The absence of retinal input disrupts the development of cholinergic brainstem projections in the mouse dorsal lateral geniculate nucleus

The dorsal lateral geniculate nucleus (dLGN) of the mouse has become a model system to study the development of thalamic circuits [1–4]. Much of our present understanding is based on studies focused on the retinogeniculate pathway, the connections between retinal ganglion cells (RGCs) and dLGN neurons. While retinal projections provide the primary excitatory drive for relay neurons of dLGN, the vast majority of input is nonretinal in origin [5], and acts to modulate the gain of retinogeniculate transmission in a state-dependent manner [6, 7]. The primary sources of nonretinal input to dLGN include glutamatergic neurons of visual cortex layer VI, GABAergic neurons of the thalamic reticular nucleus, and the cholinergic neurons from different brainstem nuclei. Despite the fact that over 90% of all synapses in dLGN arise from nonretinal sources, we know little about how and when these projections innervate dLGN, or how they interact with the arrival and refinement of retinal projections. What little we do know is based on the corticothalamic pathway [8–10]. Layer VI neurons of the neocortex begin to innervate the dorsal thalamus at perinatal ages, but corticogeniculate innervation occurs largely after the first postnatal week, after the arrival of retinal axons and their refinement into non-overlapping eye-specific domains [9]. Moreover, retinal input orchestrates the timing of corticogeniculate innervation by regulating the levels of aggrecan, a repulsive extracellular matrix molecule [11]. What remains to be tested is whether such sequencing and reliance on retinal input is part of a conserved developmental plan that governs the arrival of other nonretinal inputs to dLGN.

To address this, we examined another major nonretinal projection to dLGN, the ascending cholinergic input from the brainstem. Estimates reveal that about 25% of all synapses in dLGN arise from brainstem cholinergic nuclei [12]. These projections have a substantial influence on retinogeniculate transmission, regulating the firing mode of dLGN neurons [13], establishing network states during sleep, wakefulness, and arousal [14, 15], as well as modulating visuo-motor interactions [16]. Previous immunohistochemical studies have shown a late postnatal onset for the labeling of acetylcholine synthesizing enzyme (choline acetyltransferase, ChAT) in dLGN, which increases in density over a protracted period of development [17, 18]. However, little is known about the source, trajectory, and pattern of cholinergic innervation in the developing mouse dLGN. In several mammalian species, ascending cholinergic projections to dLGN originate from two distinct brainstem groups, the pedunculopontine and laterodorsal tegmental nuclei (PPTg, & LDTg), as well as the parabigeminal nucleus (PBG) [19–21]. The projection from PBG is especially notable since these axons course within the optic tract en route to dLGN and superior colliculus [20, 22, 23], raising the possibility that retinal axons participate in the guidance of PBG axons.

Here, we examined the postnatal development of cholinergic input to dLGN and tested whether the absence of retinogeniculate projections affects the timing and patterning of cholinergic innervation. To visualize ascending cholinergic projections, we crossed a ChAT-Cre knock-in mouse line with a Cre-dependent reporter strain (Ai9) to selectively drive expression of the fluorescent protein tdTomato (tdT) in cholinergic neurons [24–29]. To evaluate the development of cholinergic innervation of dLGN in the absence of retinal projections, we utilized a mutant mouse that lacks math5, a transcription factor necessary for RGC progenitor cell differentiation [30]. Math5^−/− mice exhibit a > 95% loss of RGCs, and the surviving RGCs fail to form an optic nerve, thus leaving the brain devoid of retinal input [31–34]. Furthermore, we used Cre-dependent viral tracing techniques to assess whether the trajectory of optic tract-associated PBG axons to dLGN was disrupted in a brain lacking retinofugal projections.

To visualize brainstem cholinergic neurons and their projections to dLGN, we crossed ChAT-Cre mice with an Ai9 reporter line [24–29]. Figure 1a shows the pattern of tdTomato (tdT) labeling in the brainstem in an adult (P60) ChAT-Cre x Ai9 mouse. Somatic labeling of tdT was seen in cholinergic neurons of brainstem nuclei reported to project to dLGN [17, 20, 22, 38], including laterodorsal tegmentum (LDTg, Fig. 1a, left), pedunculopontine tegmentum (PPTg, Fig. 1a, left & right), and parabigeminal nucleus (PBG, Fig. 1a, right). High power views of these nuclei (Fig. 1b) showed that tdT-labeled somata were present in the adult (P60) and at birth (P0), and comparable numbers of neurons were labeled. In adults, labeled projections were evident throughout the dorsal thalamus, including the ventral (vLGN) and dorsal lateral geniculate nuclei (dLGN), lateral posterior nucleus (LP), and in the pretectum (PT) (Fig. 1c, left). At P0, cholinergic innervation was sparse in these regions, and entirely lacking in dLGN (Fig. 1c, right).

Consistent with observations made in other mammalian species [20, 23] in mouse we noted that (tdT-labeled) cholinergic axons traversed through the supraoptic decussation and the optic tract (Fig. 2). The supraoptic decussation contains multiple axonal tracts of nonretinal origin, lies adjacent to the optic tract near the cerebral peduncle, and crosses midline along the ventral border of the hypothalamus just caudal and ventral to the optic chiasm [39]. To distinguish between tdT-labeled cholinergic brainstem axons and retinal axons in the optic tract, we made binocular intravitreal injections of anterograde tracer cholera toxin subunit B (CTB) conjugated to Alexa Fluor 488 (green) in an adult ChAT-Cre x Ai9 mouse. Figure 2 provides coronal views of the optic tract at the level of dLGN (Fig. 2a), cerebral peduncle (Fig. 2b), and optic chiasm (Fig. 2c). At all levels, tdT-labeled cholinergic (red) and CTB-labeled retinal (green) axons shared similar trajectories. In the dorsal thalamus (Fig. 2a), cholinergic and retinal axons coursed together in the optic tract along the dorsolateral boundary of dLGN. At more ventral levels, cholinergic axons were found within the supraoptic decussation along the lateral edge of the cerebral peduncle, as well as in the adjacent optic tract (Fig. 2b). At the base of the brain, at the level of the optic chiasm, cholinergic axons were present along midline in the supraoptic decussation and more laterally where they comingled with retinal axons in the optic tract (Fig. 2c, Fig. 10).

To determine whether retinal input is involved in regulating the development of cholinergic projections to dLGN, we compared the pattern and time course of innervation in ChAT-Cre x Ai9 mice (wild type, WT) with age-matched ChAT-Cre x Ai9 mice bred on a math5-null background (math5^−/−), a mutant strain that lacks an optic nerve as well as central retinal projections [31, 32]. In both WT and math5^−/− mice, cholinergic fibers began to innervate dLGN by P5 (Fig. 3a and b). In WT, cholinergic axons closely followed the optic tract, innervating dorsolateral edge of dLGN (Fig. 3a and A’). However, in dLGN of math5^−/− mice (4 hemispheres at P5) the aggregation of cholinergic axons along the dorsolateral edge was less striking (Fig. 3B and B’).

Figure 4 shows the progression of cholinergic innervation in WT and math5^−/− at different postnatal ages (P5, P9, P11, P14, P21, and P30). As reported previously, there was a reduction in total size of dLGN in math5^−/− mice, compared to WTs [32]. In WTs between P5-P30, cholinergic innervation progressed steadily in a dorsolateral to ventromedial manner so that by P30 the entire nucleus contained a diffuse arrangement of cholinergic arbors. Math5^−/− mice also showed an age-related increase in cholinergic innervation. However, compared to WT, math5^−/− mice exhibited alterations in the rate and route of innervation, as well as in the patterning of their axon arbors.

To compare the rate of cholinergic innervation in the WT and math5^−/− mice, we quantified the degree of cholinergic innervation as the percentage of fluorescent pixels in relation to the total number of pixels within dLGN, as a function of postnatal age. Figure 5 presents mean values for WT and math5^−/− mice between P3-P60. While both groups showed a progressive increase with age between P5–30, the rate of innervation was slower for math5^−/− between P9 and P21. At these ages, the degree of innervation was significantly lower in math5^−/− compared to age-matched WTs (two-way ANOVA, F = 2.49, Bonferroni post-hoc test, P9, p < 0.05; P11, p < 0.01; P14, p < 0.01; P18, p < 0.01; P21, p < 0.05). At P30, the degree of innervation was similar in both groups and remained stable into adulthood (P30–60).

To rule out the possibility that math5^−/− leads to a more generalized disruption in brainstem cholinergic innervation of thalamus, we examined the degree of cholinergic innervation in nuclei that do not receive direct retinal input. These include the visual and non-visual sectors of the thalamic reticular nucleus (visTRN and non-visTRN), the ventral posterolateral nucleus (VPL), and posterior nucleus (Po) of the thalamus. Figure 6 presents examples of coronal sections through these nuclei at P14 (the time of peak cholinergic disruption in dLGN, Fig. 5) for WT (Fig. 6a, left) and math5^−/− mice. (Fig. 6a, right). The squares represent a 200 × 200 μm region of interest where measurements of cholinergic innervation were obtained. Figure 6B summarizes these comparisons and shows that for each region, the degree of innervation between WT and math5^−/− was comparable (Student’s T-test, visTRN t(6) = 0.348, p = 0.740, non-visTRN t(6) = 0.372, p = 0.762, VPL t(6) = 1.912, p = 0.104, Po t(6) = 0.372, p = 0.723).

The architecture of cholinergic arbors in dLGN of math5^−/− mice also differed from WT. Examples of these differences are illustrated in Fig. 7 which depicts low and high power views of cholinergic projections in WT (Fig. 7A1 & A2) and math5^−/− (Fig. 7B1 & B2) mice at P14 and P30. In WT, low power images along with a corresponding “heat map” that plots the intensity of tdT fluorescence, revealed a diffuse and a relatively homogeneous intensity profile for areas of cholinergic innervation. High power views of these regions showed an orderly arrangement of fine-caliber axonal processes. By contrast, math5^−/− mice exhibited ectopic patches of intense fluorescence consisting of a dense tangled plexus of processes.

Another salient phenotype in all math5^−/− mice was an alteration in the trajectory of cholinergic axons, especially those which normally traverse through the dorsal thalamus along the optic tract. Figure 8 shows coronal sections of the thalamus in WT and math5^−/− mice (P14) at a caudal (Fig. 8a and b) and rostral (Fig. 8c and d) level of dLGN. In WTs (Fig. 8a and c), at both levels a large number of cholinergic axons were seen traveling along the outer border of the thalamus. High power views show these projections arise with the optic tract near the cerebral peduncle (Fig. 8A” & C″), and run dorsally along the lateral border of vLGN and dLGN (Fig. 8A’ & C′). However, in math5^−/− mice, cholinergic axons were displaced medially at caudal dLGN (Fig. 8B & B’), and were not readily apparent at a more rostral level (Fig. 8D & D’). Moreover, while cholinergic axons were present near the cerebral peduncle (Fig. 8B”), they failed to fasciculate into separate bundles (Fig. 8D”), that in WT course in the supraoptic decussation and in the optic tract.

Previous studies in rodents indicate that cholinergic axons within the optic tract arise from PBG [20, 40]. To determine whether PBG is the source of the misrouted cholinergic axons observed in math5^−/−, we selectively labeled PBG neurons in adult ChAT-Cre (WT, n = 3) and ChAT-Cre x math5^−/− (math5^−/−, n = 4) mice using injections of Cre-dependent anterograde adeno-associated viral tracer FLEX-AAV-ChIEF-tdTomato. Before conducting viral tracing experiments, we used ChAT-Cre x Ai9 reporter lines to determine whether the number of PBG neurons is affected by the deletion of math5. Figure 9 depicts a plot showing the number of PBG neurons in WT and math5^−/− ChAT-Cre x Ai9 mice. Estimates reveal no significant differences between the two groups (Student’s t-test, t(6) = 0.148, p = 0.887). On average, we found about 27 tdT labeled neurons in a sample of sections through the middle of PBG for both groups.

Figure 10 shows the pattern of axonal projections following the unilateral labeling of PBG in a WT (Fig. 10a) and math5^−/− (Fig. 10d) mouse. Examples of the injection site in PBG for WT and math5^−/− are shown in Fig. 10A’ and 10D’, along with high power views of tdT labeled PBG neurons in Fig. 10A” and D”. Dense labeling was seen throughout much of the nucleus, and estimates of somatic labeling within the middle three sections of PBG ranged between 10 and 27 neurons per case. In WT, PBG axons coursed within the ipsilateral supraoptic decussation near the cerebral peduncle (Fig. 10b), crossed to the contralateral hemisphere at ventral midline (Fig. 10c), and ascended along the lateral aspect of thalamus with the optic tract (Fig. 10B’). In math5^−/−, PBG axons followed a similar trajectory (Fig. 10E & F) until reaching contralateral ventral thalamus, where they veered in a medial direction traversing dorsally through the medial geniculate nucleus and vLGN (Fig. 10E’). Figure 11 illustrates the pattern of cholinergic arbors within dLGN of WT and math5^−/− mice after a unilateral injection into PBG. In WT, the projection to the ipsilateral dLGN was restricted to a small strip of arbors in the dorsomedial pole (Fig. 11A, A1, & A1’). By contrast these projections were more variable in math5^−/−, either missing altogether (n = 2) or showing substantial spread beyond the dorsomedial pole of dLGN. The projections in contralateral dLGN appeared to be similar in both WT and math5^−/−, with arbors distributed more evenly throughout the nucleus (Fig. 11C, & Fig. 11D).

To compare the pattern of PBG projections in WT (n = 3) and math5^−/− (n = 4) ChAT-Cre mice, we quantified the distribution of tdT-labeled arbors within dLGN. Figure 12 shows a plot of the spatial distribution of tdT-labeled PBG fibers in the ipsilateral (left) and contralateral (right) dLGN. Values represent the percentage of fluorescence in each of four quadrants (Fig. 12 inset: dorsomedial, DM; ventromedial, VM; dorsolateral, DL; ventrolateral, VL). In the ipsilateral dLGN of WT, we found a significant difference in fiber distribution between quadrants (Kruskal-Wallis, H = 8.23, p = 0.014), with the DM quadrant exhibiting the highest proportion of cholinergic labeling (mean = 72.9%). In the ipsilateral dLGN of math5^−/−, the distribution of labeling was variable across the quadrants (n = 2), or absent altogether (n = 2). It is important to note despite being absent in two cases, the corresponding contralateral labeling pattern was consistently present and no different from other mice. Since all injections were of the same volume and restricted to PBG, the absence of ipsilateral projections may simply reflect inherent variability brought about by absence of retinal input. In the contralateral dLGN of WT mice, there was a significant difference between quadrants (K-W, H = 9.97, p = 0.0003), with values in DM higher than VL quadrant (Dunn’s post-hoc test, DM vs. VL, p = 0.013). A similar pattern was noted in the contralateral dLGN of math5^−/− (K-S, H = 14.12, p < 0.0001, Dunn’s post-hoc test, DM vs. VL; p = 0.002). Finally, arbor distribution in the contralateral dLGN of WT and math5^−/− were not significantly different from each other (two-way ANOVA, F = 0.87, p = 0.47).

We used a ChAT-Cre mouse, crossed to a tdTomato reporter line (Ai9), to visualize and track the development of brainstem cholinergic projections to dLGN. Our results reveal that cholinergic brainstem innervation of mouse dLGN begins at the end of the first postnatal week, and increases slowly with age to reach an adult-like density by the end of the first postnatal month. Initially, cholinergic fibers appear in the dorsolateral region of dLGN just beneath the optic tract, then gradually progress in a ventromedial direction to form a homogeneous plexus of fibers throughout dLGN. This is consistent with earlier studies in cat and mouse, showing a slow but steady increase in ChAT immunoreactivity with postnatal age [17, 18]. Thus, cholinergic innervation of dLGN occurs well after the establishment of the retinogeniculate pathway [1]. Moreover, our results support the view that innervation of dLGN occurs in a sequential manner, with retinal projections arriving before nonretinal ones [9].

Although corticothalamic and cholinergic brainstem projections begin to arrive in dLGN at roughly the same time (P5), their rate of innervation is vastly different. For example, cortical innervation of dLGN is nearly complete between P9–12 [8, 9], yet brainstem cholinergic innervation is still sparse and restricted largely to the dorsolateral region, only reaching adult-like density of innervation by the end of the first postnatal month. Such timing indicates that the reciprocal connections between the dLGN and visual cortex are established prior to cholinergic brainstem input. The late postnatal onset of cholinergic brainstem innervation is consistent with the maturation of other reticular ascending arousal systems [41–44]. Overall this suggests that network-wide, state-dependent modulation of sensory information becomes operational well after sensory connections are established.

Our results in math5^−/− mice also support earlier studies underscoring the importance of retinal input in regulating the timing of nonretinal innervation of dLGN [9]. The absence of retinal input slowed the rate of cholinergic innervation to dLGN, though the overall degree of innervation measured in adults was preserved. It is important to note that retinal input regulates timing of nonretinal input in a bidirectional manner. Whereas the current study showed a slower rate of cholinergic innervation of dLGN in math5^−/− mice, previous studies in the same mutant demonstrate an acceleration of corticogeniculate innervation, suggesting that retinal input regulates nonretinal development through distinct molecular mechanisms. The accelerated arrival of corticogeniculate axons in math5^−/− mice results from the disruption of aggrecan, a repellant extracellular matrix molecule that prevents the premature entry of cortical axons into dLGN [11]. While the molecular mechanism underlying retinal regulation of cholinergic innervation in dLGN is unresolved, the disruption in the rate of cholinergic innervation in math5^−/− mice may reflect a reduction of trophic support during development. In other regions that receive cholinergic innervation, neurotrophins such as brain-derived neurotropic factor (BDNF) and neurotrophin-3 (NT-3), have been shown to promote the growth of cholinergic neurites [45–47]. In WT mice, BDNF and NT-3 are anterogradely transported from the retina to dLGN [48, 49]. Therefore, it is possible that a reduction in the levels of these factors may underlie the dystrophic growth of cholinergic fibers observed in math5^−/− mutants.

In addition to regulating the rate of cholinergic innervation of dLGN, we demonstrated that retinal input plays a role in establishing the thalamic trajectory of a subset of brainstem cholinergic axons. These axons, which arise from cholinergic neurons of PBG, normally run within the optic tract along the outer border of the thalamus en route to dLGN. However, in math5^−/− mice PBG axons are displaced, traveling in a diffuse manner through thalamus. This indicates that during development, PBG axons use the retinal axons of the optic tract as a scaffold to navigate through the thalamus. Such axon-axon interactions are typically mediated by cell-adhesion molecules, which promote growth along the length of an existing axon [50]. Surprisingly, our tracing studies in math5^−/− mice revealed that such interactions are not necessary for PBG axons to elongate and reach their appropriate target. We found that PBG axons continued to grow through thalamus and reach dLGN even in the absence of an optic tract. Furthermore, nuclei-specific targeting was preserved in math5^−/− mice. PBG axons continued to innervate dLGN (and superior colliculus, unpublished observations) and were not found in neighboring structures within the lateral geniculate complex. In addition, the laterality of the projection did not appear to be affected, as the bulk of PBG arbors targeted the contralateral dLGN in both WT and math5^−/− mice. Therefore, it is likely that PBG axons rely on cues expressed within the thalamus to target, innervate, and arborize in dLGN. Future experiments are needed to fully elucidate the specific cues that govern PBG axon guidance.

Though retinal input does not appear to be necessary for nuclei-specific targeting of PBG input, it plays an important role in establishing the appropriate pattern and morphology of arbors within dLGN. In WTs, PBG arbors were consistently found within the dorsomedial pole of the ipsilateral dLGN, adjacent to the optic tract. However in math5−/− mice, this pattern was perturbed, with arbors extending beyond their usual target or perhaps failing to reach the nucleus altogether. Interestingly, in both WT and math5^−/− mice, PBG projections to the contralateral dLGN were similar in distribution.

In several mammals, inputs from PBG, superior colliculus, and certain types of retinal ganglion cells converge to terminate in discrete regions of dLGN (e.g., C-laminae of carnivores, koniocellular layers of primates), suggesting the existence of a conserved visual channel involved in the coordination of visuo-motor processing [20, 21, 51–53]. The rodent dLGN contains a homologous region, known as the dorsolateral shell [54], which receives input from direction selective retinal ganglion cells (DSGCs), as well as the superficial layers of the superior colliculus [51, 55]. In WT mice, we found an ipsilateral PBG projection to dLGN that targets the dorsolateral shell but in a circumscribed area that appears to represent the upper nasal visual fields [56, 57]. While the presence of retinal input appears to be necessary for the proper targeting of PBG arbors to the ipsilateral dLGN it has little impact if any on the more widespread and diffusely organized contralateral projections. Thus, it is not yet clear how this is accomplished, or whether it involves interactions with other inputs that target the shell, such as those from DSGCs and/or the superior colliculus.

The mouse dLGN has emerged as a useful model system for exploring the development of visual thalamic circuitry. However, relatively little is known about how and when nonretinal connections are formed, or whether their timing of innervation is regulated by retinal input. In this study, we examined the developmental time course of a major ascending nonretinal input to dLGN, cholinergic brainstem nuclei, by using a transgenic mouse line to visualize cholinergic input (ChAT-Cre), and a knockout line devoid of retinal input (math5^−/−). We found that the cholinergic innervation has a protracted time course, innervating dLGN over the course of first postnatal month, with the bulk of input arriving well after retinal connections are formed. In math5^−/−, we found that the rate of cholinergic innervation is slowed compared to WTs. Furthermore, the routing and pattern of arborization of optic tract-traversing PBG cholinergic axons was altered in the ipsilateral dLGN of math5^−/− mice. Our anatomical tracing data demonstrated that while retinal input is not necessary for PBG innervation of dLGN, it plays an important role in the organization of arbors within the nucleus.