Tools for analysis and conditional deletion of subsets of sensory neurons

Background: Somatosensation depends on primary sensory neurons of the trigeminal and dorsal root ganglia (DRG). Transcriptional profiling of mouse DRG sensory neurons has defined at least 18 distinct neuronal cell types. Using an advillin promoter, we have generated a transgenic mouse line that only expresses diphtheria toxin A (DTA) in sensory neurons in the presence of Cre recombinase. This has allowed us to ablate specific neuronal subsets within the DRG using a range of established and novel Cre lines that encompass all sets of sensory neurons. Methods: A floxed-tdTomato-stop-DTA bacterial artificial chromosome (BAC) transgenic reporter line (AdvDTA) under the control of the mouse advillin DRG promoter was generated. The line was first validated using a Na v1.8 Cre and then crossed to CGRP CreER (Calca), Th CreERT2, Tmem45b Cre, Tmem233 Cre, Ntng1 Cre and TrkB CreER (Ntrk2) lines. Pain behavioural assays included Hargreaves’, hot plate, Randall-Selitto, cold plantar, partial sciatic nerve ligation and formalin tests. Results: Motor activity, as assessed by the rotarod test, was normal for all lines tested. Noxious mechanosensation was significantly reduced when either Na v1.8 positive neurons or Tmem45b positive neurons were ablated whilst acute heat pain was unaffected. In contrast, noxious mechanosensation was normal following ablation of CGRP-positive neurons but acute heat pain thresholds were significantly elevated and a reduction in nocifensive responses was observed in the second phase of the formalin test. Ablation of TrkB-positive neurons led to significant deficits in mechanical hypersensitivity in the partial sciatic nerve ligation neuropathic pain model. Conclusions: Ablation of specific DRG neuronal subsets using the AdvDTA line will be a useful resource for further functional characterization of somatosensory processing, neuro-immune interactions and chronic pain disorders.


Introduction
Somatosensory neurons have been classified into distinct sets first on the basis of action potential velocity, later with histochemical markers, and more recently by single cell transcriptional profiling [1][2][3][4] . Gasser's original observations of three sets of sensory neurons (A fibres, A delta fibres and C fibres) all of which could contribute nociceptive input, demonstrated the complexity of damage-sensing in the periphery. It was only in the 1960s that Perl demonstrated the existence of functionally specialised nociceptors 5 . The debate about polymodality as opposed to modality-specific sensory neurons has been lively. It is clear that the in vivo properties of sensory neurons are altered when they are cultured, so that polymodality is often encountered in the dish and may be artefactual 6 . The plasticity of sensory neurons caused by inflammatory mediators is also a complication.
The existence of defined transcripts that distinguish particular sets of sensory neurons provides an opportunity to analyse their function through genetic ablation studies. Recent studies suggest that the subdivisions of sensory neuron subsets identified in mice are in the main evolutionarily conserved in primates, which is of great significance for human studies 3,4,7 . Thus analysis of rodent sensory neuron function is a useful tool for a better understanding of human somatosensation. Here we have used established and new Cre recombinase lines to ablate subsets of sensory neurons and carried out a preliminary analysis of the consequences for acute pain behaviour. It was important to rule out effects on other tissues, and so we developed a floxed stop diphtheria toxin mouse driven by a sensory neuron specific promoter, advillin. Advillin is selectively expressed in all sensory neurons and Merkel cells 8 . This allowed us to ablate only sensory neurons with no effect on other tissue that may express the same transcripts as sensory neurons but in which advillin and hence diphtheria toxin is not expressed.

Ethical considerations
All experiments were performed in accordance with the UK Animals (Scientific Procedures) Act 1986 with prior approval under a Home Office project licence (PPL 70/7382). This study is reported in line with the ARRIVE guidelines 9 .

Mouse husbandry
The study took place at University College London in 2018-2021. Mice were kept on a 12-hr light/dark cycle with cage enrichment (e.g. disposable igloos) and provided with food and water ad libitum. Surgical procedures were performed by trained researchers and under aseptic conditions. For genotyping, genomic DNA was isolated from ear tissue for polymerase chain reaction (PCR) using GoTaq DNA Polymerase (Promega) (see Extended data, Supplementary Methods for primer sequences and cycling conditions 9 ). Cre ER/ERT2 mouse lines received 200 μl doses i.p. of a 1% tamoxifen solution on five consecutive days (once per day) between 8-10 weeks of age. Tamoxifen was prepared in 15% ethanol/85% sunflower oil.

Mouse lines Advillin Flox-tdTomato-Stop-DTA BAC transgenic (AdvDTA).
We generated a floxed-tdTomato-stop-DTA bacterial artificial chromosome (BAC) transgenic reporter line under the control of the mouse Advillin (Avil) promoter. The construct was generated using a recombineering protocol described in detail by Copeland et al. 10 (Extended data, Supplementary Figure 1 9 ). BAC clone RPCI23-424F19 containing part of the Advillin gene was used as a template to prepare a targeting construct. To construct the shuttle vector for recombineering the 5'-homology arm (432 bp) and the 3'-homology arm (274 bp) flanking exon 2 of the Advillin gene were cloned into a vector containing a floxTomatoDTA cassette and a Kanamycin cassette, the latter being flanked by flippase recognition target (FRT) sites. The floxTomatoDTA cassette consists of the sequence encoding tdTomato protein followed by three SV40 polyadenylation signals and flanked by the loxP sites. This sequence was fused with the sequence encoding DTA protein followed by bGH and SV40 polyA signals. The start codon in the exon 2 of the Advillin gene corresponds to the start codon of tdTomato. The completed shuttle construct was sequenced and the targeting cassette was isolated from the plasmid by AscI/PacI digest. EL250 E. coli cells transformed with the BAC clone were co-transformed with the shuttle targeting construct and recombination was induced by heat-shock according to the recombineering protocol. The induced bacterial clones were isolated by growing on Kanamycin-containing plates. The Kanamycin resistance cassette was removed by inducing Flp-recombinase expression by arabinose. The final construct was verified by PCR and analytical restriction digests. The resulting BAC containing the targeting cassette was sent to Cyagen for pronucleus injection. The founder line was bred with wild-type C57BL6/J mice to obtain germline transmission.

Cre and floxed stop tdTomato reporter lines
The following lines were provided as generous gifts: CGRP CreER11 , TrkB CreER12 (Jackson Lab stock number 027214), Th CreERT213 (Jackson Lab stock number 025614), Ntng1 Cre14 and Rosa-CAG-flox-stop-tdTomato 15 (Jackson Lab stock number 007905). The Na v 1.8 Cre line was in house and previously described 16 .
Tmem45b Cre and Tmem233 Cre lines were generated by Cyagen on a C57BL/6 background. The Tmem45b Cre line is a BAC transgenic in which a fragment containing the Cre-SV40 pA together with Frt-Kanamycin cassette-Frt was introduced in exon 2 of the murine Tmem45b gene within BAC clone RP24-249I6. The Kanamycin cassette was removed from the recombinant BAC by Flp-mediated recombination. The recombinant clone was verified by PCR and end-sequencing and then injected into fertilized eggs for transgenic mouse production. Pups were genotyped by PCR using GoTaq DNA Polymerase (Promega) (see Extended data, Supplementary Methods for primer sequences and cycling conditions 9 ). Founder 'R' was used for further breeding and characterization.
The Tmem233 Cre line is a constitutive knockin line in which the Kozak-Cre-pA cassette replaced the ATG start codon of the murine Tmem233 gene. The Cre is under the control of Tmem233 regulatory elements with homozygotes being knockouts of the Tmem233 gene. The homology arms of the targeting vector were generated by PCR using BAC clones RP23-182O14 and RP23-375I17. In the targeting vector the Neo cassette was flanked by Rox sites and DTA was used for negative selection. The final constitutive KI allele was obtained after Dre-mediated recombination with C57BL/6 ES cells used for gene targeting. Pups were genotyped by PCR using GoTaq DNA Polymerase (Promega) (see Extended data, Supplementary Methods for primer sequences and cycling conditions 9 ).

Behavioural tests
For all behavioural assays, the animals in the test groups carried both the Cre gene and the AdvDTA BAC. Animals in the control groups were negative either for the Cre gene or the AdvDTA BAC (see each figure legend for specific details). Sample sizes (1 unit=1 animal) were calculated based on our previous experience for each assay 17 and animals were excluded from the study if they became unwell. All data points were included in the statistical analyses. Animals were acclimatized to the behavioural equipment for at least 2 days prior to testing; control and test groups were assayed on the same day. Observers who performed behavioural experiments were blind to the genotype of the animals. Animals were selected and placed into the apparatus randomly by an independent experimenter. The unblinding of each group followed input of the raw behavioural data into the analysis software. Experiments were conducted using both male and female adult transgenic mice and wild-type littermates (all >7 weeks in age).
Mechanical nociceptive thresholds were measured using a modified version of the Randall Selitto test for mice (Ugo Basile) that applies pressure to the tail with a 500g cut-off 17,18 . Punctate mechanical sensitivity was measured using the up-down method for obtaining the 50% threshold using von Frey hairs (Ugo Basile stand; Aesthesio hairs), as previously described 17,19 . Thermal nociceptive thresholds were determined by measuring paw-withdrawal latency using the Hargreaves' apparatus (IITC Life Science) 17,20 with a ramp of 1.5°C/s and a 30-s cut-off and also by use of the 50°C hot-plate test (Ugo Basile) 21 . The response to noxious cold was measured using the cold plantar assay (Ugo Basile) 22 . The rotarod test (IITC Life Science) was performed as described in 23. The formalin model was performed by intraplantar injection (subcutaneously) of 20 μl of a 5% formalin solution into the left hindpaw after a 15 min acclimatization to individual observation cages. Time spent engaged in nociceptive behaviour (biting or licking the affected region) was recorded over 1 hr immediately after injection. The video recordings were sampled for 10 sec at 1 min intervals using a Python script and scored for the presence or absence of nocifensive behaviours toward the injected hind paw (licking, biting, flinching). Data is expressed as a percentage of nocifensive responses out of all responses. The early acute phase (Phase 1, 0-10 min) and the latter inflammatory phase (Phase 2, 10-60 min) were separately analysed.
Partial sciatic nerve ligation model Surgical procedures were performed under isoflurane anaesthesia (inhalation) (2-3%). A partial nerve injury in adult mice was induced by tying a tight ligature with 6-0 silk suture around approximately 1/3 to 1/2 the diameter of the sciatic nerve, similar to the approach described in rats 24 .

In situ hybridization (ISH) sample preparation and RNAscope
Animals were deeply anesthetized with pentobarbital (i.p.) and transcardially perfused with heparinized saline (0.9% NaCl) followed by 25 ml of cold 4% paraformaldehyde in phosphatebuffered saline (pH 7.4). Dorsal root ganglia were extracted from the lumbar area and post-fixed with the same fixative solution for 2 hours at 4 °C before being embedded in cryopreservative solution (30% sucrose) overnight at 4°C. Tissue samples were then placed in optimal cutting temperature (OCT) compound blocks for posterior sectioning by cryostat. 11 μm thick sections were mounted onto Superfrost Plus (Fisher Scientific) slides, allowed to freeze-dry overnight at -80 °C, for an immediate use, or were stored at −80 °C for no longer than a month for subsequent experiments.
In situ hybridization was performed using the RNAscope system (Advanced Cell Diagnostics) following a two-day protocol for fresh-frozen samples with 1hr post-fixing with 4% paraformaldehyde (PFA) in phosphate buffered saline (PBS) at 4°C and stepwise dehydration with 50%, 70% and 100% ethanol. Tissue pre-treatment consisted of hydrogen peroxide and protease IV (10 and 20 min, respectively) at room temperature (RT). Following pre-treatment, probe hybridization and detection with the Multiplex Fluorescence Kit v2 were performed according to the manufacturer's protocol.
Fluorescence was detected using Zeiss LSM 880 Airyscan microscope. Images were taken at 10x or 20x magnification with 4x averaging, airyscan processed, stitched whenever required and exported as 16-bit uncompressed tiff files for further basic editing in Adobe Lightroom v6 (Adobe) on colour calibrated iMac retina monitor. Final images were exported as jpeg files with 7,200 pix on longest side at 300 ppi.
RNA extraction and real-time qPCR RNA was extracted from dorsal root ganglia tissue using TRIzol Reagent (Life Technologies) and Purelink RNA micro kit (Thermo Fisher) and then reverse transcribed according to standard protocols. Real-time RT-PCR using technical triplicates was performed with the BioRad CXF Connect TM . PCR was carried out using the Universal SYBR Green Supermix protocol (Bio-Rad) and primers to Scn10a and Gapdh (see Extended data, Supplementary Methods for primer sequences and cycling conditions 9 ). Scn10a expression was compared with that of Gapdh measured on the same sample in parallel on the same plate, giving a CT difference (ΔCT) for Gapdh minus Scn10a. Mean and standard error were performed on the ΔCT data and converted to relative expression levels (2ˆΔCT).

Statistics
Data was analysed using GraphPad Prism 9 (GraphPad Software, Inc), and results presented as mean ± SEM with n referring to the number of samples tested per group, as indicated in figure legends. SOFA Statistics is an alternative open access package that can be used.

Results
Single cell RNA sequencing of mouse dorsal root ganglia neurons has provided an unbiased classification of neuronal subpopulations based on their genetic profiles 3,4 . Using these data, we selected marker genes that could be used to drive expression of Cre recombinase in restricted DRG neuron populations to enable further functional analyses of neuronal subsets (Table 1). We recruited existing lines and generated new lines for further analyses (see Materials and Methods): Cgrp CreER (Calca), Th CreERT2 , Scn10a Cre (Na v 1.8), Tmem45b Cre , Tmem233 Cre , Ntng1 Cre , and TrkB CreER (Ntrk2). Lines were crossed to a Rosa-CAG-flox-stop-tdTomato line and following tamoxifen administration for the CreER/CreERT2 lines, lumbar DRGs were isolated from adult mice for RNAscope testing. Figure 1 shows representative images of tdTomato expression (i.e. following Cre activity) and expression of the 'target gene' (i.e. an in situ probe against each marker gene mRNA) 9 . For each line tested, there was a large overlap between tdTomato expression and the target, indicating that Cre activity was in general consistent with expression of the endogenous marker gene. Figure 1 also shows expression of neurofilament (Nefh) mRNA, which is expressed in NF1, NF2, NF3, NF4, PEP1.2 and PEP2 neurons, according to single cell RNAseq data 3,4,25 . Comparison of Cre activity (tdTomato expression) and Nefh expression was as predicted from single cell RNAseq data (Table 1 and Figure 1). For example, Tmem45b Cre activity was present in Nefh-negative neurons; Ntng1 Cre was active in Nefh-positive neurons; and Calca Cre activity was in both Nefh-positive and -negative neurons.
Advillin Flox-tdTomato-Stop-DTA BAC transgenic The library of validated Cre lines enables specific DRG neuronal populations to be ablated via Cre-dependent expression of Diphtheria toxin A (DTA). However, as many of the Cre lines drive expression in tissues outside of the DRG, we generated a floxed-tdTomato-stop-DTA BAC transgenic reporter line under the control of DRG-enriched mouse Advillin (Avil) promoter to enable restricted ablation 26 (Extended data, Supplementary  Figure 1 9 ). The targeting construct used to generate the line consists of a tdTomato fluorescent reporter gene and transcriptional stop sequences that are flanked by loxP sites with a DTA sequence downstream. Advillin is expressed in all sensory neurons and has minimal expression elsewhere in the nervous system 27 . In Cre negative animals, the tdTomato gene is expressed with transcription of the DTA prevented by the upstream transcriptional stop sequences. In Cre positive animals, the tdTomato and transcriptional stop sequences are excised and the DTA is transcribed leading to toxin production and cell ablation.
To validate this new reporter we crossed it with the Na v 1.8 Cre line and were able to recapitulate the behavioural phenotype observed in our earlier work in which the Na v 1.8 Cre was bred to a Rosa floxed stop DTA knockin line 28 . We used the Hargreaves' test to apply a latent heat stimulus to the hind paw and recorded the latency to a nocifensive response. We also used the Randall-Selitto test to apply steadily increasing mechanical pressure to the tail until a withdrawal response was observed. We showed that mice lacking the Na v 1.8 positive population of sensory neurons have unimpaired responses to noxious heat (Figure 2A), but responses to noxious mechanical stimuli are substantially impaired ( Figure 2B) suggesting that we successfully ablated Na v 1.8 containing neurons in the BAC transgenic line. Real-time qPCR in DRG isolated from the Cre+/AdvDTA+ mice confirmed the ablation with a 97.5% decrease in Na v 1.8 mRNA levels compared to WT DRGs ( Figure 2C).

Pain behaviour tests following restricted ablation of subpopulations of DRG neurons
Following validation of the AdvDTA reporter we crossed this BAC transgenic line to five additional Cre-expressing lines with distinct expression profiles across DRG neuronal subsets ( Table 1). As expected, motor co-ordination as assessed by the rotarod test, was normal for all lines compared to Cre-negative/ AdvDTA+ littermates (single cell RNAseq data indicates that all five Cre lines are expressed outside of the parvalbuminpositive PSNF3 proprioceptive DRG neurons) ( Figure 3A). The Randall-Selitto assay on the tail showed that noxious mechanosensation was significantly reduced in the Tmem45b Cre+/ AdvDTA mice ( Figure 3B). Tmem45b is expressed in a subset of Na v 1.8 positive neurons, specifically the six non-peptidergic neuronal subsets (PSNP1-6) ( Table 1). Punctate mechanical sensitivity, as assessed using von Frey filaments, was not different to controls for the Tmem45b, Th, Ntrk2 and Ntng1 Cre+/AdvDTA mice although surprisingly, increased sensitivity was observed in the CGRP Cre+/AdvDTA mice ( Figure 3C). Noxious thermal pain thresholds were assessed by the Hargreaves' and hot plate tests and for both assays, CGRP Cre+/AdvDTA mice showed significant hyposensitivity to noxious heat ( Figures 3D and 3E). However, CGRP Cre+/AdvDTA mice did not show significant higher cold pain thresholds in the cold plantar assay, although a trend was observed ( Figure 3F). The AdvDTA reporter line was also tested in inflammatory and neuropathic pain assays. In the formalin test, no significant differences were observed for all AdvDTA+/Cre+ lines in the first phase ( Figure 4A). However, in the second phase, the AdvDTA+/CGRP Cre+ mice displayed significantly reduced nocifensive responses ( Figure 4A).
Next, we tested the AdvDTA+/Ntrk2Cre+ mice in the partial sciatic nerve ligation model to investigate the contribution of the PSNF1 and PSNF2 neuronal populations in the development of mechanical allodynia under neuropathic pain conditions. As expected mechanical hypersensitivity developed in the control littermates at 7 days and 16 days post-surgery ( Figure 4B). However, the AdvDTA+/Ntrk2Cre+ mice did not develop     mechanical hypersensitivity, highlighting the importance of the PSNF1 and PSNF2 neuronal subsets to mechanical allodynia in the partial sciatic nerve ligation (PSL) neuropathic pain model.

Discussion
Genetic tools have made a dramatic contribution to a better understanding of the nervous system. Thus human genetic studies have identified causative mutations for a variety of disorders 29 , whilst gene and cell silencing 30 and in particular activity-based reporter systems such as CANE have proved extraordinarily informative 31 .
The complexity of sensory neurons that play a key role in pain pathways has been highlighted by single cell transcriptional analysis 3 . This suggests that there are at least 18 subsets for sensory neurons in mice. Importantly very similar subsets of sensory neurons are present in primates 7 , reinforcing the view that mouse genetic studies are likely to be relevant and useful for understanding human peripheral pain pathways 32 .
Early experiments using a Na v 1.8 Cre-recombinase to release diphtheria toxin and delete many peripherin-positive sensory neurons produced significant modality specific loss of function 28 . Recent GCaMP studies are also consistent with the presence of a high proportion of modality-specific sensory neurons although there remains considerable debate on this topic 6 .
Linking transcriptional changes in pain states to particular sets of sensory neurons has proved problematic 33 . There is clearly extensive redundancy in pain mechanisms for example in temperature sensing 34 and cold sensation 35 , whilst inflammation or damage changes the expression patterns of many transcripts including those encoding neuropeptides that are known to have a role in pain pathways. In addition, cells of the skin play a role as primary sensors for many painful stimuli indirectly signalling tissue damage to sensory neurons 36 . Neuro-immune interactions also play a major role in changing the gain of the pain system, and regulating nociceptive activity 37 .
Despite these complexities, sets of sensory neurons with particular sensory modalities clearly exist, and we have assembled tools to address their specialised functional role. A major strength of this study is that we have exploited known and novel Cre lines that encompass all sensory neuron subsets, and generated an advillin floxed stop diphtheria toxin mouse that will release the toxin only in sets of sensory neurons in the presence of Cre recombinase. A limitation of the study is that there is random integration of the BAC into the genome of the reporter mice with the potential for leaky expression. An improvement would be to generate a knockin mouse line under the advillin promoter with expression of diphtheria toxin under the control of a Cre-dependent FLEx switch.
Expression patterns were analysed with tomato reporter mice and preliminary phenotyping has been carried out analysing acute pain in mice in which sets of sensory neurons have been deleted. This background information should prove useful as a platform for more complete analyses of pathways involved in pain states that are clearly mechanistically different, for example cold allodynia and chronic inflammatory pain 38 . The role of neurons normally associated with proprioception and innocuous sensation can also be addressed. Recent studies have demonstrated a significant efferent role for sensory neurons in immune responses, and conditions such as psoriasis 39 . Dissecting out the new roles of sets of peripheral sensory neurons in regulating immune responses can thus also be addressed with the assembled mouse lines.  The authors introduce a novel transgenic mouse line that allows the ablation of subpopulations of primary sensory neurons by selective expression of diptheria toxin. The advantage, compared to previous methods, is the fact that, by using the Advillin promoter, they can restrict the ablation to primary sensory neurons, avoiding possible confounding effects of eliminating other cell types also expressing the chosen Cre-driver. For example, it was recently shown that CGRP is also expressed in a subset of dorsal horn interneurons, and their selective activation or ablation has behavioral consequences (Löken et al., DOI: 10.7554/eLife.59751). 1 The authors could cite these and other relevant literature to strengthen the point for developing this new tool. With the tool in hand, the authors explored the behavioral effects of ablating different subsets of primary sensory neurons and noted some differences. I am sure these approaches will be extended in future studies.

Data availability
As a way of background, the authors briefly rant against polymodality, in the introduction and the discussion, but, I am not convinced that ablation studies can help much in establishing whether a particular neuron or a particular subpopulation is polymodal or not. The lack of a phenotype for a given submodality (e.g. mechanosensitivity) after eliminating, even a large subset of neurons, can be explained by redundancy or overlapping expression of sensors. Other approaches are needed to settle this question. A brief review on this topic, written with my colleague, Prof. Carlos Belmonte, some years ago can be found in DOI: 10.1186/1744-8069-4-14. 2 Depending on the lens you use, the view can be clear or blurry. Rather than simply showing that the expression of the reporter and the driver/target marker gene overlap ( Figure 1) reasonably well (some additional statistics may convince the more skeptic reader), I wished the authors went directly after the important goodies: how effective is the approach in ablating the different subpopulations. To be fair, they tested it with the Nav1.8 Cre line (figure 2) and the results are quite impressive. I guess that demonstrating selectivity and lack of effect on target populations would also be an important aspect to establish in the future.
In my opinion, the authors should provide further details in the description of their behavioral methods. Some of their descriptions are too sparse and would not allow the proper replication of the experimental conditions. For example, when describing their noxious cold test, they simply refer to a published article (Brenner et al., 2012, DOI: 10.1371/journal.pone.0039765). 3 However, in that study, the authors used three different glass thicknesses, which has a notable impact on the cooling rate of the surface and the withdrawal latency. Are their results comparable? Difficult to judge without knowing the thickness of the glass.
Another aspect of the behavioral analysis used I find a bit trickier. They state in the methods that "control and test groups were assayed on the same day". Presumably, this is important to minimize uncontrolled sources of variability in pain assessment in non-verbal species (DOI: 10.30802/AALAS-CM-19-000042). 4 However, in the analysis performed on figure 4 this is no longer considered and the group of control animals includes, apparently, all the controls (this group is much larger than the rest!), and the mean and SD obtained is the average for all the groups. Would the results/conclusion obtained be different if each of the groups was evaluated against the corresponding control? I don't know but it would be relevant information. An expert in statistics, which I am not, can probably provide some advice.
The labels and numbers in the figures are too small to read. It would be better if the labels are enlarged. 2.