A systematic topographical relationship between mouse lateral posterior thalamic neurons and their visual cortical projection targets

Abstract

Higher-order visual thalamus communicates broadly and bi-directionally with primary and extrastriate cortical areas in various mammals. In primates, the pulvinar is a topographically and functionally organized thalamic nucleus that is largely dedicated to visual processing. Still, a more granular connectivity map is needed to understand the role of thalamocortical loops in visually guided behavior. Similarly, the secondary visual thalamic nucleus in mice (the lateral posterior nucleus, LP) has extensive connections with cortex. To resolve the precise connectivity of these circuits, we first mapped mouse visual cortical areas using intrinsic signal optical imaging and then injected fluorescently tagged retrograde tracers (cholera toxin subunit B) into retinotopically-matched locations in various combinations of seven different visual areas. We find that LP neurons representing matched regions in visual space but projecting to different extrastriate areas are found in different topographically organized zones, with few double-labeled cells (~4–6%). In addition, V1 and extrastriate visual areas received input from the ventrolateral part of the laterodorsal nucleus of the thalamus (LDVL). These observations indicate that the thalamus provides topographically organized circuits to each mouse visual area and raise new questions about the contributions from LP and LDVL to cortical activity.

Graphical Abstract

To resolve the organization of connections from the visual thalamus (LP) to visual cortical areas in the mouse, we mapped mouse visual cortical areas using intrinsic signal optical imaging and then injected fluorescently tagged retrograde tracers into retinotopically-matched locations in various combinations of seven different visual areas. We find that LP neurons representing matched regions in visual space but projecting to different extrastriate areas are found in different topographically organized zones. These observations indicate that the thalamus provides topographically organized circuits to each mouse visual area and raise new questions about the contributions from LP to cortical activity.

Introduction

While the classic description of the visual system depicts information in parallel streams traveling up a hierarchy of cortical areas, this illustration is complicated by the fact that there is a tremendous amount of feedback to the lateral geniculate nucleus (LGN), as well as to a higher order thalamic nucleus, the pulvinar (Felleman and Van Essen, 1991). While the LGN mainly projects to just the primary visual cortex (V1), primate pulvinar is a relatively large structure that is extensively connected with various areas of visual cortex and is thought to be involved in a variety of behaviors, such as visual attention (specifically, ignoring distractors) as well as visuomotor transduction (Jones, 1985; Robinson and Petersen, 1992; Grieve et al., 2000; Saalmann and Kastner, 2011; Baldwin et al., 2017). Studies of the locations of primate pulvinar neurons projecting to different visual cortical areas demonstrate systematic topography that is orthogonal to retinoptopic maps (Shipp, 2001, 2003). Furthermore, the pulvinar also receives visual information from the superior colliculus (Stepniewska et al., 2000; Nassi and Callaway, 2009). Still, our understanding of the contributions of each of these pathways to responses in subcortical and cortical areas and to perception is limited. Investigating pulvinar function would be greatly aided by a more precise map of its connections in the mouse, where the use of genetic and viral targeting strategies and manipulations are more readily applied.

In cats and rodents, the lateral posterior nucleus of the thalamus (LP) is considered the homolog of primate pulvinar based on its reciprocal connectivity with visual, parietal, and frontal cortex and abundant input from the superior colliculus (Lent, 1982; Raczkowski and Rosenquist, 1983; Baldwin et al., 2011; Nakamura et al., 2015). In rats, hamsters, and squirrels, LP can be divided into several subdivisions (LPLR, LPLC, LPCM, LPMR), based on its cytoarchitecture and connectivity to cortical and subcortical areas (Takahashi, 1985; Abramson and Chalupa, 1988; Sukekawa, 1988; Conte et al., 2008; Kamishina et al., 2008, 2009; Reep and Corwin, 2009; Baldwin et al., 2011; Nakamura et al., 2015).

Furthermore, LP/pulvinar in rodents, cats, and primates is retinotopically organized, with different subdivisions projecting to retinotopically matched regions of cortex (Le Gros Clark, 1932; Mason, 1978; Shipp, 2003; Tohmi et al., 2014; Bennett et al., 2019). Available evidence indicates that in the mouse, horizontal retinotopy (azimuth) is represented along the medio-lateral axis of mouse LP (Roth et al., 2015; Allen et al., 2016), while elevation is mapped along the anterior-posterior axis and may consist of two separate mirror-image maps with biases in the inputs to and outputs from the anterior and posterior maps (Bennett et al., 2019). Nevertheless, the central region of LP represents retinotopic locations that overlap between the two maps (Tohmi et al., 2014; Bennett et al., 2019) and may contain neurons projecting to V1 and/or any of the extrastriate cortical areas. The few published accounts of single cell reconstructions suggest that LP neurons often project to more than one cortical region but are likely to be limited in the numbers of visual areas to which they project (Nakamura et al., 2015).

The observation that mouse visual areas have biases in the retinotopic locations they represent (Garrett et al., 2014; Zhuang et al., 2017) raises new questions about the relationships between retinotopy and projections from LP to each region. In particular, despite known biases in the visual field representations of different extrastriate areas, all areas include a representation with overlap (Garrett et al., 2014). We sought to determine whether there is topographic separation of neurons that project to different extrastriate areas from zones that have retinotopic overlap, as has been previously described for primates (Shipp, 2001). Furthermore, do neurons within the region of mouse LP that represents common visual field locations project indiscriminately to all areas, or might they have more organized projections?

To address these and other questions, we used retinotopic mapping with intrinsic signal imaging to target cortical injections of retrograde tracers to retinotopically matched locations within multiple extrastriate cortical areas of single animals; matching of retinotopic injections was further verified based on overlapping retrograde label in V1, which contains a nearly complete retinotopic map (Garrett et al., 2014; Zhuang et al., 2017). We find that projections to extrastriate areas reveal a third dorso-ventral topographic axis within LP, roughly orthogonal to the A-P and M-L axis representations of visual space. LP neurons located more dorsally project to more posterior visual cortical areas, while more ventral neurons project to more anterior areas; these topographic relationships are gradual and systematic such that intermediate visual cortical areas receive input from intermediate LP locations. Furthermore, even within regions of overlap, very few LP neurons are double-labeled, suggesting that they make relatively dedicated cortical projections. These data shed light on whether LP serves to pass information up the cortical hierarchy, or rather acts as a loop within various visual areas and contribute to growing evidence providing a framework against which various theories of thalamocortical circuit function can be evaluated (Olshausen et al., 1993; Crick and Koch, 1998; Sherman and Guillery, 1998; Grieve et al., 2000).

In addition to LP, the lateral dorsal nucleus of the thalamus (LD) is yet another “higher order” thalamic nucleus that has connections to various pathways including visual, limbic, and somatosensory (Robertson and Kaitz, 1981; Jones, 1985; Takahashi, 1985; Bezdudnaya and Keller, 2008). In the rat, the ventral lateral part of this nucleus (LDVL) receives input from both primary and secondary visual cortex, while the medial portion does not (Takahashi, 1985). Beyond this tracing study, there is remarkably little research on involvement of this relatively large nucleus in visual circuits. We find that like LP, LD also projects to all visual cortical areas in the mouse.

Materials and Methods

Headframe implantation and intrinsic signal optical imaging

All experimental procedures using live animals followed procedures approved by the Salk Institute Animal Care and Use Committee. C57BL/6 mice (RRID:MGI:5656552) (5–14 weeks old, 2 mice less than 6 weeks old) were prepared for imaging by implanting a custom metal headframe as previously described (Figure 1; Marshel et al., 2011; Garrett et al., 2014; Juavinett et al., 2017). These headframes remained on the mouse for the remainder of the experiment. Before injection, a map of visual cortex was obtained for each mouse using intrinsic signal optical imaging as previously described (Sereno et al.; Garrett et al., 2014; Juavinett et al., 2017). Using this mapping protocol, we were able to reliably target 7 different visual areas (V1, LM, RL, AL, AM, PM, and P/POR). Although visual area A receives inputs from V1, it is not retinotopically mapped and was therefore not targeted in our study (Wang and Burkhalter, 2007; Garrett et al., 2014).

Figure 1. Identification of visual areas for AF-CTB injection.

(a) Example azimuth retinotopic map. (b) Example altitude retinotopic map. (c) Visual field sign map, computed from the retinotopic maps in (a). (e) Outline of identified visual areas. Blue areas have a mirror representation of visual space, red areas have a non-mirror representation. It is often difficult to obtain retinotopic maps from P/POR, in the far posterolateral corner of the window—often P/POR injections were guided by injecting posterior to LM and lateral to V1. (e) Visual area borders overlaid on the corresponding blood vessel image acquired with a green light over the intact skull. (f) Image through the skull taken shortly after AF-CTB injection (white arrowheads) into LM (647), AM (488), and PM (555). Even through the skull, blood vessels and AF-CTB spread are visible. All figures are from Animal 3–8 (Table 1). All scale bars are 500 μm.

Cholera Toxin Subunit B injection

To trace inputs to visual cortical areas from the thalamus, we used 0.5% in filtered PB of Cholera Toxin Subunit B (CTB) conjugated to Alexa-Fluor (AF) 488, 555, or 647 (Life Technologies; AF-CTB). We found that the different AF-CTB performed comparably. The choice of AF for each injection was random and is well-balanced across animals (Table 1). Mice were anesthetized with 100 mg/kg of ketamine and 10 mg/kg xylazine cocktail via intra-peritoneal injections and mounted in a stereotax (David Kopf Instruments Model 940 series, Tujunga, CA) for surgery and stereotaxic injections. To expose the brain for injection, either a burr hole was drilled or a craniotomy was performed over visual cortex. Injections were done with air pressure using a picospritzer (General Valve Corp, Fairfield, NJ). To prevent backflow, the pipette was left in the brain for 5 minutes before and after injection.

Table 1.

Overview of all animals injected, organized by animals with three, two, or one successful injection(s) of AF-CTB. In Mouse ID, the first number is the total number of injections in that animal, the second number is the animal index. Entries indicate cortical areas injected in columns indicating the type of AF-CTB that was injected.

Mouse ID	488	555	647
3–1	RL	V1	LM
3–2	RL	LM	V1
3–3	AL	LM	V1
3–4	AM	RL	LM
3–5	PM	RL	P/POR
3–6	P/POR	PM	V1
3–7	PM	P/POR	LM
3–8	AM	PM	LM
3–9	PM	LM	AL
2–1	V1	PM	-
2–2	RL	-	AL
2–3	PM	-	AL
1–1	-	-	AL
1–2	-	-	RL
1–3	-	-	AL
1–4	-	-	AM
1–5	-	-	AM
1–6	LM	-	-

AF-CTB was injected into visual cortical areas based on functional retinotopic maps overlaid on the blood vessel pattern, using these blood vessels as landmarks (Figure 1). When multiple areas were injected, the injections were targeted to matched retinotopic locations. AF-CTB pressure injections into cortex were 20–50 nl for AL, AM, PM, and P/POR, and 40–100 nl for V1, LM and RL, with 20–40 PSI and a 20–30 μm diameter pipette (Conte et al., 2009). These conditions were optimal in order to restrict CTB to <300 μm diameter injection sites, a necessary condition for small visual areas. After recovery, mice were given water with ibuprofen (Perrigo Oral Suspension, for dose of 30 mg/kg) and housed for 4–6 days before tissue harvesting.

Histology

Brains were harvested after transcardial perfusion using phosphate-buffered saline (PBS) followed by 4% paraformaldehyde (PFA). Brains were dissected out from skulls and post-fixed with 2% PFA and 15% sucrose in PBS at 4°C overnight, and then immersed in 30% sucrose in PBS at 4°C before sectioning. Using a freezing microtome, 50 μm coronal brain sections were cut and stored in PBS with 0.01% sodium azide at 4°C. Sections were mounted on slides with Polyvinyl alcohol mounting medium containing DABCO and allowed to air-dry.

Image acquisition and processing

Each section was imaged at 10x using an Olympus BX63 Microscope with parameters adjusted based on the intensity of expression and background fluorescence. The injection sites for each animal were measured to ensure restriction to small cortical visual areas. For LM and RL injections, animals with injection site diameters greater than 500 μm for LM or RL were excluded. For AL, AM, PM, and P/POR, animals with injection site diameters greater than 300 μm were excluded.

To compute average expression patterns in LP (Figure 4B), sections at −1.9 and −2.3 AP for each animal were aligned to the Paxinos Mouse Atlas using the Landmark Correspondences macro in Fiji (Schindelin et al., 2012; Paxinos and Franklin, 2013, https://imagej.net/Landmark_Correspondences). Borders indicating the extent of expression in the thalamus were manually drawn for each animal and overlaid on the Paxinos and Franklin atlas outlines. Thalamic divisions and subcortical area borders were drawn manually, using the atlas and cell densities (as marked by DAPI) as a guide.

Figure 4.

Single injection cases and alignment to common atlas. (a) Example labeling pattern of input into each visual area tested. Scale bar = 500 um. (b) By aligning each thalamic section to a common atlas (see Materials & Methods), we can begin to see trends across animals.

To determine the number of cells projecting to multiple visual areas, we imaged the sections with at least two different AF-CTB types of retrogradely labeled cells at 20x using a Zeiss LSM780 confocal microscope. For each brain with multiple successful AF-CTB injections, every fourth section was imaged (~200 microns apart, 4–10 sections per mouse) and all cells in zones of LP with multiple tracers were manually counted. Because almost all of the regions with intermingled cells were in LPMR (see Figures 2 and 5 for illustration), these counts largely reflect the cells in LPMR, and on the LPLR/LPMR border. Percentages reported are the number of cells double-labeled with two different tracers divided by sum of the numbers of cells labeled by each of the same two tracers in LP.

Figure 2.

The injection of three different retrograde tracers into one mouse demonstrates robust label in the thalamus and overlapping label in V1. (a) Schematic of experiment. Three different colors of AF-CTB were injected into three different functionally-identified cortical areas. (b) Coronal brain section showing injection into LM (AF-CTB 647). (c) Coronal brain section showing injections into RL (red, AF-CTB 555) and AM (green, AF-CTB 488), as well as retrogradely labeled areas of the thalamus and cortex. (d) Confocal image of the thalamus, demonstrating non-overlapping label in LPLR, and intermingled but non-overlapping label in LPMR. (e) Confocal image of V1, demonstrating widespread retrograde label, with overlapping regions of cells (few of which are double-labeled). Abbreviations: AF-CTB, alexa fluor conjugated cholera toxin B; IMA, intramedullary thalamic area; dLGN, dorsal lateral geniculate nucleus; eml, external medullary lamina; Po, posterior thalamic nucleus; LPLR, laterorostral lateral posterior nucleus; LPMR, mediorostral lateral posterior nucleus; AM, anteromedial visual area; LM, lateromedial visual area; RL, rostrolateral visual area. Data from Animal 3–4. All scale bars are 500 μm.

Figure 5.

Percentages of LP cells projecting to multiple visual areas. (a) Overview of sample experiment (Animal 3–8) with three different tracer injections into PM, LM, and AM, demonstrating very little overlap of projection zones. (b) Low resolution image to demonstrate locations of cells projecting to PM, LM, and AM; field of view is marked by grey box in (a). (c) High resolution confocal images of LP showing cells projecting to PM, LM, and AM; field of view is marked by gray box in (b), white arrows point to rare double-labeled cells. (d) Sample experiment (Animal 3–1), demonstrating very little overlap of LP projections to V1, LM, and RL. (e) Low resolution image to demonstrate retrogradely labeled cells from V1, LM, and RL; field of view is marked by box in (e). (f) High resolution confocal images of LP showing cells projecting to V1, LM, and RL; field of few is marked by grey box in (e); white arrows point to rare double-labeled cells. (g) Quantification of cells that project to multiple visual areas. Percentages shown are # overlapping cells / total # cells in LP. Most combinations are based on a single animal and the following combinations included 2 animals: V1/LM, V1/PM, LM/RL, LM/AM, LM/PM. (h) Average percent overlap for each visual area with all other visual areas. Error bars are SEM. All measures of percentages of double-labeled cells are based on the total numbers of cells counted from either one or two animals that included the relevant combinations. (i) Total numbers of cell counts and animals (anim; in parentheses) for each visual area combination.

Experimental Design & Statistical Analysis

All mice used are detailed in Table 1. Mice used in this study (n=19 total) were all C57BL/6, both male and female, and aged from 5–14 weeks. With seven different targeted visual areas, >3 injections per area in different mice ensures reliability of our results. For analysis of double-labeled cells, we sectioned each brain at 50 μm and counted cells in every fourth section (additional details about related methods can be found in Histology and Image acquisition and processing). Any computed statistic (i.e. percentages in Figures 5G,H & 6F) are explained in the figure legends. We did not conduct any statistical analyses in this paper.

Figure 6.

V1 and extrastriate cortical areas receive input from LDVL. (a) Schematic of brain section, modified from Paxinos & Franklin. (b) Triple injection brains demonstrating Inputs from LDVL and LPMR. V1, RL, AM, and PM injections consistently result in substantial label in LDVL. LM, AL, and POR injections result in weaker and inconsistent label in LDVL. (c) Additional brain confirming LDVL projections to AM. (d) Additional brain demonstrating LDVL projections to V1 and PM. (e) Additional brain demonstrating LDVL projections to PM and AL. (f) Percent of brains with retrogradely labeled cells in LDVL for each injected cortical visual area.

Results

We investigated the organization of thalamocortical projections by functionally mapping visual cortex and then injecting retrograde tracers into seven different visual areas. We mapped the retinotopic organization of the visual cortex in each mouse using intrinsic signal imaging (Figure 1A–C; Juavinett et al., 2017). Using these maps, we automatically constructed borders between visual areas (Figure 1D & 1E; Garrett et al., 2014). We then used simultaneously imaged surface blood vessel patterns to target injections of an efficient retrogradely-transported neuroanatomical tracer, Cholera Toxin Subunit B, conjugated to various Alexa fluorophores (Figure 1F; AF-CTB).

We injected 18 mice with various combinations of one, two, or three different AF-CTB tracers into seven different visual cortical areas, giving 39 total injections (Table 1). When multiple tracers were injected they were injected at retinotopically matched locations, as verified by overlapping label within V1 (see below). This approach allowed us to directly compare the thalamic zones that provide input to different cortical areas from retinotopically matched locations in the same mice. It also allowed quantification of the proportions of thalamic neurons that project to multiple visual cortical areas.

We were able to achieve robust, and reliable labeling patterns using AF-CTB. Figure 2 shows an example of an experiment in which three different colors of AF-CTB (488, 555, 647) were injected into three different visual cortical areas (AM, RL, LM, respectively) in the same mouse (Figure 2A,B,C). Each visual area received input from one or multiple patches in the thalamus (Figure 2C,D). In this example, LM, AM, and RL received input from separate patches in LPLR. LM predominately received input from the area of LPLR closest to the dLGN, while the more anterior cortical area RL received input from ventral LPLR, bordering on Po. AM received input from the medial part of LPLR. Each of these areas also received relatively weaker and more widespread input from LPMR (Figure 2D).

For mice with multiple injections, we confirmed the spatial overlap of cells in V1 to assure that the injection sites were in overlapping retinotopic locations (Figure 2E). Thus, any differences in the topographic locations of labeled thalamic neurons must be attributed primarily to differences related to the areas that the thalamic cells project to rather than retinotopy. Given the small size of extrastriate cortical areas and the retinotopic extent of label observed in V1, we are confident that each AF-CTB injection here covered a large portion of visual space. While labeled cells in V1 are found in both overlapping and non-overlapping zones, as expected from the biases in visual field representations of the different extrastriate areas, the presence of overlap indicates that we successfully injected locations with overlapping retinotopic locations. Few cells in these overlapping V1 areas were double-labeled.

It is noteworthy that small differences in the V1 locations with the most dense retrograde label correspond to retinotopic distances that would predict only extremely small distances across the corresponding retinotopy in LP. For example, despite extensive overlap of AM-projecting (green) and RL-projecting (red) V1 neurons illustrated in Figure 2E, the locations of densest label are separated by about 200 microns along the mediolateral axis. Retinotopic mapping of LP projections to V1 reveals that neurons separated by 1mm medio-laterally in V1 receive input from LP neurons separated by only 100 microns (Roth et al., 2015). Thus, if the LP neurons projecting to AM versus RL in our Figure 2D were separated in position solely due to retinotopic differences in injection location, they should be separated by only about 20 microns. This is far less than the roughly 1 mm center-to-center spacing observed (Figure 2D), arguing that the different locations of LP neurons projecting to AM versus RL are not due to retinotopic mapping within LP.

Across the nine mice for which three successful cortical injections were made and three mice with two injections, we see similar patterns of retrograde labeling for each visual area (Figure 3). Within animals, we find that well-segregated zones of LP project to different visual areas for every combination of areas examined. Furthermore, there is a systematic relationship between LP projection zones and cortical areas that is conserved across animals (summarized schematically in Figure 7). As a general rule, the more posterior visual cortical areas receive input from more dorsal locations in LP, while more anterior cortical areas receive input from more ventral locations in LP. Across the mediolateral cortical axis, more lateral cortical areas receive input from more lateral LP and medial from medial. These trends are seen most clearly in the triple and double injection cases where, within each animal, the same topographic relationships are invariably observed (Figure 3) despite some variability between animals in the precise alignment to anatomical boundaries defined by the Paxinos and Franklin Atlas (Figure 4; Paxinos & Franklin, 2013).

Figure 3.

Brains with multiple injections demonstrate systematic organization of inputs from thalamus to visual cortical areas. (a) Schematic of experiment and atlas section (modified from Paxinos & Franklin, Bregma −2.3 mm). (b) Nine experiments with three injections of AF-CTB into each animal. (c) Additional experiments with two injections to further demonstrate relationships between projection regions for V1, RL, AL, and PM.

Figure 7.

Overview schematic demonstrating corresponding topographic mapping between visual cortex and LP. (a) The visual cortex of the mouse; areas investigated in the present study are colored. (b) Schematic of projection zones in the mouse thalamus colored according to the cortical area to which they project. The thalamic organization is flipped relative to cortex, with anterior visual cortex represented in the ventral thalamus, while posterior visual cortex is represented in the dorsal thalamus. Lateral and medial organization is conserved. It should be noted that this schematic reflects most of, but not all, projections. Retrograde label was seen in LP beyond the borders denoted here. This schematic is meant to reflect the bulk of these cells and is based on the maximum intensity of the retrograde label (see Fig 4b and Methods). (c) Given the systematic differences in the locations of LP cells projecting to different cortical areas, our low observed proportions of double retrograde labeling, and previous reports of cells in LP with projections to 2–3 locations, we suggest a scheme in which many cells in LP project to multiple cortical areas. However, due to the many possible combinations of visual areas, few cells are double-labeled by any particular combination. In this schematic example, each LP neuron projects to just 2 areas, as indicated by the color coding for each cell. Most cells projecting to a particular location are grouped together but they also have a secondary projection to one of the other areas.

For example, consider that the centers of the most lateral visual areas (including V1), progressing from posterior to anterior are P/POR, LM, V1, AL, and RL. Label within LP from P/POR injections (Figure 3 panels B7-B9) is located dorsal to label from injections in LM (B8), V1 (B9), and RL (B7). Similarly, LP label from LM injections is consistently more dorsal than that from injections in all other areas (except P/POR) and most notably is consistently just slightly more dorsal than LP label from V1 injections (Figure 3, panels B1, B3). Injections in the next most anterior cortical area, AL, yield LP label that is more ventral than from injections in LM (B3, B6) or V1 (B3), but more dorsal than label from an injection in RL. Injections in RL consistently result in the most ventral LP label (B1, B2, C1), most notably in a direct comparison, more ventral than label following injection in AL (C1).

On the medial side of V1, cortical injections were made into areas AM and PM. LP label following injections in area PM was consistently located more medially than label from injections in any other cortical area (see Figure 3, panels B5–6, C2–3). Furthermore, when AM and PM were both injected in the same animal (B5), label in AM is located more ventrally and laterally, consistent with the general trend for more anterior areas to receive more ventral LP input and for more medial cortical areas to receive more medial LP input. (This medial-lateral relationship is most pronounced if we consider a skewed “AP-axis” extending along the representation of the vertical meridian at the lateral edge of V1.)

Despite the clear and consistent within-animal relationships between LP projection zones and recipient cortical areas, there is scatter in the positions of LP cell clusters from same area injections aligned to the Atlas across different animals. Thus, when single injection cases are aligned to the atlas (Figure 4B) the general overall topographic trends can be observed but they are not as clear. Similar between-animal trends are also seen in the multi-injection cases (Figure 3). This variability might reflect true biological variability, such as the inability of a single idealized atlas to accurately map to animals with differences in brain structure, or systematic between animal shifts in the topographic locations of LP cells projecting to each cortical area. Alternatively, these differences might reflect experimental variability such as the plane of brain sectioning or choice of alignment between finely spaced brain sections (50 microns) and more intermittent sampling in the atlas (120 microns). Despite scatter between cases and resulting overlap in projection zones for thalamic neurons projecting to some visual areas, there was still clear segregation between LM and PM projecting neurons, with PM projecting neurons located in very anterior/medial LP, and LM projecting neurons in lateral/posterior LP. RL, AM, and AL all had two patches of input in the thalamus, divided by the external medullary lamina (eml), perhaps reflecting dual retinotopic maps within LP (Bennett et al., 2019).

While there were clear trends in the relative topographic locations of LP neurons projecting to different cortical areas, there was also clear intermingling of cells (Figures 2–4), particularly for cases involving multiple injections into neighboring cortical areas. Do these thalamic neurons nevertheless represent dedicated projection channels or do some cells project to more than one cortical area? We further analyzed cases with 2–3 different AF-CTBs into each mouse, to quantify the percent of overlap between projections to different visual areas (Figure 5). For this purpose, we analyzed sections with overlap between retrogradely labeled areas. There were very few cells that were double-labeled (~3–6% of all retrogradely labeled cells in LP; Figure 5G and 5H). Across the other visual areas, P/POR had the highest overlap (5.5%), followed by PM and V1 (5.0%). The weakest overlap was between AL and RL, with only 2.0% of all cells double-labeled. The strongest overlap was between V1 and P/POR, with 7.8% of cells double-labeled. Only zones with overlapping cells were counted in order to rule out the possibility that a lack of dual projections could be due to differences in retinotopic locations of injection in different cortical areas. Insofar as there is retinotopy in the thalamus, the quantified neurons must be from matched retinotopic locations where the cortical injections had overlapping retinotopy.

In addition, the ventrolateral part of the laterodorsal nucleus (LDVL) had significant retrograde label from extrastriate areas. Injections in V1, AL, RL, and PM consistently resulted in label in LDVL (Figure 6). LM, AM, and P/POR had inconsistent label in LDVL: 4 of 8 LM injections, 3 of 4 AM injections, and 2 of 4 P/POR injections had visible label in LDVL. In all cases in which label was visible in LDVL, close inspection across multiple brain sections revealed that the group of labeled neurons was clearly separated from and not contiguous with the label in LP. We did not see label in the dorsomedial part of the laterodorsal nucleus (LDDM) in any mouse.

Discussion

Using classic tracing techniques combined with intrinsic signal imaging, here we show that the mouse lateral posterior nucleus (LP) is topographically organized, with neurons projecting to different visual areas being located predominantly in different locations. This organization is orthogonal to retinotopic mapping within LP. Furthermore, the ventrolateral part of the laterodorsal thalamic nucleus (LDVL) projects to multiple visual cortical areas. As a comprehensive look at the organization of this underappreciated nucleus, this study lays important groundwork for future investigations into LP function in mice and provides a framework to test theories of thalamocortical circuit organization.

Extrastriate areas in the mouse are known to have biases in retinotopic field coverage (Garrett et al., 2017; Zhuang et al, 2018). We designed experiments to independently assess whether there might be a clear topography of LP projections, in addition to the known retinotopy. We find that in addition to the retinotopic maps along the A-P and M-L axes of LP (Roth et al., 2015; Bennett et al., 2019), neurons projecting to the more anterior visual areas (e.g. RL) are located more ventrally than those projecting to more posterior visual areas (e.g. P/POR; Figure 7). These patterns cannot be explained by retinotopy within LP, because they occur on a much larger spatial scale: previous studies of LP inputs to different locations within mouse V1 show that there is retinotopy within a 500 μm zone of LP (Roth et al., 2015) whereas our projection zones are often more than 500 μm apart. V1 injections that were 500 μm apart resulted in ~100 μm spread of LP label, suggesting a five-fold compression of the map in LP relative to V1 (Roth et al., 2015). It should also be noted that a topographic organization similar to what we observe in mice has also been observed in non-human primates (Stepniewska et al., 1999; Shipp, 2001, 2003; Kaas and Lyon, 2007; Baldwin et al., 2017). We speculate that this is a common feature of connections between pulvinar and cortex and would be observed in all species if suitable methods were employed. It is likely to be more readily apparent from comparisons of single injections into different animals in primates, where pulvinar is typically much larger than in rodents. The use of retinotopic cortical mapping and retinotopically matched injections into multiple cortical areas is likely to facilitate these observations in animals with smaller brains, as we have demonstrated here.

Analyses of retrograde labeling following tracer injections into two or three visual areas revealed few double labeled cells, indicating that most cells do not project to the particular combinations of areas that were injected. Taken at face value, this could be interpreted to indicate that few cells project to multiple areas. But reconstructions of single thalamic neurons reveal a range of anatomical cell types including some that have very focused cortical projections and others with more widespread projections (Clascá et al., 2012). An alternative scenario, suggested by these observations and by limited single cell reconstructions in mice (Nakamura et al., 2015; Janelia MouseLight Project) would be that LP cells project to more than one cortical region but to only a very limited subset of visual cortical areas. Or cells might project to just one of the visual areas that we sampled plus an additional area, such as the restrosplenial cortex, which we did not inject. For example, suppose that all of the LP neurons located in a particular topographic zone have a high probability of projecting to a corresponding cortical location and additionally project to just one of nine or more other possible cortical targets (Figure 7c). With such an organization, there would be very few double labeled cells resulting from any combination of two retrograde tracer injection locations, as we have observed, but single cell reconstructions would reveal multiple projections from single neurons (Nakamura et al., 2015; Janelia MouseLight Project; http://ml-neuronbrowser.janelia.org). A similar organization has been demonstrated for cortico-cortical projections of mouse V1 neurons (Han et al., 2018). In view of our observation that neurons in different topographic locations within LP have very strong biases in the visual cortical areas to which they project, we favor this interpretation.

Understanding the anatomy of thalamocortical projections can lend criticism or credit to various theories about thalamocortical function. A common label for LP/pulvinar is that of a “gate” or “router” for information as it travels through cortex, perhaps for the purposes of synchronization or multisensory processing (Olshausen et al., 1993; Crick and Koch, 1998; Shipp, 2003; Cappe et al., 2009; Ahissar and Oram, 2013). If single neurons in LP acted to synchronize disparate areas of cortex, we would expect them to project to multiple regions. Other theories of thalamic function require more precise anatomical channels, for instance to selectively route information through the cortical hierarchy (Sherman and Guillery, 2002). If LP is a router for information up the visual hierarchy, single cells in LP should receive input from cortical areas and specifically route them to higher-order areas. While our data provide new information that would constrain the precise way that these functions are implemented, they do not exclude or validate any of these hypotheses. As described above, we favor the interpretation that these circuits do not fit into either extreme model but instead have aspects that would allow them to flexibly facilitate both interactions across cortical areas and more directed routing of information.

In primate thalamus, there are two different sets of thalamocortical connections: cells in the “core” which express parvalbumin and project to specific areas of cortex, and cells in the “matrix”, which express calbindin and project broadly to superficial cortex (Jones, 1998). These two projection types, one targeted and one broad, could enable different modes of thalamic functioning, to synchronize vast areas of cortex and/or specifically route information (Jones, 1998, 2001). While we cannot distinguish between these core and matrix populations with our methods, the small proportion of double-labeled cells in our data suggests that the thalamus is poised to perform both functions – a limited number of cells may send broader information, while other cells may specifically project to only one or to a more limited set of other cortical regions. It is possible that the majority of double-labeled cells that we observed are matrix cells, but additional experiments would be needed to confirm this hypothesis.

This distinction between core and matrix pathways may be the necessary anatomical substrate for setting necessary bounds on corticothalamocortical loops, also known as “the no-strong-loops” hypothesis. This framework posits that no two brain areas can have reciprocal excitatory, driving connections (Crick and Koch, 1998). LP inputs and outputs are segregated between layers of cortex, with the exception of possible overlap in layer 5 (L5): LP receives input from L5 of cortex, a putative driving connection (Bourassa and Deschênes, 1995; Kim et al., 2015), and projects to L5 and L6 of V1 and L4, L5, and L6 of extrastriate cortex (Roth et al., 2016). According to the no-strong-loops hypothesis, LP cells receiving this driving cortical input from L5 should not project back to L5 of the same cortical region. Several visual areas, including RL, AL, and AM, receive input from two separate patches in LP. It is possible that these are separate sets of cells that receive driving versus modulatory input, that in return project to different layers or cortical areas, however this needs to be directly tested. It is also possible that these patches are actually one functional group of cells that is anatomically split by the external medullary lamina.

Furthermore, our study provides the first published evidence that the ventral part of the laterodorsal nucleus (LDVL) provides significant input to V1, AL, RL, AM, and PM, and perhaps also LM and P/POR. LDVL is known to project to cingulate, retrosplenial, and posterior parietal cortex (Jones and Leavitt, 1974; Robertson, 1977; Spiro et al., 1980). With such diffuse inputs and outputs, there is wide speculation on the functional role of LD (Bezdudnaya and Keller, 2008). Previous anatomical reports combined with our data suggest that there is a path from visual cortex through LDVL to various limbic and associative areas (Jones and Leavitt, 1974; Robertson and Kaitz, 1981). This pathway may be important for the inclusion of visual input in spatial, emotional, or decision-making behavior (Poort et al., 2015).

A critical question remains: what are the inputs to cells in LP or LDVL with projections to different visual cortical areas? Connecting our observations here to previous studies of the input zones of LP can inform our expectations based on the spatial overlap of inputs and outputs (Zhou et al., 2017). In cats, LPLR receives input from V1, while LPMR receives input from SC (Chalupa and Abramson, 1989). In rats, the distinction between V1 and SC recipient regions lies along the rostral-caudal axis: rostral LP receives cortical input, whereas caudal LP receives tectal input (Li et al., 2003). A recent study in mice suggests that LPLR receives input from one class of SC cells (Gale and Murphy, 2014), although there is other evidence that LPMR also receives input from SC (Allen Brain Institute Connectivity Atlas) as well as retinal ganglion cells (Allen et al., 2016). Future experiments utilizing retrograde viruses along with conditional transynaptic tracing viruses (Luo et al., 2018) will hopefully provide satisfying answers to these questions.