Tlr2^tm1Aki homozygous mutant mice (engineered with a constitutive mutation deleting a transmembrane domain and protein cytoplasmic tail; see Takeuchi et al., 1999) were obtained from a colleague's colony and triad-mated to create cohorts of mutant mice. We confirmed presence of this genetic lesion by PCR of DNA obtained from tail biopsy of putative Tlr2^−/− mice. WT male and female (C57BL/6) mice were obtained from Jackson labs and triad-mated to create a WT colony. Since Tlr1, Tlr6, and Tlr10 do not form homodimers, Tlr2^−/− mice cannot transduce intracellular signals evoked by exogenous or endogenous Tlr2 ligands (Zähringer et al., 2008; Erridge, 2010), and do not have a dominant negative phenotype evoked from aberrant Tlr1, Tlr6, or Tlr10 signaling. Prior to behavioral testing, mice were housed in the UNMC vivarium at a density of ≤5 per cage in a microisolator system (Lab Products Inc., Seaford DE), provided with chow (Envigo Teklad #7012) and water ad libitum, given environmental enrichment (Crinkle Paper Pouches, WF Fisher), and maintained on a 12:12 circadian lighting cycle (lights on 0600 CST). Vivarium temperatures ranged between 20 and 23°C. Mice were separated by sex at weaning. All studies were performed in full concordance with both institutional and federal regulations regarding animal care and use. The protocol was approved by the UNMC Institutional Animal Care and Use Committee (IACUC).

We performed a longitudinal assessment using dual emission x-ray absorptiometry (DEXA) for in vivo estimates of mouse adiposity. We evaluated one cohort of male WT (n = 8) and male Tlr2^−/− (n = 8) mice at age of 4–4.5 and 8–8.5 months. One mouse was lost by attrition from the 8–8.5 months old WT cohort. Before data collection, all instrumentation was calibrated to a phantom approximating mouse body composition characteristics. Briefly, mice were lightly anesthetized with isoflurane, and imaged (Piximus I, GE Lunar). Measures of body mass composition, including bone mass density (BMD), bone mineral content (BMC), bone area (BArea), tissue area (TArea), ratio of soft tissue attenuation (R_ST), total tissue mass (TTM), and percent adiposity (% fat) were calculated from images using vendor provided software (Piximus 2.10). Differences in cohort DEXA and body weight values were determined by unpaired two-sided Student t-tests, with Bonferroni correction of the critical p to control the false positive rate.

We performed a longitudinal assessment of mouse metabolic parameters using indirect calorimetry. Mouse cohorts as above. Briefly, mice were fasted overnight prior to metabolic assay. Animals were placed individually in a calorimetry chamber (8 total; Oxymax, Columbus Instruments), and tested between 12:00 and 17:00 for 1 day. Measures of metabolic rate, including maximum oxygen uptake (V˙O₂), global oxygen delivery (DO₂), oxygen output (O₂out), maximum CO₂ production, (V˙CO₂), global CO₂ removal (DCO₂), CO₂ output (CO₂out), and heat generated were calculated using vendor provided software (Oxymax for Windows 4.49). These metabolic parameters were adjusted for mouse adiposity per ANCOVA (Tschöp et al., 2012). Full details regarding our system components and operation are provided in Bonasera et al. (2017).

We performed a longitudinal assessment of mouse home cage behaviors of feeding, drinking, movement, and circadian rhythm. Mouse cohorts as above. Briefly, mice were placed in a custom-designed home cage monitoring arena (32 total) with ad libitum access to milled chow (#5058, PicoLab) and water. This system provides high spatial (within 0.5 cm) and temporal (within 1 ms) precision of all mouse behaviors. Mice are habituated to the home cage environment for 5 days before start of data collection, which then proceeds for at least 15 days (to ensure collection of at least 14 days of data for each mouse). The system collects data at all times (except brief intervals every 3–4 days for replacement of mouse food and water supplies). After the mouse is introduced to the home cage, it is not handled until the end of the experiment. Since mouse handling is a well-appreciated stressor known to alter many behaviors (Hurst and West, 2010), we thus capture mouse home cage behaviors without imposing significant external stressors. Behavioral data obtained from this system describing C57BL/6 mice are highly similar regardless of system location or investigator (Tecott, Goulding, personal communication). Data passing automated quality control procedures are then classified to determine mouse active/inactive states, mouse intake (of food and water) bouts, and mouse movement (locomotion and movement-in-place) bouts. Following classification, we determine up to 665 distinct measures of mouse feeding, drinking, movement, and circadian activity. We employ false discovery rate (FDR) statistics to minimize family-wise error rates; behaviors found significant in this manner are subjected to further analysis. Full details regarding our system hardware and software characteristics have been published (Goulding et al., 2008; Parkison et al., 2012; Bonasera et al., 2017).

We examined circadian periodicities using Lomb-Scargle analysis (Lomb, 1976; Scargle, 1982); this method detects multiple periodicities within a time series and has been validated in the setting of incompletely sampled data streams. Feeding, drinking, and movement data were binned into 6-min epochs, and significant periodicities (up to 60 h duration) were calculated using an implementation described by Van Dongen et al. (1999) and coded in MATLAB 2011b (MathWorks, Natick MA). We examined patterns of active state onset and duration using the comparison clustering algorithm described by Goulding et al. (2008) and coded in MATLAB 2011b. Briefly, active state onset and duration data from both the WT and Tlr2^−/− groups are combined. Between 2 and 50 clusters were fit to this dataset using bivariate normal distributions. We then test the null hypothesis that WT and Tlr2^−/− mice are equally represented in a given cluster by calculating a χ² statistic comparing the observed to predicted number of data points that WT and Tlr2^−/− contribute to this cluster. The overall difference between WT and Tlr2^−/− mice active state patterns is thus represented by the sum of χ² values over all clusters.

The major limitations of our home cage behavioral monitoring approach include brief epochs of data loss secondary to blocked photobeams (e.g., chow pile, or resting mouse close enough to encroach on the beam radius), licker malfunction (in particular, drips), mouse weight gain/loss during the experiment, and excessive rearing along the cage wall (both altering reported movement distances). All of these issues are identified during our data quality control phase, and removed from the dataset at that time. Since these issues occur infrequently and at random times, there are no statistical problems introduced by removing this small percentage of collected data during the quality control phase.

Most of the genotypic differences between the WT and Tlr2^−/− 4–4.5 months old cohorts involve behaviors occurring during the circadian light cycle (46 of the 73 total, Supplemental Table 1). Tlr2^−/− mice showed significantly less light cycle movement, feeding, and drinking, and budgeted significantly less time to perform these behaviors. Bout metrics for light cycle intake/movement onsets, intake/movement durations, per bout intake/movement, bout intake/movement intensities, and bout intake rates/movement speeds all show large genotypic differences as a consequence of the overall paucity of Tlr2^−/− light cycle behaviors. Inactive and active state properties (discussed in next section) were generally characterized by Tlr2^−/− mice having fewer daily active/inactive states, lower active/inactive state transition rates, but longer active/inactive state durations; all consistent with the concept that Tlr2^−/− mice have consolidated active and inactive states. Figure 1A is a volcano plot that displays two-fold changes in specific behaviors between WT and Tlr2^−/− above the horizontal dashed line; metrics where WT were significantly greater than Tlr2^−/− are depicted right of the right vertical dashed line, metrics where WT were significantly less than Tlr2^−/− are depicted left of the left vertical dashed line. In this figure, 38 of the 57 behaviors identified as having genotypic differences were observed during the light cycle.

Interestingly, at 8–8.5 months of age, these same mouse cohorts demonstrated far fewer genotypic differences in behavior. No differences were appreciated by FDR analysis (p < 0.05, Supplemental Table 1); significant differences in 6 behaviors were noted by volcano plot (Figure 1B).

As mentioned above, most of the behaviors phenotypically differing between young WT and Tlr2^−/− cohorts reflected significant changes in how these two genotypes regulate active states. We therefore decided to examine mouse active and inactive state properties in greater detail. Over the combined dark and light cycles, Tlr2^−/− mice had fewer active states per day (Figure 2A, Table 1), fewer active states per hour (Figure 2B), and a shorter active phase (Figure 2C) compared to WT mice. Active state durations trended to be longer in Tlr2^−/− mice, but this finding did not achieve statistical significance. Similarly, over the combined dark and light cycles Tlr2^−/− mice had fewer inactive states per day (Figure 2D), fewer inactive states per hour (Figure 2E), a longer inactive phase (Figure 2F), and longer duration inactive states (Figure 2G) compared to WT. These findings are consistent with Tlr2^−/− mice experiencing fewer transitions between active and inactive states throughout the circadian day, and thus having increased state stability.

This altered active/inactive state transition frequency is particularly marked during the circadian light cycle (Table 2). Compared to WT, Tlr2^−/− mice expressed fewer active states (Figure 3A), had a lower active state rate (Figure 3B), and a shorter active state duration (Figure 3C) during the circadian light cycle. Similarly, Tlr2^−/− mice had fewer inactive states (Figure 3D), a lower inactive state rate (Figure 3E), and a longer inactive state duration (Figure 3F) during the circadian light cycle. These values suggest that WT mice spend 13% of the light cycle in an active state, while Tlr2^−/− mice spend 4% of the light cycle in an active state. Not surprisingly, decreased light cycle activity duration in Tlr2^−/− mice was accompanied by decreased feeding, drinking, and movement compared to WT (Figures 3G–J respectively). Tlr2^−/− mice clearly experience fewer transitions between active/inactive states during the circadian light cycle compared to WT.

This Tlr2^−/− light cycle phenotype is well illustrated in Figure 4, which shows a dual plot actogram for a representative 4–4.5 months old WT and Tlr2^−/− mouse over 16 days. Note that the Tlr2^−/− mouse demonstrated decreased light cycle activity, and greater active state consolidation over this time period compared to WT. These findings extend to all the mice within the WT and Tlr2^−/− cohorts (Figure 5A). For all mice within a cohort across all experiment days, Figure 5A depicts active state onset on the x axis and active state duration on the y axis and clearly demonstrates that Tlr2^−/− mice express fewer active states during the light cycle compared to WT. This plot also demonstrates that short duration active states occurring throughout the circadian dark cycle are statistically underrepresented in Tlr2^−/− mice.

There were no genotypic differences between 8 and 8.5 months old WT and Tlr2^−/− mice in overall active/inactive state properties including (1) number of daily active states, (2) active state onset rates, (3) active phase durations, (4) number of daily inactive states, (5) inactive state onset rates, (6) inactive phase duration, or (7) inactive state duration. Similarly, there were no genotypic differences between 8 and 8.5 months old WT and Tlr2^−/− mice in light cycle active/inactive state properties including (1) number of light cycle active states, (2) light cycle active state onset rates, (3) light cycle active state duration, (4) number of light cycle inactive states, (5) light cycle inactive state onset rates, (6) light cycle inactive state duration, or (7) light cycle feeding, drinking, and movement. All of these metrics showed significant phenotypic differences between 4 and 4.5 months old WT and Tlr2^−/− mice.

However, 8–8.5 mo old Tlr2^−/− mice continue to demonstrate a statistically significant underrepresentation in short duration active states occurring during the dark cycle (Figure 5B), with this finding further extending to short duration active states within the light cycle. However, as apparent from comparison of similar time epochs between Figures 5A,B, this phenotype is less pronounced in the older mouse cohort.

We performed Lomb-Scargle analysis of movement, feeding, and drinking time series (from 4–4.5 to 8–8.5 months old WT and Tlr2^−/− mice) to assess periodicity of these behaviors. In both 4–4.5 months old mice and 8–8.5 months old mice, normalized power of the 24-h spectral component is significantly greater in Tlr2^−/− mice compared to WTs for movement, feeding, and drinking (Figure 6). The decrease in amplitudes of these 24-h spectral components (as well as 12-h and 8-h ultrudian spectral components) between 4–4.5 and 8–8.5 months old mice suggests that the older mice have greater variability in minute-to-minute and day-to-day performance of movement, feeding, and drinking compared with younger mice, and is consistent with our prior observations showing age-related decreases in the amplitude of these same spectral components in older C57BL/6 and BALB mice (unpublished data). Lomb-Scargle analysis suggests that Tlr2^−/− mice have less variability in movement, feeding, and drinking behaviors compared to WT mice, regardless of age.

WT mice were significantly heavier (25.2 ± 0.8 g WT; 23.1 ± 1.1 g Tlr2^−/−; p < 0.0008 at 4–4.5 months; 29.0 ± 0.9 g WT; 26.7 ± 1.7 g Tlr2^−/−; p < 0.004 at 8–8.5 months), and had a greater total tissue mass (TTM; 21.3 ± 0.8 g WT, 19.8 ± 1.0 g Tlr2^−/−; p < 0.006 at 4–4.5 months; 25.4 ± 0.85 g WT, 23.3 ± 1.5 g Tlr2^−/−; p < 0.005 at 8–8.5 months) than Tlr2^−/− mice. No significant differences in BMD, BMC, BArea, TArea, RST, or % adiposity were noted at either 4–4.5 or 8–8.5 months. We noted no significant differences in the metabolic parameters of V˙O₂, O₂out, DO₂, V˙CO₂, CO₂out, DCO₂, and heat either at rest (basal status) or with moderate activity.

Regarding home cage phenotypes, we noted no significant differences in overall food and water consumption, and overall daily movement. There were no differences in the 24 h behavioral time budgets for inactivity, feeding, drinking, locomotor, and nonlocomotor movements. Similarly, there were no differences in active state time budgets examining percentage of time devoted to feeding, drinking, locomotion, and nonlocomotor movements, nor were there differences in the probabilities of specific behaviors occurring after start of new active states or preceding the finish of ongoing active states. We also found no significant differences in the dark cycle (4–4.5 months old) or dark and light cycle (8–8.5 months old) patterns of feeding, drinking, or movement bouts, including no differences in hourly bout onset rates, bout probabilities, bout speed/intensities, bout durations, or per-bout intake/movement. Finally, we assessed anxiety-related behaviors by elevated zero maze and open field assays (Supplemental Methods), and found that Tlr2^−/− mice demonstrated increased anxiety related behaviors (increased thigmotaxis and decreased entries into open field center; Supplemental Figure 1) compared to WT mice. No genotypic differences in zone transitions or time spent within either closed or open elevated zero maze arms were appreciated.

SB is an uncompensated, ad hoc consultant to Mousera Inc., a startup company focused on high-throughput rodent behavioral monitoring technology. The other authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.