Centrifugal Innervation of the Olfactory Bulb: A Reappraisal

Abstract The inter-regional connectivity of sensory structures in the brain allows for the modulation of sensory processing in manners important for perception. In the olfactory system, odor representations in the olfactory bulb (OB) are modulated by feedback centrifugal innervation from several olfactory cortices, including the piriform cortex (PCX) and anterior olfactory nucleus (AON). Previous studies reported that an additional olfactory cortex, the olfactory tubercle (OT), also centrifugally innervates the OB and may even shape the activity of OB output neurons. In an attempt to identify the cell types of this centrifugal innervation, we performed retrograde tracing experiments in mice utilizing three unique strategies, including retrobeads, retrograde adeno-associated virus (AAV) driving a fluorescent reporter, and retrograde AAV driving Cre-expression in the Ai9-floxed transgenic reporter line. Our results replicated the standing literature and uncovered robustly labeled neurons in the ipsilateral PCX, AON, and numerous other structures known to innervate the OB. Surprisingly, consistent throughout all of our approaches, no labeled soma were observed in the OT. These findings indicate that the OT is unique among other olfactory cortices in that it does not innervate the OB, which refines our understanding of the centrifugal modulation of the OB.


Introduction
In our sensory systems, the initial steps of information processing are directed not only by bottom-up sources from the environment, but also by top-down inputs from higher-order structures (Kiselycznyk et al., 2006;Gilbert and Li, 2013;Terreros and Delano, 2015). This type of centrifugal modulation allows early sensory representations to be shaped by factors such as learning, and internal states including hunger and arousal (Lavin et al., 1959;Gervais and Pager, 1979;Sullivan et al., 1989;Wachowiak et al., 2009;Fletcher and Chen, 2010;Kato et al., 2012;Ross and Fletcher, 2018). Feedback projections may arise from neuromodulatory loci as well as cortical structures, and thus can have a wide range of effects on sensory representations and perception, including state-dependent sensory gating and experiencedependent plasticity (Shepherd, 1972;Petzold et al., 2009;Bajo et al., 2010;Devore and Linster, 2012;Smith et al., 2012;Terreros and Delano, 2015;Ogg et al., 2018). Major questions remain regarding the neural circuitry underlying centrifugal modulation.
In the olfactory system, odors detected in the epithelium are first processed in the olfactory bulb (OB; Schoppa and Urban, 2003;Ache and Young, 2005;Wachowiak and Shipley, 2006). Next, OB principal neurons called mitral and tufted cells convey information to one of several olfactory cortices, including the anterior olfactory nucleus (AON), piriform cortex (PCX), and the olfactory tubercle (OT), each of which is thought to play a specialized role in odor processing (Scott et al., 1980;Haberly, 2001;Brunjes et al., 2005;Gottfried, 2010;Wesson and Wilson, 2011;Wilson and Sullivan, 2011). Additionally, all of these structures are reported to send centrifugal inputs back to the OB (Shafa and Meisami, 1977;Kiselycznyk et al., 2006;Markopoulos et al., 2012;Otazu et al., 2015), suggesting that each may contribute, perhaps in unique manners, to the modulation of early odor representations. For example, AON inputs to the OB are activated in odor-specific and state-dependent manners . These AON inputs directly activate mitral and tufted cells and indirectly drive local inhibitory circuits, resulting in a widespread inhibition which is proposed to aid in suppressing OB background activity (Markopoulos et al., 2012). Similarly, activation of PCX inputs to the OB during odor stimulation enhances odor-evoked inhibition of mitral and tufted cells, though this occurs via different elements of the OB microcircuit (Boyd et al., 2012;Markopoulos et al., 2012). Further, PCX inputs to the OB target mitral cells, but not tufted cells (Otazu et al., 2015). Together these results indicate that feedback projections from the PCX and AON may be poised to impact odor perception, perhaps differentially.
We set off with the goal to explore the organization, cell types, and functional role of these reported OT projections to the OB. To do so we employed tracing methods in non-transgenic mice and in a floxed reporter line. Ultimately, all the tracing methods we used failed to reveal OT projections to the OB, suggesting that the OT does not centrifugally contribute to early olfactory processing, in contrast with previous reports.

Surgical procedures
Mice were anesthetized with ϳ3% isoflurane in 1 l/min O 2 and mounted into a stereotaxic frame, equipped with a heating pad to maintain body temperature at 38°C. Depth of anesthesia was confirmed by lack of toe-pinch response and meloxicam analgesic was administered subcutaneously (5 mg/kg; Putney, Inc.). After removing fur, the scalp was cleaned using betadine followed by 70% ethanol. Subcutaneous marcaine (1.7 mg/kg; Hospira, Inc.) was provided locally before midline incision. A craniotomy was made above the structure of interest, a glass micropipette containing retrobeads or AAV was lowered into the brain, and the injection was given at a rate of 2 nl/s (see below for experiment-specific details). OB injections were given at 1.5 mm anterior to the rhinal sinus, 1 mm lateral, and 1.5 mm ventral unless noted otherwise (Table 1). All OT injections were given at 1.5 mm anterior bregma, 1.2 mm lateral, and 4.8 mm ventral. Following injection, the micropipette was slowly withdrawn from the brain, the craniotomy sealed with wax, and the wound closed with Vetbond (3M Animal Care Products). The mice were returned to group housing immediately following surgery and were allowed to recover on a heating pad.
All experiments are summarized in Table 1. A total of 6 C57BL/6J mice (n ϭ 5 male, n ϭ 1 female) received unilateral injections of retrobeads in the OB. Each mouse received either 200 nl of retrobeads 1500 m ventral to the surface (n ϭ 2) or 900 nl evenly dispersed between 2400 and 800 m ventral to the surface (n ϭ 4). These differing strategies were employed to explore whether spatial targeting of the retrobeads in the OB impacted the outcome. We found that injecting 900 versus 200 nl resulted in a similar number of labeled cells in the AON (56.9 Ϯ 12.8 vs 38.8 Ϯ 6 cells; mean Ϯ SEM) and PCX (39.9 Ϯ 5.2 vs 34.7 Ϯ 5.1 cells). A total of seven male C57BL/6J mice received 200-nl unilateral injections of AAVretro-GFP in the OB. An additional two male C57Bl/6J mice received a 500-nl injection of AAVretro-GFP in the OT. A total of six male Ai9 mice received 200-nl unilateral injections of AAVretro-Cre in the OB. Later, an additional three Ai9 mice (one male, two female) received three 200-nl OB injections each (at 1 mm, 1.5 mm, and 2 mm anterior to the rhinal sinus, 1 mm lateral, and 1.5 mm ventral), for a total of 600-nl AAVretro-Cre per mouse. An additional two Ai9 mice (one male, one female) received 500-nl unilateral injections of AAVretro-Cre in the OT.

Perfusion and histology
All mice were overdosed with Fatal-plus (0.01 ml/g; Vortech Pharmaceutical, Ltd.) and perfused with 10 ml of cold saline followed by 15 ml of 10% PB formalin. Brains Figure 1. Timeline of methods and experimental groups. Three groups of mice were used in three different experimental paradigms. A, C57BL/6J mice received retrobead injections into the OB and were perfused at 2 d. B, C57BL/6J mice received AAVretro-GFP injections into the OB and were perfused at two weeks. C, Transgenic Ai9 reporter mice received AAVretro-Cre injections and were perfused at two weeks; ‫ء‬three of nine Ai9 mice received multiple injections of AAVretro-Cre throughout one OB (see Materials and Methods; Table 1). Unless otherwise noted, OB injections were given at 1.5 mm anterior the rhinal sinus, 1 mm lateral, and 1.5 mm ventral, and OT injections were given at 1.5 mm anterior bregma, 1.2 mm lateral, and 4.8 mm ventral.
were stored in 10% formalin/30% sucrose (4°C) before sectioning. All brains were frozen and alternate coronal sections were obtained with a sliding microtome at 40-m thickness and stored floating in TBS with 0.03% sodium azide. Sections containing the OB, AON, anterior PCX (aPCX), and/or OT were rinsed in deionized water and mounted on slides using Fluoromount-G containing 4',6-diamidino-2phenylindole (DAPI; Invitrogen). We selected 4 -13 sections of each brain region for quantification, ensuring that sections spanned the anterior-posterior length of each region.

Imaging
Brain areas of interest (OB, AON, aPCX, OT) were identified based on the atlas of Paxinos and Franklin (Paxinos and Franklin, 2000) and images acquired of the hemisphere ipsilateral to the OB injection. Imaging was performed with a Nikon Eclipse Ti2e fluorescent microscope at 20ϫ magnification using a Nikon 16MP DS-Qi2 monochrome CMOS camera. Images of the OT following injections of AAV locally within the OT were acquired at 40ϫ magnification. Images of the intact brain were taken using a 12 MP digital camera, and for fluorescence a Nikon AZ100 microscope at 1ϫ magnification using a Photometrics CoolSNAP DYNO CCD camera. Image acquisition settings, including gain, exposure, and light intensity, were held constant across all images and samples within treatment conditions.

Quantification and statistics
Successful injection was confirmed by observation of labeling in structures known to robustly innervate the OB (AON and aPCX; Price and Powell, 1970;Luskin and Price, 1983;Boyd et al., 2012;Markopoulos et al., 2012;Padmanabhan et al., 2016). One out of seven AAVretro-GFP mice and two out of six AAVretro-Cre mice were excluded due to a lack of fluorescence anywhere in the brain, likely due to a mechanical failure of the injection.
An ROI bounding box (500 ϫ 250 m) was overlaid within each brain region of interest (aPCX, AON, OT), with effort made to hold the location of this bounding box constant across mice. Using semi-automated thresholding methods in NIS Elements (Nikon), we identified cells within these ROIs, allowing for an unbiased estimation of cell numbers. This first involved preprocessing of the image to decrease background fluorescence and thereby enhance contrast. Then cells were identified based on their fluorescence intensity (via threshold) and their size. Lastly, detected objects were post-processed based on their area of fluorescence, resulting in the elimination of objects too small to be cells (e.g., brightly labeled fibers in the Ai9 paradigm). Due to overt differences in cell-filling across our three tracing paradigms, these methods were optimized for each method individually, but an identical method was used across all images within a paradigm. Representative results from the semi-automated cellcounting procedures for each experimental paradigm are shown in Figure 2.
For each region analyzed, multiple sections were quantified for each mouse, with some variation in the number of sections quantified for each mouse (n ϭ 16 mice, 6.75 Ϯ 0.38 sections per brain region per mouse). This variation was due to occasional histologic imperfections (e.g., bubbles in the mounting medium or torn tissue) that precluded accurate quantification and resulted in the exclusion of some samples. To avoid overrepresenting mice for which more samples were analyzed, we calculated a mean for each region, and averaged these means across mice for each region. Any ROIs resulting in values exceeding two standard deviations outside the mean for that ROI were eliminated from all quantification and statistical analyses. This largely was applied toward retrobead-treated tissue wherein some sections had abundant fluorescence resulting from residual retrobeads collecting on the microtome blade and being deposited on latter sections.

Retrobead labeling suggests lack of OT to OB innervation
In our first experimental paradigm (Fig. 1A), we injected retrobeads unilaterally in the OB of C57BL/6J mice (Fig.  3A). At the site of injection, retrobeads are endocytosed and retrogradely transported to the soma, resulting in an accumulation of retrobeads and fluorescently labeled soma. As expected, this resulted in fluorescent soma in the AON and aPCX, two structures known to innervate the OB (Price and Powell, 1970;Luskin and Price, 1983;Boyd et al., 2012;Markopoulos et al., 2012;Padmanabhan et al., 2016;Fig. 3B-E). Unexpectedly, our automated cell counting returned only a few positive OT values, from just a subset of tissue sections (Fig. 3F,G). On visual inspection of the counted "cells" in the OT, and inspection of z-stack images, we determined that these were non-neuronal fluorescent puncta (Fig. 2B), likely residual retrobeads deposited on the tissue during sectioning.

Viral tracing further supports lack of centrifugal OT to OB input
Given the surprising lack of OT fluorescence observed in the retrobead experiment, we sought to confirm our findings by using a viral labeling technique (Fig. 1B) to rule out the possibility of a false negative result. We unilaterally injected C57BL/6J mice in the OB with AAVretro-GFP to label neurons projecting to the OB (Fig. 4A). This virus drives GFP expression under control of the human synapsin promoter, and is thus capable of labeling any neuronal cell type (Kügler et al., 2003), including medium spiny neurons of the ventral striatum (McLean et al., 2014) of which the OT is largely comprised. With this independent paradigm, we observed many GFP-labeled neurons in the AON and aPCX (Fig. 4B-E), indicating that we were successful in labeling neurons that centrifugally innervate the OB. In contrast, we again observed a lack of fluorescent neurons in the OT (Fig. 4B-G).
To ensure that the virus we used is capable of infecting and driving GFP expression in OT neurons, we unilaterally injected the OT directly with AAVretro-GFP in a separate cohort of C57BL/6J mice. We observed GFP expression in OT neurons across all cell layers of the OT (Fig. 4H), including in the islands of Calleja, suggesting that the lack of GFP expression in the OT following AAVretro-GFP injection into the OB is not due to an issue of AAV tropism for OT neurons.

Cre-dependent labeling strategy confirms lack of OT to OB pathway
Finally, we employed a third strategy in an attempt to label this pathway. We used Ai9 reporter mice, which robustly express tdTomato in a Cre-dependent manner (Madisen et al., 2010). We unilaterally injected AAVretro-

Negative Results
Cre (which drives Cre expression under control of the Ef1␣ promoter and is thus capable of driving Cre expression in any mammalian cell) into the OB (Figs. 1C, 5A). This strategy resulted in robust tdTomato expression in structures known to innervate the OB, including the aPCX and AON (Fig. 5B-E). We observed a complete and striking lack of cells in the OT (Fig. 5F), which was in stark contrast to the plentiful cells observed in the AON and the aPCX (Fig. 5G).
Once again, to ensure that the virus we used is capable of infecting OT neurons, we injected AAVretro-Cre directly into the OT of a separate cohort of Ai9 mice. As for AAVretro-GFP (Fig. 4H), this approach yielded fluorophore-expressing cells throughout the OT (Fig. 5H).
Finally, we wanted to ensure that we were not inadvertently missing any regions of the OB that may be the target of OT centrifugal input. Indeed, it is possible OT projections to the OB may innervate notably small portions of the OB, which might be missed with the single bolus injections largely employed in the previous approaches (although those injections often did span OB cell layers (Figs. 4A, 5A). We also sought to ensure our retrograde OB injection approaches were capable of labeling neurons in structures other than just the aPCX and AON. To accomplish these goals, we unilaterally injected an additional cohort of Ai9 mice with AAVretro-Cre at three sites along the anterior-posterior axis in the same OB and collected tissue sections throughout the entire brain (in contrast to the previous paradigms wherein we only collected forebrain tissue and analyzed the aPCX, AON, and OT). This strategy resulted in robust tdTomato expression throughout the entire OB (Fig. 6A), which was evident even on looking at the intact brain (Fig. 6B). tdTomato expression appeared absent within the OT when viewing the ventral side of the intact brain with the naked eye and under epifluorescence (Fig. 6B). After sectioning, we again did not observe fluorescent neurons in the OT (Fig. 6C), in contrast to significant labeling in many other structures known to innervate the OB, including the aPCX (Fig. 6C), dorsal tenia tecta (Shipley and Adamek, 1984; Fig. 6C), AON (Fig. 6D), horizontal diagonal band of Broca (Ichikawa and Hirata, 1986;Devore and Linster, 2012; Fig. 6E), magnocellular preoptic area (Carson, 1984; Fig. 6E), posterior PCX (Fig. 6F), lateral hypothalamus (Shipley and Adamek, 1984;Fig. 6G), basolateral amygdala (Fig. 6H), median and dorsal raphe nuclei (Steinfeld et al., 2015;Brunert et al., 2016;Fig. 6I,J), and locus coeruleus (Shipley et al., 1985;McLean et al., 1989;Fig. 6K). Because this approach resulted in very thorough labeling of the OB, as well as Negative Results labeling in many areas known to innervate the OB, these results provide strong evidence that the OT does not send centrifugal projections to the OB (Fig. 7). Together with our results from 16 mice across three different quantitative experimental approaches (six retrobead, six AAVretro-GFP, and four AAVretro-Cre; Figs. 3-6), we observed no OB-projecting OT neurons, providing strong support for our conclusion that the OT does not project to the OB.

Discussion
Here, we demonstrate a lack of input from the OT to the OB. While we initiated this study with the goal of investigating the anatomic organization and physiologic role of this reported pathway (Shafa and Meisami, 1977), our results ultimately lead us to the conclusion that it does not exist. To minimize the chance of a false negative result, we used three independent, widely-used tracing meth-ods, each of which yielded congruent results. Our use of a retrograde labeling strategy eliminated the possibility that we may have introduced non-specific labeling due to off-target effects that can accompany anterograde tracing methods when the structure of interest is difficult to target due to small size or irregular shape, like the OT. The regions of interest were chosen in the same portion of the structures across all animals within each experiment to ensure fair sampling. Further, we set rigorous standards for sample exclusion, excluding only samples that showed no labeling in regions with very wellcharacterized, dense projections to the OB.
In only one of our experimental paradigms, the retrobead paradigm, did our semi-automated cell counting return positive cell counts in the OT. For some of these instances, we performed z-stack imaging which indicated that the cells counted were actually non-neuronal fluores-  Figure 3D. Scale bar ϭ 500 m. C, aPCX and OT GFP labeling following injection of AAVretro-GFP in the OB. Boxes indicate regions in Figure  3E,F. Scale bar ϭ 500 m. D, Enhanced view of boxed AON region in B. Scale bar ϭ 100 m. E, Enhanced view of boxed aPCX region in C. Scale bar ϭ 100 m. F, Enhanced view of boxed OT region in C. Scale bar ϭ 100 m. G, Quantification of GFP-labeled cells across all three brain regions. Each point represents one animal's mean; n ϭ 6 mice, 4 -12 sections (6.89 Ϯ 0.54; mean Ϯ SEM) each. H, GFP labeling in the OT following injection of AAVretro-GFP in the OT; n ϭ 2 mice. VP, ventral pallidum; ICj, islands of Calleja; ICj, borders were approximated based on DAPI staining. Scale bar ϭ 100 m. i.-iii., layers 1-3; D, dorsal; M, medial; L, lateral. cent puncta. It seems those artifacts were residual retrobeads that were deposited by the microtome blade onto subsequent sections (an issue we observed and noted during the tissue sectioning). This issue may also contribute to the fact that we counted slightly more cells in the aPCX in the retrobead experiments compared to the AAV experiments. In support of OT cells not innervating the OB, the results we obtained using the two viral tracing techniques indicated a complete absence of cells in the OT.
We did observe some slight differences in the numbers of cells labeled by each tracing method. For example, we observed slightly more retrobead-labeled neurons in the aPCX compared to either viral tracing method, which were likely due to differences in the efficiency of endocytosis and transport of the retrobeads compared to virallymediated GFP or Cre expression. Importantly, each method indicated the same pattern of labeling, with the densest labeling in the AON, followed by the aPCX, and no labeling in the OT. Overall, the slight differences we observed with each technique highlight the importance of using multiple strategies in our attempt to identify OBprojecting OT neurons.
The failure of all three of our primary strategies to label OT neurons projecting to the OB provides strong support for our conclusion that this population does not exist. This is further strengthened by the outcomes of our follow-up experiments, wherein we injected multiple boluses of  AAVretro-Cre widely throughout the OB of Ai9 mice. This revealed, as expected, tdTomato expression in nearly one dozen brain structures with known innervation of the OB, but not the OT (Fig. 6).
Our results are surprising, given the previous reports that suggested the existence of this pathway. In one study (Shafa and Meisami, 1977), the authors injected HRP into the rat OB. The authors concluded that there was some "light" labeling in the OT but did not provide quantitative analysis, and only presented minimal primary data. Two studies (Heimer, 1968;Gervais, 1979) revealed degeneration in the OB following lesions to the OT, suggesting that the OT sends feedback to the OB. The Heimer study (Heimer, 1968) mentions the difficulty of lesioning the OT without damaging neighboring structures, suggesting offtarget effects (namely, damage beyond the OT) may contribute to their result. The lesioning strategy used in the Gervais paper (Gervais, 1979) is subject to the same caveat. Indeed, some of their OT lesions resulted in damage to part of the lateral olfactory tract (which provides input to all olfactory cortices), thus introducing the possibility of highly non-specific effects. In both cases, even if lesions were perfectly confined to the OT, OB degeneration could result from the loss of an indirect connection to the OB via a third structure, rather than the loss of a monosynaptic connection.
A more recent study (Zhang et al., 2017b) used viral anterograde and retrograde tracing techniques to study the inputs and outputs of the OT. With an anterograde AAV injection into the OT, the authors reported labeling throughout the OB, with the majority of fluorescence localized within the glomerular layer of the OB. As with the lesioning experiments discussed above, it can be quite difficult to exclusively target an injection to the mouse OT without any leak into surrounding areas (like the aPCX which does project to the OB). Here, we avoid this caveat by using three independent retrograde labeling strategies, each of which indicates a lack of projection from the OT to the OB.
It is well established, and further supported by our data, that the AON and PCX provide centrifugal innervation to the OB (Price and Powell, 1970;Luskin and Price, 1983;Boyd et al., 2012;Markopoulos et al., 2012;Padmanabhan et al., 2016). However, our results indicate a lack of direct input from the OT to the OB. What might this indicate about the role of the OT within the olfactory system? OT neurons receive input from the OB and encode odors in a cortical-like manner, suggesting a role in odor processing (Wesson and Wilson, 2010). Further, these responses are modulated by attention (Zelano et al., 2005;Carlson et al., 2018), and it has previously been speculated that the OT may play a role in top-down, state-dependent modulation of OB activity (Gervais, 1979;Wesson and Wilson, 2011). Ultimately, our data lead us to conclude that it is very unlikely the OT directly modulates OB activity, suggesting that statedependent modulation of OB activity must result from other sources of input (e.g., AON, aPCX), or indirect input from the OT by means of other structures. Of course, our data do not rule out the highly likely possibility that the OT may indirectly, via di-or even tri-synaptic connections, influence OB activity through its projections to regions that do innervate the OB, like the aPCX or even AON. Future work could attempt to address this possibility using optogenetic and electrophysiological methods.