The locus coeruleus (LC) norepinephrine (NE) system is involved in a variety of physiological and pathophysiological processes. Refining our understanding of LC function largely relies on selective transgene expression in LC-NE neurons, allowing targeted manipulation and readout of noradrenergic neurons. Here, we performed a side-by-side comparison of the most commonly used strategies to genetically target the LC, including different cre driver lines and promoter-mediated transgene expression. We report differences between these strategies in terms of transgene expression efficacy and specificity. Parallelly, we found no behavioral alterations in cre-expressing mice of any mouse line compared to wild-type littermates. Finally, to further facilitate the investigation of LC-NE function, we created a suite of constructs, including a reporter protein, a calcium indicator, and a light-driven cation channel, whose expression is mediated by the previously described PRS×8 promoter. These constructs allow identification, monitoring, and manipulation of LC-NE activity either in wild-type mice, or in combination with tissue-specific manipulations of different cre driver lines. The results of our study are crucial for the interpretation of previous experiments using the respective targeting strategies, as well as for the design of future studies.

Efficacy and specificity of transgene expression across model systems

To visualize transgene expression in the LC-NE system, we bilaterally injected the LC with titer-matched suspensions of recombinant adeno-associated virus (rAAV2/9) encoding enhanced green fluorescent protein (eGFP; Fig 1A). In Dbh^cre, Net^cre, and Th^cre mice, cre-dependent gene expression was controlled by combining double-floxed inverted open reading frames (DIOs) [33] of the transgene with a strong, synthetic promotor (CAG) [34]. In wild-type mice (C57BL/6J), the reporter gene was expressed under the control of the synthetic PRS×8 promoter. Six weeks after injection, we analyzed eGFP expression by fluorescence microscopy of coronal sections which were immuno-stained against TH (to visualize NE-expressing neurons) and GFP (to enhance the signal originating from transgene expression; Fig 1B). To quantify transgene expression at a cellular level, cells expressing TH (TH⁺) or eGFP (eGFP⁺) were automatically segmented using the deep learning-based algorithm CellPose [35,36] (Fig 1C).

To assess the ability to selectively target the LC, we evaluated the efficacy and specificity of viral transduction of LC-NE neurons. To this end, we compared the overlap between masks segmented in the TH and GFP channels, respectively, defining cells with an overlap of ≥50% as co-expressing. Efficacy is defined as the proportion of TH⁺ cells co-expressing eGFP, while specificity reports the proportion of eGFP⁺ cells co-expressing TH.

We found differences in the efficacy between model systems (one-way ANOVA, F₂₄ = 14.71, p = 1.2 × 10⁻⁵; Fig 1D): while no significant differences in efficacy were detected between Dbh^cre, Net^cre, and PRS×8-mediated transgene expression (70.5 ± 11.8%, 79.5 ± 9.0%, 78.2 ± 12.9%; mean ± SD; Tukey’s test, p = 0.68/0.77/0.99 for Dbh^cre versus Net^cre/Dbh^cre versus PRS×8/Net^cre versus PRS×8), the efficacy of Th^cre mediated expression was significantly lower than the other model systems (33.3 ± 22.7%; Tukey’s test, p = 6 × 10⁻⁴/3 × 10⁻⁵/5 × 10⁻⁵, compared against Dbh^cre/Net^cre/PRS×8, respectively). We further noted a large variability in the efficacy of eGFP expression in Th^cre mice, with one mouse showing an efficacy of up to 75%, while the efficacies of other mice—even from the same litter—were below 20% (S1 Fig).

We also found differences in the specificity (one-way ANOVA, F₂₄ = 14.47, p = 1.4 × 10⁻⁵) between model systems (Fig 1E): Here, the specificity of Th^cre mediated expression (46.0 ± 12.1%) was significantly lower than Dbh^cre (82.2 ± 9.5%), Net^cre (71.4 ± 13.6%), and PRS×8 (65.2 ± 5.0%) mediated transgene expression (Tukey’s test, p = 7 × 10⁻⁶/ 7 × 10⁻⁴/0.01 Dbh^cre/Net^cre/PRS×8 against Th^cre). Furthermore, the Dbh^cre approach was significantly more specific than PRS×8 mediated transgene expression (Tukey’s test, p = 0.03). No significant differences were found between Dbh^cre versus Net^cre and between PRS×8 versus Net^cre.

In addition to differences in the efficacy of transgene expression and its specificity to TH⁺ neurons, we realized that also the levels of transgene expression in relation to the levels of endogenous TH expression differed across approaches. While the averaged normalized eGFP fluorescence in true positive neurons (i.e., neurons that co-expressed eGFP and TH) was expectedly higher than in false positive neurons (i.e., neurons that expressed eGFP but not TH) in Dbh^cre, Net^cre and PRS×8 mice (0.02 ± 0.03/ 0.08 ± 0.06/0.06 ± 0.03, respectively), it was lower than the eGFP fluorescence of true positive cells in Th^cre mice (−0.12 ± 0.13), indicating that the levels of cre-mediated transgene expression are functionally de-coupled from the levels of native TH expression in this mouse line (one-way ANOVA, F₂₄ = 10.1, p = 0.0002; Fig 1F). Similarly, at the cellular level, normalized eGFP fluorescence in true positive neurons was higher than in false positive neurons in Dbh^cre, Net^cre, and PRS×8 mice, while eGFP fluorescence in false positive neurons exceeded the fluorescence of true positive neurons in Th^cre mice (p < 0.1 × 10⁻⁵ for all groups, Bonferroni-corrected Wilcoxon rank sum test; S2 Fig). These findings indicate that cre recombinase activity and endogenous TH expression are decoupled in Th^cre mice, and further corroborate the heterogeneous expression patterns across approaches.

Control experiments

We performed a number of controls to validate our results and to rule out potential sources of confounding.

Ectopic expression of transgenes

In addition to differences in the specificity of neural transduction within the LC and its immediate surroundings, we also observed ectopic eGFP expression in various brain regions more distant from the LC (Fig 2). Again, we found differences between the different model systems with respect to ectopic LC transgene expression. The lowest ectopic expression was found in Dbh^cre animals, where we observed eGFP expression only in small numbers of cells in the cerebellum (<10 neurons, 3/7 animals; S8A Fig) and in the inferior colliculus (<10 neurons, 1/7 animals; Fig 2A). In Net^cre mice, we found a small number of eGFP-expressing cells in the cerebellum (<10 neurons, 4/7 mice Fig 2B), and moderate eGFP expression in the vestibular nucleus of a single mouse (~10–100 neurons, S8B Fig). In Th^cre mice, we found substantial ectopic expression in the lateral parabrachial nucleus (~10–500 neurons, 7/7 mice; S8C Fig), moderate expression in the inferior colliculus (10–100 neurons, 2/7 mice; Fig 2C) and central gray (<100 neurons, 1/7 mice; S8D Fig) and low expression in the cerebellum (<10 neurons, 1/7 mice; S8E Fig). Finally, in PRS×8 animals, we found moderate eGFP expression both in the vestibular nucleus (~10–200 neurons, 6/7 animals; S8F Fig), and in the lateral parabrachial nucleus (~10–100 neurons, 3/7 animals; Fig 2D).

Effects of promoter, serotype, and injection volume on transduction of LC-NE neurons

Next, we systematically varied the promoter controlling transgene expression, the serotype of the rAAV, and the volume of injected virus suspension, in order to investigate potential effects of these parameters on the transduction of LC-NE neurons.

Success rates of LC targeting

Occasionally, we observed complete absence of eGFP (i.e., transgene expression) in one of the two LC of bilaterally injected animals, while expression was strong in the second LC. This pattern of unilateral transgene expression was found in 3/2/1/3 out of each 5 Dbh^cre/Net^cre/Th^cre/PRS×8 mice, respectively (Fig 4A). Since we used the same micropipette to inject the viral suspension sequentially in both LC, we were concerned that this approach might have compromised the success rate of injections (e.g., due to possible inhomogeneous distribution of viral particles in the suspension). Therefore, we added two more mice to each group using a new micropipette for each LC. However, also here, we observed unilateral transgene expression in some cases (0/0/2/1 Dbh^cre/Net^cre/Th^cre/PRS×8; Fig 4B), which did not statistically differ from the previous group where the same pipette was used (p = 0.72, chi-squared test, n = 20/8 mice; Fig 4C). Thus, sequential injection with the same pipette did not explain the occasional failure of virus transduction. Importantly, infusion of the viral suspension was also visually confirmed during each injection, and hence the absence of expression in some LC could not be attributed to clogged micropipettes.

The overall success rate of injections in our study (80%, 59/74 LC) is in line with previous reports (76%–80%) [43,44], which attributed the lack of transgene expression to off-target virus administration (i.e., injection in brain regions adjacent to LC). To test this possibility, we injected Dbh^cre, Net^cre, and Th^cre mice (n = 3 each) with a so-called ‘cre-switch’ construct which consists of a DIO for tdTomato and an inverted open reading frame for eGFP (DO-tdTomato-DIO-eGFP) [45]. Hence, cell-type specific expression of eGFP in cre-positive cells allows for characterization of the molecular specificity, while expression of tdTomato in cre-negative cells allows for estimation of the viral spread (and potential off-target virus administration) at the injection site (Fig 4D). Four weeks after injection, expression of tdTomato in cre-negative cells revealed a broad viral spread in most animals, reaching from the midline to the parabrachial nucleus in the mediolateral axis, and from the fourth ventricle to the central gray in the dorsoventral axis (Fig 4E). However, also in these experiments we observed three animals (out of nine) with unilateral eGFP expression where also tdTomato was absent in the non-expressing hemisphere (Fig 4F), suggesting that absence of expression in LC was not due to off-target injection in neighboring brain tissue but rather a failure of virus infection in general.

We therefore suggest that the lack of expression in some LC might result from off-target administration into the 4th ventricle, which is in close proximity to the LC. This mistargeting could result from minor misalignment on the mediolateral axis during virus injection, resulting in more lateral virus injection in one hemisphere, but virus injection into the 4th ventricle in the other hemisphere. Indeed, when changing the injection coordinates from ±0.9 mm to ±1.1 mm lateral relative to bregma in subsequent experiments (i.e., injection of AAV2/9-hSyn-DIO-eGFP in Dbh^cre mice and injection of AAV2/2 or AAV2/9-hSyn-eGFP in WT mice; see Figs 3 and S9), we observed bilateral transgene expression in 7/7 and 12/12 LC (Fig 4F and 4G). These success rates are significantly higher than the ones observed when injecting ±0.9 mm lateral relative to bregma (Fig 4H and 4I; p = 0.032 for conditional expression, chi-squared test, n = 28/7 mice; p = 0.031 for unconditional expression, chi-squared test, n = 9/12 mice). Thus, more laterally targeted injections lead to increased transduction rates of the LC, likely by lowering the risk of injecting the virus suspension into the 4th ventricle.

Behavioral screening of cre lines

Unlike the PRS×8 approach, cre-dependent transgene expression relies on the use of mouse lines expressing cre recombinase in a molecularly defined set of cells. As phenotypic alterations have been broadly reported for neuromodulatory cre driver lines [46–48], we screened the driver lines used in this study for behavioral alterations. We applied a series of well-established behavioral paradigms to quantify spontaneous locomotion, anxiety, exploratory behavior, working memory, and spatial learning and memory.

As LC-NE activation is strongly linked to stress and anxiety [6,49], we first compared anxiety-like behavior in cre-expressing Dbh^cre, Net^cre, and Th^cre mice to their wild-type littermates. To this end, mice were left to freely explore an open field arena for 20 min (Fig 5A, left) [50]. As mice show the intrinsic tendency to avoid open areas, the time spent near the walls of the arena serves as a proxy for anxiety, with less anxious mice spending more time in the center. While not all groups displayed significant avoidance of the arena center (p = 0.004/0.06/4 × 10⁻⁴/ 4 × 10⁻⁵/0.14/6 × 10⁻⁵ for Dbh^cre/Dbh^WT/Net^cre/Net^WT/Th^cre/Th^WT, respectively; one-sample t test against chance level (16%), n = 16 mice per group), anxiogenic conditions of the paradigm could be confirmed when pooling the data of all mice (p = 6 × 10⁻¹⁵, n = 96 mice). When comparing eight males and eight females of each heterozygous cre-expressing line with a corresponding number of wild-type littermates of each mouse line, we did not detect any significant behavioral difference that could be explained by either genotype, sex, or the interactions of these two factors in any mouse line (Fig 5A, right; two-way ANOVA; genotype effects: Dbh^cre: F₂₈= 0.45, p = 0.51; Net^cre: F₂₈ = 0.54, p = 0.47; Th^cre: F₂₈ = 0.01, p = 0.91; effects of sex and interactions of sex and genotype are shown in S2 Table). Also, additional readouts used to approximate anxiety and spontaneous locomotion did not indicate any genotype-dependent behavioral difference in any mouse line (S10 Fig).

As an additional experiment approximating anxiety-related behavior, mice were left to explore an elevated plus maze (EPM) [51] for 5 min. The EPM consists of four arms radiating from a central platform, with two opposing arms being walled and the other two arms being open (Fig 5B, left). Thus, mice need to balance their natural exploratory drive with their intrinsic tendency to avoid exposed areas. At the start of the experiment, mice were placed on the central platform, and the time spent in the open and closed arms served as a proxy for anxiety, with less anxious mice spending more time in the open arms. Again, significant avoidance of the open arms could not be demonstrated in each experimental group (p = 0.08/0.002/5 × 10⁻⁴/1 × 10⁻⁴/0.01/0.23 for Dbh^cre/Dbh^WT/Net^cre/Net^WT/Th^cre/Th^WT, respectively; one-sample t test against chance level (48%), n = 16 mice per group), but the anxiogenic nature of this paradigm could be validated when pooling the data of all mice (p = 3 × 10⁻¹², n = 96 mice). Also, in the EPM, we did not detect any behavioral difference that could be explained by either genotype, sex, or the interactions of these two factors in any mouse line (Fig 5B, right; two-way ANOVA; genotype effects: Dbh^cre: F₂₈ = 1.96, p = 0.17; Net^cre: F₂₈ = 0.01, p = 0.93; Th^cre: F₂₈ = 0.32, p = 0.57; effects of sex and interactions of sex and genotype are shown in S2 Table). Similarly, no behavioral alterations were found in any additional parameters extracted from the EPM (S11 Fig). In conclusion, we did not detect any genotype-related changes in anxiety-like behavior in any mouse line tested in two distinct experimental paradigms under light anxiogenic conditions.

Recent studies have demonstrated the involvement of the LC-NE system in the formation of episodic and spatial memory [52,53], which we investigated in two additional experiments. First, we employed the Y-maze paradigm to approximate working memory in mice. In this assay, mice are placed in a maze with three symmetrically radiating arms, which they are free to explore. Based on the innate tendency to explore novel environments, mice spontaneously alternate between arms, i.e., when leaving one arm, they tend to move to the arm that was not previously visited, rather than the arm that was previously explored (Fig 5C, left; visual cues were given to distinguish the different arms) [54,55]. This was quantified as the fraction of alternations (i.e., consecutive visits of all three arms) over the total number of transitions. The fraction of alternations was above chance level in all groups (p < 0.01; Bonferroni-corrected one-sample t test), indicating intact working memory in all groups. No genotype-related behavioral changes were found (Fig 5C, right; two-way ANOVA; genotype effects: Dbh^cre: F₂₈ = 1.23, p = 0.26; Net^cre: F₂₈ = 2.48, p = 0.13; Th^cre: F₂₈ = 0.23, p = 0.63; effects of sex and interactions of sex and genotype are shown in S2 Table, additional results from Y-maze experiments are shown in S12 Fig).

Finally, to investigate spatial memory, we performed the Morris Water Maze (MWM) task [56]. In the MWM, mice are placed in a circular pool filled with opaque water that contains a hidden platform placed just below the surface that mice can access to escape the water (Fig 5D, left; visual cues for spatial orientation were placed around the pool). The MWM task was performed on three consecutive days after habituating mice to the water. On the first two days, each mouse was placed in the water four times, at pseudo-randomized positions in each trial. Decreased latencies and travel distances to reach the platform during these training sessions indicated successful spatial learning in all experimental groups (S13 Fig). On the second day, after the first two training trials, a probe trial was performed, in which the platform was removed. The time spent in the platform quadrant approximates short-term memory. On day three, a single probe trial was done, from which long-term memory can be measured. The time spent in the platform quadrant during the probe trial on day 3 was above chance level in all groups (p < 0.01; Bonferroni-corrected one-sample t test), indicating intact spatial memory formation. No genotype-dependent behavioral changes were found (Fig 5D, right; two-way ANOVA; genotype effects: Dbh^cre: F₂₈ = 1.14, p = 0.29; Net^cre: F₂₈ = 0.56, p = 0.46; Th^cre: F₂₈ = 0.34, p = 0.56; effects of sex and interactions of sex and genotype, as well as results of the short-term memory test on day two are shown in S2 Table, additional results of the MWM experiments on day 3 are shown in S14 Fig). In conclusion, we found no genotype-related behavioral differences between cre-expressing mice and wild-type littermates in Dbh^cre, Net^cre, or Th^cre mice in the four behavioral assays considered.

PRS×8-based constructs for monitoring and manipulating LC activity

Transgene expression driven by the PRS×8 promoter, which has high efficacy and good local specificity (Fig 1), is independent of cre recombinase. Thus, molecular targeting of LC-NE neurons with the PRS×8 promoter can be applied either to wild-type mice, eliminating the need for a cre driver line, or to a different cre driver line, providing access to two genetically distinct neuronal populations in the same animal [32]. In addition, PRS×8-based transgene expression can be applied to different species, as demonstrated in rats [25,28,57,58]. As there are currently very few AAV-carried PRS×8-based constructs available, we developed a set of PRS×8-based constructs to enable the monitoring and control of LC-NE activity (of note, canine Adenovirus type 2 (CAV-2) carrying PRS×8-mediated transgenes are commercially available at https://plateau-igmm.pvm.cnrs.fr/?cat=69).

To optically monitor LC-NE activity, we developed a construct encoding jGCaMP8m [59] under the control of the PRS×8 promoter. We directly compared its functionality to a conditionally expressed indicator in the contralateral LC of the same animal. To do so, we injected rAAV2/9-PRS×8-jGCaMP8m into one LC of a Dbh^cre mouse and a cre-dependent, spectrally compatible calcium indicator jRGECO1a [60] (AAV2/1-hSyn-DIO-jRGECO1a) into the contralateral LC (Fig 6A and 6B). In line with our previous results (Fig 1), we observed a specificity of 92.5% and an efficacy of 79.6% for PRS×8-jGCaMP8m, and a specificity and efficacy of 94.9% and 64.3% for hSyn-DIO-jRGECO1a. We then implanted optical fibers above the injection sites and obtained fiber photometry recordings simultaneously from both LC of awake, head-fixed mice, while monitoring the pupil diameter of the animals (Fig 6C). We observed strong correlations between the neuronal dynamics of jGCaMP8m- and jRGECO1a-expressing LC (r = 0.73, p > 0.0001; Pearson’s correlation coefficient for a recording of 20 min duration). Since the LC-NE system causally controls pupil-linked arousal [61], we aligned all signals to local peaks of the derivative of the pupil diameter that occurred in the absence of locomotion, as pupil dynamics and locomotion are tightly coupled in mice [62]. As expected, peaks in pupil size were preceded by peaks in LC activity, which confirms the biological relevance of the recorded signals (Fig 6D). Importantly, similar temporal dynamics of LC activity were observed in both channels, indicating that PRS×8- and Dbh^cre-mediated expression of calcium indicators reveals comparable functional signals of LC activity.

Importantly, when exciting the calcium indicators at their quasi-isosbestic wavelength (405 nm), no peaks were observed in either channel (Fig 6E), confirming that the observed peaks in fluorescence indeed resulted from calcium dynamics, and not from motion or hemodynamic artifacts. Additional controls were performed to exclude spectral crosstalk of the calcium indicators by restricting our recording from both LC either to the green or to the red spectral range. When aligning signals to pupil dilation, the previously observed peak in jRGECO1a fluorescence was not present in the green spectrum (Fig 6F), and, vice versa, the previously observed peak in PRS×8-jGCaMP8m fluorescence was abolished in the red spectral range (Fig 6G), confirming absence of spectral crosstalk between the two indicators.

Finally, we developed a construct encoding the red-light-sensitive cation channel ChrimsonR [63] fused to tdTomato [64] under control of the PRS×8 promoter for optogenetic activation of LC-NE neurons. After bilateral injection of rAAV2/9-PRS×8-ChrimsonR-tdTomato (efficacy: 79.1%, specificity: 55.4%) and implantation of optical fibers above the LC in a wild-type mouse (Fig 6H and 6I), we illuminated the LC (633 nm) while monitoring the pupil (Fig 6J). As expected, illumination of the LC reliably induced pupil dilation, confirming functional optogenetic activation of the LC by PRS×8-ChrimsonR-tdTomato (Fig 6K and 6L). Importantly, pupil dilations occurred in the absence of locomotion (Fig 6L, gray), confirming direct pupil control of the LC.

Ethics statement

All experiments were performed in compliance with German law according to the directive 2010/63/EU of the European Parliament on the protection of animals used for scientific purposes and in accordance with the 3R principles. Procedures were approved by local animal care committees (Behörde für Gesundheit und Verbraucherschutz, Hamburg, Germany: N033/2019; Regierungspräsidium Karlsruhe, Germany: 35-9185.81/G-213/22).

Animals

Experiments were performed on adult (2–6 months of age) C57BL/6J, Th^cre [22], Net^cre [21], and Dbh^cre [19] mice, housed in cages with nesting material with no more than five animals per cage at the animal facility of the Center for Molecular Neurobiology of the University Medical Center Hamburg-Eppendorf or at the Medical Faculty Mannheim of the University of Heidelberg. Animals were kept in a 12/12 h light/dark cycle, at 20–24 °C and 40%–65% humidity, with access to water and food ad libitum.

Molecular biology

The DIO variant for cre dependent expression of eGFP [98] was created by cloning eGFP in antisense direction, flanked by two loxP and lox2272 sites, behind the CAG promoter into an AAV2 backbone. PRS×8-based constructs were created by cloning the respective transgene behind 8 repetitions of a de novo synthetized Phox2a/Phox2b response site (PRS; based on the only PRS×8 construct (#89539) available from AddGene at the time this study started) into an AAV2 backbone. PRS×8 constructs were generated for the expression of the green light emitting reporter protein eGFP [98] (pAAV-PRS×8-eGFP; AddGene #192589), the green light emitting calcium indicator jGCaMP8m [59] (pAAV-PRS×8-jGCaMP8m; AddGene #210400), and the red-light sensitive cation channel ChrimsonR coupled to the red light emitting reporter protein tdTomato [63] (pAAV-PRS×8-ChrimsonR-tdTomato; AddGene #192587). DNA was then packed into recombinant adeno-associated virus serotype 2/9 (rAAV2/9) at the vector core facility of the University Medical Center Hamburg-Eppendorf, Germany, or by VectorBuilder.

Stereotactic injections and fiber implantation

For surgical procedures, mice were either anesthetized with Midazolam/Medetomidine/Fentanyl (MMF; 5/0.5/0.05 mg/kg body weight in NaCl, intraperitoneal), or with Isoflurane (5% for induction, ~1.5% for maintenance). In the case of isoflurane anesthesia, adequate analgesia was achieved with buprenorphine (0.1 mg/kg in NaCl, intraperitoneal). After confirming anesthesia and analgesia by the absence of the paw withdrawal reflex, the head was shaved with a fine trimmer and sterilized with iodine (MundiPharma). Mice were then fixed with an auxiliary ear bar (EB-6, Narishige Scientific) in a stereotactic frame and positioned on a heating blanket to maintain body temperature. Eyes were covered with an eye ointment (Vidisic, Bausch + Lomb) to avoid drying. Subsequently, the scalp was opened (~1 cm incision) and the skull was cleaned with NaCl (0.9%; Braun) and a bone scraper (Fine Science Tools). Bregma and lambda were identified and stereotactically aligned. A small craniotomy was then performed above the LC (AP: −5.4; ML: ± 0.9; DV: −3.6 mm relative to bregma; for later injections, the mediolateral coordinates were changed to ±1.1 mm, see Fig 4) using a dental drill (Foredom), and was constantly kept moist to prevent drying.

The viral suspension was drawn into a borosilicate glass pipette (pulled from 0.2 mm glass capillaries with a PC-100 puller; Narishige) using negative air pressure. 300 nl of titer-matched (8 × 10¹¹ gc/ml) viral suspension containing rAAV2/9 carrying DNA encoding eGFP either in a cre-dependent (DIO) manner under control of the CAG promoter (in the case of Dbh^cre, Net^cre, and Th^cre animals) or unconditional under control of the PRS×8 promoter (in the case of C57BL/6J mice) were bilaterally injected into the LC with an injection speed of ~100 nl/min using a custom-made, air-pressure driven injection system for the analysis of transgene expression in the different model systems. For viral transduction of the VTA, 200 nl of virus suspension were injected at four different injection sites (AP: −3.2; ML: ± 0.3; DV: −3.5 mm and AP: −3.2; ML: ± 0.5; DV: −3.2 mm relative to Bregma). For investigations of promoter effects, 300 nl of rAAV2/9 carrying DNA encoding eGFP in a cre-dependent manner under control of the hSyn promoter were injected in Dbh^cre mice at a titer of (8.2 × 10¹² gc/ml). For investigating the effects of viral serotype and injection volume on viral spread, 300 nl of rAAV2/2 and 300/100/50 nl of rAAV2/9 carrying DNA encoding eGFP under control of the hSyn promoter were injected in wild-type mice in a titer-matched manner at 6.8 × 10¹² gc/ml. Injections of rAAV2/9-Ef1a-DO-DIO-tdTomato-eGFP (2.3 × 10¹³ gc/ml; AddGene plasmid #37120) [45] into the LC were performed in Dbh^cre, Net^cre, and Th^cre animals. For fiber photometry experiments, we injected rAAV2/9-PRS×8-jGCaMP8m (3 × 10¹² gc/ml) and rAAV2/9-hSyn-DIO-jRGeco1a (2 × 10¹³ gc/ml) into the LC of a Dbh^cre mouse. For optogenetic manipulation of the LC, 300 nl of rAAV2/9-PRS×8-ChrimsonR-tdTomato (1 × 10¹² gc/ml) were injected into each LC of a wild-type mouse. After virus injection, the glass micropipette was kept in place for at least 5 min to avoid unnecessary viral spread and was then slowly withdrawn. In the case of histological experiments, the skin was closed with simple interrupted sutures and covered with iodine. In the case of fiber photometry experiments, optical fibers (R-FOC-BL400C-50NA: 400 µm diameter, 0.5NA; RWD) were inserted at AP: −5.4; ML: ± 1.0; DV: −3.5 mm relative to bregma using a stereotactic micromanipulator. For optogenetic experiments, optical fibers (R-FOC-BL400C-37NA: 200 µm diameter, 0.37 NA; RWD) were inserted at the same coordinates. Fiber implants and a metal headpost (00-200 500 2110; Luigs & Neumann) for animal fixation during the experiment were fixed to the roughened skull using cyanoacrylate glue (Pattex; Henkel) and dental cement (Super Bond C&B; Sun Medical). The incised skin was then glued to the dental cement to close the site, before it was covered with iodine. Mice received analgesia (Carprofen; 4 mg/kg body weight, subcutaneous), and Atipamezole/Flumazenil/Buprenorphine (FAB) was administered (2.5/0.5/0.1 mg/kg body weight, intraperitoneal) to antagonize anesthesia in the case of MMF; Isoflurane anesthetized mice were simply removed from the Isoflurane. For recovery, mice were placed in a cage which was placed on a heating pad. Meloxicam was mixed with soft food for the following two days.

Tissue collection

For perfusions, mice were deeply anesthetized with Ketamine/Xylazine (180/24 mg/kg body weight, intraperitoneal), which was confirmed by the absence of the paw withdrawal reflex. Mice were fixed on a silicone pad and transcardially perfused with ~50 ml of phosphate-buffered saline (PBS) to rinse out the blood and subsequently with ~50 ml of 4% paraformaldehyde (PFA; pH 7; Carl Roth GmbH) to fix the tissue. Brains were explanted and post-fixed in 4% PFA for at least 24 h. After fixation brains were embedded in 3% agarose (Genaxxon Bioscience) and coronal slices of ~50 µm thickness were obtained using a vibratome (VT 1000S, Leica). Slices were stored in PBS at 4 °C until they were stained and mounted on microscope slides.

Immunohistochemistry

To block unspecific binding sites for antibodies, brain slices were incubated in 10% normal goat serum (NGS) and 0.3%–0.5% Triton X-100 in PBS for 2 h at room temperature. Slices were then incubated for 2 days at 4 °C in a carrier solution (2% NGS, 0.3%–0.5% Triton X-100, in PBS) containing primary antibodies (see below) to allow for sufficient target protein-primary antibody interaction. Subsequently, slices were rinsed three times for 5–10 min in PBS to remove unbound primary antibodies. Following the washing steps, slices were incubated with secondary antibodies (see below) in the above-mentioned carrier solution over night at 4 °C. Washing steps were repeated as described before in order to remove unbound secondary antibodies. Finally, slices were mounted on microscope slides using Fluoromount mounting medium (Serva, Germany).

To analyze eGFP expression in TH⁺ cells of the LC (Figs 1, 2, 3A–3D, S1, S4, S6, and S8), slices were incubated with primary antibodies against TH (1:1000; AB152; Invitrogen, Thermo Fisher Scientific) derived from rabbit and against green fluorescent protein (1:750; A10262; Invitrogen, Thermo Fisher Scientific) derived from chicken, while Alexa Fluor 546 goat anti-rabbit (1:1000; A11035; Invitrogen, Thermo Fisher Scientific) and Alexa Fluor 488 goat anti-chicken (1:1000; A11039; Invitrogen, Thermo Fisher Scientific) secondary antibodies were used. To analyze the difference between natively expressed eGFP fluorescence and antibody-stained eGFP fluorescence (S7 Fig), slices were incubated with a primary antibody against green fluorescent protein (1:750; A10262; Invitrogen, Thermo Fisher Scientific) derived from chicken and an Alexa Fluor 546 goat anti-chicken secondary antibody (1:1000; A11040; Invitrogen, Thermo Fisher Scientific). In DO-DIO-experiments (Fig 4), slices were incubated with primary antibodies against TH (1:1000; AB152; Invitrogen, Thermo Fisher Scientific) derived from rabbit and against green fluorescent protein (1:750; A10262; Invitrogen, Thermo Fisher Scientific) derived from chicken, while Alexa Fluor 647 goat anti-rabbit (1:1000; A32733; Invitrogen, Thermo Fisher Scientific) and Alexa Fluor 488 goat anti-chicken (1:1000; A11039; Invitrogen, Thermo Fisher Scientific) secondary antibodies were used. No staining was performed against tdTomato. To visualize jGCaMP8m in TH⁺ neurons of the LC (Fig 6B), slices were incubated with primary antibodies against TH (1:1000; AB152; Invitrogen, Thermo Fisher Scientific) derived from rabbit and against green fluorescent protein (1:500; A10262; Invitrogen, Thermo Fisher Scientific) derived from chicken, while Alexa Fluor 647 goat anti-rabbit (1:1000; A32733; Invitrogen, Thermo Fisher Scientific) and Alexa 488 goat anti-chicken (1:1000; A11039; Invitrogen, Thermo Fisher Scientific) were used as secondary antibodies. To visualize jRGeco1a in TH⁺ neurons of the LC (Fig 6B), slices were incubated with primary antibodies against TH (1:500; Cat. No. 213 104; Synaptic Systems) derived from guinea pig and against DsRed (1:500; Cat. No. 632496; TaKaRa Bio) derived from rabbit, while Alexa Fluor 647 goat anti-guinea pig (1:200; A21450; Invitrogen, Thermo Fisher Scientific) and Alexa 546 goat anti-rabbit (1:1000; A11035; Invitrogen, Thermo Fisher Scientific) were used as secondary antibodies. To visualize ChrimsonR-tdTomato in TH⁺ neurons of the LC (Fig 6I), slices were incubated with primary antibodies against TH (1:1000; AB152; Invitrogen, Thermo Fisher Scientific) derived from rabbit and against mCherry (1:500; Cat. No. 632543; TaKaRa Bio) derived from mouse, while Alexa Fluor 647 goat anti-rabbit (1:1000; A32733; Invitrogen, Thermo Fisher Scientific) and Alexa Fluor 488 goat anti-mouse (1:1000; A32723; Invitrogen, Thermo Fisher Scientific) were used as secondary antibodies.

Image acquisition

Confocal images of the LC were acquired in a Z-stack (2 µm step size) using an LSM 900 airyscan 2 microscope (Zeiss, Germany), controlled by the ZEN 3.1 imaging software, with laser intensities and gain of the photomultiplier tubes adjusted to maximize dynamic range while avoiding saturation of images. A Plan-APOCHROMAT 20X/0.8 objective was used, resulting in a field of view of 319.45 × 319.45 µm with a resolution of at least 1,744 × 1,744 pixels (183.2 nm/pixel). Some images were taken at a resolution of 2,044 × 2,044 pixels (i.e., 0.156 nm/pixel), and subsequently down-sampled to match the resolution of 1,744 × 1,744 pixels. Whole slices were either imaged using a Zeiss AxioObserver epifluorescence microscope (Zeiss, Germany) with a Plan-APOCHROMAT 20×/0.8 objective, controlled by the ZEN 3.3 imaging software (ectopic transgene expression: Figs 2 and S8, verification of LC transduction: Fig 4, leakage of viral transduction: S4 Fig) or using a Zeiss Axio Scan Z.1 epifluorescence microscope (Zeiss, Germany) with a Plan-APOCHROMAT 20×/0.8 objective, controlled by the ZEN 3.1 imaging software (viral spread: Figs 3 and S9). ImageJ [99] was used to adjust image brightness, contrast, and create maximum intensity Z-projections.

Image analysis

Transgene expression was detected at the cellular level using the deep-learning-based algorithm CellPose [35]. In the first run, we analyzed all images from all approaches semi-automatically (i.e., automatically segmented cells were manually curated). We then used the obtained cell masks to re-train CellPose [36] in order to adapt it to our dataset. We have uploaded this custom-trained CellPose model, named “CP4LC_Wissing_Eschholz_et_al” as S1 Material. In the second run, we used the re-trained model for automatic cell segmentation without any manual annotation. Segmentation was done separately for each channel, resulting in a total of 2766/3638/2774/2699 TH⁺ cells and 2293/4195/1861/3061 eGFP⁺ cells in Dbh^cre, Net^cre, Th^cre, and PRS×8-animals, respectively (33/39/31/30 brain slices; 11/12/11/10 transduced LC; n = 7/7/7/7 mice). Please note that in our study transgene expression of calcium indicators and optogenetic tools were harder to quantify in an automated way, as transgene expression was not as bright and hence the delineation of individual cells not as clear as compared to the expression of eGFP. The average number of TH⁺ cells that were found per slice was 83.8 ± 19.3, 93.3 ± 15.7, 89.5 ± 16.0, and 90.0 ± 15.3 cells (mean ± standard deviation) for Dbh^cre, Net^cre, Th^cre, and PRS×8, respectively. Cell detection was cross-validated with quantification by two different experimenters, who counted stained cells with the cell counter function of ImageJ. Cell masks of segmented cells in the eGFP and TH channels exported from CellPose as.txt-files were then checked for overlap using custom-written MATLAB scripts (The MathWorks; see data availability statement). For quantification of transduction efficacy, co-expression was checked based on the cell masks in the TH channel, and masks which had an overlap of ≥50% with the masks in the GFP channel were defined as co-expressing. Similarly, co-expression was quantified based on the cell masks in the GFP channel, in order to quantify the specificity of transduction. The number of TH⁺, GFP⁺, and TH-GFP-co-expressing cells was subsequently summed up per animal, in order to avoid biasing the results based on (i) the different amount of TH⁺ cells on individual slices and (ii) the total amount of slices taken from each animal (some slices had to be excluded due to tissue damage during the mounting process). Quantification of the effect of anti-GFP immunostaining on cell fluorescence was performed by fitting (least-squares) a binormal model of the form f_R = ɸ(bias + gain × ɸ⁻¹ (f_G)) + ε with ε ~ N(0,σ²); and where f_R and f_G denotes the fluorescence in the red (GFP-stained) and green (native eGFP) channels respectively, and ɸ corresponds to the normal cumulative distribution function. Comparisons between the two channels have been performed based on the cell masks detected in the GFP-stained channel, as we aimed to reveal additional cells upon GFP staining, which, by definition, are not detected in the native eGFP channel. To quantify viral spread, images were down-sampled to 25% of their original size using bicubic interpolation (MATLABs built-in function ‘imresize’) and smoothed using a 2D gaussian filter with a kernel width of 50 µm (‘imgaussfilt’) before fluorescence was normalized from 0 to 1. Afterward, viral spread was defined as cumulative areas of the image in which fluorescence exceeded a certain threshold, while thresholds were varied between 0 and 1 at an increment of 0.1.

Behavioral screening

All behavioral experiments were performed in a quiet, darkened room by the same (female) experimenter, blind to the animal’s genotype. Animals (genotype-, age- and sex-matched littermates, max. 2 animals of the same genotype and sex per litter) between 12 and 15 weeks of age at the start of the experiment were handled for 5 min (manually picked up and let to freely explore the experimenter’s hands and arms) every other day in order to habituate mice to the experimenter. All experiments were video-recorded with a regular (when experiments were performed under illumination) or an infrared camera (when experiments were performed in darkness). All reported measures were performed using the software Ethovision XT17 (Noldus, the Netherlands), unless stated otherwise. Open Field Test. An open field arena of 50 × 50 × 40 cm, homogeneously illuminated with white light of 25 lux, was used. At the start of the experiment, mice were placed in one corner of the box, and their movement was tracked for the following 20 min. The time spent in the center (the central 20 × 20 cm) and borders (5 cm from the walls) of the arena, the number of center crossings, and the latency to enter the arena center for the first time served as proxies for anxiety-like behavior, while the total travel distance served as a proxy for spontaneous locomotion. Elevated Plus Maze. The EPM consisted of four arms (30 × 5 cm) radiating from a central platform (5 × 5 cm), elevated 75 cm from the ground. Two opposing arms are covered by opaque walls (height: 17 cm), while the two other arms remain open. Mice were placed on the central platform and left to explore the maze for 5 min in total darkness. The time spent on opened versus closed arms, the number of entries to open arms, the fraction of open arm entries upon which mice went to the distant edge of the exposed arm, and the latency to leave the center of the maze served as proxies for anxiety-like behavior. In addition, the number of rearing events, the latency to the first rearing event, and the percentage of time spent grooming served as indirect measure for the absence of anxiety. EPM data acquisition and analysis were performed using Observer XT11 (Noldus, the Netherlands). Y-maze. The maze consisted of three symmetrically radiating arms (34 × 5 cm), surrounded by walls (height: 30 cm) made of transparent plastic, with spatial cues placed behind each wall, homogeneously illuminated with white light of 10 lux. Mice were placed in the center of the maze and left to explore the maze until 26 transitions (entry to one arm with all four paws) between neighboring arms were made, or for a maximum of 15 min. Entering an arm after having visited the two other arms before, that is entering all three arms of the maze in one sequence, was considered a complete alternation. The fraction of alternations over all transitions (a proxy for working memory) as well as the time per transition (a proxy for exploratory behavior) was manually extracted. Morris Water Maze. The water maze consisted of a circular pool (diameter: 145 cm) filled with opaque water (19–21 °C) with non-toxic white paint, surrounded by black curtains, and homogeneously illuminated with white light of 60 lux. A platform (diameter: 14 cm) was placed 1 cm below the water surface in one of the four quadrants of the maze. Four landmarks (35 × 35 cm) of different shapes and gray tones were placed on the walls of the maze to allow for orientation. Animals were familiarized with swimming, platform searching, and handling in a pre-training session using a small, rectangular water tank. Mice had to search for a platform in three trials (10–20 min inter-trial intervals), with a different platform location each time. Animals were transferred back to their home cage, where heat lamps were provided. On training days 1 and 2, animals underwent four training trials each (max. 90s), with varying starting positions. Once they climbed the platform, they were left to wait there for 10 s before being transferred back to the home cage. Animals that did not find the platform within 90 s were guided to the platform until they climbed it and left to wait on the platform for 10 s, before being transferred back to the home cage. On day 2, after the second training trial, and on day 3, a probe trial (60 s duration) without the platform was conducted. The time spent in the area where the platform was previously located, the time spent in the quadrant that previously contained the platform, the latency to the platform area, and the average minimum distance to the platform area were used to approximate successful spatial learning and memory. The total travel distance served as a measure for locomotor behavior, and the average distance to the wall was analyzed to compare thigmotactic behavior as well as searching strategies (i.e., circling around the pool).

Fiber photometry recordings

Approximately 3 weeks after surgery, animals were habituated to the experimenter, set up, and head fixation on a linear treadmill (200–100 500 2100 and 700–100 100 0010; Luigs & Neumann). Fiber photometry recordings were performed using two pyPhotometry systems [100] (Open Ephys) controlling multicolor LED light sources (pE-4000; CoolLED) connected to the input ports of dichroic cubes (FMC5_E1(400–480)_F1(500–540)_E2(555–570)_F2(580–680)_S; Doric Lenses) as reported previously [13]. In brief, excitation light for quasi-isosbestic (405/10 nm) and calcium-dependent (470/10 nm) activation of jGCaMP8m was delivered in a temporally interleaved manner into the 400–480 nm input port of the dichroic cube, while excitation light for jRGECO1a (555–570 nm) was delivered via the 555–570 nm port. We also included excitation at 405/10 nm as a control channel when using jRGECO1a, as the indicator has been shown to be largely insensitive to calcium when excited at this wavelength [101], but we note that the efficiency of jRGECO1a excitation at 405 nm is very low (which, to a milder degree, is also true for jGCaMP8m). Hence, the recorded signal, which besides indicator fluorescence also contains autofluorescence originating from fiber optics and brain tissue [102], is likely composed differently and should not be over-interpreted. Emission light between 500–540 nm (jGCaMP8m) and 580–680 nm (jRGECO1a) was measured using photodetectors (DFD_FOA_FC; Doric Lenses), and digitized by the pyPhotometry-board at 130 Hz. Mice were connected to the dichroic cubes with low-autofluorescence patch cords (MFP_400/430/3000-0.57_1m_FC-ZF1.25(F)_LAF: 400 µm diameter, 0.57 NA; Doric lenses) via zirconia mating sleeves (SLEEVE_ZR_1.25; Doric Lenses). Simultaneously, the left eye of the mouse was monitored using a monochrome camera equipped with a macro-objective (DMK 22UX249 and TMN 1.0/50; The Imaging Source) and a 780 nm long-pass filter (FGL780; Thorlabs) under infrared illumination. Single video frames were triggered at 30 Hz with a voltage signal provided by an NI-DAQ-card (PCIe-6323; National Instruments), which was in addition fed to the pyPhotometry system for data synchronization.

Fiber photometry analysis

Single-sample outliers were removed from raw traces by applying a moving median filter with a window size of three sampling points, before traces of quasi-isosbestic and ligand-dependent fluorescence were detrended by fitting a double-exponential function to the respective trace and dividing the trace by the fitted curve, (accounting for bleaching components resulting both from the indicator and from the fiber optics [103]). Traces were then band-pass filtered (1/60–30 Hz; third order Butterworth filter) using MATLAB’s ‘butter’, and ‘filtfilt’ functions in order to remove signal components resulting from remaining bleaching and high frequency noise. The difference between the corrected functional and quasi-isosbestic traces was then taken as the functional signal, ΔF. Pupil edges (horizontal, vertical, and both diagonals) were extracted from each frame using DeepLabCut [104], and the pupil diameter was defined as the average distance of opposing edges in each frame. The pupil diameter was then smoothed with a window size of 250 ms (using the MATLAB functions ‘resample’ and ‘smooth’). Animal position was recorded using the treadmill-software provided by Luigs and Neumann, smoothed by using a moving median filter with a window size of 500 ms, and speed was calculated as the derivative of animal position. For the experiments reported in this manuscript, the treadmill was largely immobilized to minimize locomotion-related pupil dilation. Yet, attempts of the animals to start locomotion could be detected, and trials in which treadmill movement exceeded 5 mm/s were excluded from further analysis. All data was re-sampled to 100 Hz for further processing. To detect periods of pupil dilation, peaks were detected in the derivative of pupil diameter using using MATLAB’s ‘findpeaks’ function, with a minimum peak prominence of the mean plus three standard deviations of the whole trace, and a minimum peak distance of 3 seconds. Photometry, pupil, and treadmill recordings were then cropped and aligned in a time window starting 2 seconds before and ending 5 seconds after each peak.

Optogenetic manipulation of LC activity

General procedures for optogenetic experiments were identical to fiber photometry experiments, except for the delivery of optogenetic stimuli. Stimuli were generated using custom-written MATLAB scripts actuating on an NI-DAQ-card (USB-6001; National Instruments), controlling a 633 nm diode laser (633-100; Omicron Laserage) housed in a laser combiner system (LightHUB; Omicron Laserage). The laser combiner was coupled to the optical fiber implants using a 1 × 2 step-index multimode fiber optic coupler (200 µm core diameter, 0.37 NA; SBP(2)_200/220/900-0.37_2m_FCM-2xZF1.25; Doric Lenses) via zirconia mating sleeves (SLEEVE_ZR_1.25; Doric Lenses) wrapped with black shrinking tubes to avoid light emission from the coupling interface. To activate ChrimsonR, pulse trains (633 nm, ~5 mW at each fiber end, 20 ms pulse duration, 20 Hz repetition rate) of 4 s duration were used. Thirty stimuli were presented at an inter-stimulus-interval of 30 s. Pupil diameter was extracted from videographic images as for fiber photometry experiments, cropped, and aligned to optogenetic stimulation in a time window starting 5 s before and ending 20 s after each stimulus onset. Pupil diameter was then normalized to the median pupil diameter during the second before stimulus onset.

Statistics

For histological data, one-way ANOVA was performed; when significant, post-hoc comparisons between groups were performed using Tukey’s honestly significant difference criterion which controls for inflated risk of type-1 error (false positives) following multiple comparisons (using MATLAB’s ‘anova1’ and ‘multcompare’ functions). For behavioral data, two-way ANOVA was performed to identify differences between genotypes, sexes, and interactions thereof, using the software Graphpad Prism. All statistical tests were two-tailed, with a type-1 error risk set to ɑ = 0.05. In all figures, error bars correspond to standard error of the mean, unless specified otherwise.

S2 Table. Statistical analysis of behavioral screening.

The table indicates results of a two-way ANOVA analyzing effects of genotype (GT), sex, or interactions thereof (×) in the mouse lines included in this study on different aspects of behavior in the in the open field test (OF), elevated plus maze (EPM), Y-maze (YM), and Morris water maze (MWM). All groups included eight male and eight female mice of each genotype. p-values below 0.05 are indicated in bold font. Note that multiple comparisons were not corrected in this table.

https://doi.org/10.1371/journal.pbio.3003228.s004

(DOCX)

S2 Fig. Decoupling of TH and eGFP expression in the LC region of Th^cre mice.

Normalized fluorescence in the green channel of neurons co-expressing TH and eGFP (i.e., true positives; left) and neurons expressing eGFP only (i.e., false positives; right) in Dbh^cre (red), Net^cre (magenta), Th^cre (orange) and PRS×8 mice (blue). While normalized fluorescence of false positive neurons was lower as compared to true positive neurons in Dbh^cre, Net^cre and PRS×8 mice, normalized fluorescence of false positive neurons exceeded true positive neurons in Th^cre mice. ****: p < 0.0001 (Bonferroni-corrected Wilcoxon rank sum test). Numerical data underlying this figure can be found in S1 Data.

https://doi.org/10.1371/journal.pbio.3003228.s006

(TIF)

S3 Fig. Independence of histology quantification on the overlap threshold of cell masks.

(a) Efficacy and specificity from the four different model systems as a function of the threshold overlap parameter (light to dark colors; ranging from 5% to 95%, in steps of 5%) which defines the minimal overlap between GFP⁺ and TH⁺ channels for a TH⁺ cell to be also labeled as GFP⁺. (b) Transduction efficiency (summarizing sensitivity and specificity in a single metric) from the four different model systems quantitatively varied as a function of the threshold overlap parameter. Importantly, conclusions were robust to this parameter choice as shown by the non-crossing of the curves. In the main text, we report values for a neutral threshold overlap parameter of 50%. Shading areas denote the standard error of the mean (n = 7).

https://doi.org/10.1371/journal.pbio.3003228.s007

(TIF)

S4 Fig. Leakage of cre-dependent virus in wild-type mice.

Immunohistochemical staining against GFP (green) and TH (magenta) in two wild-type mice injected with rAAV2/9-DIO-eGFP. eGFP expression was only observed in very few cells both within and outside of the LC.

https://doi.org/10.1371/journal.pbio.3003228.s008

(TIF)

S5 Fig. Agreement between CellPose-based and human-based cell segmentation.

(a) Correlation of cell counts; (b) Efficacy and specificity; (c) transduction efficiency between cells identified by CellPose vs. labeled by each of the two human observers (AD and CW; first 2 columns); and between the two human observers (third column). A total of five slices of five different mice were used for each of the four model systems. Dashed lines indicate identity. In each case, two measures of agreement are reported: Pearson’s correlation coefficient, denoted as r; and the distance between observers, denoted as d = √∑_I (x_i − y_i)² where x and y correspond to observers’ reports (a) or derived quantities (b, c), and i to the brain slice. Error bars denote the standard error of the mean (with n = 5).

https://doi.org/10.1371/journal.pbio.3003228.s009

(TIF)

S6 Fig. Specificity of cre-dependent transgene expression in the VTA of Th^cre mice.

(a) Injection scheme: CAG-DIO-eGFP was bilaterally injected into the VTA of Th^cre mice. (b) Example of a coronal brain slice stained against TH (magenta) to visualize dopaminergic neurons of the VTA and eGFP (green) to enhance the native fluorescence of eGFP expression. (c) Exemplary confocal images of VTA-DA neurons in Th^cre mice (top), along with corresponding cell masks as segmented by CellPose (bottom). Overlaid cell masks of both channels are shown at the bottom right, where each TH⁺ cell was labeled as GFP⁺ (successful transduction, black) when cell masks overlapped 50% or more the size of the TH⁺ cell, or GFP⁻ otherwise (missed transduction, magenta). GFP⁺ and TH⁻ (erroneous transduction, green) was counted as unspecific transduction. (d) The average specificity of GFP expression in the VTA of Th^cre mice amounted to 61.7 ± 6.8% (derived from a total of 2,469 TH⁺ and 1,627 eGFP⁺ neurons obtained from 25 images of n = 4 mice). These findings are in line with the specificity of 59 ± 1% (dashed line) reported by Lammel and colleagues [39] (based on visual inspection). Numerical data underlying panel (d) can be found in S1 Data.

https://doi.org/10.1371/journal.pbio.3003228.s010

(TIF)

S7 Fig. Comparison of native eGFP expression with anti-GFP staining.

(a) Example images of brain slices showing native eGFP fluorescence (green, top) and anti-GFP immunofluorescence with a secondary antibody in the red spectrum (magenta, center) in Dbh^cre, Net^cre, Th^cre, and PRS×8 mice (from left to right). (b) Number of neurons detected by CellPose in the native eGFP (green, x-axis) and GFP-stained (red, y-axis) channels across animals (n = 3/4/4/4 Dbh^cre/Net^cre/Th^cre/PRS×8 mice, respectively). 21.5 ± 15.3% more cells were detected in the GFP-stained channel as compared to the native eGFP channel (absolute difference: 20.5 ± 16.5 cells; t test against 0: t₁₄ = 4.8, p = 2.8 × 10⁻⁴). (c) Distribution of normalized fluorescence in anti-GFP stained channel as a function of the normalized fluorescence in the native eGFP channel across cells. Fluorescence was averaged within each cell mask (detected in the GFP-stained channel), then binned into 50 × 50 equally spaced bins, and further smoothed by a 2D 10 × 10 boxcar function. The averaged normalized brightness of immuno-stained cells per animal (0.40 ± 0.06) exceeded the averaged normalized brightness of the native eGFP-expressing cells (0.26 ± 0.06) by 55 ± 35% (absolute difference: 0.13 ± 0.076; t test against 0: t₁₄ = 6.66, p = 1.08 × 10⁻⁵). (d) To quantify the difference in fluorescence between the anti-GFP immunostaining and native eGFP fluorescence in more detail, we fitted a non-linear function describing the fluorescence of GFP-stained cells as a function of native eGFP fluorescence (after within-channel normalization, see panel c). The function is of the form f′_R = ɸ(bias + gain × ɸ⁻¹(f_G)), where f_R and f_G denotes the fluorescence in the red (GFP-stained) and green (native eGFP) channels, respectively, and ɸ corresponds to the normal cumulative distribution function. The ‘bias’ parameter quantifies the overall change in fluorescence induced by the immunostaining (top panel), while the ‘gain’ parameter accounts for a possible asymmetry in the fluorescence increase depending on the native eGFP fluorescence (bottom panel). (e) Estimated parameter values from the function presented in (d). The ‘bias’ parameter was significantly larger than 0, indicating an increase in fluorescence following GFP-staining (0.14 ± 0.19; t test against 0: t₁₄ = 2.85, p = 0.013). The ‘gain’ parameter was significantly smaller than 1, indicating that the increase in fluorescence was substantially stronger for the dimmest cells in the native eGFP channel (0.50 ± 0.17; t test against 1: t₁₂ = –11.56, p = 1.5 × 10⁻⁸). In (b), (c), and (f), error bars denote the standard error of the mean (with n = 15), while statistical significance is denoted by */**/*** for p < 0.05/0.01/0.001, respectively. Numerical data underlying panel e can be found in S1 Data.

https://doi.org/10.1371/journal.pbio.3003228.s011

(TIF)

S8 Fig. Ectopic transgene expression across model systems.

Example brain slices showing ectopic expression upon Dbh^cre (a), Net^cre (b), Th^cre (c–e), and PRS×8 (f) mediated transgene expression. Immuno-straining against TH and GFP is displayed in magenta and green respectively. Insets show inverted gray scale images for a better contrast.

https://doi.org/10.1371/journal.pbio.3003228.s012

(TIF)

S9 Fig. Effects of serotype and injection volume on viral spread.

Coronal brain slices of three wild-type mice that were bilaterally injected with 300 nl of rAAV2/2-hSyn-eGFP (top row), 300 nl of rAAV2/9-hSyn-eGFP (second row from top), 100 nl of rAAV2/9-hSyn-eGFP (third row from top), or 50 nl of rAAV2/9-hSyn-eGFP (bottom row). All slices were stained against tyrosine hydroxylase to visualize LC-NE neurons (magenta) and against eGFP to enhance the native fluorescence of transgene expression (green).

https://doi.org/10.1371/journal.pbio.3003228.s013

(TIF)

S10 Fig. Supplementary readouts obtained in the open field test.

No effect of genotype could be detected on anxiety-like behavior as approximated by (a) the number of center crossings, (b) the latency to the first center crossing, (c) the mean minimum distance to the wall, (d) the percentage of time spent in the border of the arena, nor to general locomotion behavior as approximated by (e) the total distance traveled by the mouse. All data is depicted as mean ± standard error of the mean. n.s. = not significant. n = 8 females (circles) and 8 males (triangles). Numerical data underlying this figure can be found in S1 Data.

https://doi.org/10.1371/journal.pbio.3003228.s014

(TIF)

S11 Fig. Supplementary readouts obtained in the elevated plus maze.

No effect of genotype could be detected on anxiety-like behavior as approximated by (a) the number of open arm entries, (b) the percentage of open arm entries on which the animal went to the distant edge of the arm, (c) the latency to leave the central platform, (d) the number of rearing events, (e) the latency to the first rearing event, or (f) the percentage of time spent grooming. All data is depicted as mean ± standard error of the mean. n.s. = not significant. n = 8 females (circles) and 8 males (triangles). Numerical data underlying this figure can be found in S1 Data.

https://doi.org/10.1371/journal.pbio.3003228.s015

(TIF)

S12 Fig. Supplementary readouts obtained in the Y-maze.

No effect of genotype could be detected on locomotion or explorative behavior as approximated by (a) the average time per transition between Y-maze arms. All data is depicted as mean ± standard error of the mean. n.s. = not significant. n = 8 females (circles) and 8 males (triangles). Numerical data underlying this figure can be found in S1 Data.

https://doi.org/10.1371/journal.pbio.3003228.s016

(TIF)

S13 Fig. Morris water maze during training sessions.

In each group (n = 16 mice/group), both the latency to reach the platform (a) and the distance until the platform was reached (b) decreased as a function of training session number (four sessions were performed on each day for two subsequent days), as indicated by both Pearson’s and Spearman’s correlation, indicating successful spatial learning in the water maze. Boxplots depict 0/25/50/75/100 percentiles, respectively. Dashed lines indicate the linear fit according to the method of least squares. Numerical data underlying this figure can be found in S1 Data.

https://doi.org/10.1371/journal.pbio.3003228.s017

(TIF)

S14 Fig. Supplementary readouts obtained in the Morris water maze.

No effect of genotype could be detected on spatial memory as approximated by (a) the time spent in the platform area, (b) the time needed to reach the platform, and (c) the mean minimal distance to the platform. No effect of genotype could be detected on locomotion behavior as approximated by (d) the total distance traveled. No effect of genotype could be detected on (e) the average distance to the wall, suggesting comparable thigmotactic behavior of mice. All data is depicted as mean ± standard error of the mean. n.s. = not significant. n = 8 females (circles) and 8 males (triangles). */**/***, depicted in panel a, for p < 0.05/0.01/0.001, respectively, tested against chance level. Numerical data underlying this figure can be found in S1 Data.

https://doi.org/10.1371/journal.pbio.3003228.s018

(TIF)