All experiments were carried out in accordance with guidelines of the Canadian Council on Animal Care and the UBC Animal Care Committee. Hoxb8-IRES-Cre (RRID:IMSR_JAX:035978), Hoxb8-Cre (obtained from the lab of Hanns Zeilhofer, ETH Zurich, Switzerland), Hoxb8-FlpO (C57BL/6-Tg(Hoxb8-flpo)11Oki/Kctt, sperm obtained from the European Mutant Mouse Archive) mice were crossed with appropriate TdTomato (tdt) reporter mice (Ai14, 007914, RRID: IMSR_JAX:007914; Ai65, 021875, RRDI: IMSR_JAX:021875) (Figures 1A–E). Mice were kept on a standard diet and light cycle.

One male and one female (n = 2) of the Hoxb8-IRES-Cre^tdt strain, two males and one female (n = 3) of the Hoxb8-Cre^tdt strain, and three males and one female (n = 4) of the Hoxb8-FlpO^tdt strain 8–12 weeks of age were deeply anesthetized with a 1,500 mg/kg intraperitoneal injection of urethane and subsequently perfused with heparinized saline followed by 4% paraformaldehyde (PF) in 0.1 M phosphate buffer (PB). The brains and rostral portion of the spinal cords were removed and fixed overnight in 4% PF in 0.1 MPB, after which they were equilibrated in 20% sucrose in 0.1 M PB. Four percent PF-fixed brains and cervical spinal cords of Hoxb8-T2A-FlpO^tdt (n = 2, 1 M, 1F) mice were generously provided by Dr. Victoria Abraria (Rutgers University) (Bohic et al., 2023). Photos of each specimen were obtained under UV illumination visualize the rostral boundary of reporter expression in the brain and cervical cord. The brains were then frozen in optical cutting temperature compound and stored at −80°C.

One neonatal Hoxb8-FlpO^tdt mouse pup (2d) was deeply anesthetized using isoflurane and decapitated. The head was fixed for 48 h in 4% PF in 0.1 M PB, after which the brain was removed, fixed, and equilibrated in sucrose as described above.

Four additional Hoxb8-Cre^tdt mice underwent lateral hemisection at C3 in order to injure all reporter-positive spinofugal axons on one side of the animal. Immediately prior to surgery, animals received subcutaneous injections of buprenorphine (0.01 mg/kg) and lactated ringers (0.1 mL). The animals were anesthetized with isoflurane (5% induction, 2–3% maintenance) following which eye lubricant was liberally applied, the upper back was shaved (excess hair removed with depilatory cream), and the skin surface was disinfected with ethanol and chlorhexidine. Mice received a midline incision over the upper cervical spinal cord, and a C3-C4 laminectomy was performed. Hemisections were made by first piercing the dura and cord with a 25 gauge needle just to the right of the dorsal midline vein. One blade of a pair of microscissors was then inserted into the puncture to the bottom of the spinal canal; the other was placed outside the lateral side of the left hemicord and the scissors were closed. The muscle and skin layers were closed with sutures. Mice survived for 6 days (n = 1), 2 weeks (n = 2) or 4 weeks (n = 1) during which they underwent appropriate monitoring and post-surgical care. Mice received buprenorphine (0.01 mg/kg) and lactated ringers (0.1 mL) subcutaneously three times daily for the first 3 days post-surgery.

Eight brains from intact mice (two adults of each genotype) were serially-sectioned in the coronal plane at 100 microns using a Cryostat (Micron). One male Hoxb8-Cre^tdt and one male Hoxb8-FlpO^tdt brain were also sectioned parasagittally at 100 microns. An additional male Hoxb8-FlpO^tdt mouse was sectioned coronally at 50 micron sections and thaw mounted to Superfrost Plus slides. The 2d-old mouse pup brain was also sectioned coronally at 100 microns. Four-mil (~101 mm) rectangular adhesive vinyl frames custom made on a Cricut (Explore Air) were applied to Superfrost Plus slides before thaw-mounting sections to prevent tissue compression during coverslipping. The slides were air dried for a minimum of 2 h and washed with phosphate-buffered saline (PBS, 2x, 10 min each). Slides were then delipidated overnight with Cubic-L (TCI, Cat# T3740) to which NaCl was added (to 0.45%) to prevent tissue swelling and detachment from the slides. The following day, the refractive index (RI = 1.520) was matched to glass using Cubic-R+(M) (TCI, Cat# T3741) in 2 × 20 minute applications. Finally, slides were coverslipped using 24 × 50 mm coverslips and Cubic mounting medium (TCI, Cat# M3294) or ProLong Glass mountant (ThermoFisher Cat# P36984).

Sections were permeabilized and blocked with 10% normal donkey serum and 0.2% Triton X-100 plus 0.02% sodium azide in PBS (PBSTx azide) for 1 hour before the primary antibodies were applied (Table 1) for ~48 h. Prior to application of the secondary antibodies (Table 1), the slides were washed three times (5 min each) with PBS. Secondary antibodies were applied overnight at room temperature and subsequently washed 3 times for 5 min each in PBS. The refractive index was matched to glass using Cubic-R+(M) (TCI, Cat#T3741) in one 20-min application. Slides were coverslipped using 24x50mm coverslips and the Cubic mounting media (TCI, Cat# M3294) before being sealed with nail varnish. Slides with 50 micron-thick sections were not cleared with Cubic reagents, and instead coverslipped using ProLong Glass anti-fade mounting media (Invitrogen, P36984).

The spinal cords (cervical to lumbar) were harvested from an additional five PF-perfused mice (one of each genotype and a second Hoxb8-Cre^tdt mouse) to examine neuronal recombination in the spinal cord. In this case, 50-micron frozen sections were thaw-mounted onto glass slides and counterstained with NeuroTrace^™ 640/660 Deep-Red Fluorescent Nissl Stain (1:200 for 2 h, ThermoFisher Catalog number: N21483).

Images of coronally-sectioned adult brains were taken using a 5x objective on a Zeiss LSM800 scanning confocal microscope and Zen Blue software. Each brain section was imaged in all dimensions using tiling and z-stacking (4–6 z-slices with an optical section thickness of 40–50 microns after orthogonal projections). The images were stitched during acquisition, orthogonally projected, and a single pixel filter was applied to remove digital gain artifacts. To begin registration to the Allen Mouse Atlas coronal average template, any rotation of the microscope images in the rostro-caudal plane was determined using QuPath. The Allen Mouse Atlas average coronal template was then digitally sectioned to account for this rotation (which was always less than ±2.5^o). Fiducials were selected that included areas of high TdTomato reporter intensity, major surface structures, and internal landmarks (ex: inner surfaces of the ventricles). In FIJI, 8–12 fiducials were selected on each Allen Mouse Atlas section and then the same landmarks were chosen on the corresponding microscope image (FIJI > region of interest). The thickness of the microscope images and the Allen Mouse Atlas slices were not identical, therefore adjustments were made by skipping an atlas slice or doubling up (i.e., registering multiple microscope images to the same Allen Mouse Atlas slice) to minimize divergence. In general, an effort was made to keep the fiducials in the same location and only make gradual changes as we moved through the Allen mouse Atlas along the z-axis. The images were registered to the Allen Mouse Atlas coronal average template using FIJI (moving least squares with similarity method in the registration plugin) and stacked (in tools, stack sorter). It should be noted that the slight rotation along the rostro-caudal axis of our sections (and registration to a slightly rotated average coronal template).

Images of sections for immunohistochemistry were taken using a 20x or 63x objective using a Zeiss LSM800 scanning confocal microscope and Zen Blue software for the above sections. Super-resolution microscopy (63x) was used to capture images of p73-labeled cells (Airyscan).

One intact Hoxb8-IRES-Cre^tdt mouse and one Hoxb8-Cre^tdt mouse, hemisected 28 days earlier, were deeply anesthetized with a 1,500 mg/kg intraperitoneal injection of urethane and perfused with heparinized saline followed by 4% PF in 0.1 M PB. Their brains were removed and fixed overnight in 4% PF in 0.1 M PB. Brains were washed for 6 h (3 × 2 h) in PBS on a room temperature shaker protected from light. Brains were delipidated for 7 days with Cubic-L (TCI, Cat #T3740) to which NaCl was added (to 0.45%) to prevent tissue swelling in a 37°C incubated shaker protected from light. The Cubic-L solution was changed daily for the first 4 days, after which it was changed every second day. The refractive index (RI = 1.520) was matched to glass using Cubic-R+(N) (TCI, Cat# T3983) for 2 days on a room temperature shaker protected from light. The cleared cerebella were embedded in 2.5% agarose in Cubic-R+(N) (TCI, Cat#T3983). Images of the cerebella were acquired using a Zeiss Z.1 Light sheet microscope with Zen Black software with tiling and z-stacking.

Transgenic Hoxb8-driven reporter expression was easily identified with UV illumination (Figure 1A). Fluorescence in knock-in strains extended from the tail to the nape of the neck, and included the forelimb; this was not the case in non-targeted strains, in which the forelimb was spared, and reporter expression was limited to skin below mid-torso (Figure 1A). Developmental Hoxb8 mRNA expression extends from the caudal end of the spinal cord to the brainstem (Witschi et al., 2010). This is reflected in the extent of reporter expression in the CNS of Hoxb8-IRES-Cre^tdt and Hoxb8-T2A-FlpO^tdt knock-in strains (Bohic et al., 2023 and Figures 1B–E), as demonstrated under UV illumination. In the transgenic reporter models (Hoxb8-Cre^tdt and Hoxb8-FlpO^tdt), the rostral expression boundary was close to C4 (Figures 1D,E), as expected based on previously demonstrated expression of the transgenic constructs in mouse embryos and adult reporters (Charité et al., 1995; Witschi et al., 2010). In three of the mouse lines maintained in our facility, recombination occasionally occurred throughout the body. Such mice were excluded from analyses.

With rare exceptions, all spinal cord neurons caudal to Hoxb8 expression boundaries had undergone recombination in all reporter lines (Figure 2). Tdtomato intensity was expectedly intense in the dorsal grey matter given the high neuronal density. In non-targeted transgenics, only axon terminals could be detected in the cord above C4 (Figure 2E). In the thoracic and lumbar cords (but not the cervical cord), vascular-associated cells were apparent, in line with the mesodermal distribution of Hoxb8 expression during development (Deschamps et al., 1999). We are thus able to conclude that all spinofugal projections are tdtomato-positive (but only up to C4 in non-targeted reporter strains).

Spinofugal projections were labeled in all four Hoxb8 reporter lines, but to slightly different extents depending on brain region (Figures 3–10). Primary afferent terminals (and possibly post-synaptic dorsal column axon terminals) were dense in the gracile (Gr), cuneate (Cu) and external cuneate nuclei (ECu) in all strains (Figures 3B,C), as were a few neuronal somata in the gracile nucleus (but not the other dorsal column nuclei) in targeted transgenics [Figure 3D and (Bohic et al., 2023)]. Other axons extended from ventral ascending tracts into the reticular nuclei (RN). Some neurons were apparent unilaterally in the caudal spinal nucleus of the trigeminal nerve (SPVC) in Hoxb8-IRES-Cre^tdt mice due to a slightly yawed section (Figure 3B and data not shown). Labeling was also more pronounced in the spinal tract of the trigeminal nerve in the knock-in models, but these were not trigeminal in origin, given the lack of terminal labeling in the SPVC. Some neurons of the nucleus of the solitary tract were apparent in knock-in strains (NTS, Hoxb8-IRES-Cre^tdt > Hoxb8-T2A-FlpO^tdt); these were absent in the untargeted transgenic reporter mice (Figures 3B,C).

At the level of the anterior cerebellum and facial nucleus (Figure 4), spinocerebellar axons were apparent in lobules as stripes (Figures 4B,C). This was true for lobules II, III, IV, VIII and IX, with no obvious differences in density between strains. In Hoxb8-IRES-Cre^tdt mice, but in no other strain, a population of spinofugal (or possibly bulbofugal) axons was present in the core of the facial nucleus (Figure 4C).

In the midbrain (Figures 5, 6), axon terminals of the periaqueductal grey received a dense projection of reporter-positive axons, which extended well into the thalamus. In targeted transgenic models (Hoxb8-IRES-Cre^tdt and Hoxb8-T2A-Cre^tdt), many terminals were apparent in the external nucleus of the inferior colliculus (ECIC), a region that receives input from the spinal cord and brainstem (Aitkin et al., 1978; Davidson et al., 2010). Their absence in non-targeted strains revealed thick axons that continued rostrally (see also Figure 7) toward the thalamus (Figure 7). Axons projecting to the intermediate grey of the superior colliculus formed a meshwork in all mice (Supplementary Video S1). Decussating axons were also present in deeper white matter (intermediate white, iw) of the superior colliculus; these connected the left and right deep grey of the superior colliculus, which itself was continuous with the ig (Figure 7).

The lineage-tracing apporach offered an unprecedented view of the routes taken by spinofugal axons toward the tectum and thalamus (Figure 7). A major point of diversion of anterolateral ascending tracts was at the level of the midbrain-hindbrain boundary, where spinotectal and spinothalamic axons left the inferior cerebellar peduncle, some of which projected to the PAG and the intermediate grey (ig) layer of the superior colliculus (Figure 7). These finer axons with small synaptic terminals additionally projected to two unnamed vertically-oriented bands extending from the ig dorsomedially to the aqueduct ventromedially, and to the lateral border of the PAG ventrolaterally. At this same level, there was a prominent decussation of middling-calibre axons in the intermediate white matter. The thickest axons leaving the middle cerebellar peduncle traveled dorsally in the midbrain, and curved rostrally toward the diencephalon (Figure 7). These, like those in the medial lemniscus, approached the caudal end of the ventral posterior lateral nucleus of the thalamus and the posterior intralaminar nucleus (PIL), but from a dorsal rather than a ventral direction (Figure 7D).

The large-calibre axons ran rostrally through the colliculi toward posterior complex of the thalamus (PO), clustering at the dorso-medial border of the medial geniculate nucleus, and more rostrally at the dorso-medial border of the ventral posterior medial nucleus of the thalamus (VPM) (Figure 8). In the Unified Atlas, this region is labeled as the external medullary lamina rostrally, and is continuous with the superior thalamic radiation (str) caudally. From there, they traversed the MGN and VPM, and arborized in the posterior intralaminar nucleus (PIL) and ventral posterior lateral nucleus (VPL), respectively (Figure 8). The medial part of the medial geniculate nucleus (multimodal part) was particularly heavily labeled in all strains.

The density of dorsal spinothalamic axons was much greater in knock-in strains, probably because the majority of spinothalamic axons originate in the cord above C3 (Davidson et al., 2010). In the pons at the level of the midbrain-hindbrain boundary, a population of axons of the inferior cerebellar peduncle merged with those of the medial lemniscus and ascended toward thalamic nuclei (Figures 7, 8).

A previously undescribed tract of axons emerged from the vicinity of the caudal end of the VPL or possibly the rostral end of the PIL, ran ventrally between the internal capsule and optic tract, and decussated just posterior to the optic chasm (oc) in the ventral-most part of the hypothalamus (Figures 8E,F). We call this the spinofugal hypothalamic decussating tract (shxt), and its decussation (shx). This was more prominent in targeted knock-ins, indicating a contribution from spinal neurons above C4 (Figure 6E). The zona incerta (ZI), appeared to be a site of shxt termination (double yellow arrows in Figure 5A).

While axons in the ventral posterior lateral nucleus (VPL) of the thalamus were apparent in all strains (Figure 8), labeling was slightly less extensive in the mediolateral and rostro-caudal axes in the non-targeted Hoxb8 reporters (Figure 8). Spinofugal axons in the lateral PO were also markedly less prominent in the non-targeted strains. The same was true of putative shxt terminal density in the zona incerta (ZI, Figure 8). More dorsally, labeled axons were apparent in the medial part of the PO, as well as in the parafascicular nucleus (PaF) (Figure 8).

The existence of spinofugal hypothalamic decussation implies that some thalamus-projecting axons either ascend ipsilaterally or contralaterally to their origins in the spinal cord, and then cross (or cross again) in the thalamus. We subjected a small number (4) of Hoxb8-Cre^tdt mice to lateral spinal hemisection at C3 (Figure 9) – this manipulation should remove all reporter positive axonal input to the brain on the hemisected side. Gross inspection of tdt fluorescence confirmed loss of reporter-positive terminal ipsilaterally (and to a lesser extent contralaterally) in the cervical cord (Figure 9A), pons and midbrain (Figure 9B), and caudal thalamus including the ZI (Figure 9C). While this indicates ipsilateral ascension from neurons in the uninjured side of the cord, we could not readily trace a path from the ipsilateral thalamus to the contralateral ZI. We did find, however, a surprising preservation of axons in the anterior VPL nucleus (Figure 7D) at two (and even four) weeks post-hemisection (Figures 9E,F), raising the possibility of recurrent projections from the ipsilateral anterior VPL to the more posterior contralateral ZI.

Further rostrally, reporter-positive axons were present in both knock-in strains in a discrete region of the bed nucleus of the stria terminalis (BNST) ventral to the anterior commissure (Figures 10B,C), which we suspect arise from the nucleus of the solitary tract (NTS), based on previous tracing studies in the rat (Riche et al., 1990; Lebow and Chen, 2016), and the absence of recombination in the NTS of non-targeted strains. At the rostral pole of the brain, but only in Hoxb8-Cre^tdt mice, axons of the vomeronasal nerve (VON) and their terminations in glomeruli of the accessory olfactory bulb were reporter positive (Figure 10D). Interestingly, these were more prominent in female than male specimens.

In Hoxb8-Cre^tdt mice, scattered neurons in the cortex, hippocampus and cerebellum were reporter-positive, but in no consistent way; a cerebellar Purkinje cell (PC) is shown in the bottom right panel of Figure 11A.

In addition to VON axons (Figure 10), and as reported previously (Szabo et al., 2015), a population of Hoxb8-lineage neurons was present in the anteroventral (AV) and anteromedial (AM) nuclei of the thalamus in the Hoxb8-Cre^tdt mice (Figure 11A). We were able to follow the course of their axons, at least from the AV nucleus, laterally (Figure 11B) and anteriorly (Figure 11C) through the dorsal part of the striatum, after which they converged in the cingulum bundle (cing, Figure 11D). In caudal sections at the level of the midbrain, these radiated toward the retrosplenial cortex (RSPv Figures 11A,E,G) and additionally terminated in the subiculum (SUB, Figure 9G). More anteriorly, termination zones of cingulum bundle axons could be found in the anterior cingulate cortex (AC) (Figure 11F). These results were unchanged following spinal hemisection (data not shown), and agree with projections of the AV and AM nuclei demonstrated in the Allen Mouse Brain Connectivity Atlas.¹

A consistent finding in Hoxb8-Cre^tdt mice was the presence of reporter-positive astrocytes (verified with Sox9 immunohistochemistry) (Sun et al., 2017) forming a distinct border between ventral midbrain and thalamus at the cephalic flexure (CF) (Figures 11A,H–J).

As was the case for other scattered cells, a few recombined retinal ganglion cell axons were apparent in the optic tract (Figure 12A), and optic chiasm (Figure 12B, Inset). These followed the optic tract dorsally and terminated in the dorsal lateral geniculate nucleus of the thalamus, as well as in the superficial grey of the superior colliculus in discrete tufts (Figure 12B) – these were also unaffected by C3 hemiseciton (data not shown).

In Hoxb8-FlpO^tdt mice, we identified a population of cells on the outer molecular layer of the dentate gyrus as late-surviving Cajal Retzius cells based on tadpole-like morphology and their expression of p73 (Figures 13A–D) (Anstötz and Maccaferri, 2020).

The choroid plexus of Hoxb8-FlpOtdt mice contained a dense population of tdtomato-positive PDGFRa-positive mesenchymal cells (fibroblasts) (Figures 13E,F). This contrasted with the other mouse lines, in which reporter-positive choroid plexus cells were of the monocytic lineage, expressing Iba1 (Figure 13F) and f4/80 (data not shown).

Previous studies using Hoxb8 reporter mouse lines have highlighted the presence of Hoxb8-lineage microglia, thought to comprise one-third of all adult microglia in the mouse brain (Chen et al., 2010; Szabo et al., 2015; Bohic et al., 2023). Microglia were present in all Hoxb8 reporters assessed here, but differed markedly in abundance. Hoxb8-lineage microglia were most numerous in the cortex of Hoxb8-IRES-Cre^tdt mice followed by Hoxb8-Cre^tdt mice (Most readily apparent if Figures 8D, 10B,C). Interestingly, even fewer microglia were present in the other knock-in model (Hoxb8-T2A-FlpO^tdt) (Bohic et al., 2023). There were almost no Hoxb8-lineage microglia in Hoxb8-FlpO^tdt mice.

The overarching aim of this study was to determine the feasibility of using Hoxb8 reporter mice to study the distal anatomy of spinofugal projections, with a longer-term goal of understanding their plasticity in the wake of neurotrauma such as spinal cord injury.

We have used the YSK Unified Anatomical Atlas² (Chon et al., 2019) to map axons of spinal origin onto specific regions throughout the brain. This atlas was made in an effort to resolve discrepancies between the Allen Common Cordinate Framework (CCF) (Wang et al., 2020) and the Franklin-Paxinos mouse brain atlas (Paxinos and Franklin, 2019). Another issue is the relationship between axonal projections and cytoarchitectonically-defined nuclei: spinofugal projections are not necessarily overlaid onto the entirety of particular nuclei. The cerebellm is a case in point: subdivisions of spinofugal input are not apparent from gross granule cell cytoarchitecture, and therefore not included in standard atlases (Ozol and Hawkes, 1997; Gebre et al., 2012). Additionally, in more laminated structures such as the superior colliculus, dendrites (whiech receive the bulk of synaptic input) extend into distant layers. These findings beg the development of a spinofugal projection atlas; this work represents an important step in that direction.

The genetic labeling technique used here reveals the presence of a previously undescribed tract of axons arising from the spinal cord, decussating in the ventral hypothalamus, with the zona incerta as a likely zone of termination. Bulk lineage tracing also emphasizes less compact spinofugal tracts running through neuropil and not normally apparent using general myelin stains. This feature sheds light on a textbook “black box” that exists between axonal tracts in the spinal cord and their terminal fields in supraspinal structures. Figure 3 illustrates a bottleneck of ascending ventral and lateral tracts, anterior to which multiple ascending tracts diverge from the middle cerebellar peduncle (Figure 7). This point of defasciculation and/or branching occurs over just a few hundred microns in the rostro-caudal direction, and gives rise to at least five branches: a compact tract running caudally toward the cerebellar decussation (the inferior cerebellar peduncle proper), a diffuse dorsal tract of thick axons coursing rostrally toward the thalamus (the spinothalamic tract), a second diffuse spinotectal tract running dorsolaterally and dorsomedially with a prominent decussation in the superior colliculus, and two more diffuse tract projecting to the PAG (medially) and ECIC (laterally) in the inferior colliculus. The border between midbrain and hindbrain is established at this location early in development (the midbrain-hindbrain boundary, MHB, aka isthmus) and where an important fibroblast growth factor (FGF) and Wnt-secreting signaling centre emerges which patterns the surrounding tectum, pons, and cerebellum. These morphogens, in addition to dictating the fate of responsive cells, can act on growth cones to shape axonal trajectories (Charron and Tessier-Lavigne, 2005; Zou and Lyuksyutova, 2007). Hoxb8 reporter mice should serve as useful tools to track growth of spinofugal axon tracts during embryogenesis and their responses to manipulations perturbing isthmic signaling.

The presence of labeled axons in the medial lemniscus (ml), particularly in the non-targeted Hoxb8 reporters (Figures 5–7) was somewhat surprising. The ml is generally thought to consist only of post-synaptic projections from the dorsal column nuclei, which relay low threshold cutaneous and joint positional information to the thalamus. The absence of recombination above C4 in Hoxb8-Cre^tdt and Hoxb8-FlpO^tdt mice (and their degeneration following C3 hemisection in the Hoxb8-Cre^tdt mice) (Figures 3, 9) means that many of these axons arise from more caudal spinal neurons. This may represent either a population of post-synaptic dorsal column axons traversing the dorsal column nuclei (but remaining ipsilateral given their degeneration following hemisection, Figure 7) or, more likely, a divergence of ascending axons rostral to the anterolateral bottleneck in the medulla – our observations indicate that this occurs somewhere between the brainstem-spinal cord border and the rostroventral pons (possibly also at the MHB). The larger number of reporter positive axons in the ml of knock-in strains may additionally (or alternatively) reflect the presence of Hoxb8-lineage neurons in the gracile nucleus (Figure 3).

Many of the differences between targeted and non-targeted reporter lines were expected based on the differing rostral Hoxb8 expression boundaries (C4 in both non-targeted strains). Obvious examples include the relatively lower density of labeled axons projecting to the thalamus via the medial lemniscus in Hoxb8-Cre^tdt and Hoxb8-FlpO^tdt mice, and the absence of a projection from the nucleus of the solitary tract to the bed nucleus of the stria terminalis. These, and more striking occurrences of ectopic recombination in Hoxb8-Cre^tdt mice (anterior thalamic nuclei, astrocytes beneath the cephalic flexure, retinal ganglion and vomeronasal axons) and Hoxb8-FlpO^tdt mice (Cajal–Retzius cells in the dentate gyrus, choroid plexus fibroblasts) can most easily be explained by the absence of important negative regulatory elements in the transgene constructs and resulting sensitivity to nearby sources of diffusible signaling molecules during development (more on this below). Other differences, particularly between mice in which recombinase was targeted to the native Hoxb8 locus, are more difficult to reconcile. For example, there was an obvious difference in reporter-positive microglial density between Hoxb8-IRES-Cre^tdt and Hoxb8-T2A-FlpO^tdt mice. The presence of tdtomato-positive axons in the core of the facial motor nucleus in Hoxb8-IRES-Cre^tdt mice only, may also speak to artefactually elevated regulation of Hoxb8 protein expression. As such, some discussion of Hox gene regulation under normal circumstances is warranted.

Let us first consider the circumstances in non-targeted strains, in which the extent of reporter expression is simultaneously absent from tissues normally expressing Hoxb8 (e.g., the upper cervical cord), and expressed ectopically in more rostral structures. The utility of random transgene insertions to report transcriptional activity of a given gene is affected (either positively or negatively, depending on the question being asked) by the absence of regulatory elements controlling endogenous expression. In the case of the construct used to generate Hoxb8-Cre and Hoxb8-FlpO mice, the transgenes are no longer in the vicinity of an important retinoic acid response element (RARE) called the distal element (DE), positioned nearby between the Hoxb4 and Hoxb5 genes (Oosterveen et al., 2003; Valarché et al., 2003; Afzal et al., 2023), which is responsible for extension Hoxb8 expression from C4 to the obex.

Ectopic recombination, on the other hand, could be the result of the absence of suppressive regulatory sequences, insertion of the transgene in the vicinity of non-canonical promoters or enhancers, or temporally-altered chromatin accessibility. Whichever mechanism (or combination of mechanisms) is responsible, it still requires a stimulus for the activation of Hoxb8 expression. The dependence of Hoxb8 transcription on retinoic acid, elegantly dissected by the Deschamps and Krumlauf groups (Charité et al., 1995; Bel-Vialar et al., 2002; Forlani et al., 2003; Afzal et al., 2023), makes developmental exposure to this diffusible signaling molecule a likely cause. Indeed, ectopic recombination occurred close to tissues expressing retinoic synthesizing enzymes (RALDH1 and RALDH2) (olfactory epithelium, retina, craniofacial mesenchyme below the cephalic flexure), choroid plexus [Ruberte et al., 1993; Yamamoto et al., 1998; Luo et al., 2004 and atlas.brain-map.org], and/or in brain regions rendered sensitive to endogenous retinoic acid using a RARE-driven LacZ reporter (anterior thalamic nuclei, dentate gyrus) (Luo et al., 2004). If RA is the responsible driver of ectopic recombination, then timing of transgene RARE accessibility is most likely to underpin differences between Hoxb8-Cre and Hoxb8-FlpO expression.

Retinoic acid-dependent Hox gene regulation has many layers, and relief from Hoxb8 suppression must contribute to ectopic expression. Recently, Afzal et al. (2023), used CRISPR/Cas9 to generate mutations in the DE and two other RAREs (B4U and ENE) flanking the Hoxb4 coding sequence to determine their role in Hoxb regulation. With DE and ENE mutations, there was increased transcription of both Hoxb9 and Hoxb1, with a greater increase in the ENE mutant embryos (Afzal et al., 2023). The B4U mutation did not result in changes from wild-type mice with respect to Hoxb9 or Hoxb1 transcription, but Hoxb4 transcription was increased. This suggests that the DE and ENE are important in the regulation of all Hoxb genes, while B4U acts on more closely adjacent sequences in the Hoxb cluster (Afzal et al., 2023). Given that loss of control by ENE led to an increase in Hoxb9 transcription, its absence may similarly affect the Hoxb8 transgene leading to its ectopic expression.

Non-targeted reporter mice also lack a miRNA target sequence (specific for miR-196a) in the 3’ UTR of the transgene, required for restricting anterior Hoxb8 expression (Yekta et al., 2004; Hornstein et al., 2005). Without it, and under the influence of appropriate Hoxb8-inducing signals, one might expect transgene expression rostral to the brainstem.

Neither of the targeted recombinase models is expected to interfere with Hoxb8 function, and while Cre and FlpO have equivalent excision efficiencies (Birling et al., 2012), subtle differences in transcriptional or translational control might be anticipated: whereas the IRES system results in less efficient transcription of the second gene in the sequence [20–50% of the first, (Mizuguchi et al., 2000)], the near 100% efficiency of T2A self-cleavage should result in a 1:1 ratio of Hoxb8 and FlpO recombinase protein. The IRES has been reported to protect flanking sequences from miRNA-mediated inhibition of translation (Humphreys et al., 2005; Mathonnet et al., 2007; Thermann and Hentze, 2007). If this is the case for Hoxb8-IRES-Cre mice, then it may be considered a Hoxb8 over-expressor, potentially explaining the abundance of Hoxb8-lineage microglia and the unique presence of spinofugal projections to the facial nucleus’ core (Figure 3).

The differences between the four reporter strains described here raise questions about the fidelity of permanent reporter expression to endogenous Hoxb8 expression. The two targeted transgenic (knock-in) models are a case in point: there was an obvious difference in the number of reporter-positive microglia (Hoxb8-IRES-Cre^tdt > > Hoxb8-T2A-FlpO^tdt), and lateral septal neurons were only apparent in Hoxb8T2AFlpO mice (Bohic et al., 2023). One hypothesis is that this is a result of the relationship between differing lengths of the knocked-in constructs and the timing of Hox gene expression (i.e., temporal collinearity or the “Hox clock”). Increased genomic distance between the Hoxb8 sequence and the mir-196a target in the 3’UTR may result in a prolonged expression of Hoxb8 and evasion of microRNA-mediated mRNA degradation.

An approach to determining the true distribution of Hoxb8 expression might be to identify a sequence specifically transcribed in response to Hoxb8, and construct a reporter driven by its promoter. A first step towards this goal is identification of genes upregulated by Hoxb8 overexpression, and has recently been accomplished in the chick (Wilmerding et al., 2021).

Hoxb8-driven reporter expression is a viable approach to map axonal projections from the spinal cord to the brain. It offers superb visualization of trajectories otherwise inferred (spinothalamic and spinotectal tracts) or previously unknown (spinofugal hypothalamic decussating tract). Differences in transgenic constructs also provide some information on somatotopy (i.e., projections from the whole cord or from that up to C4 only). Models of choice for defining ascending pathways are the FlpO lines, given the relative paucity of Hoxb8-lineage microglia in both strains, and spatial separation of ascending pathways from the dentate gyrus in the Hoxb8-FlpO line.