Semisynthetic pH-sensitive fluorophores for imaging exocytosis and endocytosis

The GFP-based superecliptic pHluorin (SEP) enables detection of exocytosis and endocytosis, but its performance has not been duplicated in red fluorescent protein scaffolds. Here we describe ‘semisynthetic’ pH-sensitive protein conjugates that match the properties of SEP. Conjugation to genetically encoded self-labeling tags or antibodies allows visualization of both exocytosis and endocytosis, constituting new bright sensors for these key steps of synaptic transmission.

A useful extension of this technology has been the creation of pH sensors based on red fluorescent proteins (RFPs) such as mOrange2 3 , pHTomato 4 , pHoran4 5 , and pHuji 5 . Longer excitation wavelengths are less phototoxic, elicit lower levels of autofluorescence, facilitate multicolor imaging experiments, and allow concomitant use of optogenetics. Nevertheless, it has proven difficult to engineer red-shifted pHluorins that match the optimal pKa, cooperativity, and dynamic range of SEP, perhaps due to inherent limitations in RFP scaffolds. More generally, these techniques rely on overexpression of reporter proteins in SVs and the effect of overexpression is a confounding factor in interpreting experimental results 6 .
Given the limitations of pH-sensitive RFPs and the potential problems with overexpression of sensor proteins, we pursued an alternative strategy: creation of 'semisynthetic' pH indicators using organic pH-sensitive dyes attached to either expressed self-labeling tags such as the SNAP-tag 7 or antibodies that recognize native vesicular proteins (Fig. 1a). To match the performance of SEP, we required a pH-sensitive organic dye that can undergo a cooperative transition from a bright, fluorescent form at neutral pH to a nonfluorescent form at low pH. Unfortunately, the majority of pH-sensitive dyes do not meet these requirements. The archetypical small molecule pH-sensor is fluorescein (Fl, 1, Fig. 1b), which transitions between a highly fluorescent dianion (1 2-) and a less fluorescent monoanion (1 -) with a relatively low pKa value of 6.3 ( Fig. 1c) 8 . Other unsuitable synthetic pH probes include the ratiometric seminapthorhodofluor (SNARF) dyes 9 that exhibit high background, as well as cyanine and rhodamine-based pH sensors that show the opposite pH sensitivity profile 10,11 .
We recently synthesized new derivatives of fluorescein (1) where the xanthene oxygen was replaced with a gem-dimethylcarbon moiety. This work resulted in 'carbofluorescein' (CFl, 2, Fig. 1b) 12 , and the difluorinated derivative 'Virginia Orange' (VO, 3, Fig. 1b) 13 . We discovered that this oxygencarbon substitution elicited significant changes in photophysical and chemical properties of the fluorescein scaffold. Fl (1) exhibits λex/λem = 491 nm/510 nm at high pH, whereas CFl (2) and VO (3) are red-shifted with λex/λem = 544 nm/567 nm and 555 nm/581 nm, respectively. In addition to this bathochromic shift, the pH sensitivity of the dyes was markedly different. Fluorescein exhibits strong visible absorption at both pH 5.6 (vesicle pH) where the monoanion 1dominates, and pH 7.4 (extracellular pH) where the dianion form 1 2is prevalent-this can be observed by eye (Fig. 1c,d). In contrast, CFl (2) undergoes a cooperative transition between a highly colored dianion species (2 2-) and a colorless lactone form (2lactone; Fig. 1c). This is also evident visually as a solution of CFl (2) is colorless at pH 5.6, but shows robust visible absorption at pH 7.4 (Fig. 1d). Fluorescence-based titrations (Fig. 1e) gave pKa values of 6.3 and Hill coefficient (ηH) value of 0.97 for fluorescein (1), consistent with previous reports 8

. CFl
(2) and VO (3) displayed pKa values of 7.5 and 6.7, and ηH values of 1.6 and 1.5, respectively. This cooperative transition likely stems from the altered lactone-quinoid equilibrium observed in the carbon-containing analogs of fluorescein and rhodamine dyes 12 . The longer absorption and emission wavelengths, higher pKa, and the cooperative colorlesscolored transition upon increased pH make both CFl and VO attractive scaffolds for building indicators to monitor synaptic vesicle fusion events.
To allow for specific labeling of expressed proteins, we prepared the SNAP-tag ligands attached to CFl (4) or VO (5) (Fig. 1f, Supplementary Note). We tested the effects of protein conjugation on the properties of the dye by labeling SNAP-tag protein in vitro with CFl-SNAP-tag ligand 4 (Fig. 1g). We observed a shift in pKa to 7.3, and a decreased Hill coefficient (ηH = 1.2; Fig. 1h). The active site of the SNAP-tag enzyme is flanked with two Lys, one Arg, and one His (PDB structure 3KZZ, DOI: 10.2210/pdb3kzz/pdb). The resulting Coulombic interaction between these positively charged amino acid residues and the CFl label most likely explains the decrease in pKa upon conjugation 8 . This polar surface might also stabilize the open form of the dye, resulting in the decreased cooperativity of the coloredcolorless transition. Despite this lower pKa value and Hill coefficient, the fluorescence of the SNAP-tag-CFl conjugate is still completely suppressed at pH 5.6 ( Fig. 1i).
Next, we tested these SNAP-tag-based probes in living cells. Building on existing SEP-based constructs 1,2 we designed two SNAP-tag fusion proteins: (i) SNAP-tag inserted within an intra-luminal loop of the vesicle acetylcholine transporter VAChT (VAChT-SNAP), and (ii) SNAP-tag protein attached to the luminal C-terminal side of the vesicle protein VAMP2 (VAMP2-SNAP). These proteins were expressed in neuroendocrine PC12 cells, where VAChT is targeted to small synaptic-like vesicles (SSLV) while VAMP2 is found in both SSLV and large dense core vesicles 14,15 . We found that the propensity of the CFl and VO fluorophores to adopt the neutral lactone form (Fig. 1c,d) allows for efficient intracellular labeling with SNAP-tag ligands 4 or 5 without the use of other masking groups (e.g., acetate esters), which are typically required for fluorescein-based compounds (Fig. 2a). To monitor exocytosis, we depolarized cells with stimulation buffer containing high [K + ] and imaged single small vesicles as they fused with the plasma membrane using total internal reflection fluorescence (TIRF) microscopy. Cells containing VAChT constructs displayed events at high frequency. Events detected in cells co-expressing VAChT-SEP and VAChT-SNAP (labelled with CFl ligand 4) showed comparable fold increases in fluorescence at exocytosis (2.19 ± 0.07 vs. 2.40 ± 0.12) with similar decay kinetics in both the green and red channels (Fig. 2b,c). We also compared VAChT-SEP to VAChT-SNAP-VO (Fig. 2d,e) and VAChT-pHuji ( Fig. 2f,g). Like the semisynthetic indicator from CFl ligand 4, the VAChT-SNAP-VO derived from compound 5 also showed comparable performance to the SEP sensor (2.65 ± 0.10 vs. 2.60 ± 0.16 fold increase; Fig. 2d,e). However, in PC12 cells the RFP-based VAChT-pHuji sensor showed lower relative performance when compared with VAChT-SEP under the same conditions (2.01 ± 0.05 vs. 1.32 ± 0.02 fold increase; Fig. 2f,g) making events harder to detect with pHuji than with the other pH-sensitive proteins. We also observed individual fusion events using VAMP2-SEP or VAMP2-SNAP-CFl ( Supplementary Fig. 1), albeit at low frequency, perhaps due to poor incorporation of this construct in PC12 cells.
We then tested these sensors in living neurons, focusing on VAMP2 based constructs, which have been used extensively to follow SV exocytosis in neurons 1 .
We co-transfected hippocampal neurons with VAMP2-SEP and either VAMP2-pHuji or VAMP2-SNAP incubated with CFl ligand 4 or Virginia Orange ligand 5. For all the sensors, we observed a robust increase in fluorescence following electrical stimulation in fields covered with transfected axons, signaling SV exocytosis. The relative increase in fluorescence upon SV exocytosis was slightly higher for the SEP channel relative to the red-shifted fluorescent indicators, VAMP2-SNAP-VO (Fig. 3a, b), VAMP2-SNAP-CFl (Supplementary Fig. 2a,b), and VAMP2-pHuji ( Supplementary Fig. 2c,d), which behaved similarly. The kinetics of decay, which tracks endocytosis and re-acidification of the vesicle, were similar for all four labels ( Fig. 3b and Supplementary Fig. 2b,d). We also tested if the semisynthetic pH sensor system could be used in multicolor imaging experiments with GFP-based indicators. We co-transfected neurons with the GCaMP6f 16 and VAMP2-SNAP, which we labeled with CFl ligand 4. This allowed simultaneous imaging of both calcium ion transients and vesicle fusion in the same cell (Supplementary Fig. 2e-h).
Finally, we asked whether SV exocytosis could be detected without overexpression of a reporter protein. To enable imaging of endogenous vesicular proteins, we labeled a monoclonal antibody that recognizes a luminal epitope of synaptotagmin 1 (Syt1), a SV protein, with VOAc2-NHS ester (6) followed by mild deprotection of the acetate esters using hydroxylamine (Fig. 3c, Methods). This antibody has previously been used to detect endogenous Syt1 present on the plasma membrane after exocytosis in active synapses 17 . To mark vesicular Syt1, we incubated neurons with this antibody-VO conjugate (10 nM) for 3 hours in stimulation buffer, followed by extensive washing to remove the extracellularly bound protein conjugate (Fig. 3d).
The antibody labelling was done in neurons transfected with Syt1-SEP or VAMP2-SEP to compare the performance of this labeling technique with the overexpressed, genetically encoded GFP-based pH sensors. Electrical stimulation evoked a robust increase in fluorescence in axons transfected with Syt1-SEP (Fig. 3e,f) or VAMP2-SEP (Supplementary Fig. 2i), and in axons of untransfected neurons without detectable SEP. We found that the decay kinetics after stimulation was faster for the VO-antibody conjugate (8.8 ± 0.5 s) than the Syt1-SEP (12.2 ± 0.9 s, n = 41; paired ttest p < 0.0001, Fig. 3f) and VAMP2-SEP (Supplementary Fig. 2j), suggesting a difference in overexpressed vs endogenous protein behavior after SV exocytosis 18 .
Remarkably, the fluorescence transients were substantially higher in untransfected than in transfected neurons (Fig. 3f, Supplementary Fig. 2i), perhaps stemming from steric hindrance of the overexpressed SEP proteins or through quenching of the two fluorophores.
In summary, we have developed new 'semisynthetic' pH-sensitive proteins that allow for the imaging of synaptic vesicle fusion events in living cells. This sensor system combines the highly tunable properties of small molecule fluorophores with the specificity of self-labeling tags or antibodies. The SNAP-tag-based system constitutes the first genetically encoded long-wavelength pH sensor with similar or better performance than SEP in different cell types. The antibody-based pH sensor allows for imaging of vesicle fusion events without the need for overexpression of sensor proteins. Addition of other self-labeling or epitope tags by genome editing methods could allow cell-and protein-specific labeling without the need for overexpression 19 .
Future improvements of both the protein and the dye within this semi-synthetic scaffold should further enable imaging of this key biological process in increasingly complex systems.

METHODS
Methods, including statements of data availability and any associated accession codes and references, are available in the online version of the paper.

ACKNOWLEDGMENTS
We thank the cell culture core facility of IINS for preparing neuronal cultures, and

COMPETING FINANCIAL INTERESTS
The authors declare competing financial interests. L.D.L. and J.B.G. have filed patent applications whose value might be affected by this publication.   Protein chemistry. SNAP-tag protein (NEB) was labeled with excess CFl SNAP-tag ligand (4) in PBS containing 1 mM DTT. The protein concentrate was purified and concentrated using a Ziba spin desalting column (7K MWCO; ThermoFisher). The protein sample was analyzed by gel electrophoresis using a NuPAGE 4-12% Bis-Tris polyacrylamide gel (ThermoFisher) and imaged using a Typhoon Trio+ scanner (GE Healthcare); the gel was also stained with Coomassie Plus (ThermoFisher) and compared to a Mark 12 protein standard ladder (ThermoFisher). Absorbance measurements were taken in citrate (pH 5.6) or phosphate (pH 7.4) buffers described above.

Plasmid constructs
VAMP2-SEP and synaptotagmin1-SEP were kind gifts from Jürgen Klingauf. were carried out at 25 °C. TIRF microscopy was done as previously described 14,22 .
Cells were imaged on an inverted fluorescent microscope (IX-81, Olympus), equipped with a X100, 1.45 NA objective (Olympus). 488 nm and 561 nm lasers (Melles Griot) were combined and passed through a LF405/488/561/635 dichroic mirror. The laser was controlled with an acousto-optic tunable filter (Andor). Emitted light was separated using a 565 DCXR dichroic mirror on the image splitter (Photometrics), and projected through 525Q/50 and 605Q/55 filters onto the chip of a EM-CCD camera. Image acquisition was done using the Andor IQ2 software. Images were acquired sequentially with alternate 488nm and 561nm excitation at 100 ms exposure. The red and green images were aligned post-acquisition using projective image transformation as described before 14,22 . Before experiments, 100 nm yellowgreen fluorescent beads (Invitrogen) were imaged in the green and red channels, and superimposed by mapping bead positions.
Image analysis was performed using Metamorph (Molecular Devices) and custom scripts on MATLAB (Mathworks). The co-ordinates of the brightest pixel in the first frame of each fusion event in the green channel was identified by eye, and time was normalized to 0s. A circular ROI of 6 pixels (~ 990 nm) diameter and a square of 21 pixels (~ 3.5 µm) were drawn around the fusion co-ordinates. The average minimum pixel intensity in the surrounding square from 5 frames before fusion was subtracted from the intensity in the circular ROI, and the values were normalized to the frame before fusion in the green and red channels. were taken before each experiment and used to align the two channels 5 . Time lapse images were acquired at 1 or 2 Hz with integration times from 50 to 150 ms.
Image analysis was performed with custom macros in Igor Pro (Wavemetrics) using an automated detection algorithm as described previously 21 . The image from the time series showing maximum response during stimulation was subjected to an "à trous" wavelet transformation. All identified masks and calculated time courses were visually inspected for correspondence to individual functional boutons. The intensity values were normalized to the ten frames before stimulation in the green and red channels. Photobleaching in the red channels was corrected using an exponential decay fit applied on the non-responsive boutons. All data are represented as mean ± s.e.m. of n experiments.