Spontaneous activity in peripheral sensory nerves: a systematic review

Abstract In the peripheral nervous system, spontaneous activity in sensory neurons is considered to be one of the 2 main drivers of chronic pain states, alongside neuronal sensitization. Despite this, the precise nature and timing of this spontaneous activity in neuropathic pain is not well-established. Here, we have performed a systematic search and data extraction of existing electrophysiological literature to shed light on which fibre types have been shown to maintain spontaneous activity and over what time frame. We examined both in vivo recordings of preclinical models of neuropathic pain, as well as microneurography recordings in humans. Our analyses reveal that there is broad agreement on the presence of spontaneous activity in neuropathic pain conditions, even months after injury or years after onset of neuropathic symptoms in humans. However, because of the highly specialised nature of the electrophysiological methods used to measure spontaneous activity, there is also a high degree of variability and uncertainty around these results. Specifically, there are very few directly controlled experiments, with less directly comparable data between human and animals. Given that spontaneous peripheral neuron activity is considered to be a key mechanistic feature of chronic pain conditions, it may be beneficial to conduct further experiments in this space.


Introduction
Unlike cortical neurons, somatosensory afferents do not exhibit spontaneous activity under normal physiological conditions but only during pathological states.For example, one of the primary features of neuropathic pain in animal models is spontaneous activity in all major fibre types, including large myelinated Ab, thinly myelinated Ad, and unmyelinated C fibres. 14Similarly, in humans who live with neuropathy, spontaneous peripheral neuron activity can be measured using microneurography. 13,20,26ased on these observations, numerous analgesic drug development strategies are aimed at reducing this abnormal pathological activity. 31o inform these strategies, it is important to establish a consensus view of existing literature that goes beyond usual narrative summaries.We have conducted a review with systematic search and data extraction, a less biased method that can also provide insight into research quality. 24e focused on experimental designs that can directly interrogate spontaneous neuron firing in whole organisms, ie, in vivo electrophysiology and microneurography.Our first aim was to identify and collate all the relevant studies in this area into one, easily accessible spreadsheet.Our second aim was to use the data we extracted to answer a few simple but crucial questions about spontaneous activity in neuropathy: (1) Which fibre types are mainly reported to be spontaneously active?Microneurography studies frequently report spontaneous activity in C fibres in neuropathy patients, eg, Refs.13,19,27.Conversely, in vivo electrophysiology in animal models has traditionally pointed towards the importance of spontaneous activity in Ab fibres, with Marshall Devor advocating that they are the main drivers of subsequent central sensitisation and pain. 82) How long does spontaneous activity persist?The tacit assumption of many in the field is that spontaneous activity is long-lasting and can drive ongoing pain.Indeed, in chronic neuropathy patients, spontaneous C-fibre firing can be observed, which correlates with spontaneous pain. 13The picture emerging from preclinical in vivo electrophysiology is much more mixed.Many report the bulk of activity within the first 2 weeks of nerve injury, 22,28 although others have argued that long-lasting activity emerges in unmyelinated fibres 3 to 4 weeks after an insult. 163) Are muscle afferents especially prone to sensitisation as opposed to skin afferents, as has been claimed? 17If true, this could provide one potential mechanistic explanation for why deep muscle pain is a common complaint in chronic pain conditions.The data we extracted supported the presence of ongoing spontaneous activity in both A and C fibres.However, there was a Sponsorships or competing interests that may be relevant to content are disclosed at the end of this article.
high level of heterogeneity and only very few, directly controlled studies.Microneurography studies were rare and spread across a divergent range of pain conditions.We conclude that the strength of available evidence in this area does not quite match the levels of confidence and ubiquity with which we all like to assert that spontaneous activity is one of the origins and main drivers of chronic pain.

Protocol registration
We registered a systematic review protocol for our study on the Open Science Framework on June 8, 2020.However, as abstract screening and data extraction proceeded, it became clear that key articles had been missed using the search string that was originally registered, and that the literature was even more heterogeneous than initially anticipated.Significant deviations from the protocol therefore had to be undertaken, including development of a new search strategy (Box 2), to ensure inclusion of all the relevant literature.Moreover, data extraction was simplified to enable processing of the information within the timeframe of the project.Below contains a description of the methods that were eventually adopted.

Systematic search
We conducted a systematic search of the existing literature using PubMed, to identify and collate evidence of spontaneous activity in sensory nerve fibres after peripheral injury.Abstracts were imported into EndNote for subsequent screening steps.
The search terms we initially used on June 8, 2020 and logged as part of our systematic review protocol are presented as a string in Box 1. Medical Subject Headings (MeSH), ie, specialised search strings designed for indexing, were used to exclude certain fields of research and limit the number of hits obtained.Once it became apparent that key articles were missing, another round of abstract screening and data extraction was undertaken.We added to already included articles with 2 additional search strings: that provided in Box 2 and the very broad string of "microneurography AND (spontaneous OR ongoing)," which was designed to ensure that no relevant human microneurography articles would be missed.The second search was performed on May 5, 2021.
The main differences between the search strings in Box 1 and Box 2 were (1) the inclusion of "neuroma" and "microsympathectomy" before the first "AND," (2) the inclusion of additional terms after the first "AND," specifically "nerve impulse," "microneurography," "neuropathy," "sympathectomy," "nerve afferent," "sympathetic efferent," "electrophysiology AND pain"; and (3) the use of different MeSH terms for exclusion, eg, we added "brain diseases" and removed "gene expression regulation." Our search strings and eventual data extraction relied on the terms "spontaneous," "ectopic," and "ongoing."We did not distinguish between the root causes of the spontaneous, ectopic, or ongoing activity.For example, is the activity "spontaneous" in the sense of being generated endogenously by the neuron, ie, as a result of membrane oscillations or changes in sodium channel function, or is it actually triggered by external factors, like chemical mediators released in the environment of injured neurons, and therefore, arguably not truly "spontaneous." 2 We did not distinguish between these within this review.

Title and abstract screening
Title and abstract screening was performed using the Systematic Review Facility (SyRF; http://syrf.org.uk/), an open-source systematic review platform. 1Before screening, duplicate search results were removed by EndNote.If EndNote did not discard duplicates during the import stage, reviewers removed any remaining duplicate articles.The remaining unique references were then screened for eligibility against the below inclusion/ exclusion criteria using the title and abstract text.If a decision Box 1. Search string registered in the original protocol, used for the first round of data collection.Note that the PubMed database does not distinguish between American and English spelling, ie, the term "fibre" will find both "fiber" and "fibre."Box 2. Second search string used to for the second round of data collection.

Inclusion and exclusion criteria
During abstract screening of articles concerning nonhuman data, we included all primary articles that appeared to present any electrophysiology data of spontaneous peripheral neuron activity resulting from nerve injury.We excluded review articles or articles that only showed schematic drawings of electrophysiology recordings rather than actual data.References to textbooks were excluded.During subsequent data extraction, we included only full-text articles, written in English, that performed in vivo electrophysiological recordings of peripheral sensory nerves in models of neuropathic pain.We did not include articles examining spontaneous activity in inflammatory models like Complete Freund's Adjuvant because they are not generally considered to cause neuropathic pain.We also did not include diabetic models.Arguably, diabetes could cause neuropathy, but there were only 4 articles that attempted to study spontaneous activity in animal models of diabetes, 3 of which used the streptozotocin model that is not considered to have particularly good face validity. 11uring abstract screening of articles concerning human data, we included only original articles, which seemed to contain microneurography recordings.Here, data of patients with any sort of chronic pain were included.Reviews were excluded.For data extraction, the full text of the article had to be available in English.
Qualitative data collection (eg, species, sex, type of neuropathic pain model, and human disease/pain type) was performed on all the articles included for full data extraction.However, quantitative data, eg, on n-numbers and spontaneous activity, was only obtained if the articles also provided raw data in a format that permitted us to calculate the percentage of spontaneously active fibres recorded in a particular condition.

Data extraction and management
Once screened, all references and their corresponding abstracts returned from the searches were downloaded from SyRF and compiled into an Excel Database (Supplementary Table, available on OSF, available at http://links.lww.com/PAIN/B958).This database contains the list of included articles and specifically information on title, authors, journal, and year of publication.Two reviewers (D.C. and F.D.) independently went through each item, decided on full-text inclusion, and if suitable, extracted data.Disagreements between reviewers were resolved through a tiebreaker (G.G.).Given how different in vivo electrophysiology and microneurography are as techniques, different parameters were extracted for nonhuman and human data, as described in the following: For non-human data: reasons for exclusion (if applicable), study purpose, animal species, animal sex (where "unclear" was used to denote articles in which no information was provided), injury type (eg, sciatic nerve ligation), and model type.Options for the latter included "nonregenerating" (eg, nerve ligation), "regenerating" (eg, nerve crush), "nontraumatic" (eg, chemotherapyinduced neuropathy), and "other" (eg, rhizotomy).We also collected information on the type of nerve that was being injured (eg, "sciatic"), as well as whether the nerve in question contained cutaneous afferents, muscle afferents, or a mixture of both.Data were extracted on fibre type (eg, "A fibre"), how fibre type was determined (eg, using conduction velocity), and how much time after injury the recordings were conducted: ,24 hours (acute), ,8 days, weeks (ie, 8-21 days), .21days (chronic).We noted where the activity was recorded from, eg, in teased L5 dorsal root, and how many individual animals or filaments were used.The number or percentage of spontaneously active fibres recorded per group was extracted.Finally, we also scored the "level of bias" of each article, where bias refers to how likely the recorded incidence of spontaneous activity would have been impacted by how the data were acquired.Specifically, C fibres are known to sensitise and start firing spontaneously in response to sustained activation, eg, as a result of repetitive receptive field testing. 5We ranked articles based on our interpretation of how likely recording bias would have resulted, given the authors' description of their experiments.Articles were scored "low bias" when authors specifically mentioned taking care not to sensitise fibres or clearly avoided repetitive stimulation of fibres before recording of spontaneous activity; articles were scored "medium bias" when some initial receptive field testing or electrical stimulation had taken place; and finally, articles were scored "high bias," if the authors specifically focused their analyses on those fibres that were spontaneously active or conducted extensive receptive field stimulation.
For human data, we recorded the overall purpose, the sex of the participants, the injury or disease, the number of patients from whom spontaneous activity data were recorded, and whether they experienced painful or nonpainful neuropathy or whether they were healthy controls.Where available, we also recorded how many individuals were reported to experience spontaneous activity, whether this was associated with spontaneous pain, and whether any quantitative sensory testing was performed.Finally, we extracted the number of filaments that were being recorded from and how many of them were spontaneously active, again grouped into painful and nonpainful neuropathy groups vs healthy controls.Acquisition bias was not considered for microneurography recordings.
In all instances, data extraction proceeded by experiment, with some articles containing several experiments, and thus several pieces of information, each provided with their own unique experiment number.An experiment was defined based on whether it generated its own instance of spontaneous activity data, eg, spontaneous activity measured at a different time point, at a different site, or under a different condition was logged separately.
Because control data were lacking in most experiments, we chose to plot the percentage of spontaneously active fibres that were being reported.For statistics, we performed nonparametric tests, but given that this was not a hypothesis-testing endeavour and given the highly heterogeneous nature of the underlying data, we invite the reader to focus on the median and quartile ranges provided rather than on probability statistics. 29

Results
A study flow diagram starting from database search to extraction is illustrated in Figure 1.We obtained 1170, 1218, and 110 results on PubMed using the initial, second, and third search strings, respectively.After removal of 565 duplicates, primary screens based on title and abstract reduced the number of articles to 336 unique articles on nonhuman data and 57 articles on human microneurography data.
During full-text screening, we excluded 190 articles for a variety of reasons specified in Figure 1 and the "reason_meta" column in Supplementary Table 1 (available at http://links.lww.com/PAIN/B958).Of the 187 articles that were subsequently included, 74 in vivo electrophysiology and 23 microneurography articles were taken for full quantitative data extraction, whereas the remaining 90 were processed qualitatively.For the former, we extracted spontaneous activity data from individual experiments.For the latter, we extracted other information, eg, on species, sex, condition, and/or nerve injury models.
Among the nonhuman experiments, most studies were performed in male animals (62.6%, n 5 92), whereas 12.2% used either only female or both sexes.The remaining 12.9% did not provide any information on sex.In human microneurography studies, the picture was reversed, with 67.5% of the work (n 5 27) using both genders, 22.5% only females (n 5 9), 7.5% only males (n 5 3), and only 1 study not specifying (Fig. 3).

High variability in experimental design and reporting practises between studies
Among the articles we studied, there was significant heterogeneity in what was being reported.For example, nearly half of all articles we included (73 animals and 17 human studies, 48% of the total 187) could not be used for quantitative analysis.Most commonly, they failed to report the total number of fibres measured or the raw number of spontaneously active fibres, making it impossible to calculate the proportion of neurons with spontaneous firing.Also, many studies were designed to only record from spontaneously active units.
Of the 97 articles that could be used for full quantification, there was high heterogeneity among animal studies in terms of recording location and the site of injury (Table 1).Moreover, most studies in animals were on nonregenerating traumatic nerve injury models (76%), whereas human microneurography studies were relatively evenly split between recordings from healthy volunteers, those living with chronic primary pain conditions like fibromyalgia ("other") and those living with nontraumatic neuropathy, eg, as a result of diabetes or chemotherapy (see Supplementary Table, tab "human," column "disease condition" for the full list, available at http://links.lww.com/PAIN/B958).There was only one article on amputees, arguably a nonregenerating traumatic injury, and none that would mirror what passes as "regenerative" in animals, eg, nerve crush.
Another major source of heterogeneity was that many articles did not include a control group, but simply measured the percentage of spontaneously active neurons in a pain condition.This was true for both in vivo electrophysiology on animal models, and even more so for microneurography data.Thus, 79% of microneurography studies did not include concurrent recordings from more than one condition, eg, a given article might only include data from neuropathy patients or only use historically generated control data.Similarly, 42% of animal studies did not include any control condition, whether it be sham, naïve, or vehicle-treated animals.There were also many studies that used the same control group for a series of experimental comparisons or did not provide quantifiable data on the level of spontaneous activity in their controls.In total, there were therefore only 29 articles that allowed for direct comparison between injury and control groups (39%).
Finally, regarding reporting practises, in vivo electrophysiology studies, as a rule, considered individual fibres or filaments as the unit of measurement used in discussions and subsequent statistical analyses.The total animals that were used to generate the filament data would either not be reported at all (24% of cases) or reported, but on a summary level.For example, an article might cite that they used n 5 20 rats in total, but it would not be made clear which filaments were recorded in which animal.Much information on interanimal variability has therefore been lost from the literature examining spontaneous activity.

Quantitative spontaneous activity data
Within the 74 articles containing quantitative data from nonhuman models, we extracted 126 experiments conducted on A fibres and 88 experiments on nociceptors, including C and Ad fibres.For 21 experiments, the fibre type was not reported.The included studies used a large number of different neuropathic pain models, requiring 20 different terms to allow full categorisation, including spinal nerve ligation, chronic constriction injury, nerve crush, transection, nerve ligation (of the peripheral branch), neuroma, and chemotherapy-induced neuropathy (see column "injury_type" in Supplementary Table, available at http://links.lww.com/PAIN/B958).
Classifying these various neuropathic pain models into 4 broader categories showed that 76% of experiments were conducted on nonregenerative models (eg, nerve ligation), 18% on regenerative models (eg, nerve crush), 4% on nontraumatic neuropathic pain models (induced by chemotherapy, neuritis, or ischemia), and 1% on other injury models including rhizotomy, dorsal column transection, and disc rupture (Table 1 & Supplementary Table, available at http://links.lww.com/PAIN/B958 column "Model Type").All recorded model categories showed an increase in the median percentage of spontaneously active fibres compared with controls (Fig. 4 and Table 2), with the difference reaching statistical significance for nonregenerating and nontraumatic categories (Mann-Whitney tests, q 5 0.00001 and q 5 0.0002, respectively).In the regenerating group, the mean rank of the injury group was higher (18.34) compared with the control group (12.81), but this difference was not statistically significant when examined with a Mann-Whitney t test.
It is important to bear in mind that the controls in Figure 4 are not study-matched, so the results will be prone to batch effects,  and only the largest effect sizes will be detectable.If only studies with their own control condition are considered, the number of experiments eligible for inclusion drop from 130 down to 42 (Fig. 5).The overall conclusions remain unchanged although, with the difference between control and neuropathy reaching statistical significance for nonregenerating and nontraumatic categories (Table 2, Mann-Whitney tests, q 5 0.00001 and q 5 0.000044, respectively).In the regenerating group, the effect was once again less clear, although the mean rank of the injury group was still higher (8.43) compared with the control group (6.58).In humans, of those studies included for full quantitative data extraction, there were 5 experiments on A fibres, 22 on C nociceptors and 1 that was unclear.Seven articles included individuals with chronic primary pain (eg, complex regional pain syndrome or fibromyalgia), 9 included those with neuropathies of some kind (eg, as a result of diabetes), 7 included healthy volunteers, and 1 study was performed on patients with pain as a result of tooth decay.The proportion of spontaneously active C fibres was higher in those individuals living with painful neuropathy compared with those living with nonpainful neuropathies and healthy controls (Fig. 6).
A Kruskal-Wallis test between the 3 groups revealed a statistically significant difference, specifically between painful neuropathy groups and healthy controls at adj.P 5 0.0028.Only 2 of the studies plotted in Figure 6 had their own internal controls, 13,21 with one of them reporting increased firing in painful neuropathy (27% of fibres vs 13% of fibres 13 ), whereas the other one only reported increased firing in neuropathy vs control, regardless of pain status (4% of fibres in healthy controls vs 7% and 11% in painful vs nonpainful neuropathy, respectively). 21ncluding only studies with their own concurrent control groups in the same article, we can extract data from only 5 studies on painful peripheral neuropathy, fibromyalgia, and pain because of tooth decay.Spontaneous activity was significantly increased in pain conditions compared with nonpainful controls, as summarised in Figure 7 (Mann-Whitney 2-tailed t tests, P 5 0.0159).
Table 1 High heterogeneity between the studies with quantifiable data on spontaneous activity. ."Other" in animal models is a ventral rhizotomy; "other" in human conditions encompasses chronic primary pain conditions (fibromyalgia, chronic regional pain syndrome, and erythromelalgia) as well as pain because of tooth decay.DRG, dorsal root ganglia.

Impact of bias, fibre type, and time on spontaneous activity
To answer the question of how bias in the recording setup affected the percent of spontaneous activity recorded in the different fibre types, we split our nonhuman dataset according to our low-, medium-, and high-bias categories.We limited our analysis to nonregenerating traumatic neuropathic models and examined either A fibres (Fig. 8) or Ad and C fibres (Fig. 9).The latter group are predominantly nociceptors; however, not all articles confirmed responses in the noxious range with receptive field stimulation.Moreover, much of the early literature did not further classify A fibres into Ab and Ad categories.Thus, the data in Figure 8 may include activity from some nociceptive Ad fibres.
As expected, the percentage of spontaneously active fibres was higher in the medium-and high-bias groups, and this effect was particularly pronounced for Ad and C fibres-putative nociceptors and therefore prone to sensitisation (Table 3).Accordingly, a mixed effects model revealed a main effect of injury for both A fibre and nociceptor experiments, but only a main effect of bias for nociceptor experiments: fixed effect of injury (A fibres): F(1, 86) 5 9.885, P 5 0.0023; fixed effect of injury (nociceptors): F(1,66) 5 9.159, P 5 0.0035; fixed effect of bias (nociceptors): F(1, 32) 5 4.407, P 5 0.049.
To investigate whether muscle fibres indeed display more spontaneous activity, as is usually claimed in the narrative literature, we split low-bias experiments conducted on regenerative and non-regenerative models according to the type of nerve that had been injured: purely cutaneous (eg, sural), purely muscle (eg, gastrocnemius), or mixed (Fig. 10).Numerically, the median of the percent spontaneous activity recorded after injury in individual experiments was indeed marginally larger in muscle nerves (median: 17.5, n 5 12) compared with cutaneous (14.0, n 5 12) and mixed nerves (8.0, n 5 101), see also Table 3.Moreover, a Kruskal-Wallis test showed that this difference was significant comparing mixed vs muscle nerves after injury (adj.P value 5 0.029), whereas the difference between mixed vs cutaneous nerves was not (adj.P value 5 0.23).
Finally, we split our data according to the amount of time that had elapsed after injury (Fig. 11A).The vast majority of control experiments recorded ,10% spontaneous activity.After nerve damage, spontaneous activity was observed in all fibre types, although it was most prominently reported ,8 days after injury for nociceptors.This may be partly because of experiments on C and Ad nociceptors becoming less common at later time points.
Pooling across fibre types, a Kruskal-Wallis test revealed a significant increase in spontaneous activity after injury for 3 of the groups compared with the control group (Fig. 11B): ,8 days (adj.P , 0.0001), between 8 days and 3 weeks (adj.P 5 0.013), and .3weeks (adj.P 5 0.031).See Table 4 for medians and quartile ranges.
Examining the same time course but including only articles with their own internal controls revealed a similar picture, although it becomes clear that most data were generated ,8 days after nerve injury (Fig. 12 and Table 4).

Discussion
Using our systematic search strategy and screening procedure with predefined inclusion and exclusion criteria, we present quantitative data from 147 articles performing in vivo electrophysiological recordings of sensory neuron spontaneous activity in animals after peripheral nerve trauma.We also collected data from 40 human microneurography experiments.Our dataset is provided in full, in an easily accessible format, in a Supplementary Table, available at http://links.lww.com/PAIN/B958.We found that all studies are very heterogeneous, in terms of injury type, the location of the injury, recording technique, and reporting standards.Moreover, the majority of articles did not include adequate control data, with only 39% of nonhuman in vivo electrophysiology and 21% of human microneurography studies recording spontaneous activity in both nonpainful and painful conditions.Finally, many datasets are small scale in sample size, and few report spontaneous activity data on a "by animal" rather than a "by fibre" basis.This obscures the variability inherent in in vivo electrophysiological recordings, which can be illustrated by considering some articles that do report spontaneous activity data "by animal" (Fig. 13).
These studies indicate that there is very large variation between animals, even within a particular laboratory, eg, with the percentage of spontaneously active fibres in sciatic nerve ranging from 3% to 30% at around 2 weeks after neuroma induction. 28his highlights the serious limitation of so many articles lacking control data: without "internal comparators" for individual experimental runs, it remains very difficult to estimate the true effect size of neuropathy on spontaneous peripheral neuron activity.Similarly, there is very big variation between laboratories; some of this may be because of the location of the injury, with Tal and Devor reporting that they found less spontaneous activity generated by infraorbital nerve neuromas upon direct comparison (Fig. 13C).However, variation can still be observed when different laboratories report spontaneous activity after comparable injuries in comparable nerves (ie, neuromas in the sciatic nerve or its branches, Figs.13A): some studies recorded spontaneous activity in about 10% of fibres within 40 days of a neuroma, 3,23 whereas others recorded incidence of spontaneous activity far exceeding 10% in at least half of all animals. 28s basic scientists, we often choose to seek the explanation for such variability in biology.For example, we might argue that the differences seen between the 4 studies plotted in Figure 13 are because of species differences.And yet, a more parsimonious view is that we cannot yet draw firm conclusions on biology: our review indicates that studies to date may be significantly underpowered given the large range of values that can be obtained across cohorts of animals, even within a single laboratory or an experimental paradigm.Thus, we may be able to conclude that spontaneous activity emerges in painful conditions, but a large degree of uncertainty remains around the extent and timing of this activity.This is apparent when we try and consider the questions we posed in our introduction.

Do both A and C fibres show spontaneous activity in painful neuropathic conditions and how long does their spontaneous activity last?
Our analysis indicates that in both nonhuman models and in patients, painful neuropathy generally tended to induce an increase in the percentage of neurons that were being reported as spontaneously active, over and above a basal level of activity that was reported to range from 0% to 10% in animals and 0% to 4% in people.This was true across pain conditions and a range of neuropathy models, as well as A and C fibres (although in humans, only 3 studies examined A fibre spontaneous activity in a pain state, article numbers 95, 150, and 240 in Supplementary Table, available at http://links.lww.com/PAIN/B958).It also goes   PAIN ® some way to support the findings reported by Kleggetveit et al. 13 that people with painful neuropathy show increased spontaneous firing compared with those with nonpainful neuropathy (Fig. 6).
Animal data to date further indicate that there may be a peak time point, with spontaneous activity appearing at its maximum, particularly in nociceptive fibres, around 8 days after injury.Certainly, more than 3 weeks after injury, most studies reported comparatively less spontaneous activity in all types of sensory neurons.Superficially, these preclinical data appear in line with what has been the prevailing narrative in the preclinical field: Afibre spontaneous activity is more prominent and outlasts C-fibre spontaneous activity. 8Variable explanations are put forward for this phenomenon, ranging from "a few damaged hyperexcitable nociceptors are enough to drive central sensitisation" to "a few undamaged, hyperexcitable nociceptors are enough to drive central sensitisation" and finally "Ab fibre spontaneous activity is what is actually driving central sensitisation."The hypothesis implicating undamaged fibres has remained controversial because of the difficulty of conclusively demonstrating that no fibres in the supposedly intact neighbouring spinal nerves have been accidently damaged during complex spinal nerve transection and ligation surgeries. 9owever, it is important to note that our time series analyses are hampered by the lack of control data.Indeed, if we restrict our selection to those articles that studied specifically nociceptors more than 3 weeks after injury, there is only a single low-bias study (in monkey) that contains data of a noninjured control nerve. 15Three other studies that included controls and recorded from injured fibres at such late time points either recorded from nonnociceptive A fibres 30 or did not clearly define the fibres they recorded from. 7,10Hence, as it stands, we feel that our systematic review data throw up a fourth possibility regarding the timing of spontaneous activity: we simply may not yet have collected sufficient, well-controlled experimental data to determine whether chronic spontaneous firing in C fibres is present in models of neuropathy.This might also be the reason for the apparent disconnect between animal and human literature: spontaneous activity in C fibres is routinely observed in patients with painful peripheral neuropathy.

Are muscle afferents especially prone to sensitisation?
This was a view put forward as a result of direct comparison between spontaneous activity recorded from muscle vs cutaneous afferents. 17And indeed, it is one that our systematic review data further support in an unbiased fashion.We found that the median percentage of spontaneous activity recorded in  muscle nerves was significantly larger than that in mixed nerves.However, it is important to consider that studies injuring pure muscle or pure cutaneous nerves were much rarer (n 5 12 each vs n 5 101 for mixed nerves).As a result, there still is a certain degree of uncertainty attached to this finding.Nevertheless, conceptually, this is potentially very important.If deep tissue afferents are indeed more prone to spontaneous activity or show a higher incidence of spontaneous activity after damage or in disease, it could explain why certain individuals with chronic pain report pain that is deep in origin. 6,18

Limitations
Our study had several limitations, which broadly fall into 2 categories: those related to systematic review methods and those related to the kind of literature we aimed to summarise.
In systematic review methods, because of time constraints, we had to make some choices that deviate from best practice, even within the preclinical systematic review field.Specifically, we only used a single database (PubMed) for our searches, which means that we inevitably would have missed some articles; we also excluded non-English articles or those for which the full text could not be obtained easily through current subscriptions at our university; and finally, our review is not a "living review," ie, any new articles that would have appeared since May 2021 will not be included.
Some other methodological limitations arose from the difficulties inherently linked to trying to conduct a systematic review of a very heterogeneous set of preclinical literature.The considerable challenge associated with designing search strings and data extraction strategies for such a diverse set of articles meant that we had to deviate quite significantly from the protocol we originally registered.Specifically, we had to design a whole new search string when we realised that our first one was too limited, with certain MeSH terms filtering out key articles in the area we aimed to cover.Moreover, we initially were much too ambitious in our data extraction strategy and ultimately had to limit the data we derived to make the work-load manageable for 2 people within the time frame of this research project.For this last reason, we focused our analysis on in vivo electrophysiological studies and excluded studies on spontaneous activity in inflammatory states.The latter is arguably not a bad choice from a conceptual point of view because neuronal firing in these models is less likely to be "truly spontaneous." 2 Beyond methodological constraints, our dataset is limited by the nature of the articles we reviewed.For example, we were unable to summarise our data in more conventional review formats, eg, forest plots of effect sizes.This was mainly  because of the majority of studies not containing control groups, making it impossible to calculate the relative increase in spontaneous activity between case and control for many experiments.We also had to contend with the fact that most studies record their results on a "by-fibre" rather than a "byanimal" basis.As a result, animal sample sizes in this area are often not very informative.
Finally, given the historical nature of much of the literature we summarised, there was almost no standardisation of recording setups, recording protocols, or reporting standards.This was true for both animal and human studies.For example, there are many differences in how microneurography recordings are conducted across studies, ranging from differences in recording systems, recording sites (eg, peroneal nerve at ankle vs knee  4).

Table 4
Medians with lower and upper quartiles of the values plotted in Figures 11B,12  level), and analysis approaches.Particularly crucial for the latter is the quality of any spike sorting that, if undertaken, requires excellent signal-to-noise ratio, with preferably single or few fibres recorded simultaneously.

Conclusions
We aimed to use systematic search and data extraction methods to shed light on what is known about spontaneous activity in peripheral sensory nerves from in vivo electrophysiology and microneurography studies.We conclude that the literature suggests that spontaneous activity is regularly detected in both non-nociceptive A fibres and nociceptors after injury in animals and in human patients with ongoing neuropathic pain.There also appeared to be indications that spontaneous activity was more prominent in muscle afferents and in painful rather than nonpainful neuropathy.Our review also highlights surprising uncertainty in our knowledge beyond these very basic statements.Because of highly heterogeneous recording protocols, lack of controls or reports on interanimal variation, we were unable to estimate effect sizes or make strong statements regarding the emergence and duration of spontaneous activity in different fibre types.
Moreover, our data suggest that many apparent disconnects between human and rodent recordings may simply be because of comparing models and timelines that are not equivalent.For example, 76% of all rodent work is conducted on nonregenerating traumatic nerve injury, whereas there is only a single microneurography study that might arguably be equivalent (on amputees who might or might not suffer from neuromas).Similarly, only a single well-controlled animal study exists that examines specifically C-fibre spontaneous activity more than 3 weeks after nerve injury   animal-level data on spontaneous activity in sensory neurons at chronic time points after nerve injury.In all graphs, each dot is an animal.(A) Transection & ligation of superficial peroneal nerves in cats to generate neuromas. 3(B) Transection and ligation of the inferior alveolar nerve in ferrets to generate neuromas. 4(C) Transection and ligation to generate sciatic vs infraorbital neuromas in rats. 28(D) Sciatic nerve transection to cause neuroma formation in mice.Closed circles denote the total percentage of spontaneous afferent activity in individual mice. 23ith in vivo electrophysiology.Indeed, when humans and rodents were directly compared using microneurography, species differences appear to diminish. 25,26The incidence of spontaneous activity in all 1134 nociceptors recorded in various rat models of neuropathy was found to be at 19.8%, whereas in humans with small fibre neuropathy, out of 102 fibres, 20.6% were spontaneously active. 26Moreover, just like in human, microneurography could identify spontaneously active C fibres in rats many months after nerve injury, with the longest time point assessed being 287 days postsurgery. 25e therefore must end with a very clear plea for the generation of more, and more standardized, electrophysiological data in this area.Novel, higher-throughput proxy measures of sensory neuron activity, such as in vivo calcium imaging, could also be used. 12Although we recognise that these types of experiments are extremely technically demanding and time-consuming, they are also vital for us to understand perhaps one of the most immediate causes of chronic pain-that of ongoing or spontaneous activity in peripheral sensory neurons.

Figure 2 .
Figure 2. Rat and human data make up the majority of the dataset.

Figure 3 .
Figure3.The majority of data collected from nonhuman studies (n 5 147) were male-only, whereas human studies (n 5 40) included both sexes."Unclear" was used as a category in cases where no information on sex was provided.

Figure 4 .
Figure 4.The percentage of spontaneously active fibres was increased across all experimental categories.Plotted here is the percentage of spontaneously active fibres recorded at any time after injury in experiments categorised as "low bias."Experiments that injured pure muscle afferents were excluded.Each dot is an experiment, and the lines represent the medians for each group (see Table 2 for values).Nonregenerating: n 5 22 for control and n 5 88 for injury; regenerating: n 5 8 for control and n 5 25 for injury; nontraumatic (neuritis & chemotherapy-induced neuropathy because of oxaliplatin, paclitaxel, and vincristine): n 5 12 for control and n 5 14 for injury; and other (rhizotomy & dorsal column transection): n 5 1 for control and n 5 3 for injury.

Figure 5 .
Figure 5.The percentage of spontaneously active fibres considering only studies with their own internal controls.Plotted here is the percentage of spontaneously active fibres recorded at any time after injury in experiments categorised as "low bias" that included their own controls.Experiments that injured pure muscle afferents were excluded.Each dot is an experiment, with the lines connecting control and neuropathy conditions (see Table2for median values).Nonregenerating: n 5 22; regenerating: n 5 7; nontraumatic (neuritis & chemotherapy-induced neuropathy because of oxaliplatin, paclitaxel, and vincristine): n 5 12; and other (L5 ventral rhizotomy): n 5 1.

Figure 6 .
Figure 6.The percentage of spontaneously active C fibres was higher in those individuals living with painful neuropathy.Each dot is data derived from an individual experiment.Median values indicated by the line are listed in the following (confidence intervals and n numbers provided in brackets): painful neuropathy: 34 (11-50, n 5 9); nonpainful neuropathy: 12 (11-13, n 5 2); healthy controls: 2 (0-7, n 5 5).The dot in red is recorded from muscle C fibres, and the one in dark red from aging healthy volunteers.

Figure 7 .
Figure 7.The percentage of spontaneously active C fibres from human microneurography studies with concurrent control recordings.Each dot is data derived from an individual experiment.The line connects recordings made in control vs painful conditions.Data derive from 5 articles (2 studying patients with fibromyalgia, 2 studying patients with peripheral neuropathy, and 1 studying tooth decay): article numbers 32, 244, 246, 250, and 254 in Supplementary Table, available at http://links.lww.com/PAIN/B958).

Figure 8 .
Figure 8.A fibres show increased spontaneous activity after injury when compared with control recordings.The data included in the plot are derived from experiments on A fibres recorded after injury of either mixed or cutaneous nerves in nonregenerative models.The line indicates the median values (provided in Table3).The control group includes data from naïve, vehicle-treated and sham-injured animals.N numbers for control & injury groups: low bias n 5 9 and 48; medium bias n 5 2 and 21; high bias n 5 1 and 11.

Figure 9 .
Figure 9. Putative nociceptors show spontaneous activity after injury when compared with control recordings.The data included in the plot are derived from experiments on C fibres and Ad fibres recorded after injury of either mixed or cutaneous nerves in nonregenerative models.The line indicates the median values (provided in Table3).The control group includes data from naïve, vehicle-treated and sham-injured animals.N numbers for control & injury groups: low bias n 5 10 and 29; medium bias n 5 5 and 20; high bias n5 3 and 5.

Figure 10 .
Figure 10.The percentage of spontaneously active fibres was higher after injury in muscle rather than cutaneous and mixed nerves.Each dot is data derived from an individual experiment.Only low-bias experiments were included.The line indicates the median values (provided in Table 3).The control group includes data from naïve, vehicle-treated and sham-injured animals.N numbers for control & injury groups: mixed nerves n 5 27 and 101; cutaneous nerves n 5 2 and 12; muscle nerves n 5 1 and 12.

Figure 11 .
Figure 11.Postinjury, some spontaneous activity could be observed across all time points and fibre types, with a particularly pronounced cluster in (C) and Ad nociceptors ,8 days after injury.Each dot represents data from an experiment.Only low-bias experiments were included.Injury data on mixed or cutaneous nerves, as well as from either regenerative or nonregenerative models were included.The dotted box includes experiments with ,10% spontaneous activity, which encompassed most of the control data, derived from naïve, vehicle-treated or sham-injured animals.(A) data split according to fibre types; (B) fibre types pooled across time points."control" n 5 29, "acute" n 5 11, ",8 days" n 5 39, "weeks" n 5 33, "3 weeks1" n 5 30.Lines represent the median (values provided in Table4). .
percentage of spontaneous activity.

Figure 12 .
Figure 12.Most of the spontaneous activity data with direct control group comparisons are recorded within 8 days of injury.Each dot represents data from an experiment, lines represent medians and upper and lower quartiles (values provided in Table4).Only low-bias experiments were included.Injury data on mixed, cutaneous or muscle nerves, as well as from either regenerative or nonregenerative models were included."acute" n 5 3 experiments, ",8 days" n 5 18 experiments, "weeks" n 5 4 experiments, "3 wks1" n 5 5 experiments; from n 5 15 studies).

Figure 13 .
Figure 13.Figures derived from articlesreporting animal-level data on spontaneous activity in sensory neurons at chronic time points after nerve injury.In all graphs, each dot is an animal.(A) Transection & ligation of superficial peroneal nerves in cats to generate neuromas.3(B) Transection and ligation of the inferior alveolar nerve in ferrets to generate neuromas.4(C) Transection and ligation to generate sciatic vs infraorbital neuromas in rats.28 (D) Sciatic nerve transection to cause neuroma formation in mice.Closed circles denote the total percentage of spontaneous afferent activity in individual mice.23 Figure 13.Figures derived from articlesreporting animal-level data on spontaneous activity in sensory neurons at chronic time points after nerve injury.In all graphs, each dot is an animal.(A) Transection & ligation of superficial peroneal nerves in cats to generate neuromas.3(B) Transection and ligation of the inferior alveolar nerve in ferrets to generate neuromas.4(C) Transection and ligation to generate sciatic vs infraorbital neuromas in rats.28 (D) Sciatic nerve transection to cause neuroma formation in mice.Closed circles denote the total percentage of spontaneous afferent activity in individual mice.23 The 97 studies we included for full quantification varied greatly in their experimental design, with varying n numbers, injury/disease type and location.Many of them also failed to provide recordings from control groups.NB: The table includes information on 2 separate experiments from one of the n 5 23 human studies because they were conducted on 2 patients with divergent syndromes (study 250 in Supplementary Table, available at http://links.lww.com/PAIN/B958, on one individual with fibromyalgia and one with small fibre neuropathy)

Table 2
Medians with lower and upper quartiles of the values plotted inFigures 4.

Table 3
Medians with lower and upper quartiles of the values plotted inFigures 8-10.