1. Introduction

Chronic pain is a leading cause of disability and disease burden worldwide.44,57 Developing analgesics with better efficacy and safety profiles remains a high priority. Many novel analgesics with promising preclinical results failed to translate into the clinic.4,38,43,60 This raises concerns about the validity of animal pain research, particularly the clinical and ethological relevance of the models and whether outcome measures used are reflecting the clinical construct they claim to measure.

Stimulus-evoked limb withdrawal responses (eg, monofilaments test) are widely used as surrogate outcome measures to quantify nociception in rodents.13 However, these stimulus-evoked behavioural outcomes have limitations: first, they are only useful in assessing certain sensory phenotypes associated with gain of function, allodynia, and hyperalgesia.43 They cannot address spontaneous pain and pain in clinical phenotypes relating to sensory loss of function; hence, they do not fully reflect the construct (ie, pain) being measured. Second, they are prone to generating false positive or negative results. Rodents are prey species and can mask behaviours that make them appear weak or vulnerable during the stimulus-evoked paradigms. Rodents can also associate premature withdrawal with less stimulation and human interaction.13 Furthermore, these paradigms cannot distinguish analgesic effects from adverse effects such as sedation, and the subjective assessment of stimulus-evoked behaviours can potentially threaten a study's predictive validity further. Finally, they cannot provide information on how pain affects the emotional and physical function of an animal. To capture different aspects of pain and improve the validity of preclinical pain research, stimulus-evoked behavioural outcomes should be assessed in combination with other ethologically relevant outcome measures.

Using complex ethologically relevant behaviours as a form of non-evoked pain-related outcome measures has become increasingly popular in recent years.1,34,36,47,61 Ethologically relevant behaviours can provide insights into how an animal's physical wellbeing and its affective state can be affected by pain. These behaviours are not pain specific, and can be perturbed by various stress factors and disease conditions, so it is crucial to contextualise these ethologically relevant behaviours to pain. Researchers can achieve pain contextualisation by showing that changes in these behaviours are caused by disease models associated with pain and that the changes can be reversed by administering known analgesics.

Burrowing is an ethological behaviour observed in some rodent species.3,10 Rodents excavate underground holes and tunnels to construct habitation. In laboratory rodent strains, burrowing is also a highly motivated social behaviour with a self-rewarding component.11 Deficits in burrowing behaviour correlate with various perturbations, including pain, and are quantified by measuring the weight of substrate displaced from an artificial burrow. The risk of handling–induced stress-related false positive or negative results is mitigated as animals are left alone during the assessment. Reduced rodent burrowing behaviour has been observed in numerous disease models associated with persistent pain1,5,18,45,49 and has been validated in a prospective multicentre study.59 Studies have demonstrated that clinically used analgesics attenuated burrowing deficits caused by experimental persistent pain, supporting the predictive validity of the test.18,27,29,45 Furthermore, burrowing is an ethologically fundamental activity, particularly for rats, as deficits in such behaviour can negatively affect their chance of survival in the wild or “quality of life” under domestication.40,42 Studies have demonstrated that laboratory-bred rat strains also readily burrow when they are placed in a more naturalistic environment.42,52 Given laboratory-bred rodents spontaneously exhibit burrowing, this behaviour has good face validity and is considered comparable with the “activities of daily living” in humans. Therefore, measuring changes in burrowing behaviour could help to address the global impact of pain on rodents.

Finally, the wide usage of monofilaments tests in rodent pain research inspired us to assess the association between monofilament-evoked limb withdrawal and burrowing outcomes.

1.1. Aims and objectives

This systematic review aimed to (1) assess whether rodent burrowing behaviour is influenced by rodent models associated with persistent pain and analgesic drug interventions, (2) explore study design characteristics and assess their impact on burrowing outcomes, (3) perform a risk of bias assessment to evaluate studies' methodological quality, (4) identify the presence of publication bias and determine its direction and magnitude, and (5) assess the correlation between monofilament-evoked limb withdrawal and burrowing outcomes in the same cohort of animals.

2. Methods

The review protocol was registered on PROSPERO (CRD42020172320; full protocol: https://www.crd.york.ac.uk/prospero/display_record.php?RecordID=172320). The only protocol deviation is that the rationale for conducting power analysis was changed (see 2.6.5. Power analysis).

2.1. Search strategy

We systematically searched EMBASE using Ovid, PubMed, Scopus, and Web of Science on March 23, 2020, and September 29, 2020, with no restrictions on languages and date of publication. The full search strategy for each database is provided in Supplemental Digital Content 1 (available at https://links.lww.com/PAIN/B603).

Duplicates of retrieved studies were removed using EndNote. In addition, reference lists of eligible studies were manually searched to identify studies missed by the database search.

2.2. Eligibility criteria 2.2.1. Inclusion criteria (1) Population—in vivo rodent models of disease associated with persistent pain (ie, induced chemically, surgically, or genetically and developed over a period of an hour, weeks, or months). (2) Intervention—any clinically approved or novel analgesics used to interfere with nociception. (3) Comparison—a cohort of control animals. (i) For animal modelling experiments (ie, assessed effects of persistent pain-related disease models on rodent burrowing behaviour), a control population was defined as sham or naive. If sham and naive controls were not reported, the baseline measurements of the same animals before model induction were regarded as control. (ii) For studies that used transgenic rodents to study persistent pain, a wild-type control was required. (iii) For drug intervention experiments (ie, assessed effects of pharmacological interventions on rodent burrowing behaviour), a vehicle control was required. (4) Outcome—burrowing outcomes.

For the meta-analysis, we required a study to report the following data: (1) the mean burrowing outcome, (2) its variance (ie, SD or SEM), and (3) the number of animals per group.

2.2.2. Exclusion criteria

Non-rodent in vivo studies and studies that investigated acute nociception (ie, measure reflex withdrawal responses to mechanical and thermal stimulus in non–disease-induced animals) were excluded. Studies were also excluded if the burrowing outcome was not reported or there was not an appropriate control group. Studies that were not primary research articles were excluded.

2.3. Study selection

Screening of retrieved studies were completed on the Systematic Review Facility (SyRF)2 platform. Studies were screened against the inclusion criteria twice based on (1) titles and abstracts and (2) full texts by 2 independent reviewers (X.Y.Z. and A.B.). Discrepancies were resolved by a third independent reviewer (N.S.).

2.4. Data extraction

Data extraction was conducted concurrently to the full-text screening stage on SyRF by 2 independent reviewers (X.Y.Z. and A.B.).

2.4.1. Data collection

Study-level data were extracted (Table 1), and studies that were eligible for meta-analysis had experimental data extracted (Table 2). The primary outcome of interest was any outcome metric that denoted burrowing behaviour. The secondary outcome of interest was monofilament-evoked limb withdrawal assessed in the same cohort of animals. Continuous data were extracted independent of the unit of measurement. Digital ruler software (WebPlotDigitizer) was used to manually extract graphically presented data. When multiple time points were reported, the time point of the maximum effect was extracted. If the type of variance (ie, SEM or SD) was not reported, it was characterised as SD (ie, to give the most conservative estimate). The most conservative estimate was extracted when sample size data were given as a range. When key information was unclear or not reported (ie, sample size and variance), the corresponding author was contacted. If the author did not respond or was unable to provide the information, the study was recorded as having missing data and was excluded from the meta-analysis.

Table 1 - Study-level data extracted from each included study. Study level Bibliographic detail
  First author
  Year of publication
  Title Reporting quality
  Reporting guidelines, such as the ARRIVE, were developed for the purpose to improve the reporting of animal research. The following items were extracted:
  Reference following a reporting guideline for in vivo experimentation
  Provide evidence of reporting in accordance with the chosen guideline Abstract spin
  Spin is defined as intentional and unintentional reporting practices that mislead the readers by misinterpreting the true effects so that conclusions are perceived in a favourable
  light. We used the following criteria from the “Protocol of intervention to reduce spin in the abstract conclusion” (registered on Open Science Framework: https://osf.io/49r5c/)
  to assess abstract spin:
  Report information that is not supported by evidence or in accordance with the study results
  Report interpretation that is not consistent with the study design or results Methodological citing
  Often experiments are conducted in line with previously reported protocols. We extracted the cited publication(s) for the burrowing assessment protocol from each study to
  assess variations in outcome measurement.
  In addition, we determined whether the burrowing outcome metric was the same as described by Andrews et al.10: weight displaced, which was the first study reported of using
  burrowing behaviour as a pain-related outcome measure, and whether authors provided justifications for using alternative burrowing outcome metrics. Acclimatisation and animal husbandry
  Time period of acclimatisation to housing environment following transportation
  Housing environment
  Light–dark cycle
  Feeding regime Burrowing assessment characteristics
  Experimental environment
  Characteristics of the artificial burrow tube (ie, colour, size, and material)
  Type of substrate
  Training or social facilitation
  Predefined baseline burrowing threshold as an inclusion criterion Curated content
  Whether the study assessed the model effect on animal's motor activity
  Whether the study assessed the drug treatment effect on animal's motor activity
  Whether the analgesic dose was determined by conducting pilot experiments in naive animals of which the burrowing behaviours were not affected
Table 2 - Experiment-level data extracted from each included study. Experiment level Animal
  Species
  Strain
  Sex
  Animal supplier
  Age (at the start of experiments)
  Weight Disease model
  Method of model induction
  Perioperative analgesic(s) given before or during or after model induction Intervention
  Dose
  Route of administration
  No. of administrations
  Time between drug treatment and model induction
  Time between drug treatment and burrowing assessment Outcome measure assessment
   Primary outcome: burrowing
  Habituation time
  Assessment duration
  Direction of effect
  No. of trials and time separation between trials
  Time between the model induction and the first assessment
  Time between the model induction and the last assessment
  Time between the first treatment and the first assessment
  Sex of the investigator
  Presence of the investigator during assessments
   Secondary outcome: limb withdrawal evoked by monofilaments
  Habituation time
  Method of assessment (ie, duration, force, and area of application)
  Direction of effect
  No. of trials and time separation between trials
  Time between the model induction and the first assessment
  Time between the model induction and the last assessment
  Time between the first treatment and the first assessment Numerical outcome data
  Unit
  Mean outcome
  Variance
  No. of animals per group
  No. of groups served by the control group
2.4.2. Risk of bias assessment

Risk of bias was evaluated by using the adapted version of the CAMARADES checklist and SYRCLE Risk of Bias tool,26,32 which assessed the reporting of 6 methodological quality criteria: random group allocation, allocation concealment, blinding of outcome assessment, sample size calculation, predefined animal inclusion criteria, and animal exclusions. Reviewers stated whether each criterion was reported with a description of the method that the study used. A separate rating was given to each item according to the following criteria: low risk (accepted methods and were adequately described), high risk (inappropriate methods that did not efficiently mitigate bias), and unclear risk (the methodological quality criterion was not reported or details of methods were insufficiently reported). Reporting of potential conflict of interests and compliance of animal welfare regulations were also extracted but were not included in the overall risk of bias.

2.5. Reconciliation

After data extraction, reconciliation was performed by a third independent reviewer (M.D.-d.C. and A.-M.H.). For graphically presented data, the third reviewer calculated the standardised mean difference (SMD) effect sizes of individual comparisons for the 2 reviewers. When individual comparisons differed by <10%, the third reviewer took an average of the 2 means and variance measures. When they differed by >10%, the outcome data were required to be extracted by the third reviewer.

2.6. Data analyses

X.Y.Z. and A.B. conducted the meta-analysis by following the guidelines described by Vesterinen et al.56 Burrowing outcome data were first separated according to the analytic approach reported by the protocol of the original study: intention-to-treat (ITT) analysis (animal exclusion was applied before experiments) and per protocol analysis, where animal exclusion was applied after experiments (ie, during analysis). Burrowing outcome data were primarily analysed using the ITT approach (ie, animal exclusion during the training phase); we, therefore, focused on the interpretation of ITT data in this article. Per protocol burrowing data are available at https://osf.io/96hmw/. Burrowing data were further separated by the type of experiment (ie, animal modelling or intervention experiments). The number of independent cohort-level effect sizes (k) required for each meta-analysis is ≥10. When k is <10, a descriptive summary was presented. Subgroup analyses were conducted to investigate how study characteristics influences effect sizes. All analyses were conducted using R statistical packages: dmetar (version 0.0.9), meta (version 4.15.1), and metafor (version 2.4.0).

2.6.1. Effect size calculation

An effect size was calculated for each individual comparison, which was defined as a cohort of animals receiving treatment vs a control group using the Hedges' g SMD method. The use of sham control data was prioritised over naive control and then baseline of the same animals during effect size calculations. All sample size was corrected by dividing the reported number of animals in the control group by the number of treatment groups it served to obtain a “true number of control animals.” Effect sizes were weighted using the inverse variance method to reflect the contribution of each comparison with the total effect estimate. When more than 1 outcome metric of the same behavioural outcome was reported from the same cohort of animals, a single nested effect size, which denotes a summary effect of the cohort, was calculated. Cohort-level effect sizes were pooled using a random-effects model as it considers within-study and between-study variances. The restricted maximum-likelihood method was used to estimate the variance of the distribution of true effect sizes.55 The Hartung–Knapp–Sidik–Jonkman method was also applied to adjust confidence intervals (CIs).20,21,48

2.6.2. Heterogeneity

Heterogeneity was assessed by Cochran Q and I2 tests. A P value was calculated for Q, giving an indication of whether all cohort-level effect sizes shared a common effect size (P > 0.05) or not (P < 0.05). The I2 test calculates the proportion of total variance between studies that is due to true differences in effect sizes as opposed to chance. I2 values were interpreted according to the definition given by Higgins and Thompsons24: 0% to 25% indicates very low heterogeneity, 25% to 50% indicates low heterogeneity, 50% to 75% indicates moderate heterogeneity, and >75% indicates high heterogeneity.

2.6.3. Subgroup analyses

Stratified meta-analyses for categorical variables were performed according to rodent species, strain, sex, model type, drug class, type of burrowing substrate, type of burrowing measurement, and methodological quality criteria. Multivariate meta-regressions were planned to identify other factors that influence the burrowing outcome and, however, were not possible because of the low k number from each variable.

2.6.4. Publication bias

Funnel plots were generated to visually inspect plot asymmetry. Standardised mean differences were plotted against sample size–based precision estimates (1/√N).62 Egger's regression test provided a statistical assessment of the presence of publication bias. Trim-and-fill analysis attempted to correct funnel plot asymmetry by imputing the theoretically missing studies to enable a recalculation of the effect size.

2.6.5. Power analysis

Power analysis was originally planned to compare the number of animals required for burrowing and monofilaments tests. However, because the 2 tests measure different pain-associated behaviours, we conducted a power analysis to illustrate how researchers can use our metadata. We performed a power calculation for the burrowing outcome in rats induced with complete Freund's adjuvant (CFA). The most conservative estimate of the 95% CI of the pooled effect size was used to calculate the number of animals required. Calculations were based on the 2-sample 2-sided t test, with 80% power and a significance level of 0.05 (G*Power version: 3.1.9.7).

2.6.6. Sensitivity tests

Sensitivity analyses were conducted to ascertain the robustness of our findings and to investigate whether a single study or group of studies have skewed the analysis. The following tests were performed:

(1) Baujat plot (2) Single study exclusion sensitivity (3) Cumulative study exclusion sensitivity (4) Excluding studies with high risk of bias

A sensitivity analysis based on “excluding studies that reported burrowing as a primary outcome measure from those reporting it as a secondary outcome measure” could not be performed because only 5 studies declared this information.

Several post hoc (not planned in the registered protocol) analyses were also conducted:

2.6.7. Correlation of burrowing and mechanically evoked limb withdrawal behavioural outcomes

Cohort-level comparisons of trauma-induced neuropathy models that assessed both burrowing and limb withdrawal were used to investigate correlation. A line was fitted using the least square method with subsequent R2 calculation.

2.6.8. Dose-response relationships

Logarithmic transformation of different reported analgesics doses (mg/kg) was plotted against SMD effect sizes. Dose-response relationships were investigated in disease modelled and naive or sham animals. To avoid the confounding effect of repeated administration, single administration cohort-level comparisons were used. A post hoc analysis was admitted (not planned in the protocol) to calculate the significance level using the unpaired 2-tailed t test for each cohort-level comparison using the extracted mean and standard error of the control and intervention groups.30

2.6.9. Drug effect on naive animals

Subgroup analysis based on intervention class was conducted to assess the effect of drug interventions on burrowing behaviour in naive rats.

3. Results 3.1. Study selection

A total of 710 publications were retrieved; of which, 74 studies were included after title and abstract screening. Full-text screening identified 48 studies for this review (Fig. 1). Of which, 3 studies were missing key information for meta-analysis. Among the 45 studies which were included in meta-analysis, there are 3 multicentre studies,1,45,59 and, thus, we extracted the data for each individual participant laboratory. A report is defined as experimental data obtained by an individual research group. In total, data from 54 reports were extracted.

Figure 1.: A flow diagram of publications identified through 2 separate systematic searches of 4 electronic databases (EMBASE, PubMed, Web of Science, and Scopus), which were conducted on March 23 and September 29, 2020. The diagram illustrates the number of records (n) at deduplication, screening, and study eligibility for both qualitative and quantitative analyses. r denotes the number of reports, which is defined as experimental data obtained by an individual research group within a study. Reported in accordance with the PRISMA 2020 guideline.353.2. Study characteristics

Meta-analysis of the 45 studies included a total of 3232 rodents (1590 in animal modelling and 1642 in intervention experiments). Burrowing behaviour was reported as investigated in 33 different rodent models associated with persistent pain. The models are listed according to the model type (16 model types) in Table 3; inflammation (27%, k = 53), arthropathy (23%, k = 44), and trauma-induced neuropathy (15%, k = 29) were the most frequently reported. Furthermore, 27 drug interventions were used to investigate the treatment effect on burrowing outcome in rodents modelled with persistent pain. The drugs are listed by their mechanism of action (14 drug classes) in Table 4; gabapentinoids (26%, k = 28), nonsteroidal anti-inflammatory drugs (NSAIDs) (25%, k = 27), and opioids (22%, k = 24) were the most frequently tested. Mice were used in 21% of experiments (k = 41), and 79% (k = 154) used rats. Moreover, 74% (k = 145) used male animals, 17% (k = 33) used female animals, 2% (k = 3) used mixed sexes, and 7% (k = 14) did not report the sex of the animals used.

Table 3 - Summary of the model types used in animal modelling and drug intervention experiments of burrowing and monofilaments tests. Model type Model name Burrowing experiments Limb withdrawal evoked by monofilaments experiments No. of studies No. of reports No. of k No. of rats No. of mice No. of studies No. of reports No. of k No. of rats No. of mice Arthropathy Monosodium iodoacetate–induced osteoarthritis (intra-articular) 1 1 24 298 — — — — — — Complete Freund's Adjuvant (intra-articular) 3 3 16 222 — — — — — — Collagen-induced arthritis 1 1 3 48 — 1 1 3 40 — Medial meniscectomy–induced osteoarthritis 1 1 1 19 — 1 1 1 19 — Neuropathy: antiretroviral-induced Stavudine-induced neuropathy 1 1 1 15 — — — — — — Neuropathy: chemotherapy-induced Bortezomib (cancer chemotherapy)-induced neuropathy 1 1 1 13 — — — — — — Paclitaxel (cancer chemotherapy)-induced neuropathy 1 1 1 12 — — — — — — Neuropathy: diabetic-induced Zucker diabetic fatty–induced neuropathy 1 1 11 203 — 1 1 1 29 — Streptozotocin-induced neuropathy 1 1 2 39 — — — — — — Neuropathy: trauma injury Spared nerve injury 3 3 13 130 104 3 3 9 131 37 Chronic constriction injury 2 3 11 204 — 1 1 1 37 — Tibial nerve transection 2 2 2 12 — 1 1 1 12 — L5 spinal nerve transection 1 1 1 21 — 1 1 1 21 — Partial sciatic nerve ligation 1 1 1 20 — 1 1 1 20 — Unilateral ligation of the infraorbital nerve 1 1 1 20 — 1 1 1 20 — Spinal cord injury Spinal cord contusion (thoracolumbar) 1 1 6 154 — 1 1 4 83 — Nociplastic pain Reserpine-induced fibromyalgia-like 1 1 4 — 44 1 1 3 — 48 Lumbar intervertebral disc degeneration Surgical disruption of nucleus pulposus (L4/5, L5/6, and L6/S1) 1 1 1 — 39 1 1 1 — 39 Visceral inflammation Dextran sulphate sodium–induced colitis 3 3 6 40 56 — — — — — Cerulein-induced pancreatitis 1 1 1 — 16 — — — — — Mucositis Fluorouracil- induced mucositis 1 1 4 72 — — — — — — Irradiation-induced oral mucositis 1 1 1 — 16 — — — — — Complex regional pain syndrome Closed distal tibia fracture with casting 1 1 2 — 29 2 2 2 — 32 Dental injury Surgically induced dental cavity 1 1 1 — 18 1 1 1 — 18 Cancer Azoxymethane or dextran sulphate sodium–induced colitis or colitis-associated colorectal cancer 2 2 5 — 72 — — — — — Syngeneic orthotopic pancreatic tumour (6606PDA cancer cell line) 1 1 1 — 52 — — — — — Syngeneic breast cancer metastases to the bone (MRMT-1-Luc2 cancer cell line) 1 1 1 50 — — — — — — Syngeneic breast cancer metastases to the bone (4T1-Luc2 cancer cell line) 1 1 1 — 33 — — — — — Inflammation Complete Freund's Adjuvant (intraplantar) 5 13 53 837 34 — — — — — Ultraviolet B and heat rekindling–induced inflammation 1 1 1 16 — 1 1 1 16 — Migraine Glyceryl trinitrate injection (intraperitoneal) 2 2 5 10 52 — — — — — Procedure-associated pain One-side sham embryo transfer (female) or 1-side sham vasectomy (male) 6 7 8 — 139 — — — — — Paclitaxel injection (intravenous or intraperitoneal) 2 2 4 73 — 2 2 4 66 — Total 195 2528 704 35 494 174

The total number of studies and reports are not provided as summation will surpass the true total (45 studies and 54 reports) because of multiple disease models being investigated per study and reports.

k, independent cohort-level effect size.

Table 4 - Summary of the drug classes used to assess the effect on burrowing and limb withdrawal behaviours in rodent disease model–associated persistent pain.