Direct Parabrachial–Cortical Connectivity

Abstract

The parabrachial nucleus (PB) in the upper brain stem tegmentum includes several neuronal subpopulations with a wide variety of connections and functions. A subpopulation of PB neurons projects axons directly to the cerebral cortex, and limbic areas of the cerebral cortex send a return projection directly to the PB. We used retrograde and Cre-dependent anterograde tracing to identify genetic markers and characterize this PB–cortical interconnectivity in mice. Cortical projections originate from glutamatergic PB neurons that contain Lmx1b (81%), estrogen receptor alpha (26%), and Satb2 (20%), plus mRNA for the neuropeptides cholecystokinin (Cck, 48%) and calcitonin gene-related peptide (Calca, 13%), with minimal contribution from FoxP2+ PB neurons (2%). Axons from the PB produce an extensive terminal field in an unmyelinated region of the insular cortex, extending caudally into the entorhinal cortex, and arcing rostrally through the dorsolateral prefrontal cortex, with a secondary terminal field in the medial prefrontal cortex. In return, layer 5 neurons in the insular cortex and other prefrontal areas, along with a dense cluster of cells dorsal to the claustrum, send a descending projection to subregions of the PB that contain cortically projecting neurons. This information forms the neuroanatomical basis for testing PB–cortical interconnectivity in arousal and interoception.

Introduction

The parabrachial nucleus (PB) is a large complex of neurons surrounding the superior cerebellar peduncle in the upper pons and lower midbrain tegmentum. Several lines of evidence suggest that this brain stem region may be critical for maintaining consciousness. For example, some PB neurons increase their activity during wakefulness relative to sleep (Saito et al. 1977), and stimulating PB neurons promotes wakefulness (Muindi et al. 2016; Qiu et al. 2016; Kaur et al. 2017; Luo et al. 2018; Wang et al. 2019). Further, microinjecting an anesthetic drug in this region in rats reliably induces loss of consciousness (Devor and Zalkind 2001), and lesions here can produce coma (Fuller et al. 2011). In humans, this brain stem region is the only site in the nervous system where focal lesions cause coma (Parvizi and Damasio 2003; Fischer et al. 2016). The human PB region exhibits resting-state fMRI connectivity with two limbic regions of the cerebral cortex: the anterior insular (frontoinsular) cortex and the pregenual anterior cingulate cortex (Fischer et al. 2016). In rodents, homologous cortical areas receive axonal projections directly from the PB (Saper and Loewy 1980; Lasiter et al. 1982; Shipley and Sanders 1982; Saper 1982a; Fulwiler and Saper 1984a).

It is remarkable that any neurons within the PB project axons directly to the cerebral cortex, given that brain stem neurons that influence the cerebral cortex typically do so via synaptic relays in subcortical regions, like the thalamus or basal forebrain. Direct axonal projections from the brain stem to the cerebral cortex arise primarily from monoaminergic neurons in the locus coeruleus and midbrain raphe, which project diffusely to many brain regions, not to specific cortical areas (Jones et al. 1977; Azmitia and Segal 1978). To our knowledge, the PB is the only nonmonoaminergic brain stem nucleus that sends a substantial axonal projection directly to the cerebral cortex.

Within the PB, however, the identity of cortically projecting neurons remains unknown. This knowledge gap limits our ability to use cell-type–specific techniques to study the function of cortically projecting neurons, without also perturbing the wide variety of functions controlled by other neurons in the PB, which contains a tangle of subpopulations implicated in many different appetitive, autonomic, viscerosensory, and other interoceptive functions. This region of the brain stem was first studied due to its effects on breathing (Tang 1953; Chamberlin and Saper 1994), then gustatory signaling (Norgren and Leonard 1971), cardiovascular control (Chamberlin and Saper 1992), pain (Bernard and Besson 1990; Bernard and Besson 1995), thermoregulation (Nakamura and Morrison 2008; Nakamura and Morrison 2010), itch (Mu et al. 2017), thirst (Edwards and Johnson 1991), sodium appetite (Menani et al. 1996; Geerling and Loewy 2007), and hunger (Higgs and Cooper 1996; Wu et al. 2009; Geerling et al. 2010; Li et al. 2019).

Genetically targeted, cell-type–specific techniques have allowed recent investigators to match specific functions to subpopulations of PB neurons that express specific genetic markers. As examples, Calca-expressing PB neurons drive hypercapnic arousal (Kaur et al. 2017) and signal danger (Campos et al. 2018); subsets of Pdyn-expressing neurons activate in response to skin temperature (Geerling et al. 2016), pain (Hermanson, Telkov, et al. 1998) or sodium depletion (Lee et al. 2019); Oxtr-expressing neurons promote fluid satiation (Ryan et al. 2017); and rostral, Cck-expressing neurons influence the counterregulatory response to hypoglycemia (Garfield et al. 2014). With diverse, life-critical functions mediated by intermingled populations of neurons, it is often difficult to interpret the results of conventional experimental methods in this small brain stem region. Fortunately, the genetic diversity of PB neurons provides an opportunity to use conditional genetic targeting to selectively label, activate, inhibit, and record each class of neurons. Thus, learning the genetic identity of cortically projecting PB neurons would allow us to target and study their connectivity and functions.

Here, we use retrograde tracing in mice to identify cell-type–specific markers for cortically projecting PB neurons. We then use Cre-conditional, anterograde synaptic labeling to identify and compare the cortical distribution of projections from each PB subpopulation. Finally, we use anterograde and retrograde tracing to characterize the extensive, reciprocal projections from the cerebral cortex back to the PB. These approaches reveal massive, direct interconnectivity between the PB and limbic areas, spanning the full length of the cerebral cortex, from the frontal pole back to the entorhinal cortex.

Materials and Methods

Mice

All mice were group-housed in a temperature and humidity-controlled room with a 12-h light/dark cycle. They had free access to water and standard rodent chow (Envigo 7013). We used male and female mice from a mixed, predominantly C57BL6/J background along with several knockin-Cre and Cre-reporter strains (Table 1). All experiments were conducted in accordance with the guidelines of the Institutional Animal Care and Use Committees at the University of Iowa.

Table 1.

Cre driver and reporter mice used in this study

Strain	Reference	Source information	Key gene
Vglut2-IRES-Cre (Slc17a6-IRES-Cre)	Vong et al. (2011)	Jax 016963; https://www.jax.org/strain/016963	IRES-Cre inserted downstream of the stop codon of Slc17a6 on chromosome 7
Foxp2-IRES-Cre	Rousso et al. (2016)	Richard Palmiter lab, University of Washington (Jax 030541; https://www.jax.org/strain/030541)	IRES-Cre inserted after termination codon of the mouse Foxp2 gene
Pdyn-IRES-Cre	Krashes et al. (2014)	Jax 027958; https://www.jax.org/strain/027958	IRES-Cre inserted downstream of the endogenous Pdyn (prodynorphin) gene
Cck-IRES-Cre	Taniguchi et al. (2011)	Jax 012706; https://www.jax.org/strain/012706	IRES-Cre inserted in the 3’–untranslated region (3–UTR) of the cholecystokinin locus
Esr1-2A-Cre	Lee et al. (2014)	Jax 017911; https://www.jax.org/strain/017911	attP-flanked 2A oligopeptide and Cre recombinase gene in the 3–UTR of the estrogen receptor 1 alpha gene (Esr1 or ERα)
Th-IRES-Cre	Lindeberg et al. (2004)	EMMA 00254; https://www.infrafrontier.eu/	The Cre recombinase was cloned following an IRES sequence. An frt-flanked neomycin selection cassette was added and the construct cloned in the 3′ untranslated end of the TH gene
Calca-Cre (Calca-tm1.1-Cre-EGFP)	Carter et al. (2013)	Richard Palmiter, University of Washington (shared by Andrew Russo, University of Iowa)	The targeting vector used for homologous recombination contained a cre/GFP fusion sequence (cre fused in frame with EGFP at the C-terminus), N-terminal nuclear localization signal and myc tag, and frt-flanked PGK-neo marker. Cre was inserted at the initiation ATG codon in exon 2 of the gene. Chimeric animals were bred with Flper mice to remove the PGK-neo selection marker
R26-LSL-L10GFP reporter	(Krashes et al. 2014)	Available from originating investigators; http://www.informatics.jax.org/allele/MGI:5559562	Floxed transcription STOP cassette followed by EGFP/Rpl10 fusion reporter gene under control of the CAG promoter targeted to the Gt(ROSA)26Sor locus

Stereotaxic Injections

Mice were anesthetized with isoflurane (0.5–2.0%) and placed in a stereotactic frame (Kopf 1900 or 940). We made a midline incision and retracted the skin to expose the bregma and the skull over the brain. Through a pulled glass micropipette (20–30–μm inner diameter), we injected Fluorogold (FG), cholera toxin subunit B (CTb), or an AAV (Table 2). Most injection volumes were in the range of 20–100 nL, but some in the insular cortex were as large as 400 nL. Each injection was made over a 5-min period using picoliter air puffs through a solenoid valve (Clippard EV 24 V DC) pulsed by a Grass stimulator. The pipette was left in place for an additional 3 min and then withdrawn slowly before closing the skin with Vetbond (3 M). Meloxicam (1 mg/1 kg) was provided for postoperative analgesia. AAV-injected mice were allowed to survive for 3–5 weeks after surgery to allow optimal production of Cre-conditional proteins. CTb- and FG-injected mice were allowed to survive 3–6 days for retrograde transport. Three mice included in this paper (CTb cases 00549, 00714, 00715) were also analyzed in another study (Gasparini et al. 2019).

Table 2.

Injected tracers used in this study

Injected tracer	Abbreviation	Origin	Preparation
Fluorogold	FG	Fluorochrome, LLC	2% solution in sterile 0.9% saline
Cholera toxin subunit B	CTb	List Biological Laboratories Product #104 (Spangler 1992)	0.1% solution in dH20
AAV8-hEf1a-DIO-synaptophysin-mCherry	AAV8-DIO-EF1a-Syp-mCherry	Dr Rachel Neve at the Massachusetts Institute of Technology McGovern Institute for Brain Research Viral Vector Core (Garfield et al. 2014)
AAV8-hEf1a-synaptophysin-mCherry	AAV8-hEF1a-Syp-mCherry	Dr Rachel Neve at the Massachusetts Institute of Technology McGovern Institute for Brain Research Viral Vector Core

Stereotaxic injection coordinates for the PB ranged from 5.0 to 5.3 mm caudal, 0.9 to 1.4 mm (right) lateral, and 3.6 to 3.9 mm deep to the bregma. Injection coordinates for the locus coeruleus were −5.2 caudal, 0.95 (right) lateral, and 3.8–3.9 mm deep to the bregma. Coordinates for rostral, middle, and caudal injections into the insular cortex ranged from +1.8 rostral to −0.8 caudal, 2.5 to 4.0 (right) lateral, and 3.5 to 4.0 mm deep to the bregma. Injection coordinates for the medial prefrontal cortex (mPFC) were +2.0 mm rostral to the bregma, ranging 0.4–0.5 mm (right) lateral and 2.8–3.3 mm deep to the bregma. Injection coordinates for the ventromedial hypothalamus were 1.4 mm caudal, 0.5 lateral, and 5.8 mm deep to the bregma.

Perfusion and Tissue Sections

Mice were anesthetized with a mixture of ketamine-xylazine (i.p. 150–15 mg/kg, dissolved in sterile 0.9% saline) then perfused transcardially with phosphate-buffered saline (PBS), followed by 10% formalin-PBS (SF100–20, Fisher Scientific). After perfusion, the brain was removed and fixed overnight in 10% formalin-PBS. We sectioned the brains into 40-μm-thick coronal slices using a freezing microtome and collected sections into separate, 1-in-3 tissue series. Most of the sections were stored in cryoprotectant solution and stored at −20 °C until further processing. Some sections were immediately processed.

Immunohistology

For immunofluorescence labeling, we removed the tissue sections from the cryoprotectant and rinsed them in PBS before loading them into a solution with 2, 3, or 4 primary antisera. These antisera were added serially to a PBS solution of 0.25% Triton X-100 (BP151–500, Fisher Scientific), 2% normal donkey serum (NDS, 017–000-121, Jackson ImmunoResearch), and 0.05% sodium azide (14 314, Alfa Aesar) as a preservative (PBT-NDS-azide). We incubated these sections overnight at room temperature on a tissue shaker. Afterward, the sections were washed in PBS and incubated for 2 h at room temperature in PBT-NDS-azide solution containing the corresponding species-specific donkey secondary antibodies. These secondary antibodies were conjugated to Cy3, Cy5, Alexa Fluor 488, or biotin (Jackson ImmunoResearch #s 711-065-152, 711-165-152, 711-175-152, 705–065-147, 705-545-147, 713-545-147, 706-545-148, 706-165-148, 706-065-148, 715-065-15, 712-165-153; each diluted 1:1000 or 1:500). If biotin was used, they were incubated for an additional 2 h in streptavidin-Cy5 (#SA1011; Invitrogen) or streptavidin-Pacific Blue (#S11222, Life Technologies) prepared in PBT-NDS-azide. These sections were then washed 3X in PBS and mounted on glass slides (#2575-plus; Brain Research Laboratories) and coverslipped using Vectashield with DAPI (Vector Labs), Vectashield without DAPI, or Cytoseal 60 (#8310–4, Thermo Scientific). Slides were stored in slide folders at 4 °C.

For bright-field labeling (immunohistochemistry), we removed tissue sections from cryoprotectant, rinsed them in PBS, and then incubated them in 0.3% hydrogen peroxide (#H325–100, Fisher Scientific) and in PBT for 30 min to quench endogenous peroxidase activity. After 3 washes in PBS, we loaded sections into PBT-NDS-azide containing either rabbit anti-CTb or rabbit anti-dsRed overnight at room temperature on a tissue shaker. After 3 PBS washes the following morning, we incubated sections for 2 h in a solution of biotinylated donkey antirabbit (#711–065-152; Jackson) in PBT-NDS-azide. Sections were washed 3 more times and placed for 1 h in biotin-avidin complex (Vectastain ABC kit PK-6100; Vector), washed 3X in PBS, and incubated in diaminobenzidine (DAB) solution for 10 min. Our stock DAB solution was prepared by adding 100 tablets (#D-4418, Sigma, Saint Louis, MO) into 200 mL ddH2O and then filtering it. We then used 1 mL of this DAB stock solution, with 300 μL of 8% nickel chloride (#N54–500, Fisher Chemical) per 6.5 mL PBS. After 10 min in NiDAB, we added hydrogen peroxide (0.8 μL of 30% H2O2 per 1 mL PBS-DAB) and swirled sections for 2–5 min until observing black (nickel-DAB) color change. After two rapid PBS washes, we wet-mounted one or more sections, checked them in a light microscope to ensure optimal staining, and in rare cases replaced sections for up to one additional minute for additional enzymatic staining. Finally, after washing sections an additional 3X in PBS, we ordered and mounted them on glass slides. Slides were air-dried, then dehydrated in an ascending series of alcohols and xylenes, and then coverslipped with Cytoseal 60 (#8310–16 Thermo Scientific).

Nissl Counterstaining

After bright-field microscopy, all slides were Nissl-counterstained. The coverslips were removed by soaking in xylenes overnight or up to a week. After rehydration through 1-min dips in a graded series of alcohols, we rinsed the slides in water and dipped them in a 0.125% thionin solution (Fisher Scientific) for 1 min. Slides were rinsed in water until the solution cleared and then dehydrated in a series of ethanol solutions. The slides were placed in 50% EtOH, then 70% EtOH, 400 mL of 95% EtOH with 10 drops of glacial acetic acid, 95% EtOH, 100%, and 100% EtOH, and two xylene solutions. Afterwards, the slides were coverslipped with Cytoseal.

Fluorescence In Situ Hybridization

To label mRNA for Slc17a6 (vesicular glutamate transporter 2, Vglut2), Slc17a7 (vesicular glutamate transporter 1, Vglut1), Cck, Calca, or Pdyn, we used RNAscope Fluorescent Multiplex Detection Reagents (ref# 320851; Advanced Cell Diagnostics) with probes listed in Table 3. For background cytoarchitecture, we also labeled Ubc in some cases.

Table 3.

RNAscope probes used in this study

Probe	Common name	Channel	ACD catalog #	Lot #
Mm- Slc17a6-C3	Vesicular glutamate receptor 2	C3	319 171-C3	17251A
Mm-Slc17a7	Vesicular glutamate receptor 1	C1	416 631	18131A
Mm-Ubc	Ubiquitin C	C1	310 771	18010A
Mm-Ubc-C2	Ubiquitin C	C2	310 771-C2	18095B
Mm-Ubc-C3	Ubiquitin C	C3	310 771-C3	18085A
Mm-Cck	Cholecystokinin	C1	402 271	18162A
Mm-Cck-C2	Cholecystokinin	C2	402 271-C2	18017B
Mm-Pdyn	Prodynorphin	C1	318 771	16347A
Mm-Pdyn-C2	Prodynorphin	C2	318 771-C2	17290A
Mm-Pdyn-C3	Prodynorphin	C3	318 771-C3	17290A
Mm-Calca-C1	Calcitonin gene-related peptide	C1	417 961	17355B
Mm-Calca-C2	Calcitonin gene-related peptide	C2	417 961-C2	18165A

In the afternoon before hybridization, we removed sections from cryoprotectant and rinsed them in PBS at room temperature before mounting them on glass slides kept at 4 °C. In the morning, we outlined sections using a Super-HI PAP pen (Research Products Incorporated) and vacuum grease (Dow Corning) to form a hydrophobic barrier and then applied two 2 min PBS washes at room temperature. We then covered sections with Protease IV in glass petri dishes floating in a 40 °C water bath for 30 min.

After PBS 2 × 2 min washes, we incubated sections in one or more probes from Table 3 for 2 h at 40 °C. After that, we added amplification reagents 1–4 in series for 15–30 min each at 40 °C, with 2 × 2 min RNAscope Wash (ref 320 058, lot# 2001175) diluted 1:50 in ddH20 between each.

After finishing the RNAscope protocol, we either rinsed with PBS at 4 °C overnight to reduce wrinkles in the section or in some cases performed an immunofluorescence protocol to label mCherry. For the latter, we incubated sections in a PBT-azide solution with the appropriate primary antibody (rat anti-mCherry, Table 4) overnight at 4 °C. Slides were washed with PBS the next morning and then incubated in Cy3-conjugated donkey antirat (1:1000 in PBT-Az; Jackson ImmunoResearch #s 712–165-153) for 2 h at room temperature. After a final PBS wash, we coverslipped slides using Vectashield with DAPI.

Table 4.

Antisera used in this study

Antigen	Immunogen description	Source, host species, RRID	Concentration
DsRed	DsRed-Express, a variant of Discosoma sp. red fluorescent protein	Clontech, rabbit polyclonal, cat. # 632496, lot number 1509043, RRID: AB_1001313483	1:2000
mCherry	Full-length mCherry fluorescent protein	Life Sciences, rat monoclonal, cat. #M11217, lot # R1240561, RRID AB_2536611	1:2000
Cholera toxin B (CTB)	Purified choleragenoid, the B subunit of cholera toxin	List Biological, goat polyclonal, cat. #703, lot #7032A9, RRID: AB_10013220	1:10 000
Chicken ovalbumin upstream promoter transcription factor-interacting protein 2 (Ctip2)	Fusion protein corresponding to human Ctip2	Abcam, rat polyclonal, cat. #ab18465, lot # GR322373–7, RRID: AB_2064130	1:1000
TH	Purified, SDS-denatured rat pheochromocytoma TH	Millipore, mouse monoclonal, cat. # MAB318, lot: NG1802536, RRID: AB_2201528	1:2000
TH	Denatured TH from rat pheochromocytoma (denatured by sodium dodecyl sulfate)	Millipore, rabbit polyclonal, cat. #AB152, lot# 240602, RRID: AB_696697	1:10 000
Choline acetyltransferase (ChAT)	Human placental choline acetyltransferase	Millipore, goat polyclonal, cat. #AB144P, lot: JC1618187, RRID: AB_2079751	1:1000
Forkhead box protein 2 (FoxP2)	Recombinant human FOXP2 isoform 1 Ala640-Glu715	R&D Systems, sheep polyclonal, cat. #AF5647, RRID: AB_2107133	1:10 000
LIM homeobox transcription factor 1 Beta (Lmx1b)	Full-length LIM homeobox transcription factor 1 beta protein from mouse	C. Birchmeier, Max Delbruck Center for Molecular Medicine; Berlin; Germany RRID: AB_2314752	1:8000
Tryptophan hydroxylase (TpOH)	Phenylalanine hydroxylase (PH) purified from monkey liver. In formalin-fixed tissue, it only binds tryptophan hydroxylase	Millipore, mouse monoclonal, cat. # MAB5278, lot: 2891004, RRID: AB_2207684	1:2000
Estrogen receptor alpha (ERɑ)	KLH-conjugated linear peptide corresponding to the C-terminus of rat estrogen receptor alpha	Millipore, rabbit polyclonal, cat. # 06–935, lot: 2971020, RRID: AB_310305	1:16 000
Paired-like homeobox 2b (Phox2b)	BSA-coupled 15mer corresponding to the C terminus of the Phox2b protein with an added N-terminal tyrosine	H. Enomoto, School of Medicine at Kobe University; Kobe City; Japan. RRID: AB_2313690	1:12 000
Special AT-rich sequence-binding protein 2 (Satb2)	Synthetic peptide within human SATB2 (proprietary sequence)	Abcam, rabbit monoclonal, cat. # ab92446, lot: GR325015–2, RRID:AB_10563678	1:3000

Imaging, Counts, Figures, and Statistics

All slides were imaged using a 10x or 20x objective on an Olympus VS120 slide-scanning microscope. For bright-field images of NiDAB and then Nissl-counterstained sections, we used extended focal imaging (EFI) to collect and combine in-focus images from 11 focal planes through the tissue. After reviewing whole-slide images in OlyVIA or VS-ASW software (Olympus), we collected additional EFI or multifocal image stacks at higher magnifications in regions of interest. We counted double- or triple-labeled cells in EFI images or by scrolling up and down through focal planes in image stacks (12 μm stacks; 0.84 μm Z-spacing between images) to verify colocalization. In every axon-tracing case, we labeled, imaged, and examined every section from least one 1-in-3 series of sections through the full brain, from the olfactory bulbs to the cervical spinal cord.

We quantified cortical boutons immunolabeled for dsRed (Syp-mCherry) in bright-field NiDAB images in ImageJ. Images were thresholded so that 0.1% percentile lightest pixels were displayed as white and the 99.9% darkest pixels were displayed as black in OlyVIA (Supplementary Fig. 1A). Images were exported into ImageJ and then the cerebral cortex was manually segmented. A custom ImageJ script was used to perform the following automatic bouton-counting. This script transformed the grayscale image into a black and white image, using a minimum intensity threshold set at 23.4% of the value from white to black. The script then separated overlapping boutons and counted objects larger than 0.36 μm² as boutons. These size and intensity thresholds are comparable with manual bouton-counting (Supplementary Fig. 1B). We found that this algorithm consistently labels the same punctae as manual bouton-counting (Supplementary Fig. 1C) and produces similar overall counts to manual counts of 22 915 boutons across 8 brain sections (Supplementary Fig. 1D). We then used this procedure to quantify the total number of boutons in the cerebral cortex in every section from two brains from each Cre-driver strain used for Syp-mCherry tracing of PB axonal projections. Also, in three sections from each case (a rostral section at the anterior border of the corpus callosum, a middle section containing the decussation of the anterior commissure (AC), and a caudal section containing the rostral, dorsal hippocampus), we manually delimited the insular cortex and used the same algorithm to count boutons exclusively in this cortical area and then divided that by the total number of boutons in the full cerebral cortex at the same three levels.

We quantified the laminar distribution of PB–cortical boutons by taking one representative section from Vglut2-IRES-Cre mice (with AAV8-EF1a-DIO-Syp-mCherry injections into the PB, n = 4). We measured the distance (in μm) of each bouton from the closest point along the pial surface. Then, we arranged data into histograms showing the distribution of the distance of boutons from the pial surface. In the first histogram, we averaged the number of boutons at each distance across cases. In the second, we normalized the distributions across cases and averaged them. For reference, we also overlaid the average distance (in μm) of identifiable laminar boundaries from the pial surface by measuring the boundaries of the genetic markers Ctip2 and FoxP2 as well as the external capsule.

For bright-field and fluorescence images in the figures, we used Adobe Photoshop to crop raw, bitmap images from VS-ASW or OlyVIA (Olympus), adjust brightness and contrast, and combine raw fluorescence (grayscale) data for multicolor combinations. We added lettering and produced all drawings in Adobe Illustrator. Cytoarchitectural distinctions were made by comparison with images of the same section after Nissl counterstaining. Scale bars were traced atop calibrated lines in OlyVIA to produce a clean white or black line. All graphs and plots of bouton density were generated in Prism 7.04 (GraphPad Software).

Neuroanatomical Nomenclature

We used nomenclature derived from original neuroanatomical work in the peer-reviewed literature. For the PB, this literature stems from the cytoarchitecture-based taxonomy first proposed in rats by Fulwiler and Saper (1984b) and includes subsequent work clarifying subnuclear anatomy in mice (Kaur et al. 2013; Geerling et al. 2016). For the insular cortex, this literature stems from the original cytoarchitectural division of this region in rabbit and squirrel into dorsal (granular) and ventral (agranular) areas (Rose 1928) and the addition of an intermediate “dysgranular” zone in monkeys by Jones and Burton (1976), which was adapted for rats by subsequent investigators (Krettek and Price 1977; Cechetto and Saper 1987), and then adapted to the mouse brain (Dong 2008; Franklin and Paxinos 2013).

Results

Retrograde Identification of Cortically Projecting PB Neurons

We began by examining axonal projections from the PB to their densest target region in the cerebral cortex, the insular cortex (Saper and Loewy 1980; Saper 1982b; Fulwiler and Saper 1984a). To identify the cortical extent of these projections, as well as the distribution of source neurons within the PB, we made large injections of the retrograde tracer FG into the insula and into cortical areas extending rostral and caudal to it (Fig. 1A,B). All injections produced some retrograde labeling in the PB, dwindling rostrally into the caudal pedunculopontine tegmental nucleus. In all cases, we also found retrogradely labeled neurons in the ventral midbrain (substantia nigra and ventral tegmental area), dorsal raphe, and locus coeruleus, plus occasional labeling in the nucleus incertus.

Figure 1.

FG injections in the cerebral cortex produced retrograde labeling in the PB. (A) Four rostrocaudal levels through the FG injection site in one mouse (case #00577), with rostrocaudal distance in mm relative to the midline decussation of the anterior commissure (AC). (B) Rostrocaudal distribution of FG injection sites from each case (parasagittal). (C) Rostrocaudal level of injection sites compared to the number of retrogradely labeled neurons in the PB. Horizontal bars represent rostrocaudal span of the core of each injection site. (D) Retrograde labeling in the PB was largely ipsilateral. (E) Distribution of FG-labeled cells across four brain stem levels containing the PB. PB neurons are blue, while those outside the PB are black (abbreviations: DIC, dysgranular insular cortex; GIC, granular insular cortex).

In the PB, the number of retrogradely labeled neurons depended more on the location than the size of the cortical injection. FG tracer injections in the insular cortex within 1 mm rostral or caudal to the AC decussation produced a prominent cluster of retrogradely labeled neurons. Injections in the cortex more than ~ 1 mm rostral to the level of the AC decussation labeled fewer PB neurons (Fig. 1C). Retrograde labeling in the PB was predominantly ipsilateral (Fig. 1D) and clustered primarily in the medial subdivision with fewer neurons extending into the lateral subdivision (Fig. 1E).

To identify cell-type–specific markers for this cortically projecting subset of PB neurons, we began by immunolabeling the transcription factors FoxP2 and Lmx1b in brain stem sections from mice with FG injected into the insular cortex. These two transcription factors identify distinct macropopulations of neurons in the PB (Miller et al. 2012). The majority of retrogradely labeled PB neurons contained nuclear immunoreactivity for Lmx1b (81 ± 15%), and few are immunoreactive for FoxP2 (2 ± 2%) (Fig. 2B; n = 6 mice). Among a variety of PB subpopulation markers, we found that many cortically projecting PB neurons express mRNA for the neuropeptide cholecystokinin (Cck; 47 ± 7%; n = 4) and some for calcitonin gene-related peptide (Calca; 12 ± 2%; n = 4) (Fig. 2C). We also found intense nuclear immunoreactivity for estrogen receptor alpha (ERɑ) in a subpopulation of cortically projecting PB neurons (22 ± 7%; n = 2) in the external lateral subnucleus and extending into the medial PB (Fig. 2D). Intermingled caudally were neurons that express the transcription factor Satb2, which have been implicated in sweet taste signaling (Maeda et al. 2009; Fu et al. 2019), but only a minority of cortically projecting PB neurons contained nuclear immunoreactivity for this transcription factor (21 ± 5%; n = 2; not shown). Few cortically projecting PB neurons were immunoreactive for choline acetyltransferase, a marker for cholinergic neurons (ChAT; 3 ± 0.1%; n = 3), or for the transcription factor Phox2b (Phox2b; 1 ± 1%; n = 4). Injecting FG into the insular cortex of reporter mice for the vesicular glutamate transporter Vglut2 (Vglut2-IRES-Cre;R26-lsl-L10GFP) revealed that almost all retrogradely labeled neurons in the PB expressed GFP (99 ± 1%; n = 2), indicating that cortically projecting PB neurons are predominantly glutamatergic.

Figure 2.

Identification of cell-type–specific markers for PB neurons retrogradely labeled from the insular cortex. (A) Percent colabeling of retrogradely labeled PB neurons (n = 2–9 for each marker; see Results). (B) Most retrogradely labeled neurons (blue) contained nuclear immunoreactivity for Lmx1b (red) and not FoxP2 (green). (C) Roughly half contained mRNA for Cck (green) and fewer for Calca (red). (D) Some retrogradely labeled neurons in the lateral PB contained nuclear immunoreactivity for estrogen receptor α (ERα, red) (additional abbreviations: dL, dorsolateral PB subnucleus; eL, external lateral PB subnucleus; m, medial PB subnucleus; MeV, mesencephalic trigeminal nucleus and tract; scp, superior cerebellar peduncle; vsct, ventral spinocerebellar tract).

We examined additional cases with large FG injections into the mPFC, centered on the infralimbic area (not shown). Medial prefrontal injections produced fewer retrogradely labeled PB neurons (41 and 39 total PB neurons, n = 2 cases), with a distribution that skewed into the external lateral subnucleus. Similar to retrograde labeling from the insular cortex, many of these contained nuclear immunoreactivity for Lmx1b (52 ± 23%); none were immunoreactive for FoxP2.

Discrete PB Populations of Cck-Expressing Neurons

Cck-expressing neurons in the superior lateral PB subnucleus project axons to the hypothalamus (Inagaki et al. 1984; Zaborszky et al. 1984; Fulwiler and Saper 1985; Hermanson, Larhammar, and Blomqvist 1998; Garfield et al. 2014), but Cck-expressing neurons have not been reported in the medial PB. Thus, we were surprised to find Cck expression in cortically projecting neurons there. To compare the genetic identity and connectivity of these two clusters of Cck-expressing neurons, we also examined retrograde labeling in the PB after injections of FG into the ventromedial hypothalamus (Fig. 3). Hypothalamic injections (n = 2) retrogradely labeled many Cck-expressing neurons in the PB, consistent with previous reports in rats (Fig. 3C). In contrast to cortex-projecting neurons in the medial PB, hypothalamus-projecting neurons were located in the superior lateral and central lateral PB subnuclei, and they expressed more Foxp2 than Lmx1b (Fig. 3D,E). This striking dichotomy between the Cck-expressing neurons in the superior lateral PB, which project to the hypothalamus, and Cck-expressing neurons in the medial PB, which project to the insular cortex, is evident from a side-by-side comparison (Fig. 3E–J).

Figure 3.

The PB contains two different populations of neurons that express Cck, with separate output targets. (A) Summary diagram of the projections from caudal–medial and rostral–lateral Cck-expressing neurons in the PB. (B) FG injection site in the medial hypothalamus (case #01939). (C–D) Many of the retrogradely labeled (blue) neurons in the rostral-lateral PB expressed Cck (C, red) or contained nuclear immunoreactivity for FoxP2 (D, green). (E–G) FG injections in the hypothalamus (n = 2) labeled many Cck-expressing neurons in the rostral, lateral PB, but not in the caudal–medial PB. Many of these were FoxP2-immunoreactive. (H) FG injections in the cortex did not label any Cck-expressing neurons in the rostral-lateral PB and instead labeled many in the caudal-medial PB. Most contained nuclear immunoreactivity for Lmx1b and lacked FoxP2 (n = 9). The distribution of retrogradely labeled and Cck-expressing neurons in a representative case is shown at right (F-G, hypothalamic case #01939; I-J, cortex case #00563) (additional abbreviations: cL, central lateral PB subnucleus; sL, superior lateral PB subnucleus).

Anterograde Labeling of Glutamatergic PB Projections to the Cerebral Cortex

To complement our retrograde tracing, we used a highly sensitive, synaptically targeted, Cre-conditional tracer (AAV8-Ef1a-DIO-Syp-mCherry). Because most PB neurons are glutamatergic (Niu et al. 2010) and because our retrograde tracing in GFP reporter indicated that cortically projecting PB neurons are glutamatergic, we used Vglut2-IRES-Cre mice for AAV8-EF1a-DIO-Syp-mCherry anterograde tracing from the PB region (Figs 4A–D and 5A). In the PB injection site from each case (n = 4), more than 90% of Syp-mCherry–expressing neurons expressed mRNA for Slc17a6/Vglut2. In each case, neurons expressing Syp-mCherry spanned a large subregion of the PB and involved different extents of neighboring brain stem nuclei, including the cuneiform nucleus, nucleus of the lateral lemniscus, mesencephalic trigeminal nucleus, or supratrigeminal nucleus (Fig. 5B). None of these nuclei contained retrogradely labeled neurons after FG injections in the cerebral cortex, indicating that they do not contribute to the PB–cortical labeling described below.

Figure 4.

Syp-mCherry immunofluorescence (red) after injections of AAV8-hEF1a-DIO-Syp-mCherry into the PB. (A–D) Vglut2-IRES-Cre. (E–H) FoxP2-IRES-Cre. (I–L) Th-IRES-Cre.

Figure 5.

Anterograde analysis of projections from PB glutamatergic neurons. (A) Cre-conditional tracing strategy. (B) Sample injection site across four levels immunostained for Syp-mCherry (case #01280). (C) Syp-mCherry-labeled boutons in the insular cortex and mPFC, with and without Nissl counterstain. (D) Distribution of Syp-mCherry–labeled boutons across the cerebral cortex. Dashed gray lines indicate approximate dorsal and ventral boundaries of the insular cortex. (E) Different injection sites (n = 4) produce varying densities of Syp-mCherry labeling in the cerebral cortex.

The overall pattern of projections to the cortex was similar in every case. The ipsilateral insular cortex received a massive projection, with the densest concentration occupying a roughly 2 mm rostrocaudal span centered at the level of the AC decussation. In this midinsular region, a dense terminal field of Syp-mCherry–labeled boutons distributed most heavily within the intermediate (“dysgranular”) subdivision, with fewer boutons in the dorsal, “granular,” ventral, and agranular insular cortex (AIC).

Outside the insular cortex, less-dense projections extended rostrally and caudally to span the full length of the cerebral cortex, from the frontal pole back through caudal, “temporal” areas including the entorhinal cortex (Fig. 5D). The near-absence of labeling throughout the principal claustrum formed a striking contrast with denser labeling immediately dorsal to it and in the overlying insular cortex. From there, Syp-mCherry–labeled boutons extended dorsally and rostrally through layer 6b of the prefrontal cortex.

Rostrally, as labeling in the insular cortex became sparser, Syp-mCherry–labeled boutons continued into the deep layers of the dorsolateral prefrontal cortex (dlPFC). They arced dorsally and then medially over the corpus callosum, meeting a separate projection ascending through the septum to form a secondary terminal field at the ventral extreme of the mPFC. Here, Syp-mCherry–labeled boutons in the mPFC concentrated in layer 1 of the dorsal peduncular cortex and extended dorsally in deeper layers into the ventral, caudal infralimbic area (Fig. 5C). Rostral to this secondary terminal field in the mPFC, labeled boutons continued rostrally through a broad expanse of the dlPFC, up through the frontal pole (Fig. 5D).

As shown in Fig. 5E, injection site location within the PB influenced the extent of anterograde labeling in the cerebral cortex. Injections centered in the medial PB produced more extensive axonal labeling (cases 01280, 00147), with many Syp-mCherry–immunoreactive punctae (putatively synaptic boutons), while injections centered rostrally and dorsally (00017) or laterally (01281) labeled fewer cortical boutons overall.

Besides the cerebral cortex, each case had abundant axonal labeling in a variety of subcortical and brain stem projection targets (not shown), in patterns similar to prior anterograde tracing results (Saper and Loewy 1980; Fulwiler and Saper 1984b; Bernard et al. 1993; Krukoff et al. 1993; Alden et al. 1994; Bester et al. 1997; Bester et al. 1999).

Anterograde Tracing from PB Subpopulations

To complement our retrograde data and to identify the overall distribution of PB–cortical projections, we injected the same Cre-conditional tracer (AAV8-Ef1a-DIO-Syp-mCherry) in several additional strains of knockin-Cre mice to target individual subsets of PB neurons. A large subpopulation of PB glutamatergic neurons express the transcription factor FoxP2 (Miller et al. 2012; Geerling et al. 2016; Geerling et al. 2017; Verstegen et al. 2019), so we began by making large injections of this vector into the PB of Foxp2-IRES-Cre mice. Injections in this region of the brain stem (n = 4) transduced many PB neurons, with FoxP2-immunoreactive nuclei in 95% of Syp-mCherry–expressing neurons (Fig. 4E–H). This produced extensive axonal labeling in the periaqueductal gray, lateral hypothalamic area, paraventricular nucleus of the thalamus, and preoptic area (not shown), but virtually no labeling in the cerebral cortex (Fig. 6A). We also injected Pdyn-IRES-Cre mice to target the large subset of FoxP2-expressing neurons that express the neuropeptide gene prodynorphin (Miller et al. 2012; Geerling et al. 2016; Lee et al. 2019). As in Foxp2-IRES-Cre mice, this transduced many PB neurons with extensive Syp-mCherry axonal labeling in the periaqueductal gray, lateral hypothalamic area, paraventricular nucleus of the thalamus, and preoptic area, but virtually no projections to the cortex (Fig. 6B). Thus, consistent with our retrograde tracing data, the subpopulation of PB neurons defined by FoxP2 (and in part, Pdyn) does not project to the cerebral cortex.

Figure 6.

Cre-conditional anterograde tracing with Syp-mCherry in a variety of Cre-driver strains confirms cell-type specificity of PB projections to the cerebral cortex. (A–B) Injecting AAV8-hEF1a-DIO-Syp-mCherry into the PB in FoxP2-IRES-Cre and Pdyn-IRES-Cre mice produced virtually no cortical labeling. (C–D) Cck-IRES-Cre and Esr1-2A-Cre mice had dense cortical labeling, similar to Vglut2 cases. (E) Calca-Cre mice had intermediate cortical labeling. (F) Th-IRES-Cre mice with injections into the locus coeruleus had diffuse labeling throughout the cerebral cortex. (G) Raw counts of total number of boutons in the cerebral cortex (n = 2 for each genotype; each dot is one case and the line is the average). (H) Average number of boutons in the ipsilateral insular cortex across three sections from each case. (I) Fraction of boutons in the insular cortex relative to the entire cerebral cortex at the same three levels.

Conversely, injecting the same vector into the PB of Cck-IRES-Cre mice produced dense labeling in the cortex (Fig. 6A). Cases with injections that transduced many neurons in the medial PB (n = 3) produced dense cortical labeling with a distribution similar to the Vglut2-IRES-Cre cases described above, while injection sites in the lateral PB (n = 3) produced light cortical labeling. In one of these cases, Syp-mCherry in the dorsal, rostral PB (primarily in the superior lateral subnucleus) labeled a dense terminal field in the ventromedial nucleus of the hypothalamus (not shown), consistent with our retrograde tracing data identifying two separate populations of Cck-expressing neurons in the PB (Fig. 4).

Also, because we identified ERα immunoreactivity in the caudal–medial and lateral PB, including some cortically projecting PB neurons (above), we injected the PB of Esr1-2A-Cre mice. Injections that transduced neurons in the medial PB with Syp-mCherry labeled dense terminal fields in the insular and mPFC, similar to the Vglut2-IRES-Cre cases shown above and even denser than Cck-IRES-Cre cases (Fig. 6D).

Similarly, to label the cortical projections of Calca-expressing PB neurons, we injected the same vector into the PB of Calca-Cre mice. Cases with injections that transduced the core population of Calca-expressing neurons (n = 3) in the external lateral PB produced minimal labeling in the cerebral cortex, with a laminar distribution in the insular cortex that skewed superficially (Fig. 6E). Other, more medial injection sites (n = 4) transduced many neurons in the locus coeruleus (LC), in addition to the PB, consistent with Calca expression in many LC neurons (Lein et al. 2007). These cases, with Syp-mCherry-expressing LC neurons, had diffuse axonal labeling throughout the ipsilateral cerebral cortex in addition to the more focal terminal field in the insular cortex. In one case, the injection missed the PB, but still transduced LC neurons, producing diffuse cortical labeling with no focality in the insular or mPFC. These results suggest that there are two, separate populations of Calca-expressing neurons in this region—one comprised of glutamatergic PB neurons that project focally, but lightly, to superficial insular cortical layers, and the other comprised of catecholaminergic neurons in the LC, which project broadly and diffusely to the entire cerebral cortex (Jones et al. 1977).

To test this hypothesis, we also traced the axonal projections of LC neurons in Th-IRES-Cre mice (Fig. 6F). Many neurons were transduced after AAV8-Ef1a-DIO-Syp-mCherry injections into the PB/LC region of Th-IRES-Cre mice (n = 3), with immunoreactivity for the tyrosine hydroxylase (TH) in 89%. All neurons in the LC were TH-immunoreactive (Fig. 4I–L), and a small number of neurons rostral or lateral to the LC lacked immunoreactivity. These cases had diffuse Syp-mCherry axonal labeling throughout the entire ipsilateral cerebral cortex, without a concentrated terminal field in the insular, medial prefrontal, or any other cortical area (Fig. 6E).

Next, to quantitatively compare projections across neuronal subpopulations, we examined overall counts of Syp-mCherry-labeled boutons across all cortical levels in two cases from each genotype with the most extensive neuronal transduction in the PB region (Fig. 6G–I). Cortical projections from the LC (in Th-IRES-Cre cases) were far more extensive than from other cell types (Fig. 6G). However, when we restricted our analysis to just the insular cortex, we found that Vglut2-IRES-Cre mice had the strongest projections to this area, followed by Esr1-2A-Cre and Cck-IRES-Cre mice (Fig. 6H). Comparing the fraction of insular to total cortical boutons across strains (Fig. 6I) revealed that genotypes with strong PB projections to the insular cortex also had the most specific projections (Cck-IRES-Cre, Vglut2-IRES-Cre, Esr1-2A-Cre), while LC projections and cases with weak projections to the cortex, overall, had low insular specificity (Th-IRES-Cre, Foxp2-IRES-Cre, Pdyn-IRES-Cre). Calca-Cre cases had high insular specificity despite low overall cortical labeling.

Laminar Distribution of PB Terminal Field in Insular Cortex

It is difficult to discern laminar boundaries in the insular cortex, which transitions from “granular” neocortex dorsally, through a “dysgranular” zone, into a ventral, “agranular” region bordering the piriform cortex. Therefore, to better characterize axonal projections from the PB to the insular cortex, we examined molecular markers that differentiate cortical layers. Neurons in different layers of the cerebral cortex express different molecular markers, such as the transcription factors FoxP2 in layer 6 (Ferland et al. 2003), Ctip2 in layers 5 and 6 (Leid et al. 2004; Arlotta et al. 2005), and Cux1 in layers II to IV (Nieto et al. 2004). In cases with AAV8-EF1a-DIO-Syp-mCherry injections into the PB, we immunolabeled mCherry throughout the cerebral cortex, along with Ctip2 and FoxP2 in Vglut2-IRES-Cre (n = 4) and Cck-IRES-Cre (n = 2) mice, plus Cux1 and FoxP2 in adjacent sections from the Cck-IRES-Cre mice. Syp-mCherry–labeled boutons spanned the width of the cortex but heavily targeted a band superficial to layer 5, with less labeling in layers 1 and 6 (Fig. 7A).

Figure 7.

Laminar distribution of anterogradely labeled boutons in the insular cortex. (A–C) Immunoreactivity for FoxP2 (layer 6) and Ctip2 (layers 5 and 6) in relation to Syp-mCherry–labeled boutons. (D–F) Immunoreactivity for FoxP2 (layer 6) and Cux1 (layers 2–4) in an immediately adjacent section from the same brain (Cck-IRES-Cre case #01278). Dashed gray lines indicate approximate dorsal and ventral borders of the insular cortex. (G) Distribution of bouton distance from the outer surface of the brain, with layers identified from Ctip2 and FoxP2 expression. Average distance of each layer from the pial surface is represented by the vertical black bar, and standard deviation is represented by the blue gradient. (H) Average of four normalized cases, with standard deviation.

In searching for markers that may delimit the dorsal-ventral extent of cortical territory that contains labeled boutons, we uncovered a strikingly unmyelinated region within the insular cortex. Immunolabeling myelin basic protein (MBP) produces uniform labeling of radially oriented axons across layers 2–6 throughout the cerebral cortex, save for a notch in the insular cortex (Fig. 8). Syp-mCherry–labeled boutons from the PB occupied a large dorsal extent of this unmyelinated zone in the insular cortex. At levels containing the decussation of the AC and rostral to it, this terminal field straddled the dorsal boundary of the insular unmyelinated zone, and at more caudal levels, it shifted ventrally into the unmyelinated region.

Figure 8.

MBP immunoreactivity (green in B, E, and H) reveals a prominent, poorly myelinated notch corresponding to much of the insular cortex. Anterogradely labeled boutons from the PB (red in A, D, and G) at three rostrocaudal levels through the insular cortex largely overlap this poorly myelinated zone (C, F, and I). Dashed gray lines indicate approximate dorsal and ventral borders of the insular cortex.

Retrograde Identification of PB-Projecting Cortical Neurons

Many cortical neurons project axons back to the PB (Moga et al. 1990). To identify and characterize these neurons, we injected the retrograde tracer CTb into the PB (Fig. 9A; n = 3). In addition to afferent labeling in known subcortical regions, we found retrogradely labeled neurons in the insular cortex (Fig. 9B) and a variety of other cortical areas, particularly in the prefrontal region up through the frontal pole. If fact, while labeling in the insular cortex was prominent at all rostrocaudal levels (tapering caudally into the perirhinal area), prefrontal areas contained even more neurons projecting to the PB region (Fig. 9C). The majority of retrogradely labeled neurons were ipsilateral to the CTb injection in PB. All retrogradely labeled neurons in the cortex expressed vesicular glutamate transporter 1 (Vglut1/Slc17a7, Fig. 9E).

Figure 9.

Retrograde labeling from the PB. (A) Two rostrocaudal levels through an injection site of cholera toxin subunit B (CTb) in the PB (case #00549). (B) NiDAB immunolabeling for CTb in the insular cortex. Dashed white lines represent dorsoventral boundary of insular cortex. (C) Rostrocaudal distribution of retrogradely labeled neurons in the cortex (n = 3). (D) Retrogradely labeled cells (blue) in the cerebral cortex at selected brain levels (subcortical labeling not shown). (E) All retrogradely labeled neurons in the cortex express the glutamatergic marker Vglut1 (Slc17a6, green). (F–G) Retrogradely labeled layer 5 cortical neurons contain nuclear immunoreactivity for Ctip2 (green) and not FoxP2 (blue). In contrast, neurons deep to this contain varying levels of Ctip2 and FoxP2.

In the insular cortex, we found two separate groups of retrogradely labeled neurons. The larger population of retrogradely labeled neurons were distributed in layer 5 and had a triangular shape, apical dendrite, and large nucleus, which are cytoarchitectural properties of pyramidal neurons. Virtually all these neurons were Ctip2-positive and FoxP2-negative, confirming that they are layer 5 cortical pyramidal neurons (Fig. 9F–G). The other population of CTb retrogradely labeled neurons clustered immediately dorsal to the principal claustrum at all rostrocaudal levels, becoming greater in number caudally. These abutted the external capsule deep to layer 6a of the overlying cortex and lacked a pyramidal morphology, but many expressed both Ctip2 and Foxp2 (Fig. 9G).

Distribution of Cortical Terminal Field in the PB

To label the projections of cortical neurons inside the PB, we made large injections of a non–Cre-conditional anterograde tracer (AAV8-Ef1a-Synaptophysin-mCherry) into the insular or mPFC (n = 3 each; examples shown in Fig. 10A,E).

Figure 10.

Descending projections from the cerebral cortex to the PB. (A) Large injection of a non–Cre-dependent vector (AAV8-hEF1a-Syp-mCherry) into the insular cortex (case #00838). (B) NiDAB immunolabeling for Syp-mCherry in the PB. (C) Distribution of Syp-mCherry–labeled boutons (red) in an adjacent section, relative to FoxP2 (green) and Lmx1b (blue) immunoreactive neurons in PB. (D) Distribution of labeled boutons across six levels of the PB. (E) Large injection of the same, non–Cre-dependent vector into the medial prefrontal region (case #01657). (F) NiDAB immunolabeling for Syp-mCherry in the PB. (G) Distribution of Syp-mCherry–labeled boutons in an adjacent section, relative to FoxP2 and Lmx1b immunoreactive neurons in PB. (H) Distribution of labeled boutons across six levels of the PB (additional abbreviations: iL, internal lateral PB subnucleus; LC, locus coeruleus; wa, waist).

Contrasting the restricted distribution of cortically projecting PB neurons in the medial PB, axonal projections from the insular cortex to the PB distributed Syp-mCherry–labeled boutons throughout much of the medial and lateral PB subdivisions (Fig. 10D). These boutons clustered more among PB neurons that express Lmx1b than FoxP2 (Fig. 10C).

Our retrograde labeling revealed that many neurons in the prefrontal cortex project axons to the PB. This reciprocal projection of Syp-mCherry–labeled boutons from the mPFC heavily targeted the outer portion of the external lateral PB subnucleus (Fig. 10F), with lighter labeling in other PB subnuclei (Fig. 10G) as well as the locus coeruleus and dorsal raphe nucleus (not shown).

Discussion

Our results extend the initial identification of direct axonal projections between the PB and cerebral cortex in rodents (Saper and Loewy 1980; Saper 1982a) by revealing molecular markers for the subset of PB neurons with direct projections to the cerebral cortex. Cortical projections from the PB arise from an Lmx1b-expressing subset of glutamatergic neurons in the medial PB. Roughly half of these express the neuropeptide gene Cck (and to a lesser extent, Calca), with small subpopulations containing the nuclear transcription factors Satb2 and ERα. Also, we identify an unexpected breadth and density of reciprocal connectivity between the PB and limbic cortical areas (Fig. 11). These new genetic and neuroanatomical data will help focus and facilitate future cell-type–specific experiments investigating this circuit. After discussing limitations of our approach and comparing our findings to the existing neuroanatomical literature, we highlight findings that may help predict the functional role of direct PB–cortical connectivity.

Figure 11.

Summary diagram of reciprocal connectivity between the PB and cerebral cortex. Vglut2, Lmx1b, Cck, Esr1, Satb2, and Calca-expressing cells of the PB project heavily to a poorly myelinated area of the insular cortex and other limbic and prefrontal areas of the cerebral cortex. In return, many layer 5 and deep cortical neurons send axonal projections back to the PB.

Limitations

To identify the target sites of axonal projections from the PB, we used Syp-mCherry, an anterograde tracer with unprecedented sensitivity and specificity. This construct combines mCherry, a fluorescent protein tag that is easy to immunolabel with a high signal-to-noise ratio, with synaptophysin, an endogenous protein that is trafficked to presynaptic boutons. Syp-mCherry accumulates in presynaptic boutons, unlike constructs expressing an unconjugated fluorescent protein, which distribute evenly throughout the axoplasm, and in contrast to frequently used conjugates like ChR2-mCherry or hM3Dq-mCherry, which insert uniformly along the axonal membrane. That said, Syp-mCherry must traverse the axon shaft to reach a synaptic bouton, so some degree of axonal labeling is unavoidable. Thus, while our conservative design for identifying punctae is enriched for presynaptic boutons, it probably counts a small fraction of nonsynaptic punctae. For any individual puncta in a particular region, confirming structurally that it forms a synapse requires immunoelectron microscopy (Wouterlood and Jorritsma-Byham 1993), and testing whether the punctae in any given region form functional synapses requires patch-clamp electrophysiology (Petreanu et al. 2007), neither of which is practical at the scale of the present analysis.

Next, quantifying boutons at this scale (3 592 634 Syp-mCherry–labeled boutons across 20 brains) requires semiautomatic analysis. Due to the possibility of counting histologic debris or undercounting closely spaced boutons, we calibrated our method by comparing it to more than 20 000 manually counted boutons and confirmed, both quantitatively and by visual inspection, that largely the same boutons were identified by manual and automatic counting. We also visually inspected the automatically identified boutons in every section from every case presented here to ensure that the algorithm remained accurate. Our goal was to compare relative differences in the distribution of boutons across sections and between cases, so while we are confident in the proportional accuracy of our approach, the present results should not be interpreted as accurate estimates of the raw number of synapses.

Another consideration involves our Cre-conditional approach. To ensure specificity, we used only knockin strains, where Cre is expressed upon transcription of the endogenous gene. For genes with stable expression, this approach allows Cre-conditional expression in the same cells that transcribe the Cre-driver gene at all times. However, for genes with more dynamic changes in expression, such as TH (Chan and Sawchenko 1998), transient expression could allow Cre-conditional initiation of Syp-mCherry expression, which persists after transcription of the Cre-driver gene abates. This is because Cre expression any time after AAV injection produces inversion and excision of the DIO/FLEX construct, which is permanent and irreversible (Schnutgen et al. 2003; Atasoy et al. 2008). Thus, while our approach transduces cells that have expressed the Cre-driver gene at some time after injection, it is not necessarily specific to cells that produce the Cre-driver gene continuously or in detectable amounts on the day of perfusion.

Another consideration is that a gene expressed by a specific subpopulation of neurons in one brain region also may be expressed in unrelated neurons in a neighboring brain region. For example, some LC neurons expressed Syp-mCherry in Pdyn-IRES-Cre, Calca-Cre, and Esr1-2A-Cre cases with injections extending medially beyond the PB, while none expressed Syp-mCherry in any Vglut2-IRES-Cre, Cck-IRES-Cre, or Foxp2-IRES-Cre mice. Thus, it is useful to examine labeling in and around the AAV injection site in every case and in complementary Cre-driver strains. We used Th-IRES-Cre mice to compare the output projections of LC neurons and assess their contribution to diffuse cortical labeling found in Esr1-2A-Cre, Calca-Cre, or Pdyn-IRES-Cre mice that had unintended transduction of LC neurons. These comparisons confirmed that cortical projections of LC neurons, though extensive, are diffuse and that focally dense projections to limbic cortical areas arise from PB neurons that express Vglut2, but not from the LC.

Finally, while our approach uses current developmental-genetic information to target neuronal subpopulations in the PB, future findings may allow us to more comprehensively or specifically identify and target neuronal subpopulations in this intricately complex region. We began with Vglut2-IRES-Cre to label all PB neurons and then targeted more restricted subpopulation markers (Cck, Calca, Esr1, Foxp2, Pdyn). Experiments using some of these markers have identified functions for some PB neurons (Kaur et al. 2013; Garfield et al. 2014; Carter et al. 2015; Geerling et al. 2016; Kaur et al. 2017; Palmiter 2018; Lee et al. 2019), but future information likely will subdivide these subpopulations further.

Comparison with Previous Neuroanatomical Literature

Despite early descriptions of connectivity between the upper brain stem and cerebral cortex (Leonard 1969; Llamas et al. 1975; Leichnetz and Astruc 1976; Muller-Preuss and Jurgens 1976; Divac et al. 1978), direct connectivity between the PB and cortex remained poorly characterized until Saper and Loewy used the anterograde autoradiographic method to label ascending projections in rats (Saper and Loewy 1980). They described direct projections from the medial PB targeting layers III, V, and VI of the granular insular cortex (GIC), “the deeper layers of the cortex of the frontal pole,” the mPFC, and the “septo-olfactory area” ventral to it. Our results are largely consistent with theirs, except that in the insular cortex, we find presynaptic boutons heavily targeting layers II to V more than layers I or VI. This difference may reflect the sensitivity of our synaptically targeted labeling relative to the transport of a radiolabeled amino acid throughout axonal arbors.

In subsequent tracing studies, Saper rejected the claim of a PB projection to the infralimbic area of the mPFC (1982, 1982), but our results clearly identify an axon-terminal field in this location, consistent with the original report (Saper and Loewy 1980). Saper also claimed that descending projections to the PB arise from the same cortical regions that receive input from the PB (Saper 1982b). Our retrograde labeling in mice, like subsequent retrograde tracing from the rat PB (Moga et al. 1990), shows that the PB receives input from a broader array of cortical areas than it projects to.

That same year, another study in rats reported retrograde labeling in the medial PB after HRP injections into the insular cortex (Lasiter et al. 1982), and Shipley and Sanders (1982) used anterograde and retrograde transport of HRP to confirm the presence of reciprocal PB–cortical connectivity in mice. Fulwiler and Saper then delineated PB subnuclei based on cytoarchitectural features and, using their new taxonomy, showed that the neurons in the medial and external medial PB subnuclei project to the cortex (Fulwiler and Saper 1984a). Yasui et al. (1985) provided complementary evidence in cats by injecting HRP into the PB, which retrogradely labeled layer V cortical neurons in the insula and the infralimbic cortex and confirmed corticoparabrachial projections after WGA-HRP injections into each of these cortical areas. A direct axonal projection to the insular cortex was not identified after a limited number of injections into the primate PB (Pritchard et al. 2000), but descending projections from the cortex to the PB were noted incidentally by many other authors (Reep and Winans 1982a, 1982b; van der Kooy et al. 1984; Room et al. 1985; Cechetto and Saper 1987; Hurley et al. 1991; Halsell 1992; Buchanan et al. 1994; Jasmin et al. 2004; Gabbott et al. 2005; Tokita et al. 2009; Zhang et al. 2011).

Previous tracing studies have failed to identify a genetic marker for PB neurons that project to the cerebral cortex. Retrograde tracing from the insular cortex revealed that cortically projecting neurons in the PB lack FoxP2 (Shin et al. 2011), consistent with our findings. Investigators using Cre-dependent anterograde tracing from neurons in the lateral PB that express Calca (Carter et al. 2013), Oxtr (Ryan et al. 2017), or Cck (Garfield et al. 2014) did not report axonal labeling in the cerebral cortex. Similarly, using a BAC transgenic Satb2-Cre mouse, Fu et al. transduced neurons in the caudal PB with a Cre-conditional vector (2019), but did not identify axonal labeling in the cerebral cortex. Our finding that a minority of cortically projecting neurons contain Satb2 immunoreactivity shows that a subset of neurons in this PB population do project axons directly to the cortex, and their cortical projections could be analyzed using a knockin-Cre mouse for Satb2 (Palmiter 2018).

Our results identify the first set of genetic markers for PB neurons that project axons directly to the cerebral cortex. In addition to showing that cortically projecting PB neurons are glutamatergic, we found that most of them belong to one of the two major developmental subpopulations of neurons in this brain stem complex—those that derive from the roof plate neurothelium and express the transcription factor Lmx1b (Asbreuk et al. 2002; Dai et al. 2008). These neurons are distinct from the larger population of intermingled neurons that derive from Math1/Atoh1-expressing precursors in the rhombic lip neurothelium, many of which express FoxP2 (Wang et al. 2005; Gray 2008; Rose et al. 2009; Verstegen et al. 2017). The functional significance of Lmx1b subpopulations expressing neuropeptides (CCK, Calca/CGRP), a steroid receptor (ERɑ), or Satb2, remains to be determined. These genes and proteins serve as useful markers, but how they affect the function of cortically projecting PB neurons is unclear.

Functional Role of Direct PB–Cortical Connectivity

Our results highlight the cerebral cortex as one of the most prominent output targets of the PB, yet the function of this projection remains untested. In fact, after the discovery that PB neurons project to the cerebral cortex 40 years ago (Saper and Loewy 1980), the function of this connection has received virtually no attention.

Because the insula contains primary gustatory cortex, some have assumed that cortically projecting PB neurons transmit taste signals (Lasiter et al. 1982; Shipley and Sanders 1982). Supporting this possibility, a minority of cortically projecting PB neurons express Satb2, which is also found in PB neurons that relay sweet taste signals to the thalamus (Fu et al. 2019). However, 80% of cortically projecting PB neurons lack Satb2, suggesting that they may transmit a more diverse set of information. Saper (1982b) suggested a role for direct PB–cortical projections in autonomic control, also hypothesizing that this connection may subserve cortical arousal or attention, based on a classic finding in cats, where stimulating the brain stem induces cortical potentials near the dorsolateral frontal pole (Starzl et al. 1951).

Beyond this, PB–cortical connectivity received limited attention until a study in humans identified resting-state functional connectivity between the PB region and two cortical areas: the frontoinsular cortex and pregenual anterior cingulate cortex (Fischer et al. 2016). These unexpected findings suggested that the direct connectivity between these brain regions in rodents may be relevant to understanding human brain function. Also, because this finding involves a region of the human brain stem where lesions produce coma, interconnectivity between the PB and these limbic cortical areas could contribute to the ascending arousal system, as proposed by Saper (1982a). A straightforward dependence of consciousness on PB–cortical connectivity would be at odds with evidence that wakefulness persists after bilateral damage to the insula in rodents or humans (Damasio et al. 2013; Chen et al. 2016), so the forebrain neurons that sustain arousal probably form a more extensive network than just this. Nonetheless, the massive interconnectivity between these structures suggests an important functional role and demands further investigation.

Converging evidence in rodents indicates that neuronal activity in the insula encodes hedonically relevant information beyond just taste, including aversive states like dehydration, malaise, or anxiety (Cechetto and Saper 1987; Yasui et al. 1991; Yamamoto et al. 1994; Castro and Berridge 2017; Gehrlach et al. 2019). Recent experimental work indicates that the insular cortex plays an important role in assessing and responding to homeostatic threats by promoting autonomic and other behavioral responses (Livneh et al. 2017; Gehrlach et al. 2019). Likewise, abundant neuroimaging and lesion data in humans suggests a broad role for the insula in interoception and emotion (Craig 2002; Jones et al. 2010).

Studying the functional role of PB–cortical connectivity will benefit from first addressing two neuroanatomical questions that represent major knowledge gaps: 1) What is the ascending input to cortically projecting PB neurons? and 2) What are the connectivity and functions of cortical neurons that receive direct input from the PB?

Ascending Input to Cortically Projecting PB Neurons

Importantly, we do not know what information the cortically projecting PB neurons receive. In general, PB neurons receive sensory information via ascending projections from the spinal cord (Cechetto et al. 1985; Panneton and Burton 1985; Bernard et al. 1995; Feil and Herbert 1995) or the NTS (Loewy and Burton 1978; Ricardo and Koh 1978; Herbert et al. 1990), as well as descending projections from the cerebral cortex, diencephalon, and amygdala (Moga et al. 1990), and are thought to relay this information to the forebrain. However, cortically projecting neurons occupy a subregion of the PB that appears to be devoid of input from both the spinal cord and the NTS and instead receives axonal projections from the spinal trigeminal nucleus (Dallel et al. 2004) and medullary reticular formation (Herbert et al. 1990).

We have very little information about the possible identity, connectivity, or function of neurons in the medullary reticular formation that provide input to the PB. However, one intriguing line of work identified sleep-promoting GABAergic neurons in what the authors referred to as the “parafacial zone” of the medullary reticular formation, which may inhibit glutamatergic neurons in the PB (Anaclet et al. 2014). This finding raises the possibility of a core arousal circuit with reciprocal, excitatory connectivity between (a) PB-projecting neurons in the cerebral cortex and (b) cortically projecting neurons in the PB, which are subject to inhibitory control by (c) GABAergic neurons in the medullary reticular formation.

Connectivity and Functions of Cortical Neurons That Receive Direct Input from the PB

We know little about the limbic and prefrontal cortical neurons that receive input directly from the PB. Learning cell-type–specific markers for the neurons that receive synaptic input from the PB within each region would be helpful for studying their connectivity and functions. Nonetheless, some general information about the connectivity, sensory responses, and functions of these cortical areas may help formulate hypotheses about the role of direct PB–cortical connectivity.

The most extensive PB terminal field targets the insular cortex, which receives a variety of information along its extensive rostrocaudal span. Rostrally, this cortical area receives taste information indirectly, from gustatory neurons in the caudal PB whose axonal projections relay through the gustatory thalamus (Norgren and Leonard 1971; Kosar et al. 1986a). Caudally, the insular cortex receives a spectrum of viscerosensory and other sensory information, most of which is likewise thought to relay through the thalamus (Cechetto and Saper 1987). It remains unclear what type(s) of information are conveyed by the PB directly to the insula, without relaying through the thalamus.

In particular, it is unclear whether and to what extent direct PB projections overlap gustatory input from the thalamus. The PB terminal field shown here concentrates densely in the intermediate, “dysgranular” subdivision of the insular cortex, which also receives input from the gustatory and visceral relay nuclei of the thalamus (Cechetto and Saper 1987). Gustatory-responsive subregions of the insular cortex often are described in relation to landmarks on the brain surface, where the middle cerebral artery intersects the rhinal veins (Yoshimura et al. 2004; Accolla et al. 2007; Chen et al. 2011). These surface landmarks are difficult to compare with our histologic results, but from limited data in axial tissue sections from mice (Chen et al. 2011; Peng et al. 2015; Levitan et al. 2019), it appears that most gustatory-responsive cortical neurons are rostral to the direct input from the PB. Taste input in rats concentrates at similar rostral levels (Kosar et al. 1986a; Kosar et al. 1986b; Ogawa et al. 1990). Axonal projections from the PB diminish rostrally, but the mouse insula does contain a more caudal area with bitter taste responsivity, where optogenetic stimulation triggers aversive behaviors that are associated with bitter and disgusting foods (Chen et al. 2011; Peng et al. 2015). Primary gustatory responses may cluster in small hotspots within the insular cortex (Penfield and Faulk Jr 1955; Ogawa et al. 1990; Chen et al. 2011), and direct PB–cortical projections appear to fill a roughly 2 mm gap between rostral (sweet) and caudal (bitter) gustatory areas, with possible overlap in the caudal region. This zone has been posited as a gustatory association area, where neurons do not cluster by taste modality (Fletcher et al. 2017). If PB projections deliver gustatory information directly to the insular cortex, it will be important to learn how their activity interacts with taste afferents that relay through the thalamus.

Whether or not direct PB–cortical projections supply gustatory information, they likely deliver nongustatory information as well because their projections target a large swath of cortex outside the insula, from the frontal pole back to the entorhinal cortex. The caudal insular cortex, which contains the caudal extent of the PB axon terminal field, contains a general visceral sensory area where neuronal activity can be modulated by homeostatic and hedonic needs like hunger and thirst or reflect states of anxiety (Livneh et al. 2017; Gehrlach et al. 2019). This caudal region receives mechanosensory information from the stomach, heart, and lungs (Cechetto and Saper 1987), and stimulation here can alter gastric motility, blood pressure, or respiration (Yasui et al. 1991; Gehrlach et al. 2019). In humans, the posterior insula is also the primary sensory area for pain and temperature (Craig et al. 2000; Mazzola et al. 2012). Electrically stimulating the human insula produces diverse, multimodal responses, including unpleasant tastes, nausea, vomiting, abdominal sensations, changes in gastric motility, pain, thermal sensations, and paresthesias covering multiple body regions (Penfield and Faulk Jr 1955; Mazzola et al. 2017). As with taste, it remains unclear whether or how direct projections from the PB influence these sensory experiences and visceral motor responses.

We lack detailed information about the connectivity and function of the midinsular zone (between gustatory and visceral areas) that receives the bulk of direct projections from the PB. Most work on insular connectivity focused on either its rostral (gustatory) or caudal (viscerosensory) areas (Shi and Cassell 1998; Jasmin et al. 2004; Murphy and Deutch 2018; Gehrlach et al. 2019). We are not aware of any experiments focused specifically on levels near the AC decussation, where PB–cortical projections produce the massive terminal field shown here. In rats, one WGA-HRP tracer injection into this midinsular region labeled descending axons projecting to the PB and NTS as well as afferents in the thalamus, hypothalamus, amygdala, and prefrontal cortex (Saper 1982b). However, even closely intermingled populations of cortical neurons may have vastly different output projections (Gehrlach et al. 2019), so identifying the connectivity of insular cortical neurons that receive direct synaptic input from the PB will require cell-type–specific markers or trans-synaptic viral tracing.

Outside the insular cortex, it is not clear what (if anything) other PB-innervated cortical areas have in common. The PB terminal field runs the full, rostrocaudal length of the cerebral cortex, but in contrast to the multilayer-spanning terminal field in the insular cortex, it targets primarily deep layers of the prefrontal and entorhinal cortices. Beyond speculating a possible modulatory role in cortical arousal, attention, or memory, it remains unclear how direct input from the PB affects these cortical areas.

Summary

Direct, reciprocal connectivity between the PB and the cerebral cortex is extensive and derives from specific subsets of neurons in each structure. Our findings in mice provide depth and context to the original identification of these direct connections in rats and possibly to the functional connectivity that has been identified between these structures in the human brain. The cell-type–specific markers identified here will help test the functions of this unusual and understudied set of direct circuit connections between the brain stem and cerebral cortex.

Funding

National Institute of Neurological Disorders and Stroke (K08 grant NS099425 to J.C.G.); University of Iowa Center for Aging (Aging Mind and Brain Initiative to J.C.G.).

Notes

We thank Alison Hsu for her assistance in counting synaptic boutons, Richard Palmiter for sharing Foxp2- and Calca-Cre mice, Carmen Birchmeier for supplying the Lmx1b antibody, and Hideki Enomoto for supplying the Phox2b antibody. We also acknowledge Clifford Saper, Nigel Pedersen, Gordon Buchanan, and Jady Tolda for their advice and constructive criticism of early drafts of this manuscript. All procedures performed in studies involving animals were in accordance with the ethical standards of the University of Iowa. Conflict of Interest: None declared.