Next-generation multicolor indicators for in vivo imaging of norepinephrine

Abstract

Norepinephrine (NE) is a critical neuromodulator that dynamically shapes brain states and behavior. Genetically encoded fluorescent NE indicators have enabled in vivo visualization of NE release, yet their limited sensitivity and spectral flexibility constrain widespread use in complex experimental paradigms. Here we developed next-generation green and red fluorescent NE indicators, named nLightG2 and nLightR2, respectively. We systematically compared them to other state-of-the-art indicators and found that both indicators improve the detection of endogenous NE release across preparations. Using dual-color photometry, both nLightG2 and nLightR2 could reliably track physiologically relevant NE release along with neuronal activity in different brain areas. Using two-photon imaging, nLightR2 permitted simultaneous dual-color imaging of NE release and astrocytic activity in the hippocampus, while nLightG2 enabled the detection of spatiotemporally discrete NE release events in the visual cortex of awake mice. This improved toolkit will prove useful for dissecting the spatiotemporal complexity of norepinephrine signaling in the brain.

Main

Norepinephrine (NE) is a monoamine neuromodulator synthetized and released throughout the central nervous system mainly by noradrenergic neurons of the locus coeruleus (LC). NE signaling is critically involved in a number of brain functions, including the regulation of arousal¹, sensory processing², memory formation^3,4,5, aversive learning⁶ and stress responses^7,8, as well as other higher-order cognitive processes^9,10,11. Understanding the precise spatiotemporal dynamics of NE release, and how these dynamics relate to neural circuit activity across brain states and behavioral tasks, is an active area of research, yet remains technically challenging.

Traditional analytical methods such as fast-scan cyclic voltammetry¹², microdialysis¹³ and cell-based reporters (CNiFER cells¹⁴) have been used to monitor NE release in vivo. However, each technique presents important limitations: fast-scan cyclic voltammetry lacks catecholamine specificity, microdialysis offers poor temporal resolution, and cell-based reporters suffer from limited sensitivity. Moreover, all these methods lack the spatial precision necessary to dissect NE signaling in complex brain circuits.

The advent of genetically encoded fluorescent indicators (GEFIs) has greatly advanced the field, enabling direct visualization of NE release with high spatiotemporal resolution. Two main families of NE-specific GEFIs have been developed: GRAB_NE^15,16 and nLight¹⁷, based on the α2a or α1a adrenergic receptors (A2AAR and A1AAR), respectively. These indicators were engineered by fusing circularly permuted fluorescent proteins onto G-protein-coupled receptors (GPCRs) in a conformationally sensitive manner. GEFIs are now increasingly used to study NE dynamics in vivo during sleep¹⁸, learning^19,20 and cognition^21,22. Despite the transformative potential of NE-specific GEFIs, their sensitivity remains a key limitation, constraining the scope and reliability of in vivo applications. This is particularly evident for red-shifted indicators, such as the first-generation red NE indicator nLightR¹⁷, which, despite the theoretical advantages of red fluorescent proteins including reduced phototoxicity²³, deeper tissue penetration²⁴ and compatibility with multiplexed imaging alongside green indicators^{25,26,27,28,29} and optogenetic tools^30,31, has seen limited use due to suboptimal performance.

Here, we present and comprehensively benchmark the next generation of green and red NE indicators, nLightG2 and nLightR2, respectively. Both indicators display substantially enhanced brightness and dynamic range compared to previous indicators, making them robust and sensitive tools for tracking NE dynamics across diverse brain regions and behavioral contexts. Using dual-color fiber photometry, we demonstrated precise temporal coupling between LC activity and NE release during sleep, as well as a robust detection of fear learning-dependent modulation of NE release and local calcium activity in the basolateral amygdala (BLA). Furthermore, nLightR2 enabled dual-color imaging of NE release and astrocyte calcium dynamics in the dorsal hippocampus (dHPC) during virtual navigation, highlighting its utility for multiplexed studies of neuromodulatory functions. Last, in vivo two-photon imaging of nLightG2 revealed with very high sensitivity that NE signaling in the mouse visual cortex (VC) occurs within spatiotemporally discrete microdomains.

Results

Development of the next-generation nLight indicators

We and others have shown that the fluorescent reporting domain of GPCR-based GEFIs can, in some cases, be grafted onto target GPCRs of choice for modular development of new indicators^17,32. Following this approach, we previously developed nLightG and nLightR¹⁷ using reporting domains derived from the green and red fluorescent dopamine (DA) indicators dLight1.3b and RdLight1 (ref. ³³), respectively. Recent work³⁴ resulted in improved variants of dLight1.3b (termed gGRAB_DA3m) and RdLight1 (termed rGRAB_DA3m) with enhanced dynamic range and brightness. A sequence alignment between RdLight1 and rGRAB_DA3m revealed that the improvement was due to five point mutations (T237C, Q248K, H253M, L435S and A495H). Four of these mutations are located in the fluorescent reporting domain of the indicator and thus fall within the region grafted from RdLight1 onto the sperm whale A1AAR (swA1AAR) to generate nLightR (Extended Data Fig. 1a). Based on these observations, we hypothesized that some of these mutations might enhance the performance of nLightR. We tested all 31 possible combinations of the five point mutations on nLightR (M226C, Q237K, H242M, L424S and A484H) to investigate the effect of the single mutations as well as potential epistatic interactions. All 31 variants were well expressed in HEK293T cells and localized predominantly at the plasma membrane. Most (17 variants, 55%) responded with a higher dynamic range than nLightR (change in indicator fluorescence (ΔF/F₀) = 187% ± 5%, mean ± s.e.m.) to the addition of 10 μM NE (Extended Data Fig. 1b). The highest-performing variant was termed nLightR1.1 and had a dynamic range of 690% ± 10% (mean ± s.e.m.), which corresponds to a 3.7-fold improvement compared to nLightR. nLightR1.1 harbored only three (Q237K, H242M and A484H) rather than all five mutations present in rGRAB_DA3m. Despite this, it showed a higher dynamic range than the variant with all five point mutations (ΔF/F₀ = 415% ± 9%, mean ± s.e.m.). Of note, the improvement in the indicator’s dynamic range seems to be composed of a direct component mediated by individual point mutations and an epistatic component (reflected in the variants combining multiple mutations; Extended Data Fig. 1c). We then truncated the C terminus of nLightR1.1 by 55 amino acids at position Q595 (Q370 in the wild-type receptor), a site at which truncation with retention of conformational functionality was already achieved for a human receptor homolog³⁵. This truncated variant, which we termed nLightR2 (Fig. 1a), was well expressed at the plasma membrane (Fig. 1b) and had a significantly higher dynamic range than nLightR1.1 (ΔF/F₀ = 740% ± 7%, mean ± s.e.m., P = 7.87 × 10⁻⁵, two-tailed Student’s t-test with Welch’s correction; Extended Data Fig. 1d). A comparison of the basal brightness and surface expression levels (by N-terminal Flag staining) of nLightR and nLightR2 in HEK293T cells revealed that nLightR2 has a higher basal brightness and surface expression (Extended Data Fig. 1e). The apparent molecular brightness of the two indicator variants in the NE-unbound state (measured by normalizing the basal brightness by the expression level) did not significantly differ (P = 0.402, two-tailed Student’s t-test with Welch’s correction; Extended Data Fig. 1f). Consequently, the improved dynamic range of nLightR2 can be attributed to a higher brightness in the NE-bound state. We also engineered a ligand-insensitive indicator by introducing the D129A point mutation in nLightR2 (refs. ^36,37). The resulting indicator, which we named nLightR2-ctr, showed no response to NE (10 μM) but good cell surface expression (Extended Data Fig. 1g,h).

Fig. 1: In vitro properties of the next-generation nLight indicators.

a, Structural model of nLightR2 generated using AlphaFold3 (ref. ⁷²). b, Representative images of nLightR2 transiently expressed in HEK293T cells before (top) and after (middle) the addition of NE (10 μM) and the pixel-wise heat map of the dynamic range (bottom). The fluorescence intensity profile (expression) along the dashed white line is shown as a white inset in the topmost image; scale bar, 10 μm. The experiment was repeated four times with similar results. c, Same as in a but for nLightG2. d, Same as in b but for nLightG2. The experiment was repeated four times with similar results. e, Left, time trace of the fluorescence response (mean ± s.e.m) of nLightR or nLightR2 transiently expressed in HEK293T cells showing activation by NE (10 μM) and inactivation by Trz (10 μM). Right, quantification of average ΔF/F₀ following the addition of NE (10 μM) and Trz (10 μM). Data were analyzed by two-tailed Students t-test with Welch’s correction; n = 4 independent experiments with n = 28 cells (nLightR2: ****P = 1.166 × 10⁻³⁵; nLightR: ****P = 3.028 × 10⁻³⁸). f, Same as in e but for nLightG, nLightG2 and GRAB_NE2m. For the reversal of the GRAB_NE2m signal, the A2AAR-specific antagonist yohimbine (Yoh; 10 µM) was used. Data were analyzed by two-tailed Student’s t-test with Welch’s correction. Data are from four independent experiments with n = 28 cells (nLightG2: ****P = 5.635 × 10⁻²⁸; nLightG: ****P = 4.793 × 10⁻²⁶; GRAB_NE2m: ****P = 3.184 × 10⁻³²). g, Representative images of nLightR2 (left) and nLightG2 (right) expressed in primary cortical rat neurons before and after the addition of NE (10 μM) and the pixel-wise heat map of the dynamic range; scale bar, 40 μm. The experiment was repeated four times with similar results. h, Left, time trace of the fluorescence response (mean ± s.e.m) of nLightR or nLightR2 (top left) and nLightG or nLightG2 (bottom left) expressed in primary cortical rat neurons showing activation by NE (10 μM) and inactivation by Trz (10 μM). Right, quantification and statistical comparison of average ΔF/F₀ following the addition of NE (10 μM) and Trz (10 μM) for nLightR/nLightR2 (top right) and nLightR/nLightR2 (bottom right). Data were analyzed by Student’s t-test with Welch’s correction. Data are from four independent experiments with n = 4 ROIs (nLightR2: ****P = 1.355 × 10⁻⁵; nLightR: ****P = 3.199 × 10⁻⁶; nLightG2: ****P = 3.552 × 10⁻⁶; nLightG: ****P = 4.633 × 10⁻⁵). i, Fluorescence dose–response curves of nLightR2 (top) and nLightG2 (bottom) for NE and DA when expressed in primary cortical neurons. The data points were fitted with four-parameter dose–response curves to determine the EC₅₀ values. Data are from three independent experiments with n = 3 cells for nLightR2 with all concentrations of NE and DA, from three independent experiments with n = 5 cells (0.1 nM, 1 nM, 10 nM, 100 nM, 1 µM, 10 µM, 100 µM and 1 mM) or n = 3 cells (10 mM) for nLightG2 with NE or from three independent experiments with n = 3 cells (for 1 nM, 10 µM and 10 mM) or n = 4 cells (10 nM, 100 nM, 1 µM, 100 µM and 1 mM) for nLightG2 with DA. j, ∆F/F₀ signal of nLightR2 following the application of NE (5 µM) to an outside-out patch (HEK293T cell). NE was applied for 1 s by fast switching of a perfusion pipette (blue shadow on the left). k, ∆F/F₀ signal changes of single patches with nLightR2 and nLightG2 (average of four NE applications). ON and OFF kinetics were approximated by single exponential fits (black lines). Average τ values are shown. l, Quantified time constants (mean ± s.d.) from n = 3 independent experiments with n = 8 (nLightR2) or n = 11 (nLightG2) patches in total.

Source data

We then focused our attention on gGRAB_DA3m, which is built from dLight1.3b with seven point mutations and an exchange of the N-terminal secretory sequence³⁴. An amino acid sequence alignment revealed that six of these seven point mutations are located within the region grafted from dLight1.3b onto the swA1AAR when engineering nLightG (Extended Data Fig. 1i). We introduced these six point mutations (S239N, N243Y, K260G, S338L, D441N and N444F) into nLightG and tested the resulting indicator, termed nLightG1.1, in HEK293T cells following the addition of NE (10 μM). nLightG1.1 had a dynamic range of 2,139% ± 48%, corresponding to a 2.4-fold improvement compared to nLightG (ΔF/F₀ = 904% ± 17%, mean ± s.e.m.) and one of the highest dynamic ranges described for any indicator of this class so far (Extended Data Fig. 1j). Truncation of the 55 carboxyterminal amino acids at position Q594, analogously to nLightR2, resulted in a small but significant improvement of the dynamic range (ΔF/F₀ = 2,350% ± 46%, mean ± s.e.m., P = 2.89 × 10⁻³, two-tailed Student’s t-test with Welch’s correction; Extended Data Fig. 1j), and the truncated variant, termed nLightG2 (Fig. 1c), was well expressed at the plasma membrane (Fig. 1d). We then compared the expression levels and basal brightness of nLightG2 in HEK293T cells to that of its predecessor (nLightG) and another previously published¹⁶ NE indicator (GRAB_NE2m). Although the expression levels of all three indicators are comparable, nLightG and nLightG2 have a higher apparent basal brightness than GRAB_NE2m (Extended Data Fig. 1k). The apparent molecular basal brightness (apparent basal brightness normalized by expression level) of nLightG and nLightG2 does not differ significantly (P = 0.220, two-tailed Student’s t-test with Welch’s correction; Extended Data Fig. 1l). This indicates that the improved dynamic range of nLightG2 can be attributed to an increased maximal brightness. Of note, the apparent molecular basal brightness of nLightG2 is significantly higher than that of GRAB_NE2m (P = 5.918 × 10⁻⁴, two-tailed Student’s t-test with Welch’s correction; Extended Data Fig. 1l). Last, we engineered a control indicator (named nLightG2-ctr), which does not respond to NE (10 μM), by introducing the D129A mutation into nLightG2 (Extended Data Fig. 1m,n).

In vitro characterization of the indicators

Before testing nLightR2/nLightG2 in primary neurons, we performed an initial characterization in HEK293T cells focused on the dynamic range, reversibility, specificity, sensitivity and spectral properties of the indicators.

Full reversibility of indicator activation and a high molecular specificity for the ligand of interest are essential to accurately and specifically report fluctuations in ligand concentrations. We transiently expressed nLightR2 or nLightG2 in HEK293T cells and showed that indicator activation by NE (10 μM) is fully reversible using the small-molecule A1AAR antagonist trazodone (Trz; 10 μM; Fig. 1e,f). The same experiment was performed for the best previously published red and green fluorescent NE indicators (nLightR, nLightG and GRAB_NE2m), showcasing the superior dynamic range of the improved indicators. To reverse the signal of GRAB_NE2m, which is based on the A2AAR subtype, we used yohimbine (10 µM) instead of Trz. Next, we characterized the specificity profiles of nLightR2/nLightG2 by measuring the dynamic range following the application of a panel of nine small-molecule neurotransmitters (NE, DA, serotonin, γ-aminobutyric acid, histamine, glutamate, acetylcholine, adenosine and epinephrine) at a final concentration of 10 μM (Extended Data Fig. 2a,b). Except for DA (partial agonist) and epinephrine (not present in the brain), none of the tested molecules triggered a substantial fluorescent response. The selectivity for NE over DA was further determined by measuring the half-maximal effective concentration (EC₅₀) values of NE and DA for nLightR2 (EC_{50 (NE)} = 821 ± 48 nM; EC_{50 (DA)} = 24.4 ± 1.8 μM; mean ± s.e.m.; Extended Data Fig. 2c) and nLightG2 (EC_{50 (NE)} = 1,078 ± 45 nM; EC_{50 (DA)} = 46.9 ± 9.0 μM; mean ± s.e.m.; Extended Data Fig. 2d) in HEK293T cells stably expressing the indicators. For comparison, we measured the EC₅₀ values of NE and DA for GRAB_NE2m using the same experimental approach and found that GRAB_NE2m (EC_{50 (NE)} = 1,272 ± 73 nM; EC_{50 (DA)} = 20.9 ± 9.0 μM; mean ± s.e.m.; Extended Data Fig. 2e) has a lower selectivity for NE over DA than nLightG2. This is further pronounced by the fact that the relative ligand efficacy (E_max) for DA versus NE is lower for nLightG2 (E_{max (DA)}/E_{max (NE)}: 34.4% ± 1.6%, mean ± s.e.m.) than for GRAB_NE2m (E_{max (DA)}/E_{max (NE)}: 53.8% ± 0.6%, mean ± s.e.m.). To show that the next-generation indicators detect low concentrations of NE more robustly than previously available ones, we measured the fluorescent change of nLightG, nLightG2 and GRAB_NE2m or nLightR and nLightR2 following exposure to 100 nM NE (Supplementary Fig. 1a,b).

Chromophore protonation and deprotonation due to pH changes could theoretically modulate the fluorescence intensity of GEFIs and cause artifacts. To investigate this, we measured the fluorescent response of nLightG2 and nLightR2 to 10 μM NE at five different pH values (6, 6.5, 7, 7.5 and 8) and did not see a significant difference in the dynamic range (nLightG2: P = 0.318; nLightR2: P = 0.480; one-way analysis of variance (ANOVA); Supplementary Fig. 1c–h).

The spectral properties of an indicator are important for optimal excitation, signal detection and multiplexing. One-photon spectra of nLightR2 and nLightG2 expressed in HEK293T cells revealed excitation maxima at 562 nm/498 nm and emission maxima at 594 nm/514 nm in the saturated state (Extended Data Fig. 2f,g). The excitation wavelengths for which the fluorescence emission is not modulated by NE binding (that is, the isoemissive points) range from 395 nm to 415 nm for nLightR2 and 412 nm to 413 nm for nLightG2. Two-photon excitation spectra of the indicators revealed a maximal performance (given by the ratio of the two-photon brightness in the saturated versus NE-free (apo) state) at excitation wavelengths ranging from 1,020 nm to 1,120 nm for nLightR2 and from 910 nm to 980 nm for nLightG2 (Extended Data Fig. 2h,i).

Based on the promising results in HEK293T cells, we then tested our next-generation indicators in primary rat cortical neurons. nLightR2 and nLightG2 expressed well in neurons transduced with recombinant adeno-associated viruses (rAAVs) encoding the indicators under the control of the human synapsin 1 (hSyn) promoter (Fig. 1g). The activation of both indicators by NE (10 μM) can be fully reversed with Trz (10 μM) in neurons (Fig. 1h). A comparison of the dynamic range between nLightR and nLightR2 or nLightG and nLightG2 in neurons indicated a 2.8-fold improvement for nLightR2 and a 2-fold improvement for nLightG2 following the addition of NE (10 μM; Fig. 1h). As observed previously for other indicators, the dynamic range of nLightR2 (ΔF/F_{0 neurons} = 549% ± 14%, ΔF/F_{0 HEK293T} = 732% ± 8%, mean ± s.e.m.) and nLightG2 (ΔF/F_{0 neurons} = 1,655% ± 24%, ΔF/F_{0 HEK293T} = 2,319% ± 51%, mean ± s.e.m.) is lower in primary rat cortical neurons than in HEK293T cells. In primary neurons, the EC₅₀ of NE and DA for nLightR2 (EC_{50 (NE)} = 837 ± 138 nM; EC_{50 (DA)} = 26.8 ± 7.1 μM, mean ± s.e.m.) and nLightG2 (EC_{50 (NE)} = 766 ± 142 nM; EC_{50 (DA)} = 24.0 ± 5.3 μM, mean ± s.e.m.) indicated a 32-fold and 31-fold selectivity for NE, respectively (Fig. 1i). However, the apparent selectivity of both indicators is further increased by the lower efficacy (E_max) of DA than NE, which can be accounted for by calculating the ratio of E_max/EC₅₀, referred to as S₀ (refs.^38,39) or S-slope⁴⁰. The maximal efficacy of DA compared to NE in neurons is 21% for nLightR2 and 38% for nLightG2. A comparison of E_max/EC₅₀ for DA versus NE in neurons results in an apparent selectivity of 152-fold for nLightR2 and 82-fold for nLightG2.

Although a good sensitivity (EC₅₀) and signal-to-noise ratio are a prerequisite for robust sensing, the temporal resolution is important to accurately report changes in ligand concentration within the physiologically relevant timescales. To precisely investigate the activation/inactivation kinetics of nLightG2 and nLightR2, we used patch-clamp fluorometry on outside-out membrane patches of indicator-expressing HEK293T cells in combination with a double-barreled perfusion pipette and a piezoelectric actuator for rapid solution switching between 0 μM and 5 μM NE. In this setup, nLightG2 and nLightR2 exhibited subsecond activation and deactivation kinetics (nLightR2: τ_on = 52 ± 9 ms and τ_off = 646 ± 124 ms; nLightG2: τ_on = 66 ± 7 ms and τ_off = 431 ± 119 ms; mean ± s.d.; Fig. 1j–l).

GPCR-based GEFIs should not couple to downstream signaling pathways by interaction with G proteins or β-arrestins to avoid cross-talk with the physiological signaling processes of the host cells. To test if our indicators trigger an intracellular calcium increase by coupling to the G_q protein pathway (which naturally couples to A1AARs), we coexpressed nLightR2 or nLightG2 in HEK293T cells together with the calcium indicator GCaMP6s⁴¹ or jRGECO1a⁴², respectively. Activation of the GPCR-based indicators by NE (10 µM) did not trigger a measurable increase in the intracellular calcium concentration. By contrast, control experiments with the wild-type GPCR (swA1AAR) resulted in clear signals of the two calcium indicators (Extended Data Fig. 3a–d). This finding was confirmed and strengthened by a split nano luciferase (NanoLuc) complementation assay⁴³ investigating the recruitment of mini-G proteins (mini-G_s/mini-G_sq/mini-G_si^44,45) and β-arrestin2 (ref. ⁴⁶; Extended Data Fig. 3e). Stimulation by NE (10 µM) induced significantly stronger recruitment of all three mini-G-protein classes (mini-G_sq, mini-G_s, mini-G_si) to the wild-type swA1AAR than to nLightG2/nLightR2 (Students t-test with Welch’s correction: mini-G_sq and swA1AAR versus nLightG2 (****P = 4.602 × 10⁻⁷) or nLightR2 (****P = 4.322 × 10⁻⁷) mini-G_s and swA1AAR versus nLightG2 (****P = 1.568 × 10⁻⁵) or nLightR2 (****P = 1.802 × 10⁻⁵); mini-G_si and swA1AAR versus nLightG2 (****P = 2.307 × 10⁻⁵) or nLightR2 (***P = 6.508 × 10⁻⁴); Extended Data Fig. 3f,g). We used the same NanoLuc complementation assay to study the recruitment of β-arrestin2 following receptor/indicator activation by replacing the mini-G protein with β-arrestin2. Due to inconclusive information in the literature regarding β-arrestin2 recruitment to the A1AAR subtype, we used GLP1R, previously established in our hands, as a positive control instead^47,48,49,50. Stimulation of GLP1R with GLP-1 (100 nM) resulted in a significantly stronger luminescence increase than stimulation of nLightR2 and nLightG2 with NE (10 µM), indicating minimal recruitment of β-arrestin2 to the indicator following activation (β-arrestin2 and GLP1R versus nLightG2 (****P = 1.418 × 10⁻¹⁰) or nLightR2 (****P = 6.026 × 10⁻¹³); Extended Data Fig. 3f,g).

Overexpression of GPCR-based GEFIs in vivo could potentially lead to buffering of chemical neurotransmission, as ligand molecules are sequestered by indicators near the plasma membrane, thereby reducing the ligand concentrations perceived by endogenous receptors. To investigate this, we measured the NE-induced, concentration-dependent recruitment of mini-G_sq proteins to the wild-type A1AAR in the presence or absence of nLightG2 or nLightR2 using the split NanoLuc complementation assay described above. No significant difference in the dose-dependent mini-G_sq recruitment to the wild-type A1AAR could be observed in the presence of nLightG2, nLightR2 or OxLight1 (ref. ⁵¹), a GPCR-based GEFI that does not bind NE (one-way ANOVA; P = 0.341; Supplementary Fig. 1i).

Some red genetically encoded indicators (for example, the cpmApple-based jRGECO1a⁴² and RdLight⁵²) are known to exhibit photoswitching following exposure to blue light⁵³ (see explanation and experimental design in Extended Data Fig. 4a–c). We thus evaluated positive and negative photoswitching as well as photobleaching by imaging nLightR2-transfected HEK293T cells at a high frame rate (~75 Hz) using a confocal spinning-disk microscope. We normalized the positive photoswitching effect of 488-nm light to the ligand response (NE, 10 µM). In both the apo- and NE-bound states, 488-nm light (30 mW mm⁻²) increased nLightR2 fluorescence (Extended Data Fig. 4d,e). In both states, the positive photoswitching reached a plateau during 400 ms, whereas passive relaxation took about 1 s to return to baseline (Extended Data Fig. 4f), which is in line with previously reported in vivo observations using RdLight⁵². When comparing prolonged exposure to high-power (30 mW mm⁻²) 560-nm light, both jRGECO1a and nLightR2 showed a combination of fast (in the order of milliseconds) and slow decay (in the order of seconds; Extended Data Fig. 4e–h). Given that 560-nm light leads to both fluorescence excitation and negative photoswitching, we speculated that the fast component is not caused by irreversible photobleaching but rather by reversible negative photoswitching. To test that, we imaged nLightR2-expressing cells with the same high-power 560-nm light but alternating it with a 30-s-long period with no light. Even though the absolute fluorescence decreased over the course of the image blocks, every time a new block started, we observed the fast decay (Extended Data Fig. 4i). When correcting the traces for photobleaching using a second-degree exponential fit, the fast decay amplitude was similar in each block suggesting that indeed this decay is reversible negative photoswitching (Extended Data Fig. 4i,j).

Ex vivo benchmarking of the indicators

To obtain a direct side-by-side comparison of our indicators with the original parent indicators, we probed their performance in acute brain slices using two-photon microscopy (Fig. 2a). The indicators were virally expressed in the lateral hypothalamus (LH) by stereotaxic injection of hSyn promoter-driven AAV 3–4 weeks before slice preparation. First, we focused on the benchmarking of nLightR2 versus nLightR and measured ∆F/F₀ evoked by exogenous NE, applied to the slice by iontophoresis (Fig. 2a–d). Short pulses (5 ms) of NE, ejected from the iontophoretic pipette by small (20-nA) and large (200-nA) currents, evoked rapid increases in both nLightR and nLightR2 fluorescence (Fig. 2a). The peak ∆F/F₀ responses of nLightR2 fluorescence were significantly larger than those of nLightR for all ejection currents (P < 0.0001; two-way ANOVA with Sidak’s multiple comparisons test; Fig. 2b). The spatial properties of NE detection by nLightR2 were also noticeably improved compared to nLightR (Fig. 2c,d). Iontophoresis of NE evoked bright ∆F/F₀ signals (Fig. 2c), which were modeled using a Gaussian curve (Fig. 2d), demonstrating that even the smallest application of NE could be spatially quantified using the improved nLightR2 indicator. Next, we measured endogenous NE release in the LH evoked by electrical stimulation (Fig. 2e–g). Trains of electrical stimuli (5 × 40 µA) evoked slow increases of nLightR and nLightR2 fluorescence, which were significantly larger in amplitude when measured using the improved nLightR2 indicator (P = 0.0019, two-tailed Mann–Whitney U-test; Fig. 2e). In addition, the area of nLightR2 ∆F/F₀ exceeding a threshold, defined as 2.5 standard deviations (σ) of baseline signal, was significantly larger than that detected using nLightR, demonstrating its improved ability to spatially quantify endogenous NE release (P = 0.0003, U = 14, two-tailed Mann–Whitney U-test; Fig. 2f,g). Next, we compared the green NE indicators nLightG and nLightG2 (Fig. 2h). Application of exogenous NE by iontophoresis produced fluorescence responses that were larger in amplitude when measured using nLightG2 across a range of ejection currents (P < 0.0001; two-way ANOVA with Sidak’s multiple comparisons test; Fig. 2h,i). Moreover, the spatial properties of exogenous NE signals could be better detected using the improved nLightG2 indicator (Fig. 2j,k). Finally, we compared endogenous NE measurements using the green indicators (Fig. 2l). Changes in fluorescence responses to endogenous NE release elicited by trains of electrical stimuli (5 × 40 µA) were significantly larger in both peak amplitude (P = 0.0002, two-tailed unpaired t-test; Fig. 2l) and area (>2.5σ of baseline signal; P = 0.0003, two-tailed unpaired t-test; Fig. 2m,n) when measured using nLightG2 compared to nLightG. A side-by-side comparison of the indicators in response to the same puff (10 ms) of iontophoretically applied NE revealed that the two indicators have similar kinetics when expressed and imaged in brain tissue (Fig. 2o,p). Overall, these benchmarking experiments demonstrate that both nLightR2 and nLightG2 permit two-photon imaging of NE release with largely improved sensitivity over the parent indicators.

Fig. 2: Benchmarking NE indicators ex vivo with two-photon microscopy.

a, Experimental schematic showing viral expression of nLightR/nLightR2 in the LH and two-photon (2P) imaging of exogenous NE application by iontophoresis (Ionto) in ex vivo brain slices (top). Example two-photon spot photometry traces showing the time course of changes of nLightR/nLightR2 fluorescence (∆F/F₀) evoked by NE iontophoresis (ejection currents: 20 and 200 nA, duration: 100 ms; bottom). b, Summary of peak nLightR/nLightR2 ∆F/F₀ evoked by NE iontophoresis using different ejection currents (duration: 100 ms); nLightR: n = 9 slices from three mice; nLightR2: n = 11 slices from three mice; data are presented as mean ± s.e.m. (F_1,72 = 118.4, P = 7.37 × 10⁻¹⁷, two-way ANOVA: ****P = 9.80 × 10⁻⁶, t₇₂ = 5.12; ****P = 1.62 × 10⁻⁵, t₇₂ = 4.99; ****P = 3.02 × 10⁻⁶, t₇₂ = 5.42; ****P = 1.12 × 10⁻⁷, t₇₂ = 6.23; Sidak’s multiple comparisons test). c, Example 3D surface plots showing peak nLightR/nLightR2 ∆F/F₀ evoked by NE iontophoresis across 20- to 200-nA ejection currents (duration: 100 ms). d, Spatial profiles of averaged (±s.e.m. shaded) nLightR (top) and nLightR2 (bottom) ∆F/F₀ fitted using a Gaussian function evoked by NE iontophoresis; nLightR: n = 5 slices from two mice; nLightR2: n = 7 slices from two mice. e, Experimental schematic showing two-photon imaging of endogenous NE release evoked by electrical stimulation (eStim; 5 × 40 µA, 1 ms at 40 Hz) in LH brain slices (left), average (±s.e.m. shaded) nLightR/nLightR2 ∆F/F₀ measured across FOV (30 × 40 µm; middle) and summary of peak ∆F/F₀ (right); nLightR: n = 10 slices from two mice; nLightR2: n = 15 slices from three mice; data are presented as mean ± s.e.m. (**P = 0.0019, U = 21, two-tailed Mann–Whitney U-test). f, Example 3D surface plots showing peak nLightR (left) and nLightR2 (right) ∆F/F₀ evoked by electrical stimulation. g, Summary of area of nLightR/nLightR2 ∆F/F₀ evoked by electrical stimulation, exceeding 2.5σ of baseline signal; nLightR: n = 10 slices from two mice; nLightR2: n = 15 slices from three mice; data are presented as mean ± s.e.m. (***P = 0.0003, U = 14, two-tailed Mann–Whitney U-test). h, Same as in a for nLightG/nLightG2. i, Summary of peak nLightG/nLightG2 ∆F/F₀ evoked by NE iontophoresis (duration: 5 ms); nLightG: n = 14 slices from three mice; nLightG2: n = 10 slices from three mice; data are presented as mean ± s.e.m. (F_1,88 = 98.14, P = 5.59 × 10⁻¹⁶, two-way ANOVA: **P = 0.0025, t₈₈ = 3.55; ****P = 1.31 × 10⁻⁸, t₈₈ = 6.58; ****P = 1.32 × 10⁻¹², t₈₈ = 8.56; Sidak’s multiple comparisons test). j, Same as in c for nLightG/nLightG2. k,l, Same as in d and e for nLightG/nLightG2; nLightG: n = 7 slices from two mice; nLightG2: n = 7 slices from two mice; data are presented as mean ± s.e.m. (**P = 0.0012, U = 1, two-tailed Mann–Whitney U-test). m, Example 3D surface plots showing peak nLightG (left) and nLightG2 (right) ∆F/F₀ evoked by electrical stimulation. n, Same as in g for nLightG/nLightG2 (**P = 0.0012, U = 1, two-tailed Mann–Whitney U-test). o, Example nLightG2 and nLightR2 ∆F/F₀ traces in response to NE iontophoresis (ejection current: 100 nA, duration: 10 ms; left) and averaged traces (±s.e.m. shaded) normalized to peak ∆F/F₀ (right); nLightG2: n = 7 slices from two mice; nLightR2: n = 15 slices from three mice. p, Comparison of temporal kinetics of nLightG2 and nLightR2 in response to the same iontophoretic application of NE as in o (τ_ON: P = 0.49, U = 42; half-width: P = 0.78, U = 48; τ_OFF: P = 0.30, U = 37). Data are presented as mean ± s.e.m. and were analyzed by two-tailed Mann–Whitney U-test; NS, not significant.

Source data

In vivo performance of the next-generation nLight indicators

To establish the improved performance of the indicators in vivo, we started by benchmarking nLightR2 to nLightR with tail lift (TL) experiments, a behavioral paradigm that leads to strong and rapid release of NE (Extended Data Fig. 5a,b). As expected, nLightR2 outperformed nLightR, revealed by a higher peak ∆F/F₀ (P = 0.008, two-way-ANOVA: variant × event interaction; Extended Data Fig. 5c) as well as area under the curve during the TLs (P = 5.047 × 10⁻⁷, two-tailed t-test; Extended Data Fig. 5d,e). We then used optogenetics to trigger NE release and compare the sensitivity of nLightR2 to both its predecessor and the NE-insensitive nLightR2-ctr in vivo. We injected rAAV2/rAAV9 encoding for either one of the indicators into the left dHPC of Dbh-cre mice (Fig. 3a). In addition, we injected rAAV2/rAAV9 encoding Cre recombinase-dependent channelrhodopsin-2 (DIO-ChR2-eYFP) into the LC of the same hemisphere (Fig. 3b,c), allowing for optogenetic stimulation of LC-NE neurons. A few weeks after surgery, we performed fiber photometry recordings from the dHPC of mice head-fixed on a linear treadmill, while simultaneously monitoring the pupil diameter of the animal, which is tightly coupled to the evoked release of NE⁵⁴. Following optogenetic stimulation of the LC, we observed a robust increase of fluorescence in animals expressing nLightR2 when exciting the indicator at 555/30 nm (Fig. 3d). This signal was accompanied by pupil dilation and was absent in the isoemissive control channel (Fig. 3e). Averaged, z-scored responses differed across conditions (Fig. 3f; F = 6.49, P = 0.016, one-way ANOVA), with nLightR2 responses significantly exceeding responses observed with the control indicator nLightR2-ctr (3.99 ± 2.07 s.d. versus 0.56 ± 0.26 s.d., n = 6/3 mice; P = 0.024, Tukey’s test), which was not the case when using nLightR (1.31 ± 0.72 s.d.; P = 0.80, Tukey’s test against nLightR2-ctr). Importantly, no differences across indicators were detected when exciting them at 405 nm (F = 2.35, P = 0.15, one-way ANOVA), while the absence of differences in pupil dilation (F = 0.32, P = 0.73, one-way ANOVA) indicated comparable levels of optogenetic LC activation across groups. Hence, we demonstrated monitoring of optogenetically evoked NE release in the dHPC using nLightR2, which was not possible using its predecessor nLightR.

Fig. 3: Benchmarking NE indicators in vivo with fiber photometry.

a–c, Injection scheme (a) and histological verification of nLightR2 expression in the dHPC (b) and ChR2 expression in the LC (c). Data for the brain schematics obtained from ref. ⁷³ under a Creative Commons license CC BY 4.0. Data in b and c were repeated independently five times with similar results. d, Color-coded heat maps of z-scored ΔF/F₀ of nLightR2 in the dHPC following optogenetic activation of the LC. The blue bar indicates the duration of optogenetic stimulation. e, Mean ± s.e.m. of z-scored indicator fluorescence in the dHPC (top) and relative pupil diameter (bottom) in exemplary mice expressing nLightR2 (left), nLightR (center) and nLightR2-ctr (right). Blue bars indicate optogenetic LC stimulation (n = 15 trials for nLightR2 and nLightR, n = 50 trials for nLigthR2-ctr). Iso., isoemissive. f, Mean ± s.d. of optogenetically evoked indicator fluorescence (left) and pupil dilation (right) of animals expressing nLightR2 (n = 6 animals), nLightR (n = 4 animals) and nLightR2-ctr (n = 3 animals). Data were analyzed by one-way ANOVA with a Tukey’s test: nLightR2 versus nLightR (P = 0.05), nLightR2 versus nLightR2-ctr (*P = 0.024) and nLightR versus nLightR2-ctr (P = 0.80). g,h, Experimental schematics for nLightG2, nLightG and GRAB_NE2m viral injections in the pBLA. i, Behavioral task design for the cued fear conditioning paradigm, consisting of a baseline (BL) session with 20-s tone cues, an association (AS) session with the same tone cues co-terminating with 1-s footshocks and a re-exposure (RE) session with tone cues only in the absence of footshocks (n of trials in each session is 10). j,k, Baseline session. j, Heat maps showing trial-averaged z-score fluorescence responses across the ten tone trials for nLightG2 (top), nLightG (middle) and GRAB_NE2m (bottom) in the pBLA. k, Corresponding mean ± s.e.m. z-score fluorescence traces for nLightG2 (green), nLightG (gray) and GRAB_NE2m (blue). l,m, Association session. l, Heat maps of trial-averaged z-score fluorescence responses for each indicator type during ten tone–footshock pairings. m, Corresponding mean ± s.e.m. traces showing tone-evoked activity (0- to 20-s tone) co-terminating with a 1-s footshock. n,o, Re-exposure session. n, Heat maps of trial-averaged z-score fluorescence responses for each indicator type during ten tone presentations. o, Corresponding mean ± s.e.m. traces for tone-evoked responses (0- to 20-s tone) in the absence of footshock. Dashed vertical lines mark cue onset and footshock timing. p, Quantification of CS-evoked peak z-score responses in nLightG2, nLightG and GRAB_NE2m mice across all conditions (baseline, association and re-exposure). Data were analyzed by two-way ANOVA (P_session = 0.0037; P_{indicator type} = 0.0121), post hoc Tukey’s comparisons (baseline: P_{nLightG2 versus nLightG} = 0.0460, P_{nLightG2 versus GRABNE2m} = 0.1163, P_{nLightG versus GRABNE2m} = 0.9002; association: P_{nLightG2 versus nLightG} = 0.0129, P_{nLightG2 versus GRABNE2m} = 0.0061, P_{nLightG versus GRABNE2m} = 0.9553; re-exposure: P_{nLightG2 versus nLightG} = 0.0318, P_{nLightG2 versus GRABNE2m} = 0.0161, P_{nLightG versus GRABNE2m} = 0.9574). q, Quantification of footshock-evoked peak z-score responses in nLightG2, nLightG and GRAB_NE2m mice (Welch’s one-way ANOVA: P = 0.0002; post hoc Dunnett’s T3 comparisons (two-sided): P_{nLightG2 versus nLightG} = 0.0002, P_{nLightG2 versus GRABNE2m} = 0.0270, P_{nLightG versus GRABNE2m} = 0.1096; nLightG2, n = 5 mice; nLightG, n = 5 mice; GRAB_NE2m, n = 5 mice); WT, wild-type.

Source data

In our previous work, we demonstrated the robust detection of optogenetically evoked NE release in vivo using the green indicator nLightG¹⁷. To evaluate the in vivo performance of nLightG2 and compare it directly to that of the previously available indicators nLightG and GRAB_NE2m, we decided to adopt a behavioral paradigm designed to capture more subtle NE dynamics in freely moving animals. We used a cued fear conditioning task while recording NE release using fiber photometry in the posterior BLA (pBLA; Fig. 3g–i). In this paradigm, an initially neutral conditioned stimulus (CS; tone cue) is paired with an aversive unconditioned stimulus (US; mild footshock) that naturally elicits fear expression, which in rodents is typically shown as freezing behavior (that is, complete absence of motion except breathing) and is expected to be accompanied by NE release in the amygdala^55,56. After conditioning, presentation of the CS alone sufficiently evokes fear expression⁵⁷. Four to six weeks following rAAV-mediated expression of the indicators in the pBLA, we used fiber photometry to record indicator fluorescence during a baseline session (with CS presentations only), association session (with CS and US pairing) and re-exposure session (with CS presentations only; Fig. 3i). Throughout the different phases of the task, all three indicators were able to detect NE release in response to the CS and US. However, nLightG2 exhibited higher overall signal amplitudes across all sessions (Fig. 3j–q). A two-way ANOVA was conducted on the peak z-score responses time-locked to the CS (CS-triggered peaks) to evaluate how indicator type and task session influenced the magnitude of NE release during learning. This analysis revealed significant main effects of the task session (P = 0.0037), suggesting that the stage of learning influenced NE release as anticipated in an aversive learning paradigm and of the indicator type (P = 0.0121), indicating overall differences in the sensitivity among the indicators. Post hoc Tukey’s comparisons showed that nLightG2 exhibited stronger peak responses to CS presentation than both nLightG and GRAB_NE2m across the sessions (baseline: P_{nLightG2 versus}< em>P_nLightG = 0.0460, P_{nLightG2 versus GRABNE2m} = 0.1163, P_{nLightG versus GRABNE2m} = 0.9002; association: P_{nLightG2 versus nLightG} = 0.0129, P_{nLightG2 versus GRABNE2m} = 0.0061, P_{nLightG versus GRABNE2m} = 0.9553; re-exposure: P_{nLightG2 versus nLightG} = 0.0318, P_{nLightG2 versus GRABNE2m} = 0.0161, P_{nLightG versus GRABNE2m} = 0.9574; Fig. 3p). These results demonstrate the superior sensitivity of nLightG2 in detecting cue-evoked NE release before, during and after associative learning. Additionally, the footshock-evoked nLightG2 peak signals were significantly larger than those of nLightG (Dunnett’s T3 comparisons: P = 0.0002) and GRAB_NE2m (P = 0.0270), indicating that nLightG2 detects NE release following aversive stimuli with improved sensitivity (Welch’s ANOVA, P = 0.0002; Fig. 3q; n = 5 mice; nLightG, n = 5 mice; GRAB_NE2m, n = 5 mice).

To test whether the expression of the indicator could affect fear learning by buffering NE and interfering with its endogenous signaling, we used DeepLabCut (DLC) pose estimation software⁵⁸ to perform supervised video-based analysis of freezing behavior in animals expressing nLightG2 versus the NE binding-incompetent version of the indicator (nLightG2-ctr) in the pBLA (Supplementary Fig. 2a–c). A mixed-design ANOVA revealed a significant main effect of session (for full postsound epoch: P = 1.90 × 10⁻⁵; for CS trials: P = 1.73 × 10⁻⁸) but no significant effect of indicator type (P = 0.688) or condition × indicator interaction (P = 0.810). Post hoc Wilcoxon analysis showed that freezing behavior increased significantly across sessions (re-exposure versus baseline: P = 0.0117; re-exposure versus association: P = 0.0117 for both full postsound epoch and CS trials only), consistent with fear learning (Supplementary Fig. 2d,e). These results suggest that the observed behavioral changes were driven by conditioning and were not influenced by indicator expression.

In vivo multiplexed detection of NE and neural activity

NE is released during both wakefulness and sleep⁵⁹. To see whether nLightR2 could faithfully report on the variable dynamics of NE release across states of sleep and wakefulness, we focused on non-rapid eye movement (NREM) sleep. In this state, LC neurons generate vigorous activity surges roughly every ~50 s that are relevant for sleep architecture and arousability^18,60. These activity surges vary in amplitude and duration, with stronger ones often causing brief awakenings⁶⁰. To test whether nLightR2 could report NE release caused by these surges, we co-transduced LC-NE neurons of DA-β-hydroxylase-iCre (Dbh-iCre) animals with the Cre-dependent calcium indicator jGCaMP8f and hSyn-driven nLightR2 (or the mutated version nLightR2-ctr; Fig. 4a) and implanted animals with electroencephalogram/electromyography (EEG/EMG) electrodes to perform sleep–wake recordings combined with dual-color fiber photometry (Fig. 4b). nLightR2 and nLightR2-ctr expressed well with jGCaMP8f at the site of the LC, allowing for simultaneous fluorescence measurements, which we verified to be wavelength specific (Supplementary Fig. 3), through the optic fiber positioned on top of the LC (Fig. 4c). The fluorescence of nLightR2 followed LC activity on a surge-by-surge basis (Fig. 4d), whereas the mutated indicator did not generate consistent signals (Fig. 4e). When we averaged LC activity surges that occurred during continuous NREM sleep and calculated the corresponding mean nLightR2 fluorescent waveforms, it became clear that these faithfully followed LC activity variations and generated mean z-scored fluorescent increases larger than the ones of the mutated version (Fig. 4f–h; 0.26 ± 0.49 for nLightR2, 1,264 peaks from n = 7 mice; –0.06 ± 0.56 for nLightR2-ctr, 1,285 peaks from n = 4 mice; Mann–Whitney U-test, P = 4.0 × 10⁻⁵¹). Furthermore, the average cross-correlations (±10-s lag around 0) between jGCaMP8f and nLightR2 were greater than for nLightR2-ctr (Fig. 4i; P = 3.6 × 10⁻²⁷³).

Fig. 4: nLightG2 and nLightR2 are suitable to study NE release and neuronal activity via dual-color fiber photometry in vivo.

a, Surgery schematics for sleep experiments. Dbh-iCre mice were injected in the LC with a virus mix to coexpress jGCamp8f and nLightR2 or the mutated (NE-insensitive) nLightR2-ctr. b, Injected mice were implanted with two EEG electrodes (frontal and parietal + a ground over the cerebellum) and two EMG electrodes (for polysomnographic monitoring of sleep–wake behavior) and with an optic fiber over the LC (for dual-color fiber photometry). c, Representative anatomical verification of fiber location (dotted lines, left) and viral expression (middle and right). Middle, micrographs from a mouse co-injected with jGCaMP8f and nLightR2. Right, same for a mouse co-injected with jGCaMP8f and nLightR2-ctr. Zoom-ins show individual colors and overlay. These anatomical verifications were performed for three of seven nLightR2-injected mice and for two of four nLightR2-ctr-injected mice from which the data in d–i were obtained. d, Representative traces of LC dual-color fiber photometry recordings showing jGCaMP8f and nLightR2 signals (z scored) during undisturbed NREM sleep. From the top, vertical bars highlighting scored vigilance states (wake, NREM sleep), time–frequency spectrogram of EEG activity and EMG, nLightR2 (z scored) and jGCaMP8f (z scored) signals. e, As in d for nLightR2-ctr. f, Mean ± s.e.m. for nLightR2 and jGCamp8f z-scored signals around peaks of LC activity during consolidated NREM sleep (1,264 peaks, n = 7 mice). g, As in f for nLightR2-ctr (1,285 peaks, n = 4 mice). h, Swarm plots of peak values (means taken for ±10 s around the LC peak) for nLightR2 (n = 1,264 total detected peaks) and nLightR2-ctr (n = 1,285 total detected peaks) fluorescence; P = 4.0 × 10⁻⁵¹, two-sided Mann–Whitney U-test. i, Swarm plots of cross-correlations between the z-scored nLightR2 or nLightR2-ctr signal with the jGCaMP8f signal around peaks of LC activity (means taken for ±10-s lags around 0-s lag) during consolidated NREM sleep; P = 3.6 × 10⁻²⁷³, two-sided Mann–Whitney U-test with n = 1,264 for nLightR2 and n = 1,285 for nLightR2-ctr. j, Surgery schematics for cued fear conditioning experiments. Wild-type mice were injected in the aBLA with an rAAV mix of PinkyCaMP and nLightG2 or the NE-insensitive nLightG2-ctr. A fiber was implanted to record those indicators simultaneously. k, Same as in c but for PinkyCaMP + nLightG2 or nLightG2-ctr; scale bar, 1,000 µm (overview image) and 20 µm (magnification images). Images were repeated independently ten times with similar results. l, Cued fear conditioning behavioral setting to create an association between an aversive US (1-s-long mild footshock) and a neutral CS (20-s-long tone). m, Heat maps showing the mouse average per trial (rows) of a 65-s interval during CS presentations (ten trials) during baseline and re-exposure sessions for mice coexpressing either PinkyCaMP and nLighG2 (n = 8 mice) or PinkyCaMP and nLighG2-ctr (n = 4 mice). For simplicity, only nLightG2-ctr heat maps are shown. n, Mean PinkyCaMP z-score traces ±s.e.m. (n = 8 mice) of an 85-s interval during CS presentation in baseline and re-exposure sessions. o, Comparison of mean ± s.e.m. z-score nLightG2 (n = 8 mice) and nLightG2-ctr (n = 4 mice) traces in the same time interval and sessions as in n. p, z-Score peaks during CS presentation (average of ten trials per mouse) in both baseline and re-exposure sessions. The aBLA neuronal activity (PinkyCaMP) triggered by tone (CS) increased after memory association (P = 0.0421, two-tailed paired t-test). NE release (nLightG2) triggered by the CS also increased after memory association (P = 6.791 × 10⁻⁵, two-tailed paired t-test), whereas no change was seen in the nLightG2-ctr z score across sessions (P = 0.5832, two-tailed paired t-test). q, Same as in m but for the association session. The yellow square represents the US presentation. r, Same as in n but now comparing PinkyCaMP and nLightG2 signals during the association session. s,t, Same as in q and r, respectively, but now comparing PinkyCaMP to nLightG2-ctr signals.

Source data

We then proceeded to test nLightG2 in dual-color photometry experiments combining it with the bright and photostable red fluorescent calcium indicator PinkyCaMP⁶¹. To monitor learning-dependent activity changes during cued fear conditioning (Fig. 4j–l), we co-transduced excitatory/projection neurons with CaMKiia-PinkyCaMP and all neurons with hSyn-nLightG2 or hSyn-nLightG2-ctr and implanted an optic fiber for dual-color fiber photometry in the anterior BLA (aBLA; Fig. 4j). Coexpression of the two indicators was well tolerated (Fig. 4k). As expected, CS-evoked calcium peaks detected by PinkyCaMP were larger during the re-exposure session than during the baseline session. This was also the case for NE release; however, NE levels stayed elevated throughout the duration of the CS. This was not the case for the NE-insensitive mutant nLightG2-ctr, indicating that the nLightG2 signals truly reflect NE release in the aBLA and are not affected by other potential sources of artifacts (for example, hemodynamics; Fig. 4m–o). Quantification of the peaks triggered by the CS showed that both PinkyCaMP and nLightG2 significantly increased across sessions, and no difference was seen in nLightG2-ctr (*P = 0.0421, ****P = 6.791 × 10⁻⁵ and P = 0.58; two-sided, paired t-test; Fig. 4p). To characterize further how these two signals correlate and to rule out potential cross-contamination, we looked at the association session where mice were presented ten times with a footshock (US) that co-terminated with the tone. US presentation led to both a large and sharp calcium response detected by PinkyCaMP as well as a large and sustained (more than 40 s) NE release detected by nLightG2. Conversely, the US triggered a small dip in nLightG2-ctr signal, indicating the presence of a small non-NE-related component to the photometry signal (Fig. 4q–t). We performed a correlation analysis between PinkyCaMP and nLightG2 or nLightG2-ctr at the single-trial level and found that calcium and NE weakly correlate across the whole association session but strongly correlate during US presentation. Conversely, we observed no correlation between PinkyCaMP and nLightG2-ctr during most of the association session and a strong negative correlation during US presentation, in line with the non-NE-dependent nature of the control signal (Extended Data Fig. 6a–c). Together, these results showcase the multiplexing capability of the next-generation indicators using fiber photometry in vivo.

Multiplexed two-photon imaging of NE and intracellular calcium during spatial navigation

Previous work using two-photon microscopy in combination with green-emitting NE indicators¹⁷ showed that NE signaling in pairs of localized hippocampal regions is differentially correlated across behavioral conditions, suggesting that behaviorally dependent, spatially restricted subdomains of NE dynamics exist in the CA1 pyramidal layer. The development of nLighR2 provides the unprecedented opportunity to perform spectrally multiplexed experiments, in which intracellular signals measured with a green indicator are related to the extracellular concentration of NE. However, a first fundamental step toward this goal is to demonstrate that two-photon microscopy of nLightR2 signals provides sufficient sensitivity to identify the previously described behaviorally dependent spatial dynamics of NE¹⁷. Toward this goal, we expressed nLightR2 in the CA1 pyramidal layer of the HPC and imaged nLightR2 fluorescence using two-photon microscopy. We used λ = 1,040 nm to excite nLightR2, as this wavelength was observed to be effective in generating nLightR2 fluorescence emission (Extended Data Fig. 7). Awake head-fixed mice navigated in a virtual reality corridor for five sessions over consecutive days (days 1–5, one session per day, n = 4 mice; Extended Data Fig. 8a,c). Results obtained with nLightR2 largely recapitulated what we previously observed under similar experimental conditions with nLightG¹⁷. nLightR2 signal amplitude displayed a tendency to increase with the running speed of the animal, which reached significance on days 3–4 (two-sided rank-sum test; H₀, slope of the linear model equals to 0; P_run = 2.09 × 10⁻² for days 3–4, n = 4 mice; Extended Data Fig. 8h) and a tendency to increase with lick rate (Extended Data Fig. 8i), which reached significance on recording days 1–3 (two-sided rank-sum test; H₀, slope of the linear model equals to 0; P_reward = 2.09 × 10⁻² for days 1–3, n = 4 mice). Experiments with the nLightR2 control indicator (nLightR2-ctr) did not show linear dependence of NE signals over running speed and lick rate (Extended Data Fig. 9a–i).

We then leveraged the absorption/emission properties of nLightR2 to simultaneously measure NE signaling with nLightR2 in the red (595/50-nm) fluorescence channel and cellular calcium activity with GCaMP6f in the green (525/70-nm) fluorescence channel. As an important control, we first imaged nLightR2-expressing animals at 920 nm, the wavelength used to excite fluorescence in the green channel, and observed some green fluorescence emission (Extended Data Fig. 7a,b). The green fluorescence was visible following illumination at 920 nm, but much less at 1,040 nm (Extended Data Fig. 7a,b). Moreover, following 920-nm excitation, the green fluorescence was not homogeneously distributed across the cells but was concentrated in what appeared to be somatic puncta (Supplementary Fig. 4a,b). This observation could be compatible with the possibility that the nLightR2 indicator may partition in two species as previously shown for other red-shifted indicators based on the mApple fluorophore⁴². Alternatively, the green fluorescence observed following 920-nm excitation could be due to endogenous autofluorescence. In line with this hypothesis, the puncta were also observed in nontransduced cultured neurons using two-photon excitation. Additionally, those puncta had broad excitation/emission spectra and showed high green–red colocalization. This was not the case for neurons expressing nLightR2 where the brightest red pixels had low green fluorescence intensity. More importantly, these puncta did not respond to application of NE in cultured neurons (10 µM; Supplementary Fig. 4c–g). To reduce the potential impact of this green emission in our two-photon functional recordings, we spatially segregated the NE and calcium indicators by expressing nLighR2 in CA1 neurons and GCaMP6f in CA1 astrocytes using cell-type-specific promoters. Head-fixed mice coexpressing the two indicators were immersed in the same virtual reality corridor used in days 1–2 in the experiments displayed in Extended Data Fig. 8, while simultaneously recording NE and astrocytic calcium signals in the CA1 pyramidal layer (Fig. 5a–d; n = 7 sessions from four animals). Segmentation of two-photon images (Fig. 5c) was performed using cell-sized regions of interest (ROIs)³⁷ for the NE signal and a deep learning software designed to perform anatomically bound segmentation of astrocytes⁶² for GCaMP6f signal. Using event-triggered averages, we observed a sustained increase of NE when the mouse started to run (Fig. 5d,e and Supplementary Fig. 5) and a transient increase in NE signaling after crossing the reward position (Fig. 5d–g and Supplementary Fig. 6), similar to what we observed in mice expressing only nLightR2 (Extended Data Fig. 8). Astrocytic ROIs also showed behaviorally dependent GCaMP6f responses (Fig. 5i–l). Astrocytic ROIs showed prolonged positive ΔF/F₀ responses when the mouse started to run (Fig. 5i) and a more heterogeneous response after crossing the reward position (Fig. 5k). Considering pairs of ROIs, we confirmed high Pearson’s correlation values for the NE signal in pairs of ROIs when the mouse started to run (Fig. 5f and Extended Data Fig. 10a–d) and when the mouse crossed the reward position (Fig. 5h and Extended Data Fig. 10a–d). Interestingly, astrocytic GCaMP6f signals also displayed a behavioral state-dependent correlation across pairs of ROIs (Fig. 5j–l and Extended Data Fig. 10e). We next investigated the dependence of correlation among pairs of NE or pairs of astrocytic ROIs as a function of the distance between the two ROIs of the pair (pairwise distance; Extended Data Fig. 10f,g). We found that Pearson’s correlation values for NE pairs decreased as a function of pairwise distance, suggesting heterogeneous NE signaling in the spatial scale investigated here (few hundreds of microns; Extended Data Fig. 10f) in both behavioral states, that is, start of running and crossing the reward position. We also observed decreasing GCaMP6f signal correlation as a function of pairwise distance across pairs of astrocytic ROIs in the same spatial scale (Extended Data Fig. 10g). This latter result suggests that in the HPC, neighboring astrocytes share a certain level of correlation in their calcium dynamics, which, however, decreases with distance, as previously observed⁶³. Finally, we asked how astrocytic GCaMP6f activity related to NE signals. After crossing the reward position, we found that the GCaMP6f signals extracted from one astrocytic ROI correlated with NE signals extracted from the NE ROI positioned closest to the astrocytic ROI (Fig. 5m,n), whereas this was not the case when the mouse started to run (Fig. 5m,n). Moreover, we found that the correlation value between the NE signal and astrocytic GCaMP6f activity strongly depended on the behavior of the animal, with crossing the reward position inducing higher correlation values between the two signals than running (Fig. 5o). The behaviorally dependent correlation between astrocytes and NE was not a function of the distance between the astrocytic ROI and the NE ROI (Fig. 5o). These latter results demonstrate that GCaMP6f activity in astrocytes shows larger correlation values with NE signals during reward epochs than after starting a running epoch. In addition, in reward epochs, the amplitude of GCaMP6f responses in astrocytes increases with the amplitude of the NE signals in a close-by spatial location.

Fig. 5: Simultaneous two-photon imaging of local NE release and astrocytic calcium dynamics in awake mice during virtual navigation.

a, Simultaneous two-photon imaging was performed in head-fixed mice navigating a virtual corridor to collect water rewards delivered at 85 cm from the start of the corridor. b, Schematic of the HPC highlighting the imaged region (pink and green), indicating the expression of nLightR2 (neurons) and GCaMP6f (astrocytes) in the pyramidal layer. c, Representative images showing nLightR2-labeled neurons (left, magenta) and GCaMP6f-labeled astrocytes (middle, green) in the CA1 pyramidal layer. Images are normalized fluorescence intensity (F_norm) projection of recorded t-series, and contrast has been adjusted to aid visualization. nLightR2 and GCaMP6f signals were simultaneously imaged in the red and green channels, respectively, using the excitation wavelengths (λ_ex) indicated in the figure. White lines on the images indicate ROIs (for nLightR2, we show only 16 square ROIs in close proximity to one astrocyte, square ROI size ~17 μm × 17 μm (ref. ³⁷); for GCaMP6f, anatomically defined ROIs are as in ref. ⁶³). Right, the two images on the left are shown overlaid; scale bars, 50 µm; AU, arbitrary units. d, Representative functional traces of nLightR2 and GCaMP6f fluorescence extracted from the ROIs shown in c. Markers indicate either run (gold) or reward (teal) behavioral epochs. Traces in d have been smoothed with a 500-ms rolling mean filter. e, Event-triggered averages showing mean ΔF/F₀ of nLightR2 signal when the mouse started to run (Run, start of running bout at t = 0 s) for all the ROIs identified in c. f, Bottom left triangle, correlation (corr.) matrix for all traces extracted from the ROIs displayed in e. Top right triangle, corresponding hierarchical clustering. g,h, Same as in e and f, but for event-triggered averages following animals crossing the reward position (Reward, crossing time at t = 0 s). i–l, The same metrics as in e–h but referring to ΔF/F₀ computed for astrocytic ROIs and astrocytic GCaMP6f signals. m, Maximum ΔF/F₀ of GCaMP6f event-triggered average traces for astrocytic ROIs expressed as a function of the maximum ΔF/F₀ of nLightR2 event-triggered average recorded in the nLightR2 ROI nearest to the considered astrocytic ROI during either the run (gold) or reward (teal) behavioral epoch (P = 0.030 and P = 0.326 for reward and run epochs, respectively; two-tailed permutation test with 10⁴ repetitions, n = 114 pairs recorded in seven sessions from four mice). n, Mean ΔF/F₀ of GCaMP6f event-triggered average traces for astrocytic ROIs expressed as function of the mean ΔF/F₀ of nLightR2 event-triggered averages recorded in the nLightR2 ROI nearest to the considered astrocytic ROI during either the run (gold) or reward (teal) behavioral epoch (P = 0.025 and P = 0.144 for reward and run epochs, respectively; two-tailed permutation test with 10⁴ repetitions, n = 114 pairs recorded in seven sessions from four mice). o, Pearson’s correlation between nLightR2 and GCaMP6f signals computed either during the run (gold) or reward (teal) behavioral epoch as a function of pairwise distance (P = 0.001 for epoch effect (run versus reward), distance–epoch interaction not significant for all distances (Supplementary Table 1), linear mixed-effects model, n = 7 sessions from four mice). Lines and shaded areas in m–o indicate mean ± s.d. and mean ± s.e.m., respectively.

Source data

Imaging localized cortical NE dynamics in awake mice using nLightG2

To monitor NE dynamics in the cortex of awake animals and perform a head-to-head comparison of nLightG2 with the previously existing green NE indicator GRAB_NE2m and nLightG2-ctr, we expressed the indicators in layer 2/3 of the VC and performed two-photon imaging in head-fixed mice positioned on a running wheel (Fig. 6a). In nLightG2 animals, we observed robust and spatially localized NE transients, occurring both in response to looming stimuli (Fig. 6b,c) and during gray screen intervals, often linked to locomotion but at times also present in its absence (Fig. 6c and Supplementary Video 1). These transients were absent in nLightG2-ctr mice, demonstrating that they faithfully reported local NE release events (Fig. 6b,c and Supplementary Video 2). nLightG2 transients appeared as brief increases in ΔF/F₀, frequently confined to localized subregions of the field of view (FOV), forming a mosaic of transiently active microdomains (Fig. 6d–g).

Fig. 6: nLightG2 reveals spatially structured NE dynamics in the VC during sensory stimulation and behavioral state transitions in mice.

a, Schematic of viral injection in the VC and two-photon imaging setup with nLightG2 expression in layer 2/3 of the primary VC (V1) in head-fixed mice on a running wheel. b, Individual ΔF/F₀ traces from two example FOVs illustrating loom-evoked activity in nLightG2 (average across ROIs with ΔF/F₀ > 3σ; green) and nLightG2-ctr (gray) mice, highlighting a single ROI with a strong loom-evoked response in the nLightG2 animal to a single loom, whereas only a small, albeit detectable, change was observed in the control, potentially reflecting hemodynamic signals. c, Example ΔF/F₀ traces from representative nLightG2 (green), GRAB_NE2m (blue) and nLightG2-ctr (gray) animals over the course of individual recordings. Orange triangles in the traces highlight spontaneous NE events in the absence of locomotion. d, An example FOV from an nLightG2-expressing mouse. e, Same as in d for a GRAB_NE2m-expressing animal. Peak responses are indicated in red. f, Example spatial distribution of peak response latency after loom onset across tile ROIs with a response ΔF/F₀ > 3σ within a single FOV. g, Time-to-peak analysis for data plotted in f. h, Scatter plots of loom-evoked changes in ΔF/F₀ (%) per ROI, including all ROIs across all mice within each group (nLightG2 ROIs (1,075 ROIs, n = 7 mice), GRAB_NE2m (751 ROIs, n = 3 mice) and nLightG2-ctr (375 ROIs, n = 3 mice)). i, Cumulative distributions of ΔF/F₀ values in the 20-s postloom window compared to preloom baseline across all tile ROIs. nLightG2 showed a significant shift for all tile ROIs pooled across mice (two-sided Kolmogorov–Smirnov test, P = 3.8 × 10⁻⁵), whereas GRAB_NE2m (P = 0.093) and nLightG2-ctr (D = 0.093, p = 0.076) did not. j, Scatter plots of mean ΔF/F₀ during stationary versus forced running epochs from the same ROIs and animals shown in f. k, Cumulative distributions of running-evoked ΔF/F₀ changes showed a rightward shift in nLightG2 compared to GRAB_NE2m (two-sided Kolmogorov–Smirnov test, P = 4 × 10⁻⁴) and nLightG2-ctr (P = 3.2 × 10⁻³) during forced locomotion. l, GLM quantifying the variance in NE dynamics explained by looming stimuli (loom), running speed (run) and their interaction (inter) in nLightG2-expressing animals (n = 7 mice). Left, the full model explained 4.4 ± 2.3% (mean ± s.e.m., n = 7 mice) of the total fluorescence variance, whereas excluding the running term reduced explained variance to 0.37 ± 0.09%, excluding the looming term to 4.2 ± 2.2% and excluding the interaction term to 4.3 ± 2.3% (mean ± s.e.m., n = 7 mice). Right, example tiled map showing the spatial distribution of the total explained variance (ΔR²) for an nLightG2-expressing mouse; Expl., explained. m, Leave-one-out GLM analysis validating the contribution of running, looming and their interaction to explained variance in NE dynamics (mean ± s.e.m.; relative ΔR²; running: 68.0 ± 5.7%, looming: 20.7 ± 3.5%, interaction: 10.5 ± 2.8%; Friedman P = 9.1 × 10⁻⁴; Wilcoxon P = 0.016, n = 7 mice).

Source data

Looming visual stimuli occasionally elicited strong ΔF/F₀ responses in nLightG2 mice (Fig. 6c–g); although responses were inconsistent across trials and spatially restricted across ROIs, loom-evoked changes in fluorescence were observed in all nLightG2 mice. Across all tile ROIs (n = 2,200, 5 × 5 square tiles per FOV per animal), only the nLightG2 indicator showed a statistically significant shift in postloom versus preloom ΔF/F₀ distributions (Kolmogorov–Smirnov test, P = 3.1 × 10⁻⁵; Fig. 6h,i), whereas no significant changes were observed in the control variant nLightG2-ctr (P = 0.076) or GRAB_NE2m (P = 0.093; Fig. 6h,i and Supplementary Video 3). Mouse-level paired comparisons confirmed a significant loom-evoked increase for nLightG2 (paired t-test: P = 0.029; Wilcoxon: P = 0.031; n = 7 mice), with average responses across all ROIs increasing by 1.39 ± 0.49% ΔF/F₀ (mean ± s.e.m.) and spanning a broad range. By contrast, GRAB_NE2m and nLightG2-ctr exhibited only minimal changes (Fig. 6h,i). To further assess nLightG2 sensitivity, we quantified the per-ROI probability of strong loom responses, defined as the fraction of looms in which a given ROI exhibited a peak z-scored ΔF/F₀ response exceeding 5 within the 20-s postloom window. This fraction was larger in nLightG2 mice (20.1 ± 3.7%, mean ± s.e.m., n = 7 mice) than in nLightG2-ctr mice (4.0 ± 1.2%, mean ± s.e.m., n = 3; Wilcoxon: P = 0.012). In addition, the distribution of peak z scores was significantly broader in nLightG2 than in both nLightG2-ctr (Kolmogorov–Smirnov test, P = 0.00017) and GRAB_NE2m (P = 0.00081), highlighting nLightG2’s enhanced ability to resolve discrete signal fluctuations. This was consistent with greater across-ROI variability, with mean standard deviation values (calculated per mouse) being higher in nLightG2 (5.33 ± 3.01%, n = 7 mice) than in GRAB_NE2m (2.20 ± 1.07%, n = 3 mice) and nLightG2-ctr animals (0.75 ± 0.05%, n = 3 mice).

To assess behavioral state dependence, we compared ΔF/F₀ signals during stationary epochs and forced locomotion (Fig. 6j,k). In nLightG2 mice, NE signals increased significantly when animals transitioned from rest to forced locomotion (3.12 ± 0.42% ΔF/F₀ at rest versus 5.57 ± 1.54% during locomotion; Wilcoxon signed-rank test, P < 0.001; n = 7 mice). Cumulative distribution analysis of mean ΔF/F₀ across ROIs revealed a pronounced rightward shift in nLightG2 (D = 0.43, P < 1 × 10⁻⁵; n = 7 mice; Fig. 6k). GRAB_NE2m exhibited more muted modulation (1.20 ± 0.19%; n = 3 mice). Scatter plots of mean ΔF/F₀ between stationary and locomotor epochs revealed significant correlations for both nLightG2 (P = 0.0071) and GRAB_NE2m (P = 0.00075; Fig. 6j). nLightG2-ctr mice showed a moderate correlation (P = 0.041), suggesting that some state-dependent modulation is preserved even under control conditions, likely reflecting hemodynamic or nonspecific arousal-linked signals⁶⁴. Furthermore, area under the receiver operating characteristic analysis revealed that nLightG2 discriminated spontaneous locomotion from stationary epochs more effectively (0.76 ± 0.04, n = 7) than GRAB_NE2m (0.62 ± 0.01, n = 3; Mann–Whitney U-test, P = 0.017). We used a generalized linear model (GLM) to dissociate the sensory and behavioral contributions to NE dynamics in nLightG2 mice (Fig. 6l,m). Because looming stimuli can evoke running behavior, and spontaneous locomotion is itself accompanied by increases in LC activity, the GLM allowed us to separate visually driven components from those related to the running state and to quantify their respective contributions to variance in NE activity. The model included terms for looming stimuli, running speed and their interaction, enabling the variance in NE dynamics to be partitioned into components explained by each factor while accounting for shared covariance. This approach revealed how behavioral state and sensory drive jointly shape NE fluctuations in the VC. The GLM explained 4.4 ± 2.3% (mean ± s.e.m., n = 7 mice) of the total fluorescence variance (Fig. 6l). To estimate the unique contribution of each factor, we used a leave-one-out GLM approach in which each predictor was iteratively removed from the model. The resulting change in explained variance (ΔR²) reflects the variance in NE activity uniquely attributable to each term. Running accounted for the largest proportion of explained variance (68.0 ± 5.7% of ΔR²), followed by looming (20.7 ± 3.5%) and their interaction (10.5 ± 2.8%; Friedman P = 9.1 × 10⁻⁴; Wilcoxon P = 0.016), indicating that NE dynamics reflect both ongoing behavioral state and transient sensory events.

Together, these findings demonstrate that nLightG2 exhibits substantially greater sensitivity and dynamic range in vivo than GRAB_NE2m, supporting its superior performance for resolving endogenous noradrenergic signaling in vivo.

Discussion

The versatility of our next-generation NE indicators was demonstrated in vivo by their faithful reporting of NE dynamics along with neural activity during NREM sleep and cued fear conditioning. The arousal levels of wakefulness and sleep are fundamentally different; therefore, it is reasonable to expect that the mechanisms and impact of NE signaling also differ between these behavioral states. The development of nLightR2 now opens exciting opportunities for investigating how brain-wide NE dynamics control arousal fluctuations on top of globally distinct behavioral states. Moreover, not only NE but also other monoamines fluctuate during sleep states⁶⁵, which renders spectral multiplexing of fiber photometry, combining multicolor measurements with specific indicators, indispensable for progress. Furthermore, NE fluctuations can be perturbed by stress^60,66 or in neurodevelopmental disease⁶⁷. A decline of LC activity is increasingly associated with the progression of neurodegenerative diseases⁵⁹. Therefore, the possibility to monitor NE output on an event-by-event basis in NREM sleep, and to pick up minute variations of these in perturbed conditions, holds potential for developing early diagnostic measures based on the associated sleep signals. In the amygdala, nLightG2 reported a fear learning-dependent sustained increase in NE levels during the CS, which was absent from the control indicator nLightG2-ctr and was paralleled by an increased PinkyCaMP activity peak at CS onset. Considering the multifaceted roles that NE release plays, from enhancing synaptic plasticity and memory consolidation^68,69 to regulating arousal⁷⁰, attention⁹ and anxiety-like behaviors⁷¹, it can be foreseen that these tools will also prove useful to robustly dissect the interplay between neuromodulator signaling and local circuit mechanisms in adaptive behaviors. Of note, the use of NE-insensitive control indicators in this study revealed a frequent presence of NE-independent signal contaminants across preparations. We therefore recommend that end users routinely include these controls to verify the NE-dependent nature of the observed signals.

Although both nLightG2 and nLightR2 are suitable for use in most spectral multiplexing experiments, due to the intrinsic photoswitching properties of its cpmApple-based chromophore, it should be noted that nLightR2 is not suitable for a specific subset of experiments in which NE monitoring needs to be combined with intermittent pulses of strong (for example, >300 μW)⁵² blue light illumination applied on the same area, such as in experiments combining blue light optogenetic stimulation and nLightR2 photometry through the same optical fiber. Future engineering efforts should aim to develop improved nLightR2 variants that strongly reduce or eliminate photoswitching, for instance by incorporating an alternative bright red fluorescent protein such as mScarlet, as used in PinkyCaMP⁶¹.

The indicators presented in this study also provide fundamental advantages in two-photon fluorescence microscopy experiments. nLightR2 reached a level of performance similar to that of nLightG¹⁷. Additionally, two-photon imaging of nLightR2 (λ_exc = 1,040 nm) could be coupled to two-photon imaging of GCaMP6 (λ_exc = 920 nm) for high-spatial-resolution multiplexed imaging of NE dynamics and intracellular calcium signals. In such experiments, spurious fluorescence may contaminate the green channel. This artifact can be minimized considering the spatial segregation of the indicators, the low level of spurious green emission and the static versus dynamic nature of the signals. Using this configuration, we found behavioral state dependence between astrocytic calcium activity and NE signaling, with dependence being present during reward delivery but not during running behavior.

nLightG2 enabled high-resolution two-photon imaging of NE dynamics in the VC of awake mice, revealing spatially structured, transient NE signals that are selectively engaged during sensory stimulation and behavioral state changes. Unlike previous-generation indicators, nLightG2 detected sparse, microdomain-specific transients with such high sensitivity that they were clearly discernible in the raw videos without additional processing. It captured both spontaneous and stimulus-evoked NE activity, uncovering a mosaic of localized responses and highlighting the integrative nature of neuromodulatory signaling. nLightG2 outperforms GRAB_NE2m in detecting subtle neuromodulatory shifts during looming stimuli and locomotion, establishing it as a powerful tool to resolve NE dynamics across internal and behavioral states in vivo.

Methods

Molecular cloning

Mammalian expression plasmids encoding nLightG and nLightR (Addgene, 217656-217657) were used as templates to generate all variants. Site-directed mutagenesis was performed by PCR (PfuUltra II Hotstart), followed by DNA ligation or Gibson assembly⁷⁴. The GRAB_NE2m¹⁶ gene fragment (Twist Bioscience) was cloned into a pCMV backbone (Addgene, 217656) with or without a Flag tag via Gibson assembly. nLightG2, nLightR2, GLP1R and msA1AAR-SmBit constructs were generated by PCR amplification of receptor/indicator sequences and replacement of the β₂-adrenergic receptor in a B2AR-SmBit plasmid¹⁷. Viral plasmids (Viral Vector Facility of the University and ETH Zürich) encoding nLightR2/nLightG2 and nLightR2/nLightG2-ctr under an hSyn promoter were cloned by BamHI/HindIII restriction cloning. All constructs were verified by Sanger or full plasmid sequencing.

Cell culture, confocal imaging and quantification

HEK293T cells (ATCC, CRL-3216) were cultured per standard protocols⁵¹ and were authenticated by the supplier. At 40–50% confluency, cells were transfected in glass-bottom dishes or six-well plates using PolyFect (Qiagen) and used 24–48 h after transfection. Primary cortical neurons were isolated from E18 rat embryos. Specifically, cortices were dissected; rinsed in PBS + 10 mM glucose, 1 mg ml⁻¹ bovine serum albumin and 1:100 antibiotic–antimycotic; fragmented; digested in papain (37 °C, 15 min); washed in DMEM + 10% fetal calf serum + 1:100 penicillin/streptomycin and dissociated by trituration. The suspension was filtered (40 μm) and plated at 40,000–50,000 cells per well on poly-L-lysine-coated dishes (50 μg ml⁻¹) in NU medium (MEM + 15% NU serum, 2% B27, 15 mM HEPES, 0.45% glucose, 1 mM pyruvate and 2 mM GlutaMAX). After 4–6 days, neurons were transduced with AAVs (1 × 10¹⁰ genome copies per ml) and cultured 14 days in vitro. Imaging was performed at room temperature in HBSS + CaCl₂/MgCl₂ + 30 mM HEPES. For live labeling, cells were incubated for 10 min with AlexaFlour 647 M1 anti-Flag (1:1,000), washed twice and imaged on a Zeiss LSM 800 with 488/561-nm lasers using ×40 or ×63 oil objectives; ligands were manually applied. ∆F/F₀ was quantified in Fiji (ImageJ) by ROI selection on isolated membranes: ∆F/F₀ = (F_t − F₀) / F₀.

Generation of HEK293T cell lines stably expressing indicators

Isogenic stable lines expressing nLightR2, nLightG2 or GRABNE2m were generated in HEK293T LLP Int Blast⁷⁵ (LLP-HEK) cells. Cells were seeded in 35-mm dishes and transfected with promotorless attB_nLightR2_PuroR, attB_nLightG2_PuroR or attB_GRABNE2m_PuroR using PolyFect. Expression was induced 24 h later with 2 μg ml⁻¹ doxycycline, and recombined cells were selected 24 h after induction with 5 μg ml⁻¹ puromycin. Cells were passaged three or more times in medium containing doxycycline and puromycin before experiments.

Dose–response curves in HEK293T cells

Dose–response curves for NE and DA were measured on a Tecan M200 Pro. HEK293T cells stably expressing the indicator were grown to 80% confluency, detached with Versene, centrifuged (room temperature, 3 min, 150g) and resuspended at 3.33 × 10⁶ cells per ml in HBSS + 30 mM HEPES. Cells (150 μl per well) were plated in black 96-well plates. Ligands were added as a 2× dilution series (150 μl per well) and incubated for 15 min at room temperature. Fluorescence was measured at 480/520 nm (green) or 560/600 nm (red; bandwidths 9/20 nm). Biological replicates were normalized and fitted with a four-parameter dose–response curve using least squares regression to determine the EC₅₀.

pH sensitivity testing

LLP-HEK cells stably expressing nLightG2 or nLightR2 were seeded in glass-bottom dishes in medium with 2 μg ml⁻¹ doxycycline and 5 μg ml⁻¹ puromycin. Before imaging, cells were incubated for 5 min in HBSS + 30 mM HEPES at the desired pH. Imaging and NE application (10 μM) were performed as described above. Analysis in Fiji averaged pixel intensity over time; ROIs of the top 10% brightest pixels (plasma membrane) were used to calculate brightness and ∆F/F₀.

One-photon excitation/emission spectra

One-photon excitation/emission spectra were measured on a Tecan M200 Pro. HEK293T cells were transfected 24 h after seeding using PolyFect, detached 48 h later with Versene (Thermo Fisher), centrifuged (room temperature, 3 min, 150g) and resuspended at 3.33 × 10⁶ cells per ml in HBSS + 30 mM HEPES. Cells (150 μl per well) were plated in black 96-well plates and mixed with HBSS or 2× NE to reach a final concentration of 10 μM NE. Plates were incubated for 15 min at room temperature, and fluorescence was measured at constant excitation (480/550 nm, 10-nm bandwidth) or emission (520/620 nm, 20-nm bandwidth) while scanning in 2-nm increments. Mock-transfected cell autofluorescence was subtracted. Biological replicates were normalized to maximal fluorescence and averaged.

For neurons, transduced (nLightR2 or nLightR2 + jGCaMP6f) or untransduced cells on glass-bottom dishes were imaged by one- or two-photon microscopy, excited sequentially at 980–1,040 nm (20-nm increments) with a ×16/0.8-NA water objective (N16XLWD-PF), 1,024 × 1,024 resolution, 0.96-Hz frame rate and five-plane z stacks (10-µm steps). Photomultiplier (PMT) tube detectors with filters for green (535/50) or red (620/60) light were used. Laser power was wavelength dependent but consistent across experiments (1,040 nm: 73 mW; 1,020 nm: 90 mW; 1,000 nm: 110 mW; 980 nm: 100 mW; 960 nm: 88 mW; 940 nm: 90 mW; 920 nm: 92 mW). Images were analyzed as averaged stack projections.

Confocal spinning-disk imaging and analysis

An Olympus IXplore SpinSR10 (Yokogawa CSU-W1, pinholes 2 × 50 µm) with a UPLSAPO ×60 oil objective was used to characterize nLightR2. The FOV was cropped to 108.3 × 54.17 µm, achieving a 13.2-ms frame rate. Lights (488 nm and 560 nm) were matched to ~30 mW mm⁻². For positive photoswitching, 488 nm was on 950.4 ms/off 1,412.4 ms, with 560 nm always on. For photobleaching, 560 nm only was applied for 35 s. Negative photoswitching used ten 1,887.6-ms 560-nm blocks, interleaved with 30-s no-light epochs. Image processing and analysis were performed in MATLAB/ImageJ: (1) file organization and hyperstack creation, (2) background subtraction (100 for Hamamatsu ORCA-Fusion sCMOS), (3) average z projection, (4) thresholding, (5) ROI selection/intensity measurement, (6) photobleaching correction (second-degree exponential) and (7) trace/heat map plotting.

In vitro two-photon brightness

The two-photon brightness of nLightR2/nLightG2 was recorded using a previously reported two-photon microscope⁷⁶ equipped with a wavelength-tunable fs-laser (Chameleon Discovery, Coherent) and a ×20 Zeiss W Plan objective. HEK293T cells were transfected using PolyFect and imaged 24 h after transfection. Before imaging, the medium was switched to HBSS with CaCl₂ and MgCl₂ (Thermo Fisher) supplemented with 30 mM HEPES (Thermo Fisher) to avoid DMEM autofluorescence. The spectra were collected before and after the addition of NE (10 µM). The laser power after the objective was calibrated at every wavelength using a beam splitter, a reference photodiode (Thorlabs, PDA50B2) and a power meter under the objective (Thorlabs, S175C). During the experiment, the power at the objective was kept below 30 mW to minimize photobleaching and saturation effects⁷⁷. To account for the unknown wavelength dependence of the pulse width as well as potential imprecisions in the power calibrations, we measured a correction curve using the reported⁷⁸ spectral shapes of Rhodamine6G (1 × 10⁻⁵ M in ethanol) and LDS798 (5 × 10⁻⁵ M in CDCl₃). The emission spectra of the apo and sat forms of the indicator are very similar; thus, a calibration for the relative detection efficiency was not necessary. Because the two-photon absorption cross-section (σ²), the quantum yield (Φ) and the number of chromophores in the ground state can change following ligand binding⁷⁹, we report the relative two-photon excitation efficiency, which is related to the product of these quantities and can be calculated from our data using previously reported equations.

Luminescence complementation assay

G-protein or β-arrestin2 recruitment by the indicator was measured using a luciferase complementation assay. SwA1AAR, mouse A1AAR, human GLP1R, nLightG2 or nLightR2 was C-terminally fused to SmBiT and co-transfected (1:1 DNA ratio) with mini-G proteins⁴⁴ (mini-G_s, mini-G_i, mini-G_q) or β-arrestin2 (ref. ⁴⁶) fused to LgBiT into HEK293T cells. Cells (250,000 per well in 35-mm dishes) were transfected 24 h after seeding, collected 24 h later and resuspended at 0.5 × 10⁶ cells per ml in Fluorobrite-DMEM + 30 mM HEPES. Cells (200 μl) were mixed with 50 μl of 20-fold diluted NanoGlo substrate in live cell substrate buffer (Promega); 100 μl per well was plated in white 96-well plates and incubated for 45 min at 37 °C at 5% CO₂. Luminescence was measured on a Tecan Spark before and after the addition of 25 μl of Fluorobrite-DMEM with (well 1) or without (well 2) 10 μM ligand. For analysis, the ratio of the luminescence intensity from well 2 and well 1 was calculated over time.

Kinetic measurements using patch-clamp fluorometry

Response kinetics of nLightR2 and nLightG2 were measured by patch-clamp fluorometry^17,51,80 with fast, piezo-driven ligand application. HEK293T cells were transfected with 0.3–0.45 μg ml⁻¹ plasmid using PEI 25,000 and incubated for 48–72 h. Patch pipettes (6–10 MΩ) contained 135 mM K-gluconate, 10 mM NaCl, 2 mM MgCl₂, 1 mM EGTA and 10 mM HEPES (pH 7.4); cells were bathed in 138 mM NaCl, 1.5 mM KCl, 1.2 mM MgCl₂, 2.5 mM CaCl₂ and 10 mM HEPES (pH 7.3). Outside-out patches were excised and positioned before a Θ glass pipette (~150-μm tip) on a piezo actuator for submillisecond ligand application⁸¹ (5 μM NE, 0.35 ml per min per channel; bath flow ~4 ml min⁻¹). Epifluorescence imaging used a Leica DMi8 with a ×20/0.40-NA objective (red: excitation 562/40 nm, emission 624/40 nm (~2.6 mW mm⁻²); green: excitation 470/40 nm, emission 525/50 nm (~0.5 mW mm⁻²)). EMCCD acquisition (Evolve 512delta, MicroManager 2.0) at 100 fps (2 × 2 binning, 9-ms exposure, EM (electron-multiplying) gain 200) was TTL (transistor-transistor logic) triggered via pClamp 10.7. ΔF/F₀ maps were calculated from mean projections (50–100 frames before, 40–60 frames during NE), background subtracted and normalized. Only high signal-to-noise recordings (nLightR2 > 10, n = 8; nLightG2 > 20, n = 11) were included; ON/OFF kinetics were fitted to single exponentials after averaging four sweeps.

Animals

Animal procedures were performed in accordance with the European Community Council Directive, the UK Animals (Scientific Procedures) Act 1986 and the Animal Welfare Ordinance (TSchV 455.1) of the Swiss Federal Food Safety and Veterinary Office and were approved by the Zürich Cantonal Veterinary Office, the Institutional Animal Care and Use Committee at the University of Colorado School of Medicine, the Regierungspräsidium Karlsruhe, the University College London institutional ethics committee or the National Council on Animal Care of the Italian Ministry of Health. Primary cortical neuronal cultures were prepared from embryonic day 17 Rattus norvegicus (Wistar) embryos dissected from timed-pregnant females (Envigo). Wild-type Mus musculus (C57BL/6J; The Jackson Laboratory, 000664), heterozygous B6.Cg-Dbh^{tm3.2(cre)Pjen}/J (Dbh-cre⁸²; The Jackson Laboratory, 033951) and heterozygous C57BL/6-Tg(Dbh-iCre)1Gsc (Dbh-iCre⁸³; MMRRC, 036778-UCD) mice of both sexes, aged 6–20 weeks, were used. All mice were on a C57BL/6J genetic background and were bred in-house or obtained from the above suppliers. Animals were group-housed (two to five per cage) in individually ventilated cages under controlled environmental conditions (ambient temperature 21–24 °C, relative humidity 40–60%) on either a reversed (for optogenetic experiments) or normal 12-h light/12-h dark cycle. Food and water were available ad libitum. All animals were regularly monitored to ensure stable housing conditions and welfare.

Animal surgeries and viral injections

Surgeries were performed on adult isoflurane-anesthetized mice. For slice two-photon imaging, mice were mounted on a stereotaxic frame (Kopf Instruments) before bilateral injection (500 nl) of AAV9.hSyn.nLightG (1.2 × 10¹³ genome copies per ml), AAV9.hSyn.nLightG2 (0.8 × 10¹³ genome copies per ml), AAV9.hSyn.nLightR (0.7 × 10¹³ genome copies per ml) or AAV9.hSyn.nLightR2 (4.5 × 10¹² genome copies per ml) into the LH with a Nanoject III (Drummond Scientific; 2 nl s⁻¹) using the following coordinates (relative to bregma): anterior–posterier (AP) –0.3 mm, medial–lateral (ML) ±0.9 mm, dorsal–ventral (DV) –5.2 mm.

For in vivo fiber photometry experiments during cued fear conditioning or TL, AAV9 encoding nLightG2 (~1.3 × 10¹³ genome copies per ml) or nLightG (~ 2.3 × 10¹³ genome copies per ml), GRAB NE_2m (~6.0 × 10¹² genome copies per ml), nLightG2-ctr (~1.1 × 10¹³ genome copies per ml) or nLightR2 (~8.9 × 10¹² genome copies per ml) under the control of the hSyn promoter was unilaterally injected into one of the following regions: pBLA (–2.0 mm AP, ±3.5 mm ML, –4.5 mm DV, volume: 400 nl) and LH (–1.4 mm AP, +1.1 mm ML, –5.2 mm DV, volume: 600 nl). For dual-color photometry experiments, a 1:1 mix was made with either nLightG2 or nLightG2-ctr and CaMKiia-PinkyCaMP ( ~ 9.8 × 10¹² genome copies per ml). For these experiments, the injections were administered in the aBLA (–1.4 mm AP, ±3.5 mm ML, –4.5 mm DV, volume: 400 nl). In all cases, fiber optic cannulas were implanted 0.1–0.2 mm above the injection sites (for the LH, pBLA and aBLA; 400-μm core diameter, NA = 0.57, Doric lenses) on the same surgery day.

For photometry in the dHPC, a small incision (~1 cm) was made along the scalp, the skull was cleaned of remaining tissue, and small craniotomies (~0.5 mm in diameter) were drilled above the dHPC (2 mm posterior and 1.5 mm lateral to bregma) and LC (5.4 mm posterior and 1.1 mm lateral to bregma). A glass pipette was then slowly lowered into the CA1 region of the dHPC, and 500 nl of viral suspension containing either AAV9.hSyn.nLightR2 (8.9 × 10¹² viral genomes per ml), AAV9.hSyn.nLightR (1.4 × 10¹³ viral genomes per ml) or AAV9.hSyn.nLightR2-ctrl (8.9 × 10¹² viral genomes per ml) was injected –1.6 mm relative to bregma at a speed of ~100 nl min⁻¹. In addition, 300 nl of viral suspension containing AAV9.EF1α.DIO.hChR2(H134R)-eYFP (~1.2 × 10¹³ viral genomes per ml; AddGene, 20298, a gift from K. Deisseroth, Stanford University) was injected in the LC at a depth of –3.6 mm. After virus injection, the pipette was left in place for 2 min before withdrawal. Subsequently, fiber optic cannulas (400-µm core diameter, NA = 0.5) were slowly inserted ~100 µm above the injection coordinates and fixed with dental cement. On top, a small head holder was cemented to the skull to allow for head fixation during experiments.

The procedures for general and local anesthesia, temperature monitoring, stereotaxic craniotomies, viral injections and implantation of electrodes for sleep–wake recordings were as previously described^18,60. For dual-color fiber photometry recordings in the LC, Dbh-iCre mice received a unilateral injection into the right LC at six different depths: lateral (from bregma), 0.95 mm; AP (from bregma), –5.4 mm; DV (from the dura), –3.2, –3.0, –2.8, –2.6, –2.4 and –2.2 mm, with an equal parts mixture of an AAV9.shortCAG.dlox.jGCaMP8f(rev).dlox (titer 6.5 × 10¹² viral genomes per ml, Viral Vector Facility Zurich) with either a AAV9.hSyn.nLightR2 (titer 8.9 × 10¹² viral genomes per ml) virus or a AAV9.hSyn.nLightR2-ctr (titer 8.9 × 10¹² viral genomes per ml) virus. At every depth, a volume of 200 nl was injected with a rate of 50–100 nl min⁻¹ through a thin glass pipette (5-000-1001-X, Drummond Scientific) pulled on a vertical puller (Narishige, PP-830), initially filled with mineral oil and backfilled with the virus-containing solution. At every injection step, a 1- to 2-min pause was interleaved, and the pipette was left in place for 10 min after the last injection. Surgical implantation of EEG/EMG electrodes was then performed as previously described⁶⁰. For fiber photometry recordings, an optic fiber stub coupled to a cannula (Doric Lenses, MFC_400/430-0.66_4.0mm_ZF1.25(G)_FLT) was implanted over the right LC (lateral 0.9; AP –5.4; DV –2.7) with an insertion speed of 1 mm min⁻¹ using a Thorlabs stereotaxic holder (XCL). Fibers were glued to the skull using Loctite Schnellkleber 401. Finally, a dental cement structure (Paladur and Steady-Resin polymer, Scheu) was created incorporating all the electrodes and the optic fiber holding the entire implant into place.

For hippocampal two-photon imaging, nLightR2 or nLightR2-ctr expression in neuronal cells was achieved through unilateral injection of AAV9.hSyn.nLightR2 or AAV9.hSyn.nLightR2-ctr in the CA1 area of the HPC (stereotaxic coordinates: −1.75 mm AP, +1.35 mm ML, −1.40 mm DV). For dual-color imaging experiments, AAV9.hSyn.nLightR2 was co-injected with pZac2.1 gfaABC1D-cyto-GCaMP6f (52925-AAV5, Addgene; a gift from B. S. Khakh, University of California, Los Angeles; titer of stock solution, ~7 × 10¹² genome copies per ml; working dilution, 1:2). This latter virus was used to achieve astrocyte-specific expression of GCaMP6f. AAVs were diluted in saline, and the viral solution (volume of 800 nl) was injected at a rate of 100 nl min⁻¹ using a micropump (UltraMicroPump, World Precision Instruments). A chronic optical window was implanted after the viral injection procedure following previous methods⁶³. In brief, the optical window consisted of a thin-walled stainless steel cannula (outer diameter: 3 mm; inner diameter: 2.77 mm; length: 1.50–1.60 mm) and a round coverslip (diameter: 3.00 mm). The coverslip was glued to the lower end of the cannula using ultraviolet-curable optical epoxy (Norland optical adhesive 63, Norland). To implant the optical window, a craniotomy (stereotaxic coordinates: 2.00 mm posterior and 1.80 mm lateral to bregma) was first performed. Optical access to the HPC was then achieved with careful aspiration of the overlying cortical tissue, while continuously perfusing the parenchyma with HEPES-buffered saline. The optical window was positioned above the external capsule, and a thin layer of instant gel adhesive (Loctite 454 Instant Adhesive, Henkel) was placed between the skull surrounding the craniotomy and the surface of the optical window. A customized titanium head plate was finally sealed to the skull using instant gel adhesive and dental cement (Super-Bond, Sun Medical). At the end of the implant procedure, antibiotics (Baytril, Bayer) were administered to mice through intraperitoneal injection.

For cortical two-photon imaging, AAV9.hSyn.nLightG2 (1 × 10¹³ genome copies per ml), AAV9.hSyn.nLightG2-ctr (1 × 10¹³ genome copies per ml, nonbinding control variant) or AAV9.hSyn.GRAB_NE2m (6 × 10¹² genome copies per ml) was injected unilaterally into the primary VC (~600 nl at a depth of 150–200 µm). A circular craniotomy (5 mm in diameter) was sealed with a glass coverslip, and a headplate was fixed to the skull for head fixation during imaging.

Acute brain slice preparation and imaging

Coronal LH brain slices (240 μm) were prepared from AAV-injected mice 3–4 weeks after injection. Mice were anesthetized with isoflurane and perfused with ice-cold cutting solution (75 mM NaCl, 50 mM sucrose, 6 mM MgCl₂, 2.5 mM KCl, 1.2 mM NaH₂PO₄, 0.1 mM CaCl₂, 25 mM NaHCO₃, 2.5 mM D-glucose and 5% CO₂/O₂). Brains were sectioned (VT1200S, Leica) and incubated in artificial cerebrospinal fluid (126 mM NaCl, 2.5 mM KCl, 1.2 mM MgCl₂, 1.2 mM NaH₂PO₄, 2.5 mM CaCl₂, 21.4 mM NaHCO₃, 11.1 mM D-glucose and 10 μM MK-801; 34 °C, 5% CO₂/O₂) for ≥45 min before two-photon imaging. nLight fluorescence was recorded using a custom BX51WI 2PLSM (Olympus) with a ×60/1.0-NA water objective, pulsed Ti:Sapphire laser (920/1,040 nm), x–y galvanometers and GaAsP PMTs (H10770PA-40, Hamamatsu). Spot photometry⁸⁴ (ROI 157 nm, 2 kHz) or rasterized movies (FOV 30 × 40 or 80 × 110 μm, 3 Hz) were analyzed in ImageJ. NE was applied exogenously via iontophoresis (1 M NE, 5- to 100-ms pulses, 20–200 nA, retention −10 to −20 nA) or endogenously via electrical stimulation (5 × 40 μA, 1 ms, 40 Hz, interstimilus interval 1–2 min). ∆F/F₀ was calculated relative to baseline (F₀: 300 ms pre-NE or average of ten frames). Spot photometry data were analyzed in AxoGraph, and rasterized movies were analyzed in ImageJ. Spatial ∆F/F₀ was fitted to a Gaussian (radial profiles, 20° increments, threshold 2σ). Endogenous NE release area exceeded 2.5σ of baseline. ON/OFF kinetics were determined from low-pass-filtered (50-Hz) traces by single exponential fits.

Cued fear conditioning paradigm

Behavioral experiments began ≥4 weeks after surgery. After 2 days of habituation to handling, tethering, the chamber and sound cues, mice were placed in the fear conditioning chamber (Med Associates) with patch cords for photometry (baseline: 10-min recording followed by ten CS+ tones (5 kHz, 20 s, intertrial interval (ITI) 120 s); association: same context, 10-min baseline followed by ten CS+ tones co-terminating with 0.5-mA, 1-s footshocks (ITI 120 s); re-exposure: similar to baseline, with an extra 5-min baseline outside the chamber before the 10-min baseline inside before CS+ exposure).

Pose estimation with DLC

Freezing behavior was quantified from locomotor speed extracted via DLC pose estimation^58,85. Speed was computed as frame-by-frame Euclidean distance between consecutive x–y positions, averaged across tracked body parts (nose, headcenter, bodycenter, tailbase and tailtip). Freezing events were defined as periods with a speed of <2 cm s⁻¹ for ≥1 s; percent time freezing was the fraction of frames below this threshold per time window (tone or after sound). DLC v3.0.0rc1 with a ResNet-50 backbone was trained on 20 labeled frames from three videos (95% train, one shuffle; train error 2.17 pixels, test 6.36 pixels; P_cutoff 0.6 → 6.52 pixels). Videos were acquired at 25 fps; low-confidence frames (likelihood of <0.8) were filtered, and coordinates were transformed to centimeters using a homography matrix. Analyses were performed in Python, integrating DLC output with trial alignment. Freezing was assessed during cue periods (20-s tones) and full postsound epochs (tones + 2-min ITI). Normality was tested via Shapiro–Wilk; due to small, non-normal samples, a nonparametric mixed-design ANOVA (Pingouin) was used with condition (baseline, association or re-exposure) within subject and indicator (nLightG2 versus control) between subject. Significant effects were followed by nonparametric post hoc tests (pairwise_tests parametric = False), Wilcoxon signed-rank or Mann–Whitney U-tests, Bonferroni corrected. Effect sizes are reported as generalized η², and data are visualized as mean ± s.e.m. with individual points; significance (P < 0.05, Bonferroni adjusted) is indicated by asterisks.

Fiber photometry recordings

Fiber photometry recordings in the LH or aBLA/pBLA used a Doric iFMC6 system (controlled by Doric Neuroscience Studio v6.1.2.0) with a low-autofluorescence patch cord (400 μm, 0.57 NA). nLightG2/nLightG were excited at 465 nm and nLightR/nLightR2/PinkyCaMP at 560 nm, and 405 nm was used as control. Excitation signals were sinusoidally modulated (208 and 572 Hz for 405/465 nm; 333 Hz for 560 nm), demodulated online and low-pass filtered at 12 Hz.

For dHPC recordings (~3 weeks after surgery), animals were habituated to handling, the setup and head fixation. Fiber photometry was performed with pyPhotometry⁸⁶ using 562.5/15-nm (nLightR2) and 405/10-nm (control) LEDs through a dichroic cube, with emission collected via Doric detectors. Excitation intensities at the brain tip were 170–220 μW. Pupil diameter was recorded simultaneously with a monochrome camera (DMK 33UX249) through a macro-objective, illuminated with 850-nm infrared and dim blue LED. LC optogenetic stimuli (20 ms, 20 Hz, 4 s) were delivered via MATLAB-controlled digital-to-analog converter (DAC) driving a fiber-coupled 488-nm laser (3–4 mW).

Sleep recordings in the home cage used a Doric system (Fluorescence Mini cube with 405-, 475- and 560-nm LEDs, ilFMC6-G3, FPC). Excitation LEDs were modulated at 208.615, 572.205 and 333.786 Hz, delivered via the optic fiber, with emission demodulated and acquired via the console (tip power: 12.5 μW (405 nm), 25 μW (475 and 560 nm) to minimize bleaching during 6–12 h recordings).

Data analysis of photometry recordings during fear conditioning experiments

Raw photometry data were analyzed using custom MATLAB scripts. Signals were first denoised with low-pass Butterworth filters (green: 10 Hz, order 2; red: 5 Hz, order 3; PinkyCaMP: 5 Hz, order 2). Photobleaching was corrected via double exponential detrending, and motion artifacts were corrected by linearly fitting the 405-nm control signal to the functional signal and subtracting the estimated motion component. Signals were then normalized by the bleaching fit and multiplied by 100 for ΔF/F₀ (%). For TL trials, the baseline was defined as 20 s before the event. The z scores were computed by subtracting the mean of 15 s before the cue and dividing by its standard deviation. Trial averages were computed per mouse and then across the cohort; ΔF/F₀ or z-score signals were smoothed with an 80-point moving mean. Peak z scores were taken during tone or cue–footshock periods. Cumulative signal was quantified by area under the curve (AUC; trapz function) on mean trial z scores, and cohort averages were plotted.

Data analysis of photometry recordings during optogenetic stimulation

Pupil diameter was extracted from videos using DeepLabCut⁵⁸ and custom MATLAB scripts¹⁷ and then low-pass filtered (third-order Butterworth, 10 Hz). Photometry (560-nm) traces were processed with a three-point moving median to remove single-sample outliers, corrected for photobleaching via double exponential least squares fit and low-pass filtered (third order, 10 Hz). The 405-nm excited photometry trace was treated in the same way to serve as a control channel, but given that contributions to the overall fluorescent signal (fiber autofluorescence, native tissue fluorescence and indicator fluorescence) likely differ for both channels⁸⁷, the 560-nm evoked fluorescence was not corrected using the 405-nm excited control trace. Preprocessed traces were cropped around each stimulation (10/30 s before/after), z scored for photometry (mean subtracted, divided by prestimulus standard deviation) and normalized for pupil (divided by median prestimulus diameter). AUC was calculated for averaged z-scored photometry and pupil traces during optogenetic stimulation.

Data analysis of photometry recordings during sleep experiments

The sleep data were scored in 4-s epochs using the EEG/EMG signals according to standard procedures using a custom semiautomated MATLAB-based software (VeryScore 2.0)^18,60. ΔF/F₀ signals were extracted as follows. For the green emission (jGCaMP8f), F₀ was defined by the fitted calcium-independent isoemissive signal (violet excitation), following ref. ⁸⁸. For red emission (nLightR2 or nLightR2-ctr), F₀ was derived from a fifth-degree polynomial fit to the signal collected under 560-nm excitation⁸⁹. Representative examples were generated from long (≥90 s) NREMS bouts, including microarousals, for which simultaneous EEG, EMG and fiber photometry signals were collected. All photometry signals were z scored using SciPy’s zscore function. EEG spectrograms were computed by Gabor–Morlet wavelet transforms of bipolarized EEG signals. The EMG signal was bipolarized and high-pass filtered (Chebyshev, 0.25-Hz cutoff), and its lower envelope was retained. LC-associated peaks in the jGCaMP8f signal were identified by low-pass filtering (0.5 Hz, finite input response (FIR) = 100) and correction using the lower envelope filtered at 0.1 Hz. Peaks with prominence above the 60th percentile were detected using SciPy’s ‘find_peaks’ function, with additional exclusion criteria based on amplitude and proximity to wake episodes⁶⁰. For cross-correlations between jGCaMP8f and nLightR2 (or nLightR2-ctr), peaks not associated with microarousals (<12-s awakenings) were retained. Corresponding 100-s signal windows centered on each peak were isolated, z scored and cross-correlated (SciPy’s ‘correlate’ function). Mean values were extracted within ±10-s lags. Data distributions were tested for normality and homoscedasticity; when normality was not met, comparisons were performed using Mann–Whitney U-tests. For assessing temporal differences between jGCaMP8f and nLightR2 (or nLightR2-ctr) around activity peaks, z-scored traces within ±10 s of detected peaks were averaged and compared statistically.

Two-photon in vivo imaging in the HPC

Two-photon in vivo imaging was performed in awake, head-fixed mice using a previously described custom setup^17,63. nLightR2 and GCaMP6f signals were recorded with a galvo-resonant scanhead (Ultima Investigator, Bruker), a ×16/0.8-NA objective (Nikon) and multialkali photomultipliers. To identify optimal nLightR2 excitation, wavelengths from 920 to 1,060 nm (10- to 20-nm steps) were tested with a tunable Chameleon Ultra II laser at constant 50-mW power, recording 2-min t-series in the hippocampal pyramidal layer in both green and red channels while mice ran or stayed still. For experiments, nLightR2 was excited at 1,040 nm and GCaMP6f at 920 nm using independent Ti:Sapphire lasers, with beam divergence adjusted via ×2 telescopes and alignment mirrors to ensure proper back-aperture filling and <1 μm focal displacement, verified with 100-nm fluorescent beads. Imaging in the CA1 used ~80 mW per beam; emission signals (525/70 nm and 595/50 nm) were collected, amplified, digitized (12-bit) and recorded at ~30 Hz, 512 × 512 pixels, 0.53-μm pixel size and 3× optical zoom. Mice were habituated 1 month after surgery, specifically two brief handling sessions followed by five progressively longer head fixation sessions (up to 45 min) on a custom wheel (8-cm radius, 9-cm width), allowing free movement.

Experiments in virtual reality

We used Blender (v.2.78c) to create a virtual reality corridor (180 cm long and 9 cm wide) displayed across five screens, with three white textures (vertical lines, mesh, circles) on a black background. The character’s viewpoint spanned 220° horizontally and 80° vertically. Animal movement on a custom wheel with an optical rotary encoder⁶³ (Avago, AEDB-9140-A14) was rendered 1:1 in the virtual corridor, with signals processed by an Arduino Uno R3. Water rewards (~4-µl drops) were delivered via a solenoid-controlled lick port, and licks were monitored by a capacitive sensor. Imaging and behavioral setups were synchronized via the start-of-frame TTL signal. Mice were teleported to the corridor start after reaching the endpoint; each crossing was a trial with a 5-s ITI, and trials exceeding 120 s were terminated. One to 2 weeks before virtual reality experiments, mice underwent water scheduling (~1 ml per day) to maintain 80–90% of prescheduling weight. Mice were trained to receive rewards after running on the wheel over approximately five sessions with pseudorandom water delivery. Two-photon imaging in the CA1 pyramidal layer (80–150 µm depth) began on the first virtual reality exposure. For single-color experiments, FOV recordings lasted multiple consecutive days (nLightR2: days 1–5; nLightR2-ctr: days 1–2), with water reward locations gradually shifted from 85 cm to 145 cm. Dual-color experiments (nLightR2 + GCaMP6f) used a fixed reward at 85 cm. Each imaging session included three to four t-series (7,500–12,500 frames; ~250–415 s) interleaved with 3- to 4-min breaks, after which mice returned to their home cage.

Analysis for two-photon imaging data in the HPC

t-Series were analyzed using previously reported Python v.3.6 code. Motion artifacts were corrected per t-series using their average projection as reference; multiple t-series from the same FOV were aligned via x–y drifts and concatenated into a session movie, which was finally motion corrected against its average projection as in^17,63,90. Across-day FOV alignment was performed by stacking average projections from each session, computing x–y shifts and applying them to all frames. Dual-color experiments were motion corrected separately for red (nLightR2) and green (GCaMP6f) channels. Axial displacements were assessed via the structural similarity index⁹¹ relative to session averages; frames with a structural similarity index of <0.9 were excluded (<1% of frames, except one animal excluded for <95%). Fluorescence signals were extracted either across the FOV or locally using a 32-pixel (~17-μm) grid for nLightR2 ROIs with CITE-On⁹². In dual-color experiments, astrocytic GCaMP6f signals were segmented using ASTRA⁶², with proximal processes parceled by k-means clustering (Max_Roi_Area_Proc = 50 μm²). Running epochs were defined as intervals where speed exceeded 1 cm s⁻¹, preceded by 2 s below 1 cm s⁻¹ and followed by 5 s above 1 cm s⁻¹; reward epochs were based on crossing reward positions. Event-triggered averages used a −2- to 5-s window around epoch onset. Fluorescence was expressed as ΔF/F₀ = (F_t − F₀) / F₀, with F₀ as the mean intensity in −2 to 0 s and traces filtered with a 500-ms mean filter. Pearson correlations between ROI pairs were computed (self-correlations removed), and shuffled traces, least squares models and hierarchical clustering followed prior methods. For t-series acquired at different excitation wavelengths, motion correction was applied per channel. Frame intensity distributions were analyzed (mean, 25th, 50th, 75th and 90th percentiles), and median values over 3,600 frames per t-series were used for wavelength comparisons.

Two-photon in vivo imaging in the VC

In vivo calcium imaging was performed using a Bergamo II two-photon microscope (Thorlabs) controlled by ScanImage. A Chameleon Vision Ti:sapphire laser (Coherent) tuned to 920 nm was used to excite the indicators. Fluorescence was collected through a ×16/0.8-NA water-immersion objective (Nikon). Volumetric imaging (four or eight planes, 25-µm spacing) was performed at a total frame rate of 30 Hz (3.75 or 7.5 Hz per plane), with each image acquired at 512 × 512 resolution. Laser power ranged from 30 to 50 mW at the sample and was matched across indicators. Each animal was imaged over two to four sessions, spaced over multiple days to ensure sufficient sampling of both spontaneous and evoked activity. Mice were head fixed on a freely rotating running wheel. Locomotion was continuously tracked using an optical encoder (Kuebler). Visual stimuli were presented on iPad (Apple) LCD monitors positioned in front and laterally of the mouse. Looming stimuli were interleaved with 5 min of gray screen intervals. Forced locomotion trials were conducted by gently rotating the wheel while mice were head fixed.

Analysis for two-photon imaging data in the cortex

Raw time series data were corrected for lateral motion using a multipass rigid alignment approach. A reference frame was constructed from the top-correlated subset of stationary frames, and shifts were computed using phase correlation. Aligned frames were tiled into nonoverlapping square ROIs (tiles), and per-tile fluorescence was extracted by averaging pixel intensity within each tile. ΔF/F₀ was computed as using F₀ as a rolling baseline, calculated per tile to preserve local dynamics. The z-scored responses were computed using the baseline mean and standard deviation from a 10-s prestimulus window. Spontaneous and stimulus-evoked events were quantified per tile and per animal. Metrics included mean and peak ΔF/F₀ (raw and z scored) in a 20-s window after loom onset, standard deviation and range of responses and time-to-peak from stimulus onset. To characterize modulation across behavioral states, we computed regression slopes between stationary and running mean ΔF/F₀ responses across tiles. Group-level comparisons were performed using Kolmogorov–Smirnov tests (tile-level distributions), paired/unpaired t-tests and Wilcoxon tests (mouse-level summaries) and Cliff’s δ and Cohen’s d for effect size estimation. Raw time series data were corrected for lateral motion using a multipass rigid alignment approach based on phase correlation. A reference frame was generated from the most correlated subset of stationary frames, and shifts were applied to all frames accordingly. Motion-corrected movies were divided into nonoverlapping square ROIs (tiles), and per-tile fluorescence traces were obtained by averaging pixel intensities within each tile. ΔF/F₀ was computed using a rolling baseline (F₀) defined per tile to preserve local dynamics. The z-scored responses were calculated using the baseline mean and standard deviation from a 10-s prestimulus window. Spontaneous and stimulus-evoked events were quantified per tile and per animal. Extracted metrics included mean and peak ΔF/F₀ (raw and z scored) in a 20-s poststimulus window, response standard deviation and range and latency to peak from stimulus onset. To characterize modulation across behavioral states, regression slopes were computed between stationary and running mean ΔF/F₀ responses across tiles. Group-level comparisons were performed using Kolmogorov–Smirnov tests for tile-level distributions, paired or unpaired t-tests and Wilcoxon signed-rank tests for mouse-level summaries and Cliff’s δ and Cohen’s d for effect size estimation. To disentangle the relative influence of sensory and behavioral factors on noradrenergic activity, a GLM was fitted to the per-tile ΔF/F₀ time series of nLightG2-expressing mice. The model included regressors for looming stimulus epochs, running speed and their interaction. The full model was formulated as:

$$F(t)={\beta }_{0}+{\beta }_{1}\,L(t)+{\beta }_{2}\,R(t)+{\beta }_{3}\,[L(t)\times R(t)]+\varepsilon (t)$$

where F(t) denotes fluorescence at time t, L(t) represents binary loom epochs, R(t) represents continuous running speed, and ε(t) represents the residual error. Model fits were evaluated using the coefficient of determination (R²), which quantifies the proportion of variance in fluorescence explained by the model. The relative contribution of each predictor (ΔR²) was calculated as the reduction in explained variance when that predictor was removed from the model (leave-one-out approach). R² and ΔR² values were computed per tile, averaged per animal and visualized as tiled heat maps to assess spatial distributions of sensory and behavioral modulation. Statistical comparisons across predictors were performed using Friedman and Wilcoxon tests (two-sided) across mice.

Immunohistochemistry

To verify adequate transgene expression, animals used for optogenetic experiments were transcardially perfused (50 ml of PBS and 50 ml of 4% paraformaldehyde in PBS) under deep, terminal ketamine/xylazine anesthesia (180/24 mg per kg (body weight), intraperitoneal). Brains were extracted, postfixed in 4% paraformaldehyde (4 °C for 24 h), embedded in 3% agarose and sliced into ~50-µm coronal slices using a vibratome (Microm HM 650V, Thermo Scientific). Nonspecific binding sites were blocked (10% normal goat serum and 0.3% Triton X-100 in PBS) for 2 h at room temperature before incubating LC slices with primary antibodies (chicken anti-GFP (1:500; A10262, Thermo Fisher Scientific) to visualize ChR2–eYFP and rabbit anti-TH (1:1,000; AB152, Merck Millipore) to visualize tyrosine hydroxylase-expressing neurons in the LC) in carrier solution (2% normal goat serum and 0.3% Triton X-100 in PBS, 48 h, 4 °C). Slices were then washed in PBS (three times for 10 min each) before incubation with secondary antibodies (Alexa Fluor 488 goat anti-chicken (A11039, Thermo Fisher Scientific) and Alexa Fluor 647 goat anti-rabbit (A32733, Thermo Fisher Scientific); both diluted 1:1,000 in carrier solution) overnight at 4 °C. Subsequently, slices were washed again (three times for 10 min in PBS) before mounting them on microscope slides using Roti-Mount Fluor Care (HP19.1, Carl Roth). dHPC slices were mounted on microscope slides unstained. Slices were imaged with an epifluorescence microscope (Axio Scan Z1, Zeiss; Plan-Apochromat ×10/0.45-NA M27, Zeiss, for overview images; Plan-Apochromat ×40/0.95-NA Korr M27 for detailed LC images), and images were processed using ImageJ (Fiji).

For dual-color sleep experiments in the LC, mice were perfused with 4% paraformaldehyde after pentobarbital overdose. Brains were sectioned at 50 µm using a manually guided freezing microtome. LC sections were washed (three times with 0.3% Triton X-100 in PBS) and incubated 1–1.5 h with mouse anti-GFP (1:150; Roche, 1814460) and rabbit anti-mCherry (1:1,000; ab167453, Abcam), followed by goat anti-mouse Alexa Fluor 488 (1:1,000; Jackson ImmunoResearch, 115-545-003) and donkey anti-rabbit Alexa Fluor 594 (1:1,000; Thermo Fisher Scientific, R37119) secondary antibodies. Sections were rinsed with 0.1 M phosphate buffer, mounted on slides and fixed with Mowiol. Images were acquired with three microscopes: a Nikon SMZ25 stereomicroscope (DS-Ri2) for brightfield fiber localization, a Nikon Ti2 spinning disk (DS-Qi2) for whole-hemisphere viral expression and a Leica Stellaris 8 (K5 sCMOS) for LC fluorescence.

Mice used for in vivo two-photon imaging were intracardially perfused with 4% paraformaldehyde in PBS after deep anesthesia with urethane. Brains were then fixed overnight at 4 °C and subsequently cut into coronal slices of 50 μm thickness. Cell nuclei were counterstained with Hoechst 33342 (50 μM), incubating the slices for 20 min at room temperature. Samples were mounted on glass slides using Fluoromount (Sigma-Aldrich) and coverslipped. Fluorescence images were acquired with a Leica SP5 inverted confocal microscope (×40/1.25-NA oil immersion objective, Leica). Fluorescence excitation was performed sequentially using 561-, 488- and 405-nm laser lines. Fluorescence detection was performed using spectral PMTs with bandpass settings at 525/70 nm, 595/50 nm and 455/70 nm for green, red and blue channels, respectively. Red and green channels were acquired simultaneously during either 561-nm or 488-nm excitation, emulating the detection configuration used in two-photon multiplexed imaging experiments. Excitation power was maintained constant for both 561- and 488-nm lasers. The blue channel was acquired separately. Images were acquired with the pinhole set to 1 Airy unit and line average set to 2, as 512 × 512 pixel frames with 0.197-µm pixel size.

Statistical analysis

Statistical analyses were performed using GraphPad Prism or custom Python scripts. Two-tailed unpaired t-tests, two-way ANOVAs or nonparametric equivalents (Mann–Whitney U-test, when normality was violated per Shapiro–Wilk test) were used. Post hoc comparisons used Sidak’s, Tukey’s or Dunnett’s tests as appropriate. Linear mixed-effects models were computed with Statsmodels v0.9.0. Data are presented as mean ± s.e.m., with significance denoted as *P < 0.05, **P < 0.01, ***P < 0.001 and ****P < 0.0001; exact values are reported in Supplementary Tables 1–3. Histology confirmation for photometry experiments was conducted in at least two slices per mouse. Micrographs in figures show a representative image of each group.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.