov et al.³⁶ and Shu et al.³⁷ We found that sGPC2, sGPC3, and BphP1 have extinction coefficients of 87,200, 89,100, and $\mathrm{40,400}{\mathrm{M}}^{-1}·{\mathrm{cm}}^{-1}$, respectively.

We have also plotted the full absorption spectra of both the orange and far-red states of sGPC2 and sGPC3, Fig. 4. The absorption spectra of the orange and far-red states of BphP1 have been previously published.³⁰ For convenience, the absorption spectra for BphP1 are represented in Fig. 4 along with the absorption spectrum of the nonphotoswitching contrast agent, mIFP. mIFP will be relevant in the experimental section of this paper. Absorption spectra were recorded on a DU-800 UV–Visible spectrophotometer (Beckman). Absorption measurements were performed with a 1-cm quartz microcell cuvette. The absorption spectra of sGPC2 and sGPC3 are similar, as both have orange state and far-red state peaks around 630 and 700 nm, respectively. They differ in that sGPC2 has its highest absorption peak in its orange state, whereas sGPC3 has its highest absorption peak in its far-red state. In comparison, BphP1 has red state and far-red state peaks around 670 and 760 nm, respectively. For simplicity, we refer to the red state of BphP1 as the orange state. Experimentally determined difference spectra are used for difference-spectra demixing by taking the peak chromoprotein signal intensities from difference images taken at different wavelengths.

Fig. 4

Normalized absorption spectra of sGPC2, sGPC3, BphP1, and mIFP.

We compared the photobleaching properties of BphP1 with sGPC2 and sGPC3 by imaging them through six photoconversion cycles. PA images of three tubes, submersed in water-containing purified protein samples of sGPC2 ($89.53\mu \mathrm{M}$), sGPC3 ($133.84\mu \mathrm{M}$), and BphP1 ($89.17\mu \mathrm{M}$) were taken at wavelengths near their respective peaks of 710 ($4.3\mathrm{mJ}/{\mathrm{cm}}^2$ fluence) and 780 nm ($5.5\mathrm{mJ}/{\mathrm{cm}}^2$ fluence). The 608-nm wavelength used to photoconvert from the orange to the far-red state had a fluence of $3.6\mathrm{mJ}/{\mathrm{cm}}^2$. We used the photoconversion imaging cycle shown in Fig. 3 ($t_{\text{cycle}}=40.4\mathrm{s}$, 202 pulses per half cycle). We averaged the images acquired from each cycle. We assume that photobleaching would affect all images in the cycle proportionately despite decreasing imaging intensity due to photoconversion. We normalized the signal intensity by protein concentration and accounted for different fluence values at different wavelengths by plotting the normalized signal intensity by the accumulated fluence (AF) exposure. Figure 5 shows the photobleaching curves of sGPC2, sGPC3, and BphP1. We found that sGPC2 and sGPC3 photobleached 2.8 and 1.5 times faster than BphP1, respectively.

Fig. 5

Photobleaching curves of sGPC2, sGPC3, and BphP1.

3.1.3.

Photoconversion rate

We compared the photoconversion rates of sGPC2, sGPC3, and BphP1. We used the custom PA imaging system shown in Fig. 2. PA images of three tubes, submersed in water containing purified protein samples of sGPC2 ($89.53\mu \mathrm{M}$), sGPC3 ($133.84\mu \mathrm{M}$), and BphP1 ($89.17\mu \mathrm{M}$), were taken at wavelengths near their respective peaks of 710 ($4.3\mathrm{mJ}/{\mathrm{cm}}^2$ fluence) and 780 nm ($5.5\mathrm{mJ}/{\mathrm{cm}}^2$ fluence) as well as at 730 nm ($5.5\mathrm{mJ}/{\mathrm{cm}}^2$ fluence). We used the photoconverting imaging cycle shown in Fig. 3 ($t_{\text{cycle}}=4.4\mathrm{s}$, 22 pulses per half cycle) and were able to compare the photoconversion rates of sGPC2, sGPC3, and BphP1 as they were photoconverted back in the second half of the imaging cycle. Figure 6 shows the photoconversion as a function of AF at the wavelengths near their peak absorption wavelength. In that figure, the signal-to-noise ratios (SNRs) are solid lines with the standard error plotted above and below it in dotted red lines. The black dashed lines show where the proteins are 80% photoconverted. It is noteworthy that there seems to be some instability in the photoconversion of sGPC2 and sGPC3 at 710 nm. We believe that this is due to relatively high levels of absorption by the orange state at 710 nm, which is simultaneously converting some of sGPC2 and sGPC3 back to the far-red state. This seems to be supported by the fact that we found higher photoconversion stability of sGPC2 and sGPC3 when doing the same experiment at 730 nm, Fig. 7, where there is less absorption by the orange state. A similar instability is seen in BphP1 when the experiment is repeated at 730 nm, Fig. 7, where there is higher orange-state absorption. Despite improved stability at 730 nm for sGPC2 and sGPC3, 730 nm would not be preferable for imaging due to the relatively low change in SNR. It is also noteworthy that the dark recovery rate (rate of natural conversion from the orange to far-red state) of sGPC3 is substantially faster than that of sGPC2. However, this does not seem to affect the photoconversion rate at a given fluence.

Fig. 6

Photoconversion rate comparison between sGPC2 at 710, sGPC3 at 710, and BphP1 at 780 nm. Samples are normalized to 100 uM purified protein concentration. The intensity is plotted as a solid line with the standard error plotted as dotted red lines. The dashed black lines show where the proteins are 80% photoconverted.

Fig. 7

Photoconversion rate comparison between sGPC2 at 730, sGPC3 at 730, and BphP1 at 730 nm. Samples are normalized to $100\mu \mathrm{M}$ purified protein concentration. The intensity is plotted as a solid line with the standard error plotted as dotted red lines. The dashed black lines show where the proteins are 80% photoconverted.

Table 1 shows the photoconversion rates as well as the change in SNR and AF at 80% of the photoconverted state for sGPC2, sGPC3, and BphP1 in Fig. 6. The photoconversion rates with respect to percent photoconversion per $\mathrm{mJ}/{\mathrm{cm}}^2$ are 2.4, 1.6, and 0.6 for sGPC2, sGPC3, and BphP1, respectively. The photoconversion rates with respect to change in SNR per $\mathrm{mJ}/{\mathrm{cm}}^2$ are 0.86, 0.63, and 0.27 for sGPC2, sGPC3, and BphP1, respectively. sGPC2 has a 3.2 times faster photoconversion rate than BphP1 when measured by a change in SNR and has a four times faster photoconversion rate when measured by photoconversion percentage. This means that at the same fluence, sGPC2 will switch absorption states four times as fast as BphP1. BphP1 has the largest change in SNR between photoconverted states of 37.8 in comparison with 28.7 and 30.7 for sGPC2 and sGPC3, respectively. Thus, a higher CNR improvement is possible with BphP1.

Table 1

Comparison of the change in SNR and photoconversion rate. The comparison of the change in SNR and AF is taken at 80% of the fully converted state for sGPC2, sGPC3, and BphP1. Different measures of the photoconversion rates are shown.

Protein	80% photoconverted		Photoconversion rate
	ΔSNR	AF (mJ/cm2)	(ΔSNR per mJ/cm2)	(% per mJ/cm2)
sGPC2	28.7	34	0.86	2.4
sGPC3	30.7	49	0.63	1.6
BphP1	37.8	141	0.27	0.6

3.2.

Deep Photoacoustic Tube Phantom Imaging

We demonstrate the ability to obtain multiplexed PA difference-spectra demixed images deep in tissue using tube phantoms. We injected $100\mathrm{mg}/\mathrm{mL}$ concentrations of E. coli expressing sGPC2, sGPC3, BphP1, and mIFP through PFA medical microtubing (BB310-18, Scientific Commodities, Inc.) and insert them $\sim 10\text{-}\mathrm{mm}$ deep in raw chicken breast. mIFP is used as a nonphotoconverting contrast agent in addition to the oxygenated and deoxygenated hemoglobin present. We used our custom PA imaging system (Fig. 2) and the imaging cycle is shown in Fig. 3, with $t_{\text{cycle}}=20.4\mathrm{s}$ and 102 pulses per half cycle. The experiment was repeated at seven different image acquisition wavelengths including 700 ($9.2\mathrm{mJ}/{\mathrm{cm}}^2$), 710 ($10.7\mathrm{mJ}/{\mathrm{cm}}^2$), 720 ($10.5\mathrm{mJ}/{\mathrm{cm}}^2$), 730 ($10.7\mathrm{mJ}/{\mathrm{cm}}^2$), 740 ($10.9\mathrm{mJ}/{\mathrm{cm}}^2$), 750 ($11.3\mathrm{mJ}/{\mathrm{cm}}^2$), and 760 nm ($11.3\mathrm{mJ}/{\mathrm{cm}}^2$). It would have been preferred to also image at the 780-nm peak of BphP1, but we were unable due to low fluence at the 780-nm wavelength. The imaging wavelengths were also used to simultaneously convert from the far-red to the orange state. The 608-nm wavelength used to photoconvert from the orange to the far-red state had a fluence of $5.4\mathrm{mJ}/{\mathrm{cm}}^2$. We compare difference-spectra demixing with traditional spectral demixing.

Figure 8 shows the PA and spectral demixing results. Figure 8(a) shows the PA difference image taken at 710 nm overlaid on an US background. Figure 8(b) shows the PA image taken at 710 nm. Figures 8(c) and 8(d) show the PA difference image taken at 710 and 760 nm, respectively. Figures 8(e) and 8(f) show the traditional PA demixed image and the PA difference-spectra demixed image, respectively. It should be noted that sGPC2 and sGPC3 are demixed together because they have near identical absorption spectra.

Fig. 8

Deep tube phantom images of sGPC2, sGPC3, BphP1, and mIFP in chicken breast. (a) PA difference image taken at 710 nm and overlaid on an US image, (b) PA image taken at 710 nm, (c) PA difference image taken at 710 nm, (d) PA difference image taken at 760 nm, (e) traditionally demixed PA image using seven wavelengths (700, 710, 720, 730, 740, 750, and 760 nm) and overlaid on an US background, and (f) difference-spectra demixed PA image using the same seven wavelengths and overlaid on a US background.

PA images are obtained at a depth of around 10 mm in tissue. In Fig. 8(b), signal from the tubes and background blood is visible in addition to the protein signals, but taking the difference image in Fig. 8(c) removes the background and tube signal while leaving the sGPC2 and sGPC3 signal behind. Similarly in Fig. 8(d), the BphP1 signal remains while the other signals are mostly removed. When we compare the traditionally demixed PA image in Fig. 8(e) with the difference-spectra demixed image in Fig. 8(f), we see that the difference-spectra demixed image is better able to distinguish the concentrations of SGPC2/sGPC3 and BphP1. We leave out the demixed oxygenated and deoxygenated hemoglobin in Fig. 8(e) to reduce clutter. mIFP and hemoglobin are not present in the difference-spectra demixed image in Fig. 8(f), because it is subtracted in the difference image. The overall improvement in the difference-spectra demixed image results from the removal of mIFP. It is difficult to distinguish between sGPC2 and mIFP with traditional demixing due to their similar absorption spectra and PA signal intensity. It is also noteworthy that in Fig. 8(e), traditional demixing failed to properly demix BphP1 as it was confused with an area of blood outside the highlighted region. Although the optical properties of sGPC2 and sGPC3 are the same, differences in expression in E.coli bacteria may account for the differences in PA signal intensities seen in Fig. 8.

Comparing the difference image of sGPC2/sGPC3 in Fig. 8(c) to the traditional PA image in Fig. 8(b), we found a 4.2-fold CNR improvement at a 720-nm imaging wavelength, taking blood as the contrast signal. When comparing the difference-spectra demixed image of sGPC2/sGPC3 with traditional demixing, we observe a 1.3-fold CNR improvement using difference-spectra demixing. In addition, difference-spectra demixing significantly reduced extraneous signal cross talk for sGPC2/sGPC3 and BphP1 by 17.9% and 4.8%, respectively.

3.3.

In Vivo Photoacoustic Imaging

We next demonstrate our approach in vivo. We used our custom PA imaging system (Fig. 2) and the imaging cycle shown in Fig. 3 with $t_{\text{cycle}}=2.4\mathrm{s}$ and 12 pulses per half cycle. The experiment was repeated at seven different image acquisition wavelengths as explained in Sec. 3.2. The 608-nm wavelength was again used to photoconvert from the orange to the far-red state ($5.4\mathrm{mJ}/{\mathrm{cm}}^2$). We injected $100\mathrm{mg}/\mathrm{mL}$ concentrations of E. coli expressing sGPC2, BphP1, and mIFP into the hind flank of a hairless skid mouse (Nu/Nu, Charles River). Either sGPC2 or sGPC3 would have been suitable as a photoswitchable reporter for this experiment. mIFP is used as a nonphotoconverting contrast agent in addition to the oxygenated and deoxygenated hemoglobin present in vivo. The difference image removes signal from both the mIFP and hemoglobin. In addition, we compare difference-spectra demixing with traditional spectral demixing. All animal experiments were approved by the Animal Care and Use Committee at the University of Alberta.

Figure 9 shows the PA and PA difference image results. Figure 9(a) shows the PA difference image taken at 710 nm overlaid on an US background. Figure 9(b) shows the PA image taken at 710 nm. Figures 9(c) and 9(d) show the PA difference image taken at 710 and 760 nm, respectively. Figure 10 shows the traditional and difference-spectra demixing results overlaid on an US background. Figures 10(a) and 10(b) show the traditional PA demixed image and the PA difference-spectra demixed image, respectively. The raw difference-spectra and traditional spectrally demixed images can be seen in Fig. 11.

Fig. 9

In vivo PA and PA difference images of sGPC2, BphP1, and mIFP injected into a mouse. (a) PA difference image taken at 710 nm and overlaid on the US image, (b) PA image taken at 710 nm, (c) PA difference image taken at 710 nm, and (d) PA difference image taken at 760 nm.

Fig. 10

In vivo demixed images of sGPC2, BphP1, and mIFP injected into a mouse. (a) traditionally demixed PA image using seven wavelengths (700, 710, 720, 730, 740, 750, and 760 nm) and overlaid on an US background and (b) difference-spectra demixed PA image using the same seven wavelengths and overlaid on an US background.

Fig. 11

The raw difference-spectra demixed in vivo images of (a) sGPC2 and (b) BphP1 as well as the traditional spectrally demixed in vivo images of (c) sGPC2, (d) BphP1, (e) deoxygenated hemoglobin (Hb), (f) oxygenated hemoglobin (${\mathrm{HbO}}_2$), and (g) mIFP are shown.

In Fig. 9(b), a large background signal can be observed in comparison with the sGPC2 signal, but the difference image in Fig. 9(c) removes the background signal. Figure 9(d) shows the removal of background signal around BphP1 at 760 nm. Comparing the demixed PA image in Fig. 10(a) with the difference-spectra demixed image in Fig. 10(b), we see that the difference-spectra demixed image is again better able to distinguish the concentrations of sGPC2 and BphP1 due to the removal of mIFP, which is difficult to demix due to having a similar absorption spectrum as sGPC2. We leave out the demixed oxygenated (HbO2) and deoxygenated (Hb) hemoglobin in Fig. 10(a) to reduce clutter. mIFP and hemoglobin are not present in the difference-spectra demixed image in Fig. 10(b), because it is subtracted in the difference images.

Comparing the difference image of sGPC2 to the traditional PA image in Fig. 9(c) to the traditional PA image in Fig. 9(b), we found a 21.6-fold CNR improvement at a 720-nm imaging wavelength and taking mIFP as the reference signal. The CNRs of the difference-spectra demixed images of sGPC2 and BphP1 were larger by a magnitude of 21.6 and 24.0, respectively, in comparison with the CNRs of the traditional spectrally demixed images of sGPC2 and BphP1. In addition, difference-spectra demixing significantly reduced extraneous cross talk for sGPC2 and BphP1 by 8.2% and 40.2%, respectively.