Abstract
The habenula integrates sensory stimuli and reward information to regulate the release of neuromodulators with broad effects on brain state and behavior. One stimulus that affects habenula activity is light, but how it does so is unknown. Here, we address this question using larval zebrafish. Calcium imaging shows that light evokes widespread activity in habenula neurons, coupled with a prominent early response in the dorsal left neuropil. Injection of a lipophilic dye into this region retrogradely labels a retino-recipient thalamic nucleus. Anterograde tracing of the thalamus demonstrates a projection to the habenula, while optogenetic and lesion experiments confirm functional connectivity. An analysis of the mouse mesoscale connectome indicates that a visual nucleus in the thalamus, the ventral lateral geniculate nucleus, projects to the habenula in this species also. Together, these data suggest the existence of a conserved thalamo-habenula projection that enables light to affect habenula activity in vertebrates.
Introduction
The habenula is an evolutionarily conserved structure (Stephenson-Jones et al., 2012) that influences multiple behaviors, ranging from fear (Agetsuma et al., 2010; Lee et al., 2010; Zhang et al., 2016), to learning (Matsumoto and Hikosaka, 2007; 2009; Amo et al., 2014), addiction (Fowler et al., 2011), sleep (Aizawa et al., 2013), aggression (Chou et al., 2016; Golden et al., 2016) and performance under stress (Thornton and Davies, 1991). It acts by regulating the release of broadly-acting neuromodulators such as serotonin, dopamine, epinephrine and histamine (Wang and Aghajanian, 1977; Morley et al., 1985; Jhou et al., 2009; Quina et al., 2014). To precisely control these neuromodulators, such that behavior is appropriate for a given context, the habenula integrates diverse information including internal state, sensory stimuli and reward value.
These different types of information reach the habenula via different pathways. Internal states such as the circadian clock may be conveyed from the hypothalamus, for example by hypocretin-secreting neurons (Appelbaum et al., 2009). Negative reward causes the entopeduncular nucleus, or internal segment of the globus pallidus, to send excitatory input to the habenula (Hong and Hikosaka, 2008). Sensory stimuli such as odours evoke activity in the zebrafish habenula (Dreosti et al., 2014; Krishnan et al., 2014) via a direct pathway from the olfactory bulb (Miyasaka et al., 2009). Another stimulus that causes activity in the habenula is light, as has been demonstrated in rat (Zhao and Rusak, 2005), pigeon (Semm and Demaine, 1984) and zebrafish (Dreosti et al., 2014). How light affects the habenula is not known, implying the existence of an input pathway that is not yet well defined.
The habenula is divided into two major regions based on pattern of connectivity. In mammals, these are called the medial and lateral habenula, while in fish these are the dorsal and ventral habenula. In larval zebrafish, short pulses of red light cause asymmetric depolarization of the dorsal habenula, with more cells in the left side showing response; the response in the ventral habenula is symmetric (Dreosti et al., 2014). This response is dependent on the eyes (Dreosti et al., 2014), but no direct pathway from the retina to the habenula has been documented (Burrill and Easter, 1994; Robles et al., 2014). By retrograde tracing in adult zebrafish, Turner et al (Turner et al., 2016) proposed that the habenula receives input from the nucleus rostrolateralis, a thalamic nucleus with retinal input that is found in fish (Butler and Saidel, 2003; Saidel, 2013). However, no connectivity with the retina was shown, and because the thalamic nucleus does not asymmetrically innervate the left habenula and potential artifacts in labeling, the source of light-evoked activity in the habenula could not be determined (Turner et al., 2016). Here, using a combination of imaging, tracing and manipulation, we demonstrate that light-evoked activity in the habenula of zebrafish larva is mediated by a thalamic nucleus with retinal input, i.e. the putative nucleus rostrolateralis. We then ask whether a similar pathway could exist in mammals.
Results
The habenula displays a broad and dynamic response to irradiance change
The zebrafish habenula consists of neurons surrounding neuropils that are innervated by afferent neurons (Hendricks and Jesuthasan, 2007; Miyasaka et al., 2009; Amo et al., 2014; Turner et al., 2016). To gain insight into the neural circuits that enable light to influence the habenula, we first characterized habenula activity evoked by pulses of light. Two-photon imaging was performed on a transgenic zebrafish line expressing the calcium indicator GCaMP3 throughout the habenula (Krishnan et al., 2014) (Figure 1A). Resonant-scanning, combined with piezo-driven focusing, was used to record the activity of cells at multiple focal planes (Figure 1B, C). With a step size of 10 μm, so that each cell would be sampled only once, most of habenula could be covered with 5 planes at a rate of 1 Hz. Habenula activity was monitored as the larva was exposed to 20-second pulses of blue light. We used relatively long pulses, rather than brief flashes, to allow responses to transition as well as steady state to be identified. Pixel-wise analysis in one 7-day old fish indicates that evoked activity – both transient and sustained - occurred throughout the habenula in response to increase and decrease in irradiance and to light and darkness (Figure 1D, E). The spatio-temporal pattern of activity was reproducible across several cycles, as shown by the trajectory of the system through state space (Figure 1F).
To assess if these responses were reproducible across multiple fish, we imaged the habenula in 6 fish. Habenula neurons were segmented (Figure 2A-C; total of 4986 cells, with an average of 831 ± 53 cells (95% CI) per fish) and their activity was clustered by k-means. Cluster centers were classified by response type. Transient and sustained responses to increase and decrease in irradiance could be reliably evoked (Figure 2D-G). The mean percentage of responding cells per fish (±95% CI) were ON: 30.98 ± 9.94%; OFF: 19.03 ± 3.93%; Inhibitory: 7.93 ± 3.50%. Correlating the cells corresponding to the different response types revealed that, in general, neurons that were excited by an increase of irradiance did not fire to a decrease (Figure 2H, I). Some neurons that were inhibited by light did, however, show a response at light offset (Figure 2F, blue trace; see also Figure 1E, red trace). These observations confirm that the activity of zebrafish habenula neurons is affected by change in irradiance, and that in addition to excitation there is inhibition by light, as well as excitation to loss of light.
The s1011tGAL4 line drives GCaMP3 expression throughout the habenula, and the response in all focal planes imaged suggests that all domains of the habenula respond to change in light. To further test this, we imaged fish where specific domains could be identified. Using the narp promoter (Agetsuma et al., 2010) (Figure 3A-C), responses to light ON and OFF could be detected in the lateral subdomain of the dorsal habenula (dHbL) (Figure 3 D, G, H; n = 5 fish). As we lacked a driver that is specific for the medial subdomain of the dorsal habenula (dHbM), we assessed activity here by analyzing narp-negative regions of the dorsal habenula in narp:GAL4, UAS:DsRed, elavl3:GCaMP6f fish (Figure 3E, G, I; n = 5 fish). Additionally, we analyzed the interpeduncular nucleus (IPN), which receives input from all regions of the dorsal habenula (deCarvalho et al., 2014). Evoked activity was seen in all domains of the interpeduncular nucleus (Figure 3K, L), consistent with all domains of the dorsal habenula showing a response to change in irradiance. To assess the ventral habenula, we imaged a line with a calcium indicator under the control of the dao promoter (Amo et al., 2014). Again, evoked activity was seen (Figure 3F, G, J; n = 8 fish). These results indicate that all domains of the habenula in larval zebrafish respond to change in illumination.
To further characterize light evoked activity, we used high speed widefield microscopy. With 200 Hz imaging, a strong and rapid increase in fluorescence was detected in a discrete region in the dorsal left habenula (Figure 4A-F; n = 6 fish), as well as bilaterally in the thalamic region, soon after the onset of visible light. Two-photon microscopy at 13 Hz suggests that the rapid increase in the habenula occurs in the dorsal left neuropil (Figure 4G-I; n = 4 fish). The increase suggests that there is asymmetric activation of the dorsal habenula by light, potentially caused by a stronger input to the dorsal left neuropil. The bilateral activation in the thalamic region argues against the habenula asymmetry being an artifact of asymmetric illumination or reporter expression.
Taken together, these results suggest that the habenula afferents providing information about illumination should have the following properties. Firstly, they should terminate broadly within the habenula. Secondly, there should be stronger activity at light onset in terminals innervating the dorsal left neuropil. Thirdly, they should respond to increase or decrease in irradiance. Fourthly, they should cause excitation and inhibition, and may thus include excitatory and inhibitory neurons
The thalamus provides input to the habenula
We next searched for afferents that could provide such properties in larval zebrafish. We focused initially on inputs that could account for the asymmetry in the dorsal habenula. The entopeduncular nucleus (EN) is the major source of habenula afferents in teleosts (Yañez and Anadon, 1994), including zebrafish (Amo et al., 2014; Turner et al., 2016). This nucleus is labeled in the Et(sqKR11) line (Lee et al., 2010), providing a simple way of visualizing EN afferents to the habenula. Some labeled fibers were detected in the dorsal left neuropil, indicating that the EN does provide some innervation to this region of the habenula (Figure 5A). However, labeling in this neuropil was relatively sparse, compared with other neuropils. We therefore hypothesized that there may be other inputs, from an anatomically distinct population, that targets this neuropil. To test this, the lipophilic tracer DiD was injected into the dorsal left neuropil (n = 6 fish). In all cases, neurons in the dorsal left habenula (which extend dendrites into the neuropil), the parapineal, and a thalamic nucleus located ventrally to both habenula (Figure 5B-D) were labeled.
DiD label was not detected in any other regions of the brain, and only rarely in the entopeduncular nucleus, suggesting that the thalamus is the major source of input to the dorsal left neuropil. This is consistent with the early activity seen in the thalamus (Figure 4A-F). The label in the thalamus cannot represent anterograde label from the habenula, as tracing of projections from the habenula by expressing fluorescent proteins specifically in the habenula does not result in a projection to the thalamus (Movie 1). Moreover, the labeling of cell bodies in the thalamus (Figure 5D inset, E) indicates that this is likely to be a retrograde label. The neuropil of this thalamic nucleus contains terminals of retinal ganglion cells, as shown by DiI injection into the retina (Figure 5E). Thus, the habenula neuropil with a strong response to light is innervated by thalamic nuclei that receive retinal input. This is likely to be the nucleus rostrolateralis.
Expressing a fluorescent protein in thalamic neurons, using the s1020tGAL4 driver (Figure 5F, G), led to label of terminals throughout the habenula (Movie 2). Thus, although the dorsal left neuropil receives input from the thalamus, the thalamus projection is not restricted to this region of the habenula.
The habenula may receive glutamatergic and GABAergic input from the thalamus
As noted above, light caused both increase and decrease in activity of habenula neurons, implying that there may be excitatory and inhibitory afferents. Using an antibody to vGlut1/2, glutamatergic pre-synapses were detected in all neuropils of the habenula (Figure 5H), indicating the existence of excitatory afferents. GAD65/67 labeled puncta could also be detected in habenula neuropils (Figure 5I). In the lateral regions, corresponding to the ventral habenula, labeled streaks were detected adjacent to cell bodies. These labels were not located within habenula neurons, as they did not co-localize with cytoplasmic label provided by GCaMP3, nor did they fill the cytoplasm, implying that these puncta and streaks must reside in habenula afferents (i.e. axon terminals) such as those labeled in Movie 2. Labeled cell bodies were seen below the level of the habenula (see Movie 3). Consistent with this, GABAergic neurons could be detected in the dorsal thalamus using the transgenic line Tg(gad1b:RFP, vGlut2a:GAL4, UAS:eGFP) (Satou et al., 2013) (Figure 5J, K; Movie 4). No label was seen in the entopeduncular nucleus, which has previously been shown to be glutamatergic (Amo et al., 2014; Turner et al., 2016). These observations confirm that the thalamus contains both glutamatergic and GABAergic neurons, as described previously (Mueller, 2012), which may mediate light-evoked excitation and inhibition of habenula neurons.
Optogenetic stimulation of the thalamus drives habenula activity
The lipophilic and transgenic tracing experiments documented in Figure 5 demonstrate anatomical connectivity from the thalamus to the habenula. To test functional connectivity, we used optogenetics. Channelrhodopsin-2 (ChR2) was expressed in thalamic neurons with the s1020tGAL4 driver (Fig. 6A). Experiments were carried out on fish lacking eyes, to prevent a response to visual stimulation. Short pulses of blue light reproducibly caused an increase in fluorescence of GCaMP6f in habenula neurons of fish with expression of ChR2 in the thalamus (Figure 6B, D, G). Some response was seen in fish without ChR2 expression (Fig. 6F), suggesting that some habenula response may be due to non-ocular sensors such as deep brain photoreceptors (Matos-Cruz et al., 2011; Fernandes et al., 2013). The larger response in fish with ChR2 expression, however, suggests that there is functional connectivity between the thalamus and the habenula.
Irradiance change evokes activity in thalamus neurons
If the thalamus provides afferents mediating illumination-dependent activity in the habenula, thalamic neurons should respond to increase and decrease of illumination. To test this, calcium imaging was carried out in s1020tGAL4, UAS:GCaMP6s transgenic fish. A response to increase and decrease in illumination was detected in cell bodies in the anterior thalamus (Figure 7A-F) in all fish imaged (n = 5). Dendrites in the neuropil of the thalamus also responded to change in illumination. Increase in irradiance caused activity more dorsally, while decrease caused activity more ventrally (Figure 7G-J). Thus, the thalamus has a response to both increase and decrease of illumination.
Habenula response to light has been shown to depend on the eyes (Dreosti et al., 2014). Thus, if the thalamus mediates habenula response to light, light-evoked activity here should depend on the eyes. Indeed, the robust responses to light were lost in fish lacking eyes (Figure 7K-S), consistent with this hypothesis.
Functional asymmetry in light-evoked activity
As shown in Figure 4G-I, light evokes strong activity in the dorsal neuropil of the left habenula. As the line used in this imaging experiment contains only labeled habenula neurons, the response probabaly occur in dendrites. Given that there is no obvious anatomical asymmetry in thalamic input to the habenula (Figure 5F, G; see also Turner et al, 2016), we hypothesized that there may be functional asymmetry in activity within thalamic inputs. To test this, we imaged the terminals of thalamic axons using s1020tGAL4, UAS:GCaMP6s fish. Thalamic terminals in the dorsal left neropil showed a greater activity compared to those in the right neuropil (Figure 8). Thus, preferential light-evoked activity in thalamic afferents that project to the dorsal left neuropil may underlie the asymmetric dorsal habenula response.
Thalamic lesion inhibits habenula response to illumination change
To further test if the thalamus contributes to light evoked activity in the habenula, we lesioned the thalamic neuropil with a two-photon laser. This technique is expected to injure fibers innervating the neuropil (Semmelhack et al., 2014). The laser was targeted to the neuropil of the putative nucleus rostrolateralis, which was identified by first imaging the response to light pulses (Figure 9A, B). Lesioning led visible damage in the neuropil (Figure 9D), and to a reduction of evoked activity in the thalamus and habenula (Figure 9C, E-G, I, L, M). There was some variability in the effect (Figure 9N), possibly reflecting the limitations of this technique in enabling consistent ablation. Lesioning other targets of retinal ganglion cell axons, which are located more posteriorly, did not lead to a loss of light-evoked activity in the habenula (Figure 9G, I, J, K), indicating that this technique did not cause indiscriminate damage. These observations support the hypothesis that the putative nucleus rostrolateralis of the thalamus has a role in light-evoked activity in the habenula of larval zebrafish.
A thalamo-habenula projection in the mouse
Finally, we asked whether a projection from the thalamus to the habenula, especially from a visual nucleus, is restricted to zebrafish, or whether it could also exist in a mammal. To do this, we examined the mouse mesoscale connectome (Oh et al., 2014). A search of the connectivity database derived from AAV injections into p56 mice, using thalamus as the injection site and epithalamus as the target site, yielded 18 hits covering different thalamic nuclei. Three experiments (numbers 267538006, 525796603 and 147212977) had been targeted to the ventral lateral geniculate nucleus. In one case (Figure 10), the anterograde tracer virus had been injected into a Slc32a1-IRES-Cre mouse, which expresses Cre in GABAergic neurons. Viral tracing in the mouse thus suggests that a visual nucleus in the thalamus projects to the habenula in a mammal.
Discussion
We have investigated how illumination conditions influence activity in the habenula of larval zebrafish. The pineal, although light responsive, does not innervate the zebrafish habenula (Yáñez et al., 2009) while the parapineal, which innervates the habenula, has been shown to be dispensable (Dreosti et al., 2014). Calcium imaging suggests that the afferent neurons mediating responses to light should terminate broadly in the habenula, cause stronger activity in the dorsal left neuropil, depolarize to increase or decrease in irradiance, and potentially include excitatory and inhibitory neurons. Several observations suggest that afferent neurons with these properties reside in the thalamus.
Lipophilic tracing of the habenula and transgenic labeling of thalamic neurons demonstrate that the thalamus directly innervates the larval zebrafish habenula. Highspeed imaging with widefield microscopy suggests that there is correlated activity in the thalamus and habenula, consistent with functional connectivity. Although widefield imaging uses visible light to excite the reporter, the rate of imaging used here is faster than the rise time of GCaMP6f (Chen et al., 2013), so initial images reflect activity prior to onset of the stimulating light. Further evidence for functional connectivity is provided by optogenetic stimulation of the thalamus, which causes a response in the habenula. Moreover, lesion of the thalamus reduced light-evoked activity in the habenula. The thalamus responds to both increase and decrease in illumination, and contains both excitatory and inhibitory neurons; no other source of GABAergic inputs to the zebrafish habenula has been described. Thus, by optical recording, anatomical tracing, activation and lesion, our data suggests that the thalamus mediates the habenula responses to irradiance change in larval zebrafish.
The region of the thalamus mediating activity in the habenula appears to be the nucleus rostrolateralis, as proposed by Turner et al (2016). The neuropil here can be functionally separated into two domains, based on the response to light – excitation to light OFF in the ventral regions and excitation to light ON more dorsally. This neuropil contains two previously defined targets of retinal ganglion cells, AF2 and AF4 (Burrill and Easter, 1994). AF4 is innervated predominantly by M3 and M4 retinal ganglion cells (Robles et al., 2014), which extend their dendritic tree into the proximal layer of the inner plexiform layer and are considered ON neurons. AF2 is innervated by B1 retinal ganglion cells that have dendrites in the distal layer (Robles et al., 2014), and these may account for the OFF responses in the thalamus and habenula. This thalamic nucleus may also receive input from non-retinal sources, but this remains to be investigated. The loss of a thalamic response to light in fish lacking eyes, however, suggests that such contribution may be minor.
Light is a potent regulator of brain function. It can affect mood (Vandewalle et al., 2010), alertness (Badia et al., 1991), cognitive ability (LeGates et al., 2012) and movement (Aschoff, 1960; Burgess and Granato, 2007). These phenomena are sensitive to irradiance, not image formation, and are mediated by a number of sensors including intrinsically-sensitive retinal ganglion cells whose targets include the thalamus (Hattar et al., 2006). The ability of light to affect normal movement patterns (Burgess et al., 2010) or to disrupt mood and cognition (LeGates et al., 2012) involves neuromodulators such as serotonin, and changing irradiance affects activity in the dorsal raphe (Fite et al., 2005; Cheng et al., 2016). Based on the data here, and the well-established roles of the habenula in regulating neuromodulators, we suggest that some of these effects of light may be mediated by the thalamic projection to the habenula.
A projection from the thalamus to the habenula may be evolutionarily conserved in vertebrates. In humans and rabbits, a thalamo-habenula projection was proposed many years ago based on degeneration experiments (Marburg, 1944; , 1961). Using retrograde tracing with horseradish peroxidase, a projection from the dorsal thalamus to the habenula was reported in a lizard (Díaz and Puelles, 1992). Hints of a projection can also be seen in a tracing experiment performed in rats (Moore et al., 2000). The large-scale mouse mesoscale connectome project (Oh et al., 2014), which uses viral-based anterograde tracing, provides the most recent evidence for a thalamo-habenula projection in a mammal. This technique suggests a projection from the ventral lateral geniculate nucleus and other thalamic nuclei to the habenula. It will be interesting to determine whether a similar anatomical connection exists in humans, as this may contribute to the high functional connectivity between the thalamus and habenula reported recently (Torrisi et al., 2016).
Materials and Methods
Fish lines
Experiments were performed in accordance with guidelines issued by the Institutional Animal Care and Use Committee of the Biological Resource Centre at Biopolis, Singapore. Zebrafish (Danio rerio) lines used for this study were:Tg(UAS:GCaMP6s)sq202, SqKR11Et, GAL4s1011t, GAL4s1020t, Tg(UAS:GCaMP3)sq200, Tg(elavl3:GCaMP6f)a12200, Tg(UAS:ChR2-eYFP) (Arrenberg et al., 2009) and AB wildtype.
Tg(elavl3:GCaMP6f)a12200 was generated by PCR amplification of the GCaMP6fopen reading frame (Addgene plasmid 40755 (Chen et al., 2013)) with forward primer ataACTAGTgccaccATGGGTTCTCATCATCAT and reverse ataCCGCGGcTCACTTCGCTGTCATCATTTGTAC (restriction site and coding sequences are in upper case). This fragment was cloned into a plasmid with Tol2 arms flanking an upstream attR1-R2 cassette and the insertion site using restriction enzymes SpeI and SacII. Previously described elavl3 (HuC) cis-regulatory elements (Higashijima et al., 2003) were placed upstream via LR recombination (Invitrogen) with an attL flanked elavl3 entry clone. The resulting plasmid was then co-injected into 1-cell stage embryos at a concentration of 30 ng/μL with Tol2 transposase mRNA at a concentration of 30 ng/μL. A single founder was selected based on high and spatially broad expression. Outcrossing this founder generated 50% GCaMP6f-positive embryos, which were selected to establish the line.
Imaging
Zebrafish larvae (aged 5 - 10 dpf) were anaesthetized in mivacurium and embedded in low-melting temperature agarose (1.2-2.0 % in E3) in a glass-bottom dish (Mat Tek). They were imaged on a Nikon two-photon microscope (A1RMP), attached to a fixed stage upright microscope, using a 25x water immersion objective (NA = 1.1). The femtosecond laser (Coherent Vision II) was tuned to 920 nm for GCaMP imaging. Stacks were collected in resonant-scanning mode with a 525/50 nm bandpass emission filter and with 8x pixel averaging; single-plane images were collected in galvano-scanning mode with 2x pixel averaging. The sample size was based on (Dreosti et al., 2014).
Light stimuli were generated by 5 mm blue LEDs (458 nm peak emission). They were powered by a 5 V TTL signal from a control computer and synchronized with image capture using a National Instruments DAQ board, controlled by the Nikon Elements software. Light intensity at the sample was 0.13 mW/cm2.
For widefield microscopy, excitation was provided by LEDs (Cairn OptoLED) at 470 nm. Images were captured on a Zeiss Axio Examiner with a 20x water immersion objective, using a Flash4 camera (Hamamatsu) controlled by MetaMorph. After background subtraction, change in fluorescence was measured using the deltaF-up command in Fiji.
Data analysis
Initial Data Preprocessing
Raw images obtained were first registered to correct for any vertical/horizontal movement artifacts using cross correlation. Then, a median filter of size 3 was applied to remove noise. A darker region outside the region of interest was chosen as the background and subtracted from the image to remove any background noise. Non linear trends in the data were detrended using polynomials of order 2-5. Data was then normalized into Z-scores by subtracting the overall mean and dividing by the standard deviation. A rolling window average was then used to smooth noisy traces where necessary. Where possible, cells were segmented or images were directly analysed as pixels (see Below).
Correlation to light evoked activity
Temporal traces from pixels (Thalamus or Thalamic afferents) or segmented cells (in the habenula) were classified as responding to light ON or OFF by calculating their correlation coefficient to a square wave that is 1 when the light is ON or when light is OFF and 0 during other time periods. High correlation to these traces indicated that the pixel or cell is responding to light ON or OFF respectively.
Pixel based analysis of the habenula and thalamus
To show the spatial and temporal distribution of light evoked activity in the habenula and thalamus, the Thunder platform (Freeman et al., 2014) was used for fast pixel based clustering and factorization.
PCA
Principal Component Analysis (PCA) was used to obtain a low dimensional representation of the population. In normal fish, in the habenula (Figure 1D) and thalamus (Figure 7M), the first two Principal Components (PC) easily picked up the evoked responses to light ON and OFF. The explained variance, though small, decayed rapidly after the first component (Figure 1D, variance explained PC1 12.92%, PC2 6.12% and PC3 4.57%; first 20 PCs 55.60%; Figure 7M, variance explained PC1 11.87%, PC2 3.61% and PC3 2.67%; first 20 PCs 40.82%) and hence the first two PCs were chosen to plot the data in low dimension. They showed reproducible state changes in the population to changes in irradiance. In eye lesioned fish (Figure 7P) however, neither of the first twenty PCs (variance explained PC1 4.92%, PC2 4.24% and PC3 3.26%; first 20 PCs 39.53%) showed any discernable correlation to light evoked activity (The inset in Figure 7P). The first two are plotted in Figure 7P for comparison with controls. Correlation coefficients plotted in insets of Figure 7M and 7P were obtained by correlating the PCs with a trace that was 1 when light was ON and 0 otherwise.
K-means
K-means clustering was performed to identify pixels with similar responses profiles. Given the uncertainty of k-means to the optimal cluster number, an iterative approach was used to separate pixels relating to evoked responses versus pixels that don’t (here referred to as noise clusters). The number of clusters were chosen to reveal as many stimulus related clusters as possible, until there was little change in the number and types of stimulus related clusters and increase in noise related clusters. Noise clusters were then removed from the spatial and temporal plots for clarity. It is to be noted that this analysis does have its drawbacks when used to analyse data with variable variance and cannot not capture all possible types of responses. Multiple runs were made to ensure that noisy pixels were not clustered into evoked clusters. Where applicable cell segmentation and manual classification was used. For example, we were not able to separate inhibitory clusters from off custers in Figure 3E-G. This was revealed when the cells were segmented, and their temporal traces inspected. In all cases, K-means cluster center showing evoked responses to light ON were colored in shades of blue and those showing responses to light OFF were colored in shades of red. Pixels belonging to the cluster were colored similarly and superimposed on an average image of the plane analysed. In different datasets, on average, this analysis provided an optimal k of 6-10; 3-4 clusters that didn’t correspond to evoked activity were not included while plotting.
For data following eye lesion (FIgure 7) and thalamic neuropil/AF9/AF7 lesion (Figure 9), k-means was performed to differentiate responses between the controls and the lesion. Hence, number of clusters were chosen such that cluster center adequately differentiated responses before and after lesion. Cluster center not responding to light were also plotted here. To verify the results of k-means, the presence of light response was verified by examining the pixels (Figure 8 Q-S) or segmenting the cells (Figure 9 J-N) in both control and lesioned animals.
The scripts used for analysis are provided at http://dx.doi.org/10.5061/dryad.q0171.
Cell segmentation
Each stack was scaled 2x in imageJ, then maximally projected to a single image, which was then subjected to a minimum filter and unsharp mask to sharpen the boundary of cells. ROIs were identified using the “find maxima…” command, as a way to localize regional darkest point as the center of each ROI. The boundary of the ROI was outlined by “analyze particle…” that connects bright pixels into mosaic-like tessellated plane, encircling each darkest point. Each ROI was then numbered sequentially using the ImageJ ROI Manager and mapped back to the original despeckled image stack. Manual segmentation was done here to delete extraneous ROIs outside the habenula and to encircle cells that were not detected by the algorithm (<10% of total ROIs). In the last step, “Set measurements…” and “measure” in ImageJ provided the mean fluorescence value of all pixels within each ROI across the entire image stack and the x-y coordinates of each ROI. Time-lapse series in which z drifting occurred were excluded, as in this case ROIs could not be defined.
K-means on segmented cells
For Figures 2D-F, k-means was performed from cells segmented by the semi automated algorithm described above. The purpose is to determine heterogeneity of temporal responses to changes in irradiance, accurately classify cells into ON, OFF and Inhibitory responses and perform correlation between them. Analysis was done on traces from 4986 habenula cells from 6 fish. Traces were detrended, smoothed and normalized to z-scores using baseline as the time before the first blue light. Traces that did not reach a Z-score of 2 during the period of irradiance change were classified as not having an evoked response and not included in the clustering analysis. 2456 of 4986 cells were thus removed. K-means was first run with an arbitrary k = 60. This generated a wide range of clusters capturing the temporal heterogeneity of the responses. The clusters were then divided into ON, OFF, Inhibitory and no evoked response. Neurons belonging to each cluster were correlated among each other and to the cluster center. If they had an evoked response, they were correlated with other clusters and assigned to one with the highest correlation. Otherwise they were classified into the no evoked response category; 138 such cells were reclassified. Traces of cells belonging to ON, OFF and Inhibitory clusters are plotted as a heatmap in Figure 2G and their correlation coefficients in Figure 2H. Similarly, cells segmented from multiple fish were classified to reveal responses in different habenula regions as plotted in Figure 3H-J.
Boxplots
Boxplots in Figures 7Q, 8D-E and 9N were plotted to show the full distribution of the data. The box in the boxplot ranges from the first quartile to the third quartile, and the box shows the interquartile range (IQR). The line across the box is the median of the data. The whiskers extend to 1.5*IQR on either side of the box. Anything above this range are defined as outliers and plotted as black diamonds in the plots.
Neural tracing
DiD (Life Technologies) was dissolved in 50 μl ethanol to make a saturated solution. This was heated to 55˚C for 5 minutes prior to injection into the fish that had been fixed in 4% paraformaldehyde. Fish were mounted in 1.2% low melting temperature agarose dissolved in PBS. The dye was pressure injected into the habenula under a compound microscope (Leica DM LFS), using a 20X water immersion objective. For labeling the retina, a saturated solution of DiI in chloroform was used. Injections were carried out under a stereo microscope (Zeiss Stemi 2000). After injections, fish were stored at 4˚C overnight to allow tracing, and then imaged with a 40x water immersion objective on a Zeiss LSM 710 confocal microscope.
Antibody label
Larvae were fixed in 4% para-formaldehyde/PBS overnight at 4˚C. They were then rinsed in PBS. The brains were dissected out, and permeabalized using 1% BSA (fraction V; Sigma), 0.1% DMSO and 0.1% Triton X-100. The antibodies used here, anti-vGlut1/2 (Synaptic Systems 135503, RRID:AB_1279466; 1:100) and anti-GAD65/67 (Abcam ab11070, RRID:AB_297722; 1:500), have previously been used in zebrafish (Wyart et al., 2009; Lee et al., 2010). The brains were incubated in the primary antibody overnight, rinsed several times in PBS, then incubated in secondary antibody (Alexa 488 goat anti-rabbit; 1:1000). After washing, these were mounted in 1.2% agarose/PBS. Imaging was carried out using a Zeiss LSM 510 laser scanning confocal microscope, with a 40x water immersion objective.
Optogenetic stimulation
5 dpf Tg(s1020GAL4, UAS:ChR2-eYFP, elavl3:GCaMP6f) larvae were used. The eyes were removed using fine tungsten needles in fish that were anesthetized with MS222. This procedure was carried out in Ringers saline. Fish were then mounted in 1.2% agarose in Ringers saline, and imaged using two-photon microscopy as described above, at 1 Hz. Optical stimulation was carried out using a 50 µm fiber optic probe (Doric Lenses), placed approximately 20 µm from the thalamus. The 465 nm LED (Doric) was driven with a current of 900 mA, 30 seconds after the start of imaging. 10 pulses were provided, with a pulse duration of 25 milliseconds and a frequency between 1 and 8 Hz. Each fish was exposed to at least 3 pulse trains. For Figure 6B-C, the average of the first 29 frames was used as a reference. The ratio of all frames relative to this reference was obtained using FIJI. The analysis to generate Figure 6G was blind to the genotype.
Enucleation
5 day-old fish were anaesthetized in Ringer’s saline containing MS222. The eyes were removed using electrolytically sharpened tungsten needles. Fish were allowed to recover for several hours in anesthetic-free saline. Activity recorded 2 - 4 hours after eye removal. To enable lateral imaging of the thalamus (Figure 7H-J), one eye was removed using this method.
Laser ablation
Tg(Elavl3:GCaMP6f) larvae were anaesthetized and then mounted in 2% low-melting temperature agarose. Lesions were created with the femto-second laser tuned to 960 nm and fixed on a single point. Several pulses, each lasting 100 - 500 msec, were used. Lesioning was monitored by time-lapse imaging before and after each pulse, and was terminated when there was a localized increase in GCaMP6f fluorescence. The formation of a bubble indicates that lesioning involved plasma formation (Venugopalan et al., 2002). These can be imaged in the red channel of the two-photon microscope. Sample size was chosen based on (Aizenberg and Schuman, 2011). Animals with bleeding in the brain after lesioning, due to bursting of a blood vessel in the thalamis, were discarded.
Analysis of the mouse mesoscale connectome
The mouse connectivity atlas (http://connectivity.brain-map.org/) was searched in “”Target Search” mode, using “TH” as the source structure and “EPI” as the target structure. The minimum target volume listed was 0.005 mm3. For each hit, serial transverse sections containing the habenula, which had been imaged by two-photon microscopy, were screened manually to verify that there were labelled fibers within the habenula.
Author contributions
Experiments were designed by CRK, QL and SJ. CRK carried out two-photon imaging. SK developed software and analyzed imaging data. DGCH and IHB generated the elavl3:GCaMP6f transgenic line. SJ performed wide-field imaging, analysis, dye tracing, optogenetic manipulation and wrote the manuscript.
Movie 1. Habenula neurons do not project to the thalamus. 3D rendition of habenula projection in a zebrafish larva, visualized by expression of RFP under the narp promoter (red) and eGFP under the brn3a promoter (green). There is a clear projection to the interpeduncular nucleus (IPN), but not to the thalamus.
Movie 2. Thalamic neurons project to the habenula. Z-stack of a s1020t:GAL4, UAS: Kaede, elavl3:GCaMP6f fish. Thalamic neurons are shown in red, and they can be seen to project to the neuropils of the habenula. Red label also appears in streaks in the lateral habenula. Anterior is to the top.
Movie 3. GAD65/67 label in a zebrafish larva. z-stack of a s1011t:GAL4, UAS:GCaMP3 transgenic fish, after immuno-labelling with an antibody to GAD65/67 (magenta). The stack goes from dorsal to ventral. GAD65/67 label is visible in neuropils of the habenula; puncta can be seen between cells in the lateral regions of the habenula in more ventral planes. GAD65/67 labeled cells are visible in the deep focal planes, but these do not express GCaMP3. The location of GAD65/67 expressing cells correlates with the thalamus. S1011Et drives GAL4 expression in the habenula, medial pallium and anterior-lateral pallium. This is a dorsal view, with anterior to the left.
Movie 4. Z-stack of 6 day old gad1b:RFP, elavl3:GCaMP6f fish. GABAergic neurons (magenta) are visible in the thalamus, below the habenula. Anterior is to the left. The stack goes from dorsal to ventral.
Acknowledgements
This work was funded by core funding from the Institute of Molecular and Cell Biology and a grant from the Ministry of Education, Singapore. SK and QL were supported by fellowships from the National University of Singapore. DGCH was supported by NIH grant 5T32HL007901. We thank Caroline Kibat for technical assistance and Claire Wyart for providing the UAS:ChR2-eYFP transgenic line.