The role of somatosensory innervation of adipose tissues

Adipose tissues communicate with the central nervous system to maintain whole-body energy homeostasis. The mainstream view is that circulating hormones secreted by the fat convey the metabolic state to the brain, which integrates peripheral information and regulates adipocyte function through noradrenergic sympathetic output1. Moreover, somatosensory neurons of the dorsal root ganglia innervate adipose tissue2. However, the lack of genetic tools to selectively target these neurons has limited understanding of their physiological importance. Here we developed viral, genetic and imaging strategies to manipulate sensory nerves in an organ-specific manner in mice. This enabled us to visualize the entire axonal projection of dorsal root ganglia from the soma to subcutaneous adipocytes, establishing the anatomical underpinnings of adipose sensory innervation. Functionally, selective sensory ablation in adipose tissue enhanced the lipogenic and thermogenetic transcriptional programs, resulting in an enlarged fat pad, enrichment of beige adipocytes and elevated body temperature under thermoneutral conditions. The sensory-ablation-induced phenotypes required intact sympathetic function. We postulate that beige-fat-innervating sensory neurons modulate adipocyte function by acting as a brake on the sympathetic system. These results reveal an important role of the innervation by dorsal root ganglia of adipose tissues, and could enable future studies to examine the role of sensory innervation of disparate interoceptive systems.

Adipose tissues communicate with the central nervous system to maintain whole-body energy homeostasis. The mainstream view is that circulating hormones secreted by the fat convey the metabolic state to the brain, which integrates peripheral information and regulates adipocyte function through noradrenergic sympathetic output 1 . Moreover, somatosensory neurons of the dorsal root ganglia innervate adipose tissue 2 . However, the lack of genetic tools to selectively target these neurons has limited understanding of their physiological importance. Here we developed viral, genetic and imaging strategies to manipulate sensory nerves in an organ-specific manner in mice. This enabled us to visualize the entire axonal projection of dorsal root ganglia from the soma to subcutaneous adipocytes, establishing the anatomical underpinnings of adipose sensory innervation. Functionally, selective sensory ablation in adipose tissue enhanced the lipogenic and thermogenetic transcriptional programs, resulting in an enlarged fat pad, enrichment of beige adipocytes and elevated body temperature under thermoneutral conditions. The sensory-ablation-induced phenotypes required intact sympathetic function. We postulate that beige-fat-innervating sensory neurons modulate adipocyte function by acting as a brake on the sympathetic system. These results reveal an important role of the innervation by dorsal root ganglia of adipose tissues, and could enable future studies to examine the role of sensory innervation of disparate interoceptive systems.
Mammalian adipose tissue is highly innervated. This innervation has been primarily studied for its efferent functions through tyrosine hydroxylase (TH)-expressing, noradrenaline-secreting sympathetic fibres 3,4 . These sympathetic fibres act on β-adrenergic receptors and have a well-recognized role in regulating thermogenesis and lipid metabolism in both brown and beige fat 5 .
The afferent function of adipose innervation is much less understood. Adipose tissue is one of the few internal organs that receive little vagal sensory innervation 2 . By contrast, somatosensory fibres originating from the dorsal root ganglia (DRGs)-best known for skin and muscle sensation-have been reported to innervate adipose tissue in rats and hamsters 2,6 , but the extent and importance of this innervation remain to be determined across species 7 . Although pioneering work using herpes viral tracing has elegantly mapped the central projection of fat afferent 8 , the functional importance of this sensory innervation remains less understood because traditional chemical or surgical denervation of sensory fibres resulted in only minor phenotypes in the hamsters 2,9 . As we begin to appreciate the cellular and molecular heterogeneity of DRG neurons 10 and adipose tissue 5 , the specificity of earlier classic denervation approaches needs to be re-evaluated. For example, whereas surgical denervation cannot discriminate between sensory and sympathetic fibres as they travel together in bundles, capsaicin denervation, which was thought to selectively ablate sensory fibres in fat, can target non-neuronal transient receptor potential vanilloid (TRPV1)-expressing cells [11][12][13] . Moreover, capsaicin-mediated denervation is clearly biased towards the heat-sensitive and nociceptive TRPV1-expressing neurons (Aδ-and C-fibres) 14,15 , potentially dampening the loss-of-function phenotypes 9 . By contrast, it is now known that a prominent non-peptidergic DRG population also expresses TH 16 , calling into question the use of TH as a selective marker for sympathetic fibres in fat. These caveats motivated us to develop new imaging, molecular and circuit-specific tools to examine sensory innervation in adipose tissues.

Direct visualization and characterization of the DRG projections to fat
In conventional tracing studies, viruses or dyes are injected into the fat to be retrogradely transported back to the DRG somas and assessed by histology. This indirect measurement is susceptible to varying tracer efficiency across hosts, potentially contributing to the discrepancy observed across animal species [6][7][8] . Ideally, direct visualization of the entire projection from DRG soma to the target organs, for example, through an axon-filling fluorophore, would provide the most reliable anatomical proof. However, the peripheral branch of the mouse DRG travels several centimetres before reaching its targets, making it impossible to visualize using conventional histology. We recently developed Article HYBRiD, specifically designed for en bloc fluorescence visualization of large tissues 17 , paving the way to directly characterize the intact sensory innervation from DRGs to fat.
Another barrier to selectively labelling sensory innervation in the fat is that common pan-DRG Cre transgenic mice (such as Pirt-Cre, Scn10a-Cre, Advillin-CreERT2 18 ) also target sympathetic neurons (Extended Data Fig. 1a). We therefore adopted an intraganglionic DRG surgery in mice to directly inject a recombinant adeno-associated virus (AAV) expressing fluorescent protein into individual DRGs without transducing sympathetic ganglia ( Fig. 1a and Extended Data Fig. 1b). For the rest of the study, we focused on the inguinal white adipose tissue (iWAT), a beige fat pad with well-established roles in mouse physiology 19 . We injected a fluorescent-protein-expressing AAV into the thoracolumbar DRGs (vertebral level T13 and L1 (T13/L1) 6 ), to target all projecting fibres from these two ganglia. After en bloc HYBRiD clearing and light-sheet imaging of the whole torso (Fig. 1a), the entire projection from DRG soma to iWAT travelling across 1.2 cm can be resolved (Fig. 1b-d and Supplementary Video 1), unequivocally demonstrating that thoracolumbar DRGs directly innervate iWAT (Extended Data Fig. 1c).
Furthermore, retrograde cholera toxin subunit B (CTB) labelling was also used to confirm the imaging findings. As expected 6,8,20,21 , iWAT receives sensory innervation and sympathetic innervation mainly from T11-L3 DRGs and paravertebral sympathetic chain ganglia (SChGs), respectively; but not from nodose ganglion (vagal nerve) or collateral sympathetic ganglia (Extended Data Fig. 2a,b). Fat-projecting DRGs consist of multiple neuron types with an enrichment of peptidergic and myelinated fibres (Extended Data Fig. 2h,i). Given the proximity between iWAT and skin (the primary DRG target), we examined whether fat innervation is part of the cutaneous sensory circuit. Dual-colour CTB labelling from iWAT and neighbouring flank skin showed that these two organs are innervated by two non-overlapping DRG populations (Fig. 1e,f). Visceral fat (epidydimal white adipose tissue (eWAT)) also receives sensory innervation at comparable vertebral levels but from a separate population (Extended Data Fig. 2d,e). Together, these 3D imaging and retrograde labelling data demonstrate that distinct neurons in the thoracolumbar DRGs robustly project to adipose tissues.
The volumetric images enabled us to examine the morphology and topology of somatosensory terminals in adipose tissue in a high level of detail (Fig. 1g-j). The sensory fibres in iWAT can be divided into at  least two main types: (1) larger bundles travelling along the vasculature ( Fig. 1j and Extended Data Fig. 3e) and (2) parenchymal innervation, in which the sensory nerve terminates in close apposition to adipocytes (Fig. 1g,h and Extended Data Fig. 3a,d). TH has long been regarded as a sympathetic marker and a surrogate for adipose sympathetic innervation 4,7 . For the larger bundles, sensory fibres travel together with TH + sympathetic fibres along the vasculature but rarely wrap the vessel as the latter typically do 4 ( Fig. 1j and Extended Data Fig. 3e). Notably, in the parenchymal portion, nearly 40% of thin sensory terminals close to adipocytes are immunopositive for TH (Fig. 1h,i and Extended Data Fig. 3d), challenging the traditional view that TH is an exclusive sympathetic marker in fat and, importantly, suggesting that earlier studies based on TH might be confounded, at least partially, by the sensory innervation. Thus, with these findings in mind, it is imperative to establish the specific functions of sensory innervation in adipose tissue.

Selectively targeting adipose sensory innervation
Sympathetic output in adipose tissues works through β-adrenergic receptors, enabling the use of catecholaminergic neurotoxins (such as 6-hydroxydopamine (6-OHDA)) and adrenergic ligands to specifically manipulate their activities. By contrast, sensory fibres are more diverse and respond to different sensory modalities and therefore cannot be manipulated based on a single signalling pathway. Thus, a neuron-specific, projection-defined genetic approach is necessary to study sensory innervation in adipose tissue. A combination of axonal target injection of retrograde Cre and somatic expression of a Cre-dependent payload has been widely used to manipulate projection-specific circuits in the brain. However, legacy peripheral viral tracers such as pseudorabies virus and herpes simplex virus are highly toxic, restricting their use beyond acute anatomical mapping. In search for newer and safer viral vectors suitable for long-term functional manipulations, we found that AAV9 exhibited a high retrograde potential from iWAT to DRGs (Extended Data Fig. 4a). We adopted a published viral engineering pipeline 22 to generate randomized mutants of AAV9 (Extended Data Fig. 4b,c). Although our initial intent was to improve retrograde efficiency from fat to DRGs, the evolved new retrograde vector optimized for organ tracing (or ROOT) is more desirable mainly for its significantly reduced off-target expression such as in SChGs, contralateral DRGs and the liver (Fig. 2a,b and Extended Data Fig. 4d-g).
ROOT provides an opportunity to specifically ablate the sensory innervation in fat-we injected Cre-dependent diphtheria toxin subunit A (DTA) construct (mCherry-flex-DTA) into the T13/L1 DRGs bilaterally while injecting Cre-or YFP-expressing ROOT unilaterally in the iWATs (Fig. 2c). On the basis of CTB quantification, we noticed a decrease of approximately 40% in fat-projecting neurons in Cre + ipsilateral DRGs compared with the control side ( Fig. 2d-f), comparable to the efficiency achieved by previous viral DTA ablations 23 . No difference was observed in the SChGs ( Fig. 2d and Extended Data Fig. 5a). Importantly, the structure and function of sensory terminals in the flank skin remained intact (Extended Data Fig. 5b-e). Together, our projection-defined, partial sensory ablation in the iWAT provides a specific loss-of-function model to study the adipose-innervating DRGs.

Fat gene program changes after sensory ablation
We next investigated how the loss of sensory innervation affects the molecular programs of the iWAT. We compared the sensory-ablated iWAT (Cre + ) with the non-ablated contralateral fat pad (YFP + ) within the same animal. RNA-sequencing (RNA-seq) analysis was performed in the fat pads 3-4 weeks after viral injection ( Fig. 3a-d). Unbiased Gene Ontology analysis showed that both fatty acid and lipid metabolism and cold-induced thermogenesis pathways (Fig. 3d) were enriched by sensory ablation. At the individual-gene level, well-established thermogenic brown/beige cell markers were significantly upregulated and Acacb (2.6-fold, P = 1.1 × 10 −4 ), also showed elevated expression (Fig. 3b,c). The concurrent activation of opposing lipid oxidation and lipogenesis pathways is a unique but well-documented mechanism in the adipose tissue to ensure fuel availability for heat production under cold or β-adrenergic stimulation 25,26 . The mRNA expression of these genes, together with that of the gene encoding ChREBPβ (Mlxipl; called 'ChREBPβ' here), a transcriptional master regulator of DNL 27,28 , was confirmed by quantitative PCR (qPCR), showing increases in the ablated iWAT ( Fig. 3e) but not in non-targeted eWAT or interscapular brown adipose tissues (iBAT) (Extended Data Fig. 6a,b).

Sensory regulation of adipose physiology
We next examined how the sensory-ablation-induced gene changes affect adipose functions (Fig. 4a-c). We observed higher phosphorylation of hormone-sensitive lipase (HSL) and enrichment of multilocular beige adipocytes in the unilateral sensory-ablated iWAT ( Fig. 4c and Extended Data Fig. 7d-f), consistent with upregulation of the thermogenic program (Fig. 3). This resembles the beiging of iWAT after cold exposure or β-adrenergic agonism. However, under these conditions, the wild-type fat pads normally shrink in size, presumably due to stronger lipid utilization than DNL. By contrast, we observed an increase in fat mass of the sensory-ablated iWAT compared with the contralateral controls (Fig. 4a,b and Extended Data Fig. 7a-c), suggesting that the sensory innervation not only counteracts sympathetic  Statistical analysis was performed using two-tailed paired t-tests. f,g, Transcriptional analysis of iWAT with sensory ablation and sympathetic ablation. f, Schematic of sensory and sympathetic double ablation. Mice with Cre-dependent unilateral sensory ablation were subjected to bilateral sympathetic 6-OHDA denervation. g, RT-qPCR analysis of thermogenic and lipogenic genes in the iWAT with or without sympathetic denervation. n = 6 mice per group. Statistical analysis was performed using two-way analysis of variance with Sidak's multiple-comparisons test.
activity, but may also have a role in coordinating these two opposing downstream pathways. Unilateral ablation enabled accurate assessment of adipose phenotypes within the same animal. To determine how adipose sensory innervation affects whole-body physiology, we further ablated sensory innervation in the iWAT bilaterally. Consistent with the unilateral ablation, we observed an enrichment of beige adipocytes (Fig. 4d) and upregulated lipogenic and thermogenic genes in iWAT in the ablated animals ( Fig. 4e and Extended Data Fig. 8a,b). These animals were housed at murine thermoneutrality (30 °C) 30 (Fig. 4f) to eliminate confounding effects from other thermoregulatory mechanisms (such as background brown fat activity). Fat sensory ablation did not lead to significant changes in body weight, food intake, temperature sensitivity or systemic sympathetic tone (Fig. 4g,h and Extended Data Fig. 8c-i), but exhibited elevated core body temperature compared with the controls (Fig. 4i), consistent with an enhanced thermogenesis program in the iWAT. Interestingly, the difference in body temperature was normalized at 22 °C (Extended Data Fig. 8c), suggesting that the elevated body temperature at thermoneutrality was not due to a deficit in central thermoregulatory mechanisms (pyrexia) 31,32 . After challenge with a high-fat diet, despite showing a small difference in body weight, the mice with adipose sensory ablation exhibited markedly improved glucose tolerance compared with the control mice (Extended Data Fig. 8j-n). Such disproportional changes in glucose tolerance and body weight resemble the phenotype reported in PRDM16-induced beiging in transgenic models 33,34 , suggesting that adipose sensory ablation could protect mice from diet-induced glucose intolerance through beige-fat-related mechanisms.

Discussion
The conventional view posits that adipose signals transmit to the brain through slow, diffusive circulating hormones. Here the unequivocal anatomical proof of sensory innervation in fat provides a circuitry basis for a potential fast, spatially encoded neural transmission from the peripheral organs to the brain. Indeed, we go on to show evidence that DRG sensory neurons act as an inhibitory break on the local sympathetic function, reminiscent of the role of vagal baroreceptors in modulating blood pressure 35,36 . These findings fill an important gap of how the central nervous system monitors and orchestrates adipose functions, highlighting the importance of an underappreciated branch of brain-body communication.
It remains to be determined how sensory pathways mechanistically interact with sympathetic signalling. It has been suggested that capsaicin denervation of iWAT altered sympathetic output in the distal iBAT through central circuits 8,37 ; however, this observation might be confounded by the likely increase in sympathetic tone 38 due to peripheral capsaicin administration. By contrast, with projection-specific ablation, our adipose phenotypes were strictly limited in the ipsilateral iWAT but not in the distal iBAT or eWAT (Figs. 3 and 4 and Extended Data Figs. 5 and 6), suggesting a local specificity of the sensory-sympathetic interaction (Fig. 3f,g). Sensory modulation of sympathetic activity has been documented in other organs 39,40 at the spinal or supraspinal levels [41][42][43] . We did not observe a significant change in total noradrenaline content in sensory-ablated fat (Extended Data Fig. 8i), suggesting that sensory activity may act on sympathetic signalling downstream of the adrenergic receptors, although we cannot rule out potential temporospatial noradrenaline changes, which cannot be resolved by a single, bulk noradrenaline measurement. Moreover, remodelling of sympathetic fibres is a hallmark of adipose adaptation to chronic metabolic challenges, and it remains to be tested whether sensory innervation undergoes a similar process, especially in different strains of mice as well as across species.
Our study raises many questions. The somatosensory nervous system, which is characterized by clusters of first-order neurons that mainly reside in the DRG, has great molecular heterogeneity as profiled by recent single-cell transcriptomes 10,44 . Which DRG subtypes innervate fat and whether distinct subtypes have different functions (that is, regulating thermogenesis versus lipogenesis) is currently unclear. This could potentially be determined in the future by coupling ROOT-based retrograde targeting with single-cell RNA-seq to establish the identities the fat-innervating neurons. This information could also help to identify the interoceptive signal that fat-innervating neurons sense. Previous research showed that infusion of exogeneous leptin and free fatty acids could activate DRGs by FOS staining or ex vivo recording 45,46 ; however, the identification of endogenous signals (chemical or physical) will probably require a full understanding of the putative receptor expression using the organ-targeted single-cell RNA-seq approach suggested above. Moreover, the nature of neural transmission also suggests that these endogenous activities probably occur at second or millisecond scales. Thus, a matching ability to readout projection-specific DRG activities in vivo would be required to identify the triggering signals, for example, using emerging long-term DRG calcium imaging techniques 47

Fig. 4 | Ablation of iWAT DRGs changes the morphological and physiological properties of the iWAT. a,b, Representative images (a) and quantification of fat mass (b) of iWAT with
Cre-dependent unilateral sensory ablation. n = 11 mice. Statistical analysis was performed using two-tailed paired t-tests. c, Histology of iWAT with Cre-dependent unilateral sensory ablation. d, Histology of iWAT with Cre-dependent bilateral sensory ablation. Each panel is from a different mouse. e, RT-qPCR analysis of iWAT with Cre-dependent bilateral sensory ablation. n = 4 mice per group. Statistical analysis was performed using two-tailed unpaired t-tests. f-i, Physiological measurements after Cre-dependent bilateral sensory ablation. f, Schematic of bilateral sensory ablation, and the timeline of physiological measurement. g, Two-temperature choice assay (30 °C versus 18 °C) of Cre − (n = 6) and Cre + (n = 6) mice. h, Heart rate of Cre − (n = 6) and Cre + (n = 6) mice. i, Rectal temperature at thermoneutrality of Cre − (n = 9) and Cre + (n = 10) mice. For g-i, data are mean ± s.e.m. Statistical analysis was performed using two-tailed unpaired t-tests with Welch's correction (g-i). For c and d, scale bars, 100 μm.

Article
Our research also has implications beyond the sensory innervation of fat. Interoception has been mostly studied from the perspective of cranial ganglia of vagal origin 49 . Increasing evidence indicates the DRG innervation of internal organs also has a crucial role in interoception 50 . The discoveries made here could serve as a proof-of-principle for investigating the role of DRG neurons in a variety of other internal organs.

Online content
Any methods, additional references, Nature Research reporting summaries, source data, extended data, supplementary information, acknowledgements, peer review information; details of author contributions and competing interests; and statements of data and code availability are available at https://doi.org/10.1038/s41586-022-05137-7.

In vivo selection of ROOT
Plasmids. The plasmids used for in vivo selection were adapted from the previous publication 54 . rAAV-pUBC-sfGFP-Cap and AAV2/9-REP-AAP was generated from pUBC-mCherry-rAB (Addgene, 115239), pUCmini-iCAP-PHP.S (Addgene, 103006), pAAV2/8 (Addgene, 112864). No in-cis-Lox module or transgenic Cre lines were used given the anatomical separation of DRGs and iWAT. The ROOT capsid library was generated by inserting random heptamers using NNK degenerate primers (Integrated DNA technologies) between the 588 and 589 sites of AAV9 by Gibson assembly as previously described 54 .
AAV capsid library production. The viral libraries were produced as previously described 53,54 . In brief, only 10 ng of rAAV-pUBC-sfGFP-Cap library DNA was transfected in HEK293FT (Invitrogen R70007) cells per 150 mm plate, and the virus was collected after 60 h for purification.
DNA recovery and sequencing. The resulting AAV capsid library was injected bilaterally into the iWAT of C57Bl/6J male mice at 10 9 viral genomes (VGs) per fat pad. rAAV genomes were recovered from T12-L3 DRGs from injected mice two weeks after injection using the DNeasy Blood and Tissue kit (Qiagen). The recovered viral DNA was amplified for two rounds against a fragment containing the heptamer insertion with Q5 high-fidelity polymerase (New England Biolabs). The amplified products were cleaned up and processed for complete amplicon sequencing at Massachusetts General Hospital.
NGS data alignment and processing. Raw FASTQ files from NGS runs were aligned to an AAV9-template DNA fragment containing the 21 bp diversified region between amino acids 588 and 589 using SAMtools (v.1.10). The abundance of each 21 bp sequence in all of the recovered sequences with the heptamer insertion was quantified.

Surgery
Mice were anaesthetized with isoflurane (4% for induction and 1.5-2% for maintenance), with their skin at the surgical area shaved and hair removed and sterilized using ethanol and iodine. After surgery, the mice were given a subcutaneous injection of flunixin and topical antibiotic ointment for post-operative care.

Retrograde tracer labelling of sensory neurons. For injection into the iWAT, a lateral incision was made on the flank skin of each side.
For injection into the eWAT, a lateral incision was made on the lower abdominal wall. For injection into the iBAT, a midline incision was made in the interscapular region. For injection into the skin, an intradermal injection was performed. A Hamilton syringe with a 31G (point type 2) needle was used for all retrograde tracing. A total of 4-5 μl 0.1% CTB-488 or CTB-647 (Invitrogen) in PBS was injected per fat pad (2 μl for iBAT) or the flank skin with 8-15 injections to spread out the tracer. The abdominal wall (for eWAT injection) and skin (for iWAT, eWAT or iBAT injection) were sutured separately. Tissues were taken 3-5 days after injection to allow the dye to reach the DRG soma.
Characterization of ROOT. Injections were performed as described above. For ROOT characterization, AAV9 or ROOT-mScarlet (4 × 10 13 VGs per ml, 2 μl) in PBS with 0.001% F-68 and 0.01% FastGreen was administered unilaterally into iWAT of WT mice. Two weeks after the first surgery, 4 μl of 0.1% CTB-647 was injected into the same iWAT fat pad, and tissues were taken 3-5 days after the second injection for quantification.
6-OHDA treatment. 6-OHDA has been used previously for selective sympathetic denervation 55,56 . 6-OHDA (Tocris) (12 mg ml −1 in saline with 0.02% ascorbic acid (Sigma-Aldrich)) was prepared fresh before use and kept on ice and in the absence of light. 6-OHDA (8 μl) was injected into each iWAT fat pad. Saline with 0.02% ascorbic acid was used as a control. Tissues were taken 7-9 days after injection.
Intraganglionic DRG injection. Intraganglionic DRG injection was performed according to a previous report 57 . A midline incision was made on the dorsal skin to expose the dorsal muscles. The muscles along the vertebra were carefully separated to expose DRGs. DRGs at the T13 and L1 vertebral level were exposed, and AAV (~1 × 10 13 VGs per ml in PBS with 0.001% F-68 and 0.01% FastGreen) was injected in the ganglion (~200 nl per ganglion) with a pulled glass pipette using the Nanoliter 2020 Injector (World Precision Instrument). Care was taken to avoid damaging the surrounding vasculatures. Dorsal muscle and skin were sutured separately. PHP.S-TdTomato or PHP.S-mScarlet was used for anterograde labelling.
6-OHDA treatment and Cre-dependent sensory ablation. AAVs were injected into iWAT and DRGs to achieve Cre-dependent sensory ablation as described above. Two to three weeks after the first injection, 6-OHDA or saline was injected into the iWAT again. Tissues were extracted 7-9 days after the second injection.

En bloc HYBRiD tissue clearing and immunolabelling of iWAT
To visualize sensory nerves in mouse torso and iWAT samples, tissue samples were cleared using the HYBRiD method as described previously 17 .
Sample collection and pretreatment. Mice were terminally anaesthetized with isoflurane and intracardially perfused with ice-cold PBS and ice-cold 4% PFA in PBS with 4% sucrose (Electron Microscopy Perfusion Fixative, 1224SK). For torso samples, the skin was carefully removed leaving the iWAT attached to the muscle, the spinal cord was cut from the midline to facilitate clearing and imaging. All of the collected samples were post-fixed in 4% PFA at 4 °C for 1-2 days before being washed in PBS. The torso samples were decalcified in 10% EDTA/15% imidazole (at 4 °C for 7 days), then decolourized in 25% N,N,N′,N′-tetraki s(2-hydroxypropyl)ethylenediamine (Quadrol) (in 1× PBS) (at 37 °C for 4 days).
Refractive-index matching and mounting. Cleared or immunolabelled samples were refractive-index matched in EasyIndex (RI = 1.52, Life Canvas) and mounted in spacers (Sunjin Lab) for confocal microscopy imaging or mounted in agarose for light-sheet microscopy imaging.

Histological analysis of whole-mount samples and cryosections
Mice were terminally anaesthetized with isoflurane and intracardially perfused with PBS and 4% PFA. For flank skin samples, skin was shaved, and hair was removed using Nair.
Whole-mount imaging of ganglia and flank skin. Ganglia of interest (DRGs, SChGs and celiac/mesenteric complex) were dissected and mounted in RapiClear (Sunjin Lab) with silicone spacer (Electron Microscopy) for confocal imaging. Flank skin was dissected, post-fixed in PFA and washed three times in PBS before mounting in Fluoromount-G (Invitrogen 00-4958-02) using 0.25 mm iSpacers (Sunjin Lab) for confocal imaging.
Immunolabelling analysis of skin sections. Flank skin was dissected, post-fixed in PFA, dehydrated in 30% sucrose before being embedded in OCT, then sectioned at 25 μm and mounted on gelatin-coated slides. For immunofluorescence analysis, skin tissue sections were blocked with 5% normal donkey serum in PBS with 0.3% Triton X-100. Primary antibodies were prepared in the same blocking solution and incubated overnight (anti-βIII-tubulin (Abcam, ab18207, 1:1,000). The next day, the sections were washed in PBS and then incubated for 2 h at room temperature with secondary antibodies (anti-rabbit-647, Jackson Immuno Research, 711-606-152, 1:500), and stained with DAPI before mounted with ProLong Gold antifade mountant (Invitrogen) for confocal microscopy imaging.
Haematoxylin and eosin staining of adipose tissues. Mice were terminally anaesthetized with isoflurane. iWAT, eWAT and iBAT were extracted and post-fixed in 4% PFA overnight at 4 °C. The tissues were paraffin-embedded, sectioned at 5 μm and mounted onto glass slides. The sections were then stained with haematoxylin and eosin at Sanford Burnham Prebys histology core and imaged using a bright-field microscope.
Light-sheet microscopy. Torso or iWAT samples were mounted using 1% agarose/EasyIndex. Mounted samples were imaged inside the SmartSPIM chamber filled with EasyIndex and sealed with mineral oil on the top. Images were acquired using a ×3.6/0.2 NA objective (LifeCanvas), with a 1.79 μm, 1.79 μm, 4 μm xyz voxel size. Image acquisition was completed with bilateral illumination along the central plane of symmetry within the sample.
Bright-field microscopy. Slides stained with haematoxylin and eosin were imaged using the Keyence BZ-X710 microscope with a ×40/0.6 NA objective (CFI S Plan Fluor ELWD ADM, Nikon).

Transcriptional analysis
RNA preparation and RT-qPCR analysis. Adipose tissues were dissected between 12:00 and 14:00 and flash-frozen in liquid nitrogen. Total RNA was extracted from frozen tissue using TRIzol (Invitrogen) and RNeasy Mini kits (Qiagen). For RT-qPCR analysis, total RNA was reverse-transcribed using the Maxima H Minus First Strand cDNA Synthesis Kit (Thermo Fisher Scientific). The resultant cDNA was mixed with primers (Integrated DNA Technology) and SyGreen Blue Mix (Genesee Scientific, 17-507) for RT-qPCR using the CFX384 real-time PCR system (BioRad). Normalized mRNA expression was calculated using ΔΔC t method, using Tbp (encoding TATA-box-binding protein) mRNA as the reference gene. Statistical analysis was performed on ΔΔC t . The primer sequences (forward and reverse sequence, 5′ to 3′, respectively) were as follows: Tbp (CCTTGTACCCTTCACCAATGAC and RNA library preparation and sequencing. RNA library preparation and sequencing was performed at Scripps Genomics core. Total RNA samples were prepared into RNA-seq libraries using the NEBNext Ultra Directional RNA Library Prep Kit for Illumina according to the manufacturer's recommended protocol. In brief, 1 μg total RNA was poly(A)-selected for each sample, converted to double-stranded cDNA followed by fragmentation and ligation of sequencing adapters. The library was then PCR-amplified for 8 cycles using barcoded PCR primers, purified and size-selected using AMPure XP Beads before loading onto an Illumina NextSeq 2000 for 100 bp single-read sequencing.
RNA-seq analysis. Sequenced reads were aligned to the GRCm39 reference genome (Ensembl, v.104; http://uswest.ensembl.org/Mus_ musculus/Info/Index), and gene counts were quantified using Salmon (v.1.5.1) 59 . Differential gene expression analysis and P-value calculation were performed by DESeq2 (v.1.32.0) 58 . Gene Ontology enrichment analysis was performed using Metascape 59 with the Gene Prioritization by Evidence Counting setting. Two-temperature choice assay. The two-temperature choice test apparatus was set up as previously described 61 . Lanes were evenly divided between two different temperature plates set at 30 °C and 18 °C, and individual mice were placed in one of the lanes. The mice were given 10 min to acclimatize and were then tracked for 1 h using the EthoVision tracking system (Noldus Information Technology) in a dark room with infrared lighting. The total time spent in each temperature zone was analysed.

Behavioural and physiological assays
Core body temperature measurement. Core temperature was measured using a thermocouple rectal probe (World Precision Instruments). Room temperature core body temperature was taken when the mice in thermoneutral condition moved to room temperature for more than 24 h. Blood pressure and heart rate measurement. Blood pressure and heart rate were measured using the tail-cuff method with the CODA High Throughput Non-Invasive Blood Pressure System (Kent Scientific) as previously described 62 .
Targeted detection of norepinephrine in bulk iWAT. Frozen iWAT tissues were lysed in 4× weight of 0.1 mol l −1 perchloric acid, centrifuged and run through a 30 kDa filtration tube (Millipore). The filtrates were analysed on an Agilent 6470 Triple Quadrupole (QQQ) liquid chromatography-mass spectrometry (LC-MS) system using electrospray ionization (ESI) in positive mode. The AJS ESI source parameters were set as follows: the gas temperature was set at 250 °C with a gas flow of 12 l min −1 and the nebulizer pressure at 25 p.s.i. The sheath gas temperature was set to 300 °C with the sheath gas flow set at 12 l min −1 . The capillary voltage was set to 3,500 V. Separation of metabolites was conducted on an Agilent Eclipse Plus C18 LC column (3.5 μm, 4.6 × 100 mm, 959961-902). Mobile phases were as follows: buffer A, water with 0.1% formic acid; buffer B, acetonitrile with 0.1% formic acid. The LC gradient started at 5% B from 0 to 2 min. The gradient was then increased linearly to 5% A/95% B from 2 to 23 min. From 23 to 28 min, the gradient was maintained at 5% A/95% B; and from 28 to 29 min, the gradient went back to the starting concentration of 5% B. The flow rate was maintained at 0.7 ml min −1 throughout the run. Multiple reaction monitoring was performed for noradrenaline, looking at the transition of m/z = 170.1 as the precursor ion to the m/z = 152 fragment. The dwell time was 200, the fragmentor set at 60, the collision energy set at 4 and the cell accelerator voltage set at 4.

Metabolic studies
Mice received bilateral sensory ablation were fed a high-fat diet (HFD) (D12492, Research Diets) under thermoneutrality at 30 °C. HFD feeding was started at 1 month after surgery (aged around 11-12 weeks old). Body weight was measured every week. Fasting glucose was taken and glucose tolerance tests were performed at 9 weeks on HFD, and plasma insulin was measured at 13 weeks on HFD.
Glucose tolerance test. Mice were fasted for 4 h (08:30-12:30) before intraperitoneal injection of glucose (1 g kg −1 ). Blood glucose levels were measured at the indicated time point using the OneTouch Ultra 2 blood glucose meter.

Plasma insulin measurement.
Fasting blood samples were retro-orbitally obtained 3 h into the light cycle after fasting for 14 h. Plasma was separated in heparin-treated tubes (Microvette CB 300LH). Plasma insulin was measured with an ELISA kit (Crystal Chem, 90080) and read using the BioTek Cytation 5 Imaging Reader.

Imaging analysis and quantification
All of the images were analysed using ImageJ. 3D volume images were rendered in Imaris.
Quantification of TH + DRG nerves in the iWAT. Regions of 40 μm × 40 μm × 40 μm (x,y,z) were randomly selected and maximally projected over z using customized ImageJ scripts in the whole stacks of intraganglionically labelled iWAT with TH staining. Only regions containing positive TdTomato (DRG) signals were retained. The thickness of TdTomato-positive fibres was measured using the ImageJ straight line function across two different places in the view and averaged. Fibres with widths of less than 2.5 μm (arbitrary cut-off) were considered to be thin fibres. If the TH-647 channel showed overlap with the TdTomato signal, the view was considered to be positive. We quantified 13 images from 3 biological replicates, and a total of 77 views containing the TdTomato positive thin fibres were quantified for TH positivity.