High-Throughput Platform for Optoacoustic Probing of Genetically Encoded Calcium Ion Indicators

Summary

Functional optoacoustic (OA) imaging assisted with genetically encoded calcium ion indicators (GECIs) holds promise for imaging large-scale neuronal activity at depths and spatiotemporal resolutions not attainable with existing optical microscopic techniques. However, currently available GECIs optimized for fluorescence (FL) imaging lack sufficient contrast for OA imaging and respond at wavelengths having limited penetration into the mammalian brain. Here we present an imaging platform capable of rapid assessment and cross-validation between OA and FL responses of sensor proteins expressed in Escherichia coli colonies. The screening system features optimized pulsed light excitation combined with ultrasensitive ultrasound detection to mitigate photobleaching while further allowing the dynamic characterization of calcium ion responses with millisecond precision. Targeted probing of up to six individual colonies per second in both calcium-loaded and calcium-unloaded states was possible with the system. The new platform greatly facilitates optimization of absorption-based labels, thus setting the stage for directed evolution of OA GECIs.

Graphical Abstract

Highlights

• •New platform for rapid screening of optoacoustic calcium ion indicators
• •Coregistered fluorescence and optoacoustic recordings from bacterial colonies
• •Sensitive imaging and selective probing in calcium-loaded and calcium-unloaded states
• •Assessment of protein photobleaching resistance to pulsed laser irradiation

Analytical Chemistry; Bioengineering; Biomaterials

Introduction

Functional, volumetric, and non-invasive imaging of large neuronal populations at high spatiotemporal resolution is essential for expanding our understanding of complex brain functions (Ji et al., 2016, Prevedel et al., 2014). Fluorescence (FL) microscopy assisted with genetically encoded calcium ion indicators (GECIs) has facilitated the direct visualization of neuronal activation beyond traditional macroscopic observations looking at hemodynamic changes as indirect manifestation of neural activity (Kannan et al., 2018). State-of-the-art techniques such as multiphoton excitation fluorescence imaging have enabled the mapping of fast neuronal activation with sub-cellular resolution at hundreds of micrometers depths in the scattering brain of rodents (Chamberland et al., 2017). Extensive efforts are being directed toward optimizing quantum yield ($\sigma$), molar extinction coefficient (ε), kinetics, and photostability of GECIs as well as other calcium ion (${\text{Ca}}^{2+}$) and voltage sensors (Thestrup et al., 2014, Chen et al., 2013, Mank et al., 2006, Shen et al., 2018). Yet, intense light scattering in the brain, skin, and skull restricts optical microscopy to highly invasive observations at sub-millimeter depths and fields of view (FOVs) while imposing additional limitations on the effective spatiotemporal resolution due to the need for beam scanning (Helmchen and Denk, 2005, Mittmann et al., 2011, Stirman et al., 2016). Optoacoustic (OA) imaging overcomes the optical diffusion barrier and hence bridges the gap between microscopic and macroscopic realms (Deán-Ben et al., 2017, Wang, 2009). Being based on optical-absorption contrast, OA methods uniquely enable multi-scale imaging of hemodynamic responses across entire mammalian brains at several millimeters' depth (Ni et al., 2018, Gottschalk et al., 2016, Wang et al., 2003, Yao et al., 2015). More importantly, OA has recently been shown to be sensitive to ${\text{Ca}}^{2+}$ and voltage signaling in vivo (Deán-Ben et al., 2016, Rao et al., 2017, Zhang et al., 2017, Gottschalk et al., 2019). This unique contrast versatility of OA imaging, combined with its high temporal resolution and large scalability, is poised to advance our understanding of neuronal activity and mechanisms of neurovascular coupling in the whole rodent brain (Deán-Ben et al., 2017).

Existing GECIs and genetically encoded voltage indicators (GEVIs) are, however, sub-optimal for OA imaging because most biosensors used to visualize neuronal activity have been specifically optimized for FL imaging (Laufer et al., 2013). For instance, strong variations in $\sigma$ may not result in detectable OA responses (Weber et al., 2016). Generally, proteins with low $\sigma$ have improved OA responses and have been shown to be more photostable (Laufer et al., 2013). Sensors responding to ${\text{Ca}}^{2+}$ changes by strong corresponding changes in their ε are available, holding promise for OA neuroimaging applications (Shen et al., 2018, Zhao et al., 2011). Particularly, green fluorescent protein-derived GCaMP sensors have been shown to provide sufficient OA signal changes to be detectable in highly vascularized mammalian brains (Gottschalk et al., 2019). Red fluorescent genetically encoded ${\text{Ca}}^{2+}$ indicators for optical imaging (R-GECO) exhibit similar changes in ε when bound to ${\text{Ca}}^{2+}$ while having excitation and emission spectra in a wavelength range where scattering and auto-fluorescence tissue background is diminished (Shen et al., 2018). Yet, an optimal sensor for OA ${\text{Ca}}^{2+}$ imaging would feature strong fractional absorbance changes in response to ${\text{Ca}}^{2+}$ variations and have its peak absorption in the far-red or near-infrared spectral range where the blood absorption background also is diminished. However, currently available protein screening platforms, e.g., employing directed evolution approaches, are tailored toward optimization of FL imaging performance, making them unsuitable for screening the optimal variants for OA (absorption-based) imaging.

Herein, we present an imaging and screening platform for efficient characterization of changes in both OA and FL signal intensity upon ${\text{Ca}}^{2+}$ delivery and showcase its performance for high-throughput screening of a large number of Escherichia coli (E. coli) bacterial colonies expressing R-GECO1 proteins. The system thus enables optimization of sensor's $\sigma$, ε, and photostability and can readily be adapted to other genetically encoded biosensors.

Results

The Rapid Protein Probing Platform

The developed hybrid imaging platform allows for a rapid screening of both OA and FL properties of several hundred E. coli colonies growing on a Petri dish and expressing genetically encoded protein variants (Figure 1). In a mechanical overfly scan mode, OA imaging is performed over the entire 50 mm × 50 mm FOV with a uniform step size of 100 μm, which takes about 150 s (line scan rate of 3.34 Hz). The pulse repetition frequency (PRF) of the excitation laser reaches 5 kHz when the stages reach their maximum velocity of 500 mm s⁻¹ during the scan. Oversampling is avoided at acceleration and deceleration phases by using position-dependent triggering of the laser source (see Transparent Methods for details). Band-pass frequency filtering and envelope extraction of the acquired datasets were performed during the scan, allowing for line-by-line live preview of the acquired image data.

Figure 1.

Combined OA and FL Screening Platform for ${\text{Ca}}^{2+}$ Biosensors

A Petri dish (PD) positioned in a holder (PDH) contains cell colonies (CC) on a filter paper (FP) immersed in screening buffer (SB) for acoustic coupling. A motorized z-stage (ZS) is used to switch between OA and FL imaging modes. In the OA mode, an FOV of 50 mm × 50 mm is rapidly acquired by scanning an US transducer (UST) over the PD using two perpendicularly mounted stages (XS, YS). A multimode fiber (MMF) delivers laser pulses from a dye laser (DL), tuned to the required wavelength for OA excitation, through an opening in the transducer's active aperture. FL imaging is performed with a camera (FLC) equipped with an emission filter (EMF) whose FOV matches the size of the Petri dish. For imaging, the platform is lowered by 30 mm and the transducer is moved out of the camera's FOV. For uniform concentric illumination, excitation light was provided by a tungsten light source from below (FLS) equipped with the appropriate excitation filter (EXF) for the tested protein.

For a more subtle protein characterization, positions of all the colonies are first mapped using either an overfly OA scan or an FL wide-field image and colonies are subsequently probed individually in the OA mode. This further allows to increase screening speed, diminish bleaching effects, and/or enable signal averaging. In our first experiments, over 95% of the colonies were accurately localized with 100 μm precision from the FL image by applying a circle detection algorithm (Figure 2A, see Methods for details). Time required to travel between individual colonies was subsequently calculated taking into account stage accelerations and maximum velocities (Figure 2B). The so-called traveling salesman problem was solved to determine the optimal colony probing path using a genetic algorithm (Figure 2C, see Transparent Methods for details) (Beardwood et al., 1959). Example of an optimized path for visiting 483 colonies is shown in Figure 2D. An average probing speed of 6 colonies per second was attained when accounting for probing, data acquisition, and hardware interfacing.

Figure 2.

Path Optimization Method for Fast OA Probing of Cell Colonies in a Petri Dish

(A) Before selective OA probing, an FL image of the entire Petri dish is acquired to identify the location of the colonies based on their autofluorescence or ${\text{Ca}}^{2+}$-mediated signal.

(B) Time required to travel between individual cell colonies is calculated taking the stage acceleration and its maximum velocity into account.

(C) The probing order is optimized using iterative genetic optimization algorithm, e.g., resulting in a theoretical travel time of only 21 s for visiting 483 colonies. Note that data acquisition, hardware communication, laser triggering, and post-processing add an overhead to the actual probing times.

(D) Example of the optimized scanning path.

Optoacoustic and Fluorescence Probing of GECIs

Owing to its distinct absorption spectrum and high sensitivity to ${\text{Ca}}^{2+}$ dynamics, R-GECO1 was selected for testing signal to noise ratio (SNR), imaging speed, and ${\text{Ca}}^{2+}$ delivery protocol attained by the developed platform (calculated OA and FL properties of alternative GECIs are enlisted in Table S1). Note that the cell colonies are not visible with the naked eye in both loaded and unloaded states (Figure 3A), underlining the required high sensitivity of the OA probing platform. The OA spectrum of the purified protein correlates well with the optical absorption spectrum measured in a wavelength range between 430 nm and 680 nm (Figure 3B). High relative changes in OA and FL signals were observed when the protein binds to ${\text{Ca}}^{2+}$, indicating that the sensor is suitable for both modalities. The strong changes in OA spectra further indicate that ε is substantially altered due to ${\text{Ca}}^{2+}$ binding, which is consistent with the observed strong changes in the optical absorption spectrum. Specifically, below an excitation wavelength of 475 nm, a moderate decrease in OA signal is observed following ${\text{Ca}}^{2+}$ delivery (−33% on average). The OA signal reaches its maximum intensity near the optical excitation peak of R-GECO1 (577 nm/561 nm for ${\text{Ca}}^{2+}$ unbound/bound states, respectively). For 560 nm excitation wavelength, ${\text{Ca}}^{2+}$ delivery has boosted the OA signal intensity by 201% relatively to the ${\text{Ca}}^{2+}$-free baseline value, comparable to the signal increase in the optical absorption spectra. The protein did not bleach significantly over 50 measurement cycles.

Figure 3.

OA and FL Probing of R-GECO1 Protein

(A) Photograph of a Petri dish containing a filter paper with E. coli colonies expressing R-GECO1.

(B) The optical absorption and OA spectra of the purified protein R-GECO1 were analyzed in ${\text{Ca}}^{2+}$-unloaded and Ca²⁺-loaded states.

(C–G) (C) Screening workflow using hybrid OA and FL imaging platform. (D and E) FL image of the colonies before and after ${\text{Ca}}^{2+}$ loading, respectively. The corresponding OA images before and after ${\text{Ca}}^{2+}$ loading are shown in (F) and (G), respectively.

(H) The correlation between signal increase in the FL and OA modes was analyzed for each individual colony, revealing a relatively low value of R² = 0.38.

(I) The average OA signal increase after ${\text{Ca}}^{2+}$ delivery (305%) was significantly higher than the corresponding FL value (p < 0.001, 270%). No significant signal variations were recorded in the control group containing no protein.

OA and FL images of 500–800 individual R-GECO1-expressing colonies positioned in a single Petri dish were subsequently acquired before and after ${\text{Ca}}^{2+}$ delivery (Figures 3D–3G) according to the workflow depicted in Figure 3C. Sufficiently strong OA signals from single laser shots (SNR >20 dB) were generated by per-pulse-energies (PPEs) as low as 25 $\mathrm{\mu }\text{J}$ (6.54 mJ cm⁻²). The increase in OA and FL signal intensity correlated with R² = 0.38 (Figure 3G). E. coli colonies expressing R-GECO1 showed a strong increase in OA signal following ${\text{Ca}}^{2+}$ delivery by approximately 305% compared with ${\text{Ca}}^{2+}$-free baseline value, whereas the FL signals increased by approximately 270% compared with the baseline value after ${\text{Ca}}^{2+}$ delivery (Figures 3H and 3I). The lower signal enhancement in FL mode can be attributed to higher background autofluorescence signal of flavins (Mihalcescu et al., 2015). Control plates were imaged before and after adding 3-(N-morpholino)propanesulfonic acid (MOPS) buffer instead of ${\text{Ca}}^{2+}$-enriched MOPS buffer to evaluate the influence of photoactivation effects. The latter were previously reported to lead to approximately 20% FL signal increase for the K-GECO1 sensor at 405 nm and 488 nm after light fluence deposition of 1.76 J cm⁻² and 6.12 J cm⁻², respectively (Shen et al., 2018). The FL signal of the control plate slightly increased by 24%, whereas an opposite effect was observed in the OA mode. This finding indicates slight photoactivation of the protein during the first OA scan, which was performed between the first and second FL image acquisition. During the second OA scan, bleaching effects have dominated the photoactivation, leading to an overall 24% decrease in OA signal intensity (Figures 3H and 4A).

Figure 4.

Dynamic Monitoring of Cell Properties in OA and FL Modes

(A) Photobleaching characteristics of R-GECO1 in OA probing mode when colonies are iteratively exposed to pulsed laser radiation. To distinguish between bleaching of background biochromes and ${\text{Ca}}^{2+}$-sensitive protein, experiments were performed in ${\text{Ca}}^{2+}$-loaded (solid curves) and Ca²⁺-unloaded (dashed curves) states. Each measurement represents an average of 30 probed colonies. The inset shows histograms of the PPEs measured by the photodiode at the three used energy levels.

(B) The corresponding photobleaching effects in FL imaging mode using continuous-wave excitation light.

(C) The FL signal continuously monitored after delivering ${\text{Ca}}^{2+}$ to the cells at t = 0 min: following strong increase in the signal intensity after approximately 4 min, the signal decays toward an equilibrium state after approximately 20 min. The excitation light intensity was the same as for the FL bleaching experiments. The time window used for screening is highlighted in red.

To showcase the platform's capability to distinguish between different proteins, we further imaged K-GECO1 (Shen et al., 2018) and RCaMP1.07 (Ohkura et al., 2012) expressing colonies. OA signal increases upon ${\text{Ca}}^{2+}$ were measured to be 3.1% and 21.1% respectively, i.e., nearly one order of magnitude lower responses compared with R-GECO1. The difference between theoretically calculated values and experimentally measured signals can be ascribed to a lower resistance of K-GECO1 and RCaMP1.07 to photobleaching (Shen et al., 2018).

To provide stronger evidence for the cell culture viability following OA screening, a plate containing R-GECO1-expressing E. coli colonies was additionally imaged at a high PPE (50$\mathrm{\mu }\text{J}$, 560 nm). The colony providing highest OA signal was subsequently picked, mechanically resuspended, and transferred to a fresh plate. After overnight incubation, visual inspection clearly revealed the expected growth, showcasing cell survival after the screening procedure.

Photobleaching Effects due to Probing

The photobleaching characteristics of R-GECO1-expressing cell colonies were evaluated dynamically using the same platform. Cell colonies were exposed to pulsed laser radiation while OA signal intensity was measured as a function of the deposited energy (Figure 4A). Bleaching experiments were performed at three different PPE levels, namely, 25, 34, and 44 $\mathrm{\mu }\text{J}$ (see inset in Figure 4A), corresponding to 250–450 mJ cm⁻² cumulative laser energy density reaching the cells. Note that the bleaching affects other background chromophores besides the proteins, which results in a complex photobleaching pattern contributed by multiple fluorescent and absorbing structures each having four possible states, namely, higher-energy excited singlet state $S^{\ast }$, lower-energy singlet state S, excited triplet state $T^{\ast }$, and bleached State X (Gangola and Rosen, 1987). To distinguish between bleaching of cell compartments and ${\text{Ca}}^{2+}$ sensor, measurements were performed at both high and low ${\text{Ca}}^{2+}$ concentrations (solid and dashed lines in Figure 4A, respectively). Although bleaching of background structures is independent of ${\text{Ca}}^{2+}$ concentration and therefore identical in both cases, the protein in the ${\text{Ca}}^{2+}$-loaded state exhibits a higher absorption coefficient, leading to an increased probability for molecules to be pumped into $S^{\ast }$ (proportional to the OA signal intensity). Therefore, initial signal intensities were higher for all ${\text{Ca}}^{2+}$-loaded cases when compared with their ${\text{Ca}}^{2+}$-unloaded counterparts. Furthermore, the proteins can transition into the bleached state X from either $S^{\ast }$ or the excited triplet state $T^{\ast }$, which occurs significantly faster for the ${\text{Ca}}^{2+}$-loaded case. Owing to increasing size of X and a constant overall population size $\left(S+S\ast +T\ast +X=\text{const}.\right)$, S and $S^{\ast }$ are decaying faster, leading to a faster decrease in signal intensity for the ${\text{Ca}}^{2+}$-loaded state. A maximum signal decrease of 55% was observed for the ${\text{Ca}}^{2+}$-loaded state at PPEs of 44 $\mathrm{\mu }\text{J}$.

Photobleaching in FL mode was investigated by exposing the entire Petri dish to an unfocused continuous-wave excitation light source with cumulative energy density of 88–177mJ cm⁻² deposited onto the cells. The FL signal decreased by 10% and 12% at high and low ${\text{Ca}}^{2+}$ concentrations, respectively, after 25 min exposure time (Figure 4B).

Calcium Ion Delivery Dynamics

During ${\text{Ca}}^{2+}$ delivery, the FL signal reaches its peak intensity after 227 s, and then decays by 49% (Figure 4C). As the protein is restricted to the intracellular space, FL and OA signal intensities are directly correlated to ${\text{Ca}}^{2+}$ concentration within the bacteria. It is likely that adding ${\text{Ca}}^{2+}$ to the buffer medium leads to an initial high uptake of the bacteria before it is actively transported out of the cell interior. The ${\text{Ca}}^{2+}$ uptake dynamics are furthermore affected by the time bacteria are stored out of fridge before screening (Figure S1). To avoid unwanted influences of ion uptake dynamics, all experiments were carefully time controlled. Plates were taken from a 7° fridge and imaged immediately (t = 0 min), ${\text{Ca}}^{2+}$ was delivered at t = 5 min, and plates were imaged again at t = 10 min.

Discussion

We developed an imaging platform for combined and co-registered evaluation of key OA and FL excitation parameters of proteins expressed in bacterial colonies. The system can characterize different parameters dynamically without the need for protein purification or special treatment, i.e., by observing cells embedded in a Petri dish directly. We showcased the platform's functionality using ${\text{Ca}}^{2+}$ delivery in R-GECO1-expressing E. coli.

OA properties of bacterial cell colonies distributed over an area of up to 50 mm × 50 mm, i.e., the approximate size of a standard Petri dish, can be characterized within less than 3 min by performing a mechanical overfly scan. In addition, rapid targeted OA probing of individual colonies was performed to extract important protein characteristics, including bleaching and ${\text{Ca}}^{2+}$ delivery dynamics.

The use of a fast, wavelength-tunable dye laser for OA excitation increases probing and imaging speed of this setup by several orders of magnitude compared with previously reported design using optical parametric oscillator lasers (Li et al., 2016). Although mirror-based OA microscopy scanners can image at higher speeds, their effective FOV is diminished due to a highly non-uniform sensitivity field of an ultrasound (US) detector remaining in a constant position (Allen et al., 2018). In contrast, our mechanical scanning approach allowed for attaining uniform sensitivity field in detecting the generated US signals over large FOVs, an ideal trait for high-throughput protein screening.

High-power lasers often employed in OA tomography applications are known for their strong protein bleaching characteristics, thus making photostability an important criterion when developing biosensors for OA imaging (Gottschalk et al., 2015). Using a highly sensitive, focused US sensor in combination with an optimized OA excitation scheme significantly decreased laser exposure and thus bleaching effects while boosting the screening throughput. Yet, OA imaging induced more significant photobleaching in the probed colonies when compared with FL imaging owing to the use of high-peak-power nanosecond-duration laser pulses for OA excitation (Gottschalk et al., 2015). The overall deposited energy in our experiments was 2–3 times higher for the OA mode. In addition, accurate assessment of the bleaching effects caused by nanosecond laser pulses was facilitated by means of the targeted probing mode.

R-GECO1 exhibited a strong increase in its signal intensity following ${\text{Ca}}^{2+}$ delivery in both FL and OA modes. The background signal induced by the intrinsic cell structures was significantly lower in the OA mode leading to better contrast for ${\text{Ca}}^{2+}$ concentration. However, in an early stage of the protein screening process, the OA signal intensity might be insufficient to accurately extract OA properties of all the colonies from an overfly scan. In this case, imaging autofluorescence of bacterial cell compartments in FL mode allows for an accurate identification of colonies even if the protein is in its inactive state. After localizing the colonies based on the FL image, individual OA probing can be facilitated by means of signal averaging to increase the SNR and compare between colonies expressing mutant protein variants.

The current study focused on screening of ${\text{Ca}}^{2+}$-sensitive proteins in E. coli colonies, which were previously shown as a robust platform for directed evolution purposes (Fabritius et al., 2018). Nevertheless, additional applications can be envisioned, such as screening proteins for their dopamine responses (Patriarchi et al., 2018, Sun et al., 2018), using a different cell substrate, optimizing OA signal of chromoproteins, or study the effects of nanobioconjugates influencing ${\text{Ca}}^{2+}$ uptake using OA (Lyu et al., 2016). However, the current sensitivity and spatial resolution of the US detector are not suitable for resolving or sensing individual cells. Our screening platform can seamlessly be operated in a pulse-echo US mode (Estrada et al., 2014), and thus can be potentially employed for screening bacteria containing gas vesicles using its pulse-echo imaging capability (Bourdeau et al., 2018).

OA imaging has the unique capacity to map a large number of hemodynamic parameters using endogenous tissue contrast, such as the dynamic distribution of oxygenated and deoxygenated hemoglobin or the total blood volume (Rebling et al., 2018, Wang, 2009, Yao et al., 2015), and hence offers a more versatile contrast when compared with other neuroimaging techniques (Gozzi and Schwarz, 2016, Jonckers et al., 2011). However, optical absorption of most neuronal activity sensors peaks in the visible light range, limiting their in vivo applicability to shallow depths owing to the high absorption by blood in highly vascularized mammalian tissues at those wavelengths. The development of efficient far-red- and near-infrared-shifted sensors using the newly developed protein screening platform can thus open new vistas for the study of neuronal function and neurovascular coupling in whole mammalian brains, a long-standing goal of neuroscience.

In this work, simultaneous probing of both OA and FL properties of proteins expressed in E. coli colonies was demonstrated with our method. However, the approach can generally be applied for screening GECIs or other proteins featuring sensing or contrast modulation properties (Patriarchi et al., 2018, Zhao et al., 2011, Shen et al., 2018). For these and other proteins, optimization of maximum OA signal intensity, relative signal change, and protein dynamics are essential to take full advantage of their sensing capacities.

The devised system is therefore expected to facilitate the development of a new generation of sensors optimized for a variety of contrast-enhanced OA imaging applications using directed evolution of gene reporter proteins.

Limitations of the Study

The presented platform is optimized for screening proteins in bacterial colonies allowing for a high throughput and robust handling. In this setting, functionality of the optimized variants needs to be validated in mammalian cell cultures followed by in vivo validation.

In our current proof-of-concept experiments, the sensors employed were not yet optimized for OA imaging performance, thus exhibiting low SNR. As a result, relatively high laser intensities were necessary, leading to partial photobleaching. Optimization and comparison between different proteins should therefore rely on their combined OA and photobleaching characteristics.

In our previously published work we have shown the ability to automatically pick bacterial colonies from the dish based on their fluorescent properties (Fabritius et al., 2018). This capacity has not yet been implemented in the newly developed OA screening platform.

Methods

All methods can be found in the accompanying Transparent Methods supplemental file.

Published: December 20, 2019

Data and Code Availability

The software used for postprocessing and hardware control is available online at https://github.com/razanskylab/. Data are available upon request.