Dose-dependent determination of colitis-induced severity progression by monitoring of VWR behaviour

VWR was monitored in C57BL/6J (B6) mice that were treated with either 0%, 1%, or 1.5% dextran sulfate sodium (DSS) to induce acute intestinal inflammation. Furthermore, VWR was monitored in DSS-treated B6 mice that additionally underwent facial vein phlebotomy (for an overview of groups and n values see S1 Table). All mice were single housed in cages supplemented with a running wheel (Revolyzer 3TS system, software DASY Lab 11.0) that allowed monitoring of wheel rotations (WR₂₀) and maximum velocity (Vmax₂₀) of 20 hours/day. During the 14-day (d) adaptation phase, WR₂₀ and Vmax₂₀ increased continuously, reaching a consistent plateau after 9 days (S1 Fig). Mean WR₂₀ and Vmax₂₀ of the last 3 days of the respective adaption phases served as the baseline to calculate the relative change in %. Subsequent experimental procedures comprised faecal sampling (all groups, Fig 1A–1D); blood sampling (selected groups, Fig 1C and 1D) on d 0, d 5, and d 14; DSS treatment (d 1–d 5), and necropsy (d 14) (see S1 Table). Mice were monitored daily by clinical scoring and weighing.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Assessment of severity during acute intestinal inflammation.

(a) Determination of body weight (% change from baseline) and (b) WR₂₀ (percent change from baseline) in mice receiving 0%, 1%, or 1.5% DSS. (c) Body weight and (d) WR₂₀ over time in mice receiving 0%, 1%, or 1.5% DSS and additionally undergoing facial vein phlebotomy on d 0, d 5, and d 14. All mice of (a–d) underwent faecal sampling. For n values see S1 Table. *P < 0.05, **P < 0.01, and ***P < 0.001; colours indicate comparison between respective groups: medium grey between 0% and 1%, black between 0% and 1.5%, and light grey between 1% and 1.5% (one-way ANOVA, subsequent Tukey posthoc test or Kruskal–Wallis test followed by Dunn’s multiple comparison test); underlined asterisks indicate the comparison to baseline levels within a group (repeated measure ANOVA, subsequent Dunnett’s posthoc test or Friedman test followed by Dunn’s multiple comparison test). (e) B6 mouse demonstrating VWR behaviour in a running wheel; WR₂₀ of all mice (a–d) plotted against body weight in k-means cluster analysis with cluster borders (solid lines) and 95% confidence borders (dashed lines). (f) Cluster analysis as in (e), DSS-treated mice at d 7 individually highlighted in black; (g) the corresponding calculation of severity fractions. (h) Cluster analysis as in (e), DSS-treated mice at d 7 that were submitted to facial vein phlebotomy individually highlighted in black; (i) the corresponding calculation of severity fractions. The underlying numerical data of each figure panel are provided in the respective excel sheet of S1 Data; underlying numerical data of Fig 1F–1I are provided in the corresponding sheet of Fig 1E. B6, C57BL/6J; DSS, dextran sulfate sodium; VWR, voluntary wheel running; WR₂₀, wheel rotations during 20 hours/day.

https://doi.org/10.1371/journal.pbio.2006159.g001

Significant weight loss up to 21.6% (mean 11.9% ±1.9% SEM) on d 7 was observed in mice treated with 1.5% DSS but not 1% DSS, compared to controls (d 7, Kruskal–Wallis test statistic: 12.22, df = 2; Dunn’s test P < 0.01, Fig 1A). Accordingly, clinical scoring was solely but merely marginally increased in the 1.5% treatment group (S2A Fig). In contrast, WR₂₀ was reduced in both treatment groups, rendering the monitoring of VWR behaviour more sensitive than clinical scoring in determining disease progression in a dose-dependent manner (d 7, Kruskal–Wallis test statistic: 11.97, df = 2; Dunn’s test P < 0.01 for 1.5% versus 0% DSS group, Fig 1B). Vmax₂₀ was reduced solely in mice treated with 1.5% DSS (S3A Fig). Next, serial blood sampling by facial vein phlebotomy, a sampling procedure frequently applied in animal-based research, was performed on d 0, d 5, and d 14 in DSS-treated and control mice (see S1 Table for groups and n values). Unexpectedly, WR₂₀ was significantly reduced in DSS-treated and control mice after blood sampling (d 5, repeated measure ANOVA F_(6.221) = 21.47 [0% DSS], F_(19.84) = 21.90 [1% DSS]; Dunnett’s tests P < 0.001 compared to baseline; Friedman test statistic: 83.05 [1.5% DSS]; Dunn’s test P < 0.001 compared to baseline, Fig 1D). Additionally, blood sampling not only impacted VWR behaviour but also aggravated colitis progression, as 1% DSS-treated mice now displayed a similar course of body weight loss, WR₂₀, and Vmax₂₀ as 1.5% DSS-treated mice (Fig 1C and 1D and S3B Fig). The aggravated condition was not detected by clinical scoring (S2B Fig) but was corroborated by histological analysis (S4 Fig).

Demarcation of individual severity levels by k-means algorithm-based cluster analysis

To enable unbiased severity allocation, k-means cluster analysis based on behavioural data sets (VWR performance) and clinical data sets (body weight measurements) derived from all DSS-treated and respective control mice, including their baseline values, was determined to be suitable. Interestingly, an optimal cluster size of three clusters was obtained by scree plot analysis as well as calculation of the Bayesian information criterion (S5A and S5B Fig). Cluster stability was monitored by permutation analysis. Cluster borders were calculated to be WR₂₀ = 87.37% and WR₂₀ = 50.16%, with 95% confidence borders (83.75; 90.39) and (46.43; 53.57), respectively (Fig 1E and S5C Fig). Accordingly, three severity categories were classified as ‘severity level 0, 1, and 2’, respectively (depicted in Fig 1E). Exemplary highlighting of mice at d 7 demonstrated that all of the control mice (0% DSS) were allocated to severity level 0, whereas the distribution of 1% and 1.5% DSS-treated mice shifted toward severity levels 1 and 2 (Fig 1F). Calculation of the percental proportion of mice assigned to a particular severity category (‘severity fraction’) for each treatment regime revealed that 100% of the control mice were allocated to severity level 0 and none were assigned to severity level 2 (Fig 1G). However, this was reversed in 1.5% DSS-treated mice, as 71% of mice were allocated to severity level 2 and none to severity level 0. Highlighting of 1% and 1.5% DSS-treated mice that additionally underwent facial vein phlebotomy revealed a shift in the distribution pattern toward severity levels 1 and 2, respectively (compare Fig 1F and Fig 1H), further corroborating an aggravated condition due to this routine blood sampling procedure. Merely 38% of control mice (0% DSS) were allocated to severity level 0 but 12% to severity level 2 following routine blood sampling (Fig 1I).

Derivation of distinct severity levels in a mouse model of restraint stress affirms applicability of VWR behaviour–based k-means clustering for individual severity grading

As a next step, the applicability of the cluster model as a tool for severity categorization was tested in mice submitted to restraint stress. In this model, mice were immobilized using restraint tubes for 1 hour from d 1 to d 10. These and respective control mice underwent faecal sampling on d 0, d 7, and d 10. Clinical scoring and body weight were merely marginally altered in restraint-stressed mice (S2C Fig and Fig 2B). However, WR₂₀ was significantly reduced to approximately 50% of baseline performance from d 1 to d 10 in restraint stressed mice (repeated measure ANOVA F_(7.15) = 7.337; Dunnett’s test P < 0.05–0.001, Fig 2C). Interestingly, a drop in WR₂₀ was also observed on days of faecal sampling (d 0, d 7, and d 10) in both control and restraint-stressed mice (Fig 2C). Reduction of Vmax₂₀ in restraint-stressed mice was less pronounced than reduction of WR₂₀ (S3 Fig). Next, these data were tested in the cluster model, revealing an equal distribution of control mice into severity level 0 and 1 on d 1 (Fig 2D and 2E), which might be attributed to the impact of faecal sampling on d 0. This effect of the sampling procedure was also discernible on d 7 and d 10, whereas all control mice on d 3 were categorized into severity level 0 (Fig 2D and 2E). However, the distribution pattern in mice undergoing restraint stress markedly shifted into severity levels 1 and 2, with up to 62% of restraint-stressed mice allocated to severity level 2 on d 7 (Fig 2F and 2G).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. Assessment of severity during restraint stress.

(a) Restrained mice in their home cage. (b) Determination of body weight (n = 8) and (c) WR₂₀ (n = 8) in control and restrained mice, all of which underwent faecal sampling (d 0, d 7, d 10). For groups and n values see also S1 Table. *P < 0.05, **P < 0.01, and ***P < 0.001, comparison between groups (Mann–Whitney or unpaired t test with Welch’s correction in case of unequal variance); underlined asterisks indicate the comparison to baseline levels within a group (Friedman test followed by Dunn’s multiple comparison test). Incorporation of restraint stress data at d 1, d 3, d 7, and d 10 into the cluster model; (d) control mice with (e) the corresponding calculation of severity fractions; and (f) restraint-stressed mice with (g) the corresponding calculation of severity fractions. The underlying numerical data of each figure panel are provided in the respective excel sheet of S1 Data; underlying numerical data of Fig 2D–2G are provided in the corresponding sheet of Fig 1E.

https://doi.org/10.1371/journal.pbio.2006159.g002

Ethics statement

This study was conducted in accordance with the German law for animal protection and the European Directive, 2010/63/EU. All experiments were approved and permitted by the Lower Saxony State Office for Consumer Protection and Food Safety (LAVES, license 15/1905).

Mice and experimental set up

Ten–thirteen-week old female B6 mice were obtained from the Central Animal Facility (Hannover Medical School, Hannover, Germany). Routine health surveillance and microbiologic monitoring according to the Federation of European Laboratory Animal Associations recommendations did not reveal any evidence of infection with common murine pathogens [49,50]. Mice were maintained in a room with controlled environment (21°C–23°C; relative humidity 55% ± 5%; 14:10-hour light:dark cycle). Mice were housed in macrolon cages (360 cm²) with softwood granulates (poplar wood, AB 368P, AsBe-wood GmbH, Germany) and cleaned once per week. Pelleted diet (Altromin 1324, Lage, Germany) and autoclaved water were provided ad libitum. During the 2-week habituation to the room, animals were merely handled for cage cleaning.

For each experimental set up, a different cohort of mice was used (as specified in S1 Table). Sample size calculations were performed using the power analysis program G*Power 3.1 [51]. N values are given in S1 Table. Animals were then divided into treatment and control groups by applying a random selection procedure (drawing lots).

All mice of this study had access to running wheels. Prior to study initiation, a 2-week adaption phase to the running wheel was chosen as outlined below. In the cohorts, the experimental set up was as follows: animals were treated with DSS (0% [control], 1%, or 1.5%) from d 1 to d 5. In these mice, faecal sampling was performed on d 0, d 5, and d 14. Additional DSS-treated mice (0% [control], 1%, or 1.5%) underwent faecal sampling as well as phlebotomy on d 0, d 5, and d 14. Additional mice were used in the restraint stress model. In these groups, restraint stress was applied from d 1 to d 10. In these and respective control mice, faecal sampling was performed on d 0, d 7, and d 10.

Handling during experimental procedures was performed in reference to Sorge and colleagues solely by females [30]. Mice were handled by the tail, i.e., the mice were grasped by the base of the tail using the thumb and forefinger and then transported on the flat of the hand to support the body.

VWR

Mice were single housed in home cages with free access to a running wheel (diameter of 11.5 cm, Revolyzer 3TS system, software DASY Lab 11.0 preclinics GmbH, Germany) that allowed automatic and undisturbed 20-hour monitoring of wheel rotations (WR₂₀) and maximum velocity (Vmax₂₀, referring to the maximal number of wheel rotations per minute recorded during the 20-hour period) from 12:00 PM to 08:00 AM daily, leaving a 4-hour interval for general maintenance and experimental procedures (depending on the cohort, e.g., weighing, phlebotomy, restraint stress). To determine the steady state running performance, an adaption phase of 14 days was chosen before subsequent experiments (see also S1 Fig). During the adaption phase the health status of the animals was monitored twice per week. All B6 mice started to run as soon as they were introduced into the cage supplemented with the running wheel. The peak time of running expectedly occurred during the dark phase. For subsequent WR₂₀ and Vmax₂₀ analysis, the mean of the last 3 days of the respective adaption phases were set as the baseline to calculate relative changes (%).

Induction of DSS colitis

To fully control the onset, duration, and degree of intestinal inflammation for relating severity assessment parameters to the degree of colitis [52,53], an acute colitis model induced by DSS (mol wt 36,000–50,000; MP Biomedicals, Eschwege, Germany) was chosen. Mice of the respective cohorts (see also S1 Table) were exposed to 0% (control group), 1%, and 1.5% DSS in drinking water for 5 consecutive days (d 1–d 5) to induce a mild to moderate intestinal inflammation. Mice were weighed and monitored daily according to the clinical score described below. To prevent severe conditions, a body weight loss ≥ 20% was defined as a humane endpoint.

Restraint stress

To induce acute stress mice were inserted into restraint tubes on 10 consecutive days (d 1–d 10) for 60 minutes (from 09:00 to 10:00 AM) and placed in empty housing cages during the restraint period. Restraint tubes (23-mm internal diameter, 93-mm length) consisted of clear acrylic glass with ventilation holes (8-mm diameter) and a whole length spanning 7-mm–wide opening along the upper side of the tube. The ends of the tube were sealed on one side by a piece of acrylic glass with a slot for the mouse tail and on the other end by a screwable solid plastic ring. Mice were able to rotate around their own axis but not to move horizontally.

Clinical scoring

Clinical scoring was performed daily by the same person between 08:00 AM and 09:00 AM, as described recently [54] including the parameters stool consistency, posture, behaviour, and the appearance of eyes and fur. Clinical scoring constituted a base parameter mandatory for project authorization and was performed by an experienced veterinarian, which was not blinded to the treatment groups. In addition, body weight was determined every day.

Faecal sampling

Mice in the DSS model (0%, 1%, or 1.5% DSS) were transferred from their home cage on d 0, d 5, and d 14 and mice from the stress model were transferred from their home cage on d 0, d 7, and d 10 for a period of 2 hours to a new cage containing LabSand (Coastline Global Inc., Palo Alto, United States) to collect a bulk sample of faecal pellets.

Facial vein phlebotomy

Facial vein phlebotomy was performed in the respective cohorts (as specified in S1 Table) at d 0, d 5, and d 14, as described recently [27]. For this, mice were grabbed by the scruff of the neck to gently but firmly immobilize head, neck, and forelimbs without anaesthesia. The right lateral facial vein was then punctured with a 20-gauge needle. Phlebotomy was performed by the same trained and experienced person throughout the study. Approximately 15 μl of blood were collected with the Protein Saver Card (Whatman 903™, GE Healthcare Europe GmbH, Freiburg, Germany) to be stored as dried blood spots at room temperature for further analyses.

Histology

A ‘Swiss roll’ was prepared from the colon, as described previously [55]. Colon samples were retrieved at d 14 and fixed in neutral buffered 4% formalin, processed routinely, embedded in paraffin, sectioned at 5–6 μm, and stained with hematoxylin and eosin. Histology slides were scored, as published recently, and by grading histopathologic lesions separately for the proximal and distal colon [54,56,57]. Scoring was performed blinded to sample identity/treatment group. Evaluated parameters included the presence of infiltrating inflammatory cells (severity and maximum extent); the intestinal architecture (epithelial and mucosal); the extent of edema, erosion, and ulceration; and the involved area. Each parameter was graded from 0 (no changes) to 4 (severe changes) in the proximal and distal colon sections, achieving a maximum score of 46.

Statistics

Values are means ± standard error of the mean. All statistical analyses were performed using Graph-Pad Prism 5 and 6 software (La Jolla, California). All data were analysed with the Shapiro Wilk test for normal distribution. For parametric data, an unpaired t test with Welch’s correction in case of unequal variance or one-way analysis of variance (ANOVA) or repeated measure ANOVA was carried out. In case of ANOVA, Bartlett’s test was applied to check for homoscedasticity, and if the hypothesis of equal variance was rejected (P < 0.05), nonparametric methods were used. In inferential testing of multiple groups, p-values were adjusted for multiplicity during their individual posthoc testing procedure (Tukey test or Dunnett’s multiple comparison test). For nonparametric data, the Mann–Whitney or Wilcoxon test were performed to compare 2 groups. Other nonparametric data were analysed by the Friedman or Kruskal–Wallis test, both followed by Dunn’s multiple comparisons as posthoc test. P < 0.05 was considered significant. In all figures, * indicates P < 0.05, ** indicates P < 0.01, and *** indicates P < 0.001.

K-means algorithm-based cluster analysis

To calculate clusters in order to assess and categorize severity, the R [58] software and unsupervised k-means clustering were used [58]. Regarding the general k-means clustering procedure, all data sets were retrieved from the experimental colitis group including standardized WR₂₀ and body weight (BW). Both variables were used to calculate k-means clusters (701 × 2 data points out of n = 54 mice). Different conditions and days were pooled to include all possible states in one model. To calculate the cluster thresholds, the 701 × 2 data points were randomly divided into a training (80%) and a test set (20%). The training set was then used to calculate the thresholds. For stratification, this was repeated 100 times (with q = 0.8 × 701 = 561 permutations) at each run. Cluster thresholds were determined by calculating the median of the stratification data after filtering out extreme values; margins of 30% deviation in both directions from the median were allowed. The result was set as the global cluster threshold. This was repeated for each cluster, also resulting in 95% confidence borders (CBs; calculated by CB_{95% CI} = mean_thr ± 1.96 x SD(thr)/√561, with thr = all thresholds for each of the permutations and SD = standard deviation). The number of permutations was chosen to limit a potential overfitting of the resulting 95% CB and never exceeded the number of available data points per iteration. It was therefore considered to be fair. The 95% CBs reflect the randomness due to seeding during the clustering process and indicate a transition zone between the condition borders. Test samples in the confidence regions can be seen as ambiguous and cannot explicitly be allocated to either cluster.

For k-means optimization 2 methods, the scree plot and the Bayesian information criterion (BIC) were used, and for subsequent cluster stabilization analysis, seeding permutations were monitored. For scree plot analysis, the variation was analysed by the ‘within groups sum of squares’ at different cluster sizes. In the scree plot, three clusters were identified as the optimal size for a k-means clustering (S5A Fig). For validation, the R package Mclust [59] and the Mclust function were used to calculate the BIC. The BIC was calculated for 20 components (clusters) in 14 multivariate models. All multivariate models except EII and VII had a maximum BIC at three clusters. However, as EII and VII are both spherical models but the analysed data are rather diagonal and ellipsoidal, these models were not included in the determination of the optimal cluster size (S5B Fig). Cluster stability was also monitored by permutation analysis. For this, the median of 100 samples with 561 permutations, each with different seeding positions, were analysed. The median upper threshold at random seeding over 100 iterations was WR₂₀ = 87.37% and the lower median threshold WR₂₀ = 50.16%. Out of 100 iterations, no cluster showed outliers above or below 1% deviation from the median. Therefore, the median cluster thresholds from the random permutations can be considered stable (S5C Fig).