Xiphoid Surface Temperature Predicts Mortality in a Murine Model of Septic Shock

Abstract

Sepsis continues to be a major challenge for modern medicine. Several preclinical models were developed to study sepsis and each has strengths and weaknesses. The cecal slurry (CS) method is a practical alternative because it does not require surgery, and the infection can be dosed. However, one disadvantage is that the dosage must be determined for each CS preparation using survival studies. Our aim was to refine a survival protocol for the CS model by determining a premonitory humane endpoint that would reduce animal suffering. Mice become hypothermic in sepsis; therefore, we tested whether reductions in surface temperature (T_s), measured by non-invasive infrared thermometry, could predict eventual death. We injected 154 C57BL/6J mice with CS (0.9–1.8 mg/g) and periodically monitored T_s at the xiphoid process over 5 days. We used, as predictors, combinations of temperature thresholds (29–31°C) and times, post injection (18–36 h). A receiver-operator curve, sensitivity, and specificity were determined. A Distress Index value was calculated for the threshold conditions. The optimum detection threshold (highest Youden’s index) was found at T_s ≤ 30.5°C at 24 h (90% specific, 84% sensitive). This threshold condition reduced animal suffering by 41% while providing an accurate survival rate estimate. Using this threshold, only 13/154 mice would have died from sepsis; 67 would have been euthanized at 24 h, and only 7/154 would have been euthanized unnecessarily. In conclusion, using a humane endpoint of T_s ≤ 30.5°C at 24 h accurately predicts mortality and can effectively reduce animal suffering during CS survival protocols.

INTRODUCTION

Injection of rodents with cecal slurry (CS) is a pre-clinical model of polymicrobial peritoneal infection that results in a rapid systemic inflammatory response and can induce circulatory shock, multiple organ dysfunction and death. The technique consists of collecting cecal contents from donor animals, creating a suspension and then injecting the suspended slurry into the peritoneum of experimental animals (1–2). Cryogenic preservation techniques allow the bacterial slurry to be frozen and used in multiple experiments over time, thus avoiding variations in the bacterial content (2). However, survival experiments are required for each new preparation of slurry, to determine the appropriate dose, and then for determining the impact of interventions. Using death as an endpoint raises important ethical concerns of animal discomfort, pain and suffering and using commonly employed clinical signs of morbidity as a humane endpoint is often imprecise and subject to wide ranges of interpretation by practitioners (3) despite previous attempts to reduce bias (4). Therefore, we set out to find a practical, non-moribund end-point that would reduce suffering of our mice.

Mice become hypothermic in sepsis, proportional to the severity of infection (5–6), and changes in core temperature have previously been employed as markers of recovery (7–10). However, in most cases the experimental methods have been impractical, as core temperature was monitored by insertion of a rectal probe, which can induce additional stress, or by surgical placement of a telemetric thermometer. However, studies have successfully used near infrared (NIR) thermometry in models of fungal (10) and gram-negative bacteria (11) infections. This approach was very practical, non-invasive and promising, but it could not directly translate to polymicrobial intraperitoneal sepsis because one study required experimentally induced immunosuppression and the latter used gastric infusion of a specific marine bacterium similar to cholera. Therefore, we tested the feasibility of this NIR approach in a more general model of peritoneal infection. We found that this refinement was a practical alternative in survival protocols that greatly reduced animal suffering and yet offered an accurate means of assessing survival rates in the murine CS model of septic shock.

MATERIALS AND METHODS

Animals and temperature measurement accuracy

We studied a cohort of 154 mice (100 males and 54 females, C57BL/6J background) aged 4 to 6 months old, weighing 20–25 g. Animals were purchased from The Jackson Laboratory or bred, in house, and singly housed at the University of Florida Animal Care Facilities where they remained for the whole duration of the experiment. The temperature and relative humidity range of the room were 20–25°C and 31–67%, respectively. Animals were under a 12:12-h light-dark cycle and had access to standard chow (2918, Envigo) and automatic tap water ad libitum. All procedures were previously approved by the University of Florida Institutional Animal Care and Use Committee and reported information about the procedures conformed to the Animal Research Reporting of in vivo Experiments – ARRIVE guidelines (12).

In our lab, we study the impact of skeletal muscles on innate immune responses to septic shock and we have developed two transgenic colonies of inducible, skeletal muscle specific knockout mice. The mice used in the current study were of wild type or two different genetic mutations on C57BL/6J background. One transgenic strain was bred to have skeletal muscle specific knockout of interleukin-6, whereas another strain was designed to have skeletal muscle-specific myeloid differentiation protein 88 knockout. The observations herein reported were made while performing survival experiments to define the appropriate dose of CS.

Approximately one month prior the experiment, transgenic animals were either treated (30/154 mice) or untreated (30/154 mice) with a single intraperitoneal injection of 110 mg/g of body weight of a selective estrogen receptor modulator (Raloxifene Hydrochloride) suspended in polyethylene glycol (PEG400) to induce the CRE-LOX recombinase gene (13). Untreated animals received an intraperitoneal volume of PEG-400 (vehicle) only. Wild type mice (94/154 mice) did not receive any treatment prior to CS injections. We did not observe a clear impact of these mutations or treatments on the relationship between temperature profile after CS and survival; therefore, we combined these groups into a single cohort for analyses.

To test for accuracy of the T_s measurements against a gold standard temperature monitoring system, a group of 12 wild type mice (no CS injection) were instrumented with radio-telemetry transmitters (TA-E-Mitter; Starr life Sciences, Oakmont, PA) for real time monitoring of core temperature. For this experiment, mice were monitored both at rest and 15 min after exercising on a forced running wheel (Lafayette Instruments), in the heat, until exhaustion using methodologies described previously (14).

CS model

We induced sepsis by intraperitoneal injection of CS collected previously from donor mice. The batch of CS used in this study was prepared and stored according to the method described by Starr et al. (2). In brief, cecal contents were filtered and mixed with 15% glycerol solution in phosphate buffered saline. CS was then aliquoted in 2 ml cryovials and kept in freezing containers (Nalgene Mr. Frosty, Thermo Scientific) at −80°C. Immediately prior to injection, cryovials were thawed to room temperature, mixed well and then injected. Our initial goal was to determine the dose of CS in adult male and female mice that would induce a 40–60% survival over the course of 5 days. Trial CS dosages ranged from 0.9 to 1.8 mg/g, which corresponded to volumes of CS ranging from 225 (0.9 mg/g) to 450 µl (1.8 mg/g) for a 25 g mouse. All dosing took place at approximately 5 PM. Animals were restrained briefly by tail and scruff handling and CS injected IP into the lower left quadrant of the peritoneal cavity while the animal was held in the supine position with the head below the body. For consistency, all injections were performed by the same experienced lab member. Mice were monitored at 12 h, 18 h, 24 h, 30 h, 36 h and then every 12 hours for 5 days. Nineteen animals that met a humane endpoint (moribund criteria, absence of righting reflexes after being laid in a lateral recumbent position) were euthanized by CO₂ inhalation during recovery from CS injection.

Surface Temperature recordings

Before CS injection and at each monitoring period, NIR temperature was measured at the level of the xiphoid process using a non-contact, infrared thermometer (Etekcity Lasergrip 774), Fig. 1A. We performed parallel experiments in subgroups of non-septic mice to compare surface temperature measurements at different sites (e.g. forehead vs. xiphoid process). During the measurement at the xiphoid and at forehead, animals were restrained briefly by hand scruffing. The thermometer is equipped with a target laser, and the beam was aimed approximately 3 inches from the surface. This instrument works in a range of −50° to 380 °C. Manufacturer instructions did not indicate that calibration, beyond initial factory settings, was necessary. Results in one group of animals were compared against a more sophisticated NIR monitoring device with calibration capability (Fluke 62 Max, set with an emissivity setting = 0.95). That system also provided visual laser feedback of the outer limits of the NIR detection area as demonstrated in figure 1B.

Figure 1.

Xiphoid temperature measurements taken with infra-red thermometry Etekcity (IR1) on panel A and Fluke 62 Max (IR2) on panel B. IR1 uses one laser beam to measure surface temperature while IR2 uses 2 laser beams to define the surface detection limits measured. Emissivity was set at 0.95 on IR2. IR1 was used according to settings established by the manufacturer.

Statistical approach and calculations

Normality of data was tested by analyzing skewness and kurtosis profile of data distribution. For parametric data, we used t-test for independent samples for comparisons between forehead and xiphoid process surface mean temperatures and one-way ANOVA to compare mean surface temperature at different time points between survivors and non-survivors. Repeated measures ANOVA was used to compare means between IR thermometry and radio telemetry temperature monitoring system. Data were considered statistically significant when p<0.05.

Sensitivity, specificity and Youden’s Index were determined for surface temperatures varying from 29 to 31°C, 18–36 hours post-injection of CS. The sensitivity of the proposed cutoff values was calculated based on the proportion of animals for whom the outcome was positive (i.e. a true positive response was identified when a set of test parameters accurately predicted eventual death within 5 days, as shown in Table 1). A true negative response was when a set of parameters predicted that animals would survive the experiment. A receiver operator curve (ROC) was then created based on the outcome of these analyses. The ROC was fit to an exponential relationship using nonlinear regression software (GraphPad Prism 5). We utilized Youden’s Index (J = sensitivity + specificity – 1) as a traditional indicator to decide which value should be used to discriminate between animals according to outcomes (15). We defined the best predictor cutoff as the one displaying the highest Youden’s Index (16). We created a Distress Index (Eq. 1) for estimating the accumulated distress and for evaluating the impact of each proposed threshold on the non-surviving animals.

$$\text{Distress Index}=\sum ({\text{mice}}^{\ast }\text{hours post CS injection},\text{prior to death}).$$ Eq. 1

Table 1.

Criteria used to define true/false positives/negatives. Surface temperatures ranging from 29 to 31°C and time from 18 to 36h after inoculation of CS were included in the analysis.

Classification function	Surface Temperature	Outcome
True Positive	≤ threshold	Died
True Negative	> threshold	Survived
False Positive	≤ threshold	Survived
False Negative	> threshold	Died

RESULTS

In preliminary studies, we evaluated different locations for T_s measurements and found that T_s measured at the base of the sternum, just over the xiphoid process and heart, was the most reliable location. This location also resulted in a value closest to actual core temperature, an approach utilized previously (10). In one group of animals, we made direct comparisons between T_s measured at the xiphoid vs. the forehead (Fig 2A). The xiphoid measurements provided the most reliable and reproducible measurement.

Figure 2.

A) Surface temperature measured at the level of xiphoid process and forehead with IR thermometry. Note that xiphoid surface temperature provides less variation and significantly higher absolute values in comparison to forehead (p<0.05). B) Comparison between different IR devices (IR1 = Etekcity; IR2 = Fluke) and radio telemetry (RT) both at rest and after exercise with hyperthermia at the level of the xiphoid process. There were no significant differences between IR devices at rest and after exercise in the heat. RT displayed a significant higher temperature post-exercise with hyperthermia (p<0.05) in comparison to both IR devices.

In another group of animals, we tested xiphoid T_s using two different IR thermometers (Etekcity [IR1] vs. Fluke [IR2]) and against measurements made using implanted radio-telemetry (RT) devices in the same animals (Fig. 2B). The core temperatures measured with the two IR thermometers were not significantly different at rest. At rest, average T_s at the xiphoid was ≈2°C lower than core temperature, measured by telemetry. These animals were then tested immediately following a prolonged moderate exercise test in hyperthermia. The measurements with the two contrasting NIR systems were not significantly different from each other, but during exercise-imposed hyperthermia the gradient between xiphoid T_s and core temperature widened to nearly 3.5°C (p < 0.05).

Eighty animals died and 74 survived during the sepsis experiments, for an overall mortality rate of 52%. Regardless of the experimental group, there was a definite pattern to the frequency of deaths occurring over the first 78 h of recovery, with the large majority of animals dying between 30–48 h, as shown in Figure 3. Surface temperature profiles of survivors and non-survivors are reported in Figure 4 and follow a clear bimodal distribution between outcomes. The general temperature profiles for wild type and transgenic males and females, when treated separately, were not different from the overall combined group. Average baseline T_s before CS injection was 34.0 ± 3.0 °C and within the first 12 h, post CS injection, fell to 30.9 ± 4.0 and 26.6 ± 1.1 °C in survivors and non-survivors, respectively (p < 0.05). For survivors, T_s returned towards baseline within 18 h and remained elevated throughout the remaining experiment. For non-survivors, T_s remained lower throughout the experiment, sometimes rebounding briefly but never fully recovering. Sixty-one non-surviving animals were discovered dead in their cage, while 19 were euthanized after reaching a humane endpoint (lack of righting reflex).

Figure 3.

Number of dead animals per time point. There is a peak in number of deaths 48h after injection of CS. Overall the total number of non-survivors was 80.

Figure 4.

Xiphoid surface temperature profile for survivors (black dashed line) and non-survivors (grey dashed line) in the combined cohort (panel A), wild type mice (panel B), transgenic mice treated with raloxifene (panel C) and PEG400 (panel D), males (panel E) and females (panel F) 0–96 hours following CS injection. In all groups, there is an initial drop in surface temperature after 12 h in both survivors and non-survivors. However, after 18h surface temperature returns towards pre-injection values in the survivors whereas it remains lower for the non-survivors. Data are mean ± standard deviation.

Table 2 reports the individual sensitivity, specificity, and Youden’s index for each set of results for proposed humane endpoint criteria at specific measurement times (18 – 36 h) and threshold temperatures from 29 - 31°C. The overall outcome of the ROC analysis is summarized in Figure 5A and the calculated range of Youden’s Index values are shown in Figure 5B. The highest Youden’s index (J = 0.74) was identified at 24 h, with a threshold T_s of ≤ 30.5°C. This condition predicted mortality with 84% sensitivity and 91% specificity. Other threshold conditions with strong predictive specificity are also highlighted in Table 3 to distinguish desirable endpoint criteria that may be of equal value to Youden’s index for survival studies. Comparisons of these candidate endpoint criteria can be found in Table 3, which reports the reductions in stress, based on calculated Distress Index, that would occur in similar cohorts when select temperature and time criteria are chosen for euthanasia. If the condition with highest Youden’s index (T_s ≤ 30.5°C at 24 h) was used as the threshold, there would be a 41% reduction in mortality. However, if the threshold T_s ≤ 29°C at 24 h was used, a more accurate estimate of mortality is obtained as specificity is 96%, but with a smaller reduction in Distress Index (31.3 %) of the non-surviving mice. Condition T_s ≤ 29°C at 36 h has some advantages in that the specificity is 100% and the prediction of mortality rate is perfect, but the approach would have even less impact on reducing animal suffering (26.5%).

Table 2.

Raw data used to calculate Receiver Operator Curve.

Time	Ts (°C)	False Positive	True Positive	True Negative	False Negative	Early deaths	Sensitivity	Specificity	Younden’s Index (J)
18h	29	6	58	68	9	13	0.725	0.919	0.644
18h	29.5	7	62	67	5	13	0.775	0.905	0.680
18h	30	8	63	66	4	13	0.787	0.892	0.679
18h	30.5	9	66	65	1	13	0.825	0.878	0.703
18h	31	12	66	62	1	13	0.825	0.838	0.663
24h	29	3	61	71	6	13	0.762	0.959	0.722
24h	29.5	5	64	69	3	13	0.800	0.932	0.732
24h	30	6	65	68	2	13	0.812	0.919	0.731
24h	30.5	7	67	67	0	13	0.837	0.905	0.743
24h	31	13	67	61	0	13	0.837	0.824	0.662
30h	29	1	47	73	9	24	0.587	0.986	0.574
30h	29.5	2	49	72	7	24	0.612	0.973	0.585
30h	30	2	51	72	5	24	0.637	0.973	0.610
30h	30.5	3	53	71	3	24	0.662	0.959	0.622
30h	31	6	55	68	1	24	0.687	0.919	0.606
36h	29	0	38	74	3	39	0.475	1.000	0.475
36h	29.5	0	39	74	2	39	0.487	1.000	0.488
36h	30	0	39	74	2	39	0.487	1.000	0.488
36h	30.5	1	41	73	0	39	0.512	0.986	0.499
36h	31	1	41	73	0	39	0.512	0.986	0.499

Figure 5.

A. Receiver operator curve (ROC) for the predictive effectiveness of the set of measurement parameters studied in Table 1. The three dashed arrows refer to decision points discussed in the text. The equation of the line for the ROC curve was Sensitivity = 0.45 + 0.3851(1-e^−23*X). B. The Youden’s index calculation as a function of the false positive rate (1-specificity). The equation of the line is a 2^nd degree polynomial fit.

Table 3.

Thresholds resulting in the best predictive potential and their percentage reduction in Distress Index.

Termination of experiment based on decision choice	Mice euthanized at decision (TP)	Euthanized prematurely at decision (FP)	Mice died prior to decision	Mice eventually died later	Calculated Mortality Rate	Distress Index (mice*hours)	% Distress Reduction (mice*hours)
Ts ≤ 29°C at 24h	61	3	13	6	0.50	2142	31.3%
Ts ≤ 30.5°C at 24h (peak J)	67	7	13	0	0.56	1842	41.0%
Ts ≤ 29°C at 36h	38	0	39	3	0.50	2292	26.5%
Actual Mortality					0.52	3120

DISCUSSION

Our results demonstrate that the drop in xiphoid T_s during the first 24 h of recovery from sepsis is a strong predictor of mortality in this murine CS model. We employed a ROC characteristic approach to detect the T_s and time after induction of sepsis that most accurately predicts mortality in this model. This is the first study to refine a predictive marker for mortality in this particular preclinical model of septic shock using IR thermometry.

Early prediction of mortality during survival protocols allows for early euthanasia and reduces the level of suffering throughout the remainder of the protocol. It is thus a more humane approach to obtaining accurate survival characteristics of an animal population. As observed in Figure 3, mortality rates peak at approximately 48 h after CS administration. Based on this figure and Table 3, we calculated estimates of the “cost” in terms of stress exposure and suffering (Distress Index) in the non-surviving animals. When no criteria except lack of righting reflex were used, this study accumulated a Distress Index of 3,120 (mice * hours) in just the non-surviving mice. Note that ideally one would estimate this value for all surviving mice as well, but the surviving mice recover from their infection within 24–36 h, making it more difficult to determine the hours that they suffered during the infection.

Refinement of mortality indicators to reduce pain and suffering is not the single advantage of these results. For instance, studies involving the use of CS as a preclinical model of sepsis require survival experiments to determine the dose to be injected in subsequent trials (2). This is extremely costly in both investigator time and animal housing costs. Here we injected doses ranging from 0.9 to 1.8 mg/g of body mass in a cohort of 154 mice of both sexes and monitored the animals periodically over the course of 5 days. By employing our proposed threshold at 24 h (T_s ≤ 30.5°C or T_s ≤ 29°C at 24h, Table 2) researchers can readily estimate mortality rate within 24 h, rather than after many days. For this application, it may be desirable to utilize T_s ≤ 29°C at 24h, because it has a more ideal specificity (lower number of false positives) and predicted mortality rate, despite a smaller reduction in Distress Index.

Depending on the purpose of the experiment; however, different criteria might be more ideally suited to the needs of the trial. For instance, if the accuracy of the mortality rate is absolutely critical, choosing T_s ≤ 29°C at 36h and allowing the remaining animals to continue until death might be appropriate in some cases. It still reduces the suffering by approximately 27% and allows investigators to evaluate the shape of the remaining mortality curve. As shown in Figure 6, if the decision on eventual survival is selected at the 24 h time point, additional information is lost in the tail of the actual survival curve. This tail can contain information on treatment effects (i.e. prolonging or shortening life) that may be independent of raw survival. In this case, choosing T_s ≤ 29°C at 36 h might be ideal because the investigator could follow this tail for the remainder of the animals and thus possibly have greater fidelity for treatment effects. Nonetheless, our approach may not be useful if the purpose of the experiment is to prolong survival with a given intervention for longer than 36 h after CS injection.

Figure 6.

A mortality curve from a subgroup of 24 female transgenic mice in response to CS injection, at 1.6 mg/g. The continuous line is the raw, traditional data. The dashed line is the mortality results using the ≤30.5°C at 24 h threshold described in Table 2 and euthanizing all animals that met the criteria. Shaded area represents the net reduction in animals suffering when this criterion is employed.

Our proposed approach to reduce animal suffering and predict mortality by using NIR surface temperature is specific to the murine model of CS, but it shares some interesting features with previous reports (10–11, 17). For instance, Trammell & Toth (17) surgically implanted an intra-abdominal radiofrequency transmitter to measure body temperature and to establish its potential to predict mortality in mice exposed to multiple infectious organisms. They studied different mouse strains such as A/J, DBA/2J, C57BL/6J, BALB/cByJ and monitored body weight over the course of infection. Although no threshold for a humane endpoint was described, hypothermia was the most valuable characteristic for distinguishing mice that would survive or succumb to the infection. Importantly, they suggested that the endpoint threshold should be determined for each specific experimental model. Another study (10) used a NIR surface temperature approach in a fungal infection model in the CD1 strain of mice. They established a cut-off of 33.3°C at any time point that was 97% specific and 68% sensitive as a predictor of death. A recently published investigation suggested a threshold of 23.5°C at 18h as a predictor of mortality in a murine model of Gram-negative bacteria infection (11), which greatly reduced animal suffering in that model. We also know that the temperature profiles of mice who undergo cecal ligation and puncture often show less dramatic and less rapid reductions in temperature, with clear male and female differences (18). However, core temperatures of less than 30°C are commonly reported in severe cecal ligation models (6). Since cecal ligation and puncture is such a common and now well- described procedure, it may be of benefit to establish a similar NIR criteria in that model.

The mechanisms by which hypothermia affects mortality in septic shock in humans and rodents are still unclear, but may involve an impaired host response to infection, which can halt the ability of the immune system to prevent multi-organ damage and/or dysfunction (19). An important aspect to consider is that, in some instances, hypothermia may be considered an adaptive and protective response to sepsis (20). Rodents are endotherms and can regulate their core temperature via metabolism as suggested by a study demonstrating that hypometabolism can prevent hypoxia in rodents undergoing endotoxic shock (21). This could be paradoxically interpreted as hypothermia being an indication of good prognosis in response to infection. Nevertheless, our findings as well as reports of other research groups using different models of infection (10–11, 17, 22) indicated that hypothermia was a sensitive predictor of morbidity and mortality in response to distinct murine models of sepsis/infection and was an effective, practical, and non-invasive way to reduce animal suffering.

Body temperature of mice can be influenced by a number of factors such as ambient temperature, the time of day, the presence and type of bedding (23), and the number of cage mates (24). Additionally, as previously shown (25–26), different strains of mice can differ widely with regards to temperature responses to the same experimental model. In our current study, mice were housed individually with corncob bedding. They were also kept at the animal facility for the whole duration of the experiment, which ensured an exposure to a controlled environmental temperature and humidity. Importantly, mice thermoregulatory response to sepsis is influenced by age (27). Therefore, whether our results hold true to groups of animals within a different age span remains to be determined. These considerations must be contemplated when using NIR surface temperature as a predictor of infection severity and mortality in this murine model of CS.

In conclusion, our results support the use of xiphoid T_s as a surrogate marker of infection severity and a predictor of mortality. For most applications, we recommend a cutoff criterion of T_s ≤ 30.5°C at 24 hours, which in this study proved an accurate predictor of mortality rate within 4% of actual mortality rate. Most importantly, the use of this or similar criteria can drastically reduce the amount of animal suffering (by 41%) required for estimates of severity of infection, effects of dosages, differences in susceptibility of populations and the effects of treatment interventions.