Orbital tightening assessment to evaluate pain and physical discomfort in chlorine-exposed rats: A machine learning based measurement approach

Introduction

Chlorine is a toxic industrial chemical produced in large quantities, mostly for disinfecting drinking water, swimming pools, water reservoirs, and use in food processing systems. Chlorine is a highly noxious gas, and depending upon concentration and duration of exposure, it can cause irritation of the respiratory mucosa and acute lung injury, often leading to fatalities. Humans exposed to accidental or deliberate chlorine spills report severe pain in the exposed areas. The extent of pain can indicate the severity of injury or the toxicity of oxious chemicals in both humans and animals. Animal models of chemical toxicity are frequently evaluated for exposure severity by measuring the animals’ clinical scores, which are composite scores of the respiratory quality, stridor, and activity. ¹ Pulse oximetry is also performed to evaluate heart rate, breath rate, and oxygenation. ² Such noninvasive measurements are highly valuable for assessing post-exposure injury and longitudinal evaluation of the therapeutic efficacy of potential antidotes. Although toxic chemical exposures cause substantial injury, the degree of discomfort and pain has rarely been evaluated. A variety of methods are used to assess pain in animal models, but none are suitable for evaluating pain and physical discomfort induced by inhalation of noxious chemicals.

The respiratory system is innervated by afferent nerves that detect mechanical, chemical, and thermal stimuli from the airways and lungs, transmitting this information to the central nervous system to regulate breathing and protective reflexes. This innervation is primarily mediated by vagal afferent nerves and somatosensory afferent nerves, each contributing distinct roles. ³ Vagal afferent nerves (cranial nerve X) provide the dominant afferent innervation to the respiratory system, including larynx and pharynx, trachea and bronchi, and lung parenchyma. They contain mechanoreceptors that respond to stretch and pressure changes in the lungs and airways, which are critical for reflexes like the Hering-Breuer reflex (inhibiting inspiration during lung inflation). ⁴ Vagal afferent nerves also contain chemical receptors that respond to irritants, inflammatory mediators, and temperature changes, mediating chemical-induced reflexes such as coughing, bronchoconstriction, and mucus secretion.^3,4 The respiratory system is also innervated by somatosensory afferent nerves, especially in the upper airways and thoracic structures. Somatosensory afferent nerves innervating the pleura sense pain and stretch; those innervating the chest wall and diaphragm provide proprioceptive feedback; and those innervating the upper airways (nasal cavity, pharynx) detect irritants and temperature changes. Inhalation of noxious chemicals can activate chemical receptors on both somatosensory and vagal afferent nerve endings in the respiratory system, resulting in both painful sensation and physical discomfort.^5–7

Although chlorine-exposed mice demonstrated a reduction in their mobility as an indication of pain, not many studies attempted to address this symptom. ⁸ In the current study, we employed a machine-learning-based approach using DeepLabCut to quantify eyelid distance and palpebral fissure width in Sprague-Dawley rats to assess their responses to chlorine exposure. This method was previously established by our group to evaluate orbital tightening induced by facial pain following capsaicin injection in mice.^9,10 These measurements were taken under two conditions: (1) in rats that were not exposed to any nociceptive or chemical insult, and (2) in rats that had been exposed to chlorine gas, a known respiratory and systemic irritant associated with nociceptive and inflammatory responses. ¹¹ By comparing these two groups, we were able to quantitatively assess changes in orbital tightening associated with nociceptive distress. Our findings demonstrate that rats exposed to chlorine gas exhibited measurable reductions in both eyelid distance and palpebral fissure width, indicating pain-related changes in facial expression. These results support the utility of automated facial feature tracking as a sensitive method for detecting pain in chemically exposed rodent models.

Materials and methods

Animals and chlorine exposure: Sprague Dawley rats (250–300 g) were obtained (Envigo), housed, and cared for, and animal experiments performed according to approved protocols and University of Alabama at Birmingham Institutional Animal Care and Use Committee (IACUC) guidelines. Our studies with female rat exposure to halogens demonstrated similar injury, therefore to avoid redundancy, obtain consistent data and avoid the complexities associated with hormonal variability we utilized male rats only here. ¹² Male rats 2/cage were housed under a 12 h dim light/12 h dark cycle in standard polycarbonate cages with wood chip bedding and nesting enrichment (EnviroPak). Standard diet and tap water were provided ad libitum, the room temperature was maintained at 70°F and humidity maintained between 40% and 60%. A priori power analysis was performed using the ANOVA F-test to compute sample size. Assuming the group sizes with a power of 95% and a-error probability of 5% or less, expected mortality, and an effect size of 0.5–0.7, the number of animals required for the proposed studies was calculated. Unanesthetized rats were exposed (whole body) to 600 ppm Cl₂ (chlorine obtained from Airgas, Birmingham, AL, gas concentrations in balance air were certified within 2% by the manufacturer) for 45 min as previously described.^13–15 Rats were then returned to room air (RA) and monitored continuously for 2, 6, and 24 h, depending on the time of endpoint evaluation. To evaluate the effect of opioid analgesic chlorine exposed rats were administered 0.03 mg/kg buprenorphine intraperitoneally (IP) 30 min after exposure. Breath rate and heart rate were measured using Mouse Ox (Starr Life Sciences, Oakmont, PA) pulse oximetry as described before. ² In another set of experiments, chlorine exposures were done with a thin layer of petroleum jelly (Vaseline) applied around the eyes of the rats. The eyes were gently wiped after exposure to remove the petroleum jelly, and the videos were made at predetermined time points. Rats were euthanized at the end of the experiment in accordance with the 2013 American Veterinary Medical Association Panel on Euthanasia Guidelines to allow for specimen collection.

Setup of a sheltering tube for orbital tightening assessment: To enable consistent and unobtrusive observation of orofacial pain responses, we employed a sheltering tube setup previously adapted for behavioral assessments such as the von Frey, acetone spray, and optogenetic tests. ⁹ In the current study, this method was utilized to evaluate orbital tightening (OT) in rats. Each animal was placed into a cylindrical plastic tube with one end closed, functioning as a shelter during testing. The internal diameter of each tube was approximately 2.5 inches, with an effective depth adjusted to 6 inches to accommodate differences in animal size. This adjustment was achieved by inserting modified lids at the base of the tube. The tube itself was fixed to a stable stand on the laboratory bench and angled upward at roughly 50 degrees to ensure a clear and stable view of the animal’s face. The setup was positioned 30 cm above the bench surface. To reduce stress and encourage natural behavior during testing, animals underwent a multi-day habituation protocol. This included approximately 30 min of acclimation in their home cages placed on the lab bench, followed by 45 min of exposure to the sheltering tubes per day, over two consecutive days. On the day of behavioral recording, animals were again given 30 min to acclimate in their home cages before being placed into the sheltering tubes. Once the animals were comfortably positioned with their heads naturally facing the tube opening, video recording began. Facial behavior was video recorded using a high-resolution webcam (1920 × 1080 pixels) at 30 frames per second. The camera was positioned 10 cm from the tube opening, aligned with the animal’s head height, and angled to the right to optimize the view of one side of the face. Each session typically lasted 5 min, yielding a total of 9000 frames per session. After each use, the sheltering tubes were thoroughly cleaned with either laboratory-grade detergent or 70% ethanol, then rinsed with water to prepare them for the next set of experiments.

Deep machine learning and eyelid labeling: To analyze facial features associated with pain, video recordings of Sprague-Dawley rats positioned in sheltering tubes were captured at 1920 × 1080 pixels and 30 frames per second, with each session lasting approximately 5 min per animal. The software DeepLabCut v2.3.0 (https://github.com/DeepLabCut/DeepLabCut), a Python-based open-source tool for markerless pose estimation, was used to process and analyze the video data.^16,17

To begin model training, 20 representative video frames were extracted from each of 5 rats, focusing on clear views of the right eye region. On each selected frame, key anatomical landmarks – including the superior and inferior eyelid margins as well as the medial and lateral canthi – were manually annotated. These labeled points served as training data for the neural network.

DeepLabCut used this labeled dataset to train a convolutional neural network through 100,000 iterations, producing a customized model capable of automatically detecting and tracking the defined eyelid features across video frames. Once trained, the model was applied to all experimental video data to track the labeled regions across time. The output consisted of x-y coordinate positions of each labeled feature, exported in CSV format. These data were then processed in Excel to compute two primary quantitative metrics of interest: eyelid distance (vertical distance between the upper and lower eyelid margins) and palpebral fissure width (horizontal distance between the medial and lateral canthi) for the right eye.

Data Processing and Measurement of Eyelid Metrics: The coordinate data generated by DeepLabCut were exported as CSV files, which were subsequently opened and processed in Microsoft Excel. Each row in the dataset corresponded to a video frame and included the following information: the frame number, names of the labeled anatomical points (e.g. upper eyelid margin), the x and y coordinates of each point, and the associated confidence score or detection likelihood.

Using these coordinates, we calculated two key measurements for each frame:

These measurements were computed directly within Excel using standard distance formulas applied to the coordinate values. A new spreadsheet was created to organize the processed data, including columns for frame number, eyelid distance, palpebral fissure width, and detection likelihood values for each of the four annotated landmarks. To ensure data reliability, we applied an inclusion criterion: only frames in which all four key points had a detection likelihood of 0.95 or higher were retained. Frames that mistakenly labeled the left eye instead of the right were also excluded from the final analysis. After filtering the data, pixel-based distances were converted to millimeter units using a previously determined scale factor, enabling consistent quantitative comparisons across animals.

Statistical Analysis: After completing the calculations for eyelid distance and palpebral fissure width in Excel, the cleaned and filtered dataset was imported into GraphPad Prism version 10.1.2 for statistical evaluation. Using this software, histograms were generated for both eyelid distance and palpebral fissure width, with a bin size of 0.1 mm to visualize the distributions. The mean values for each metric were computed, and comparisons between experimental groups were performed using either a one-way ANOVA or a two-tailed Student’s t-test, depending on the experimental design and number of groups involved. Results are reported as mean ± SEM (standard error of the mean). Statistical significance was denoted as follows: ns = not significant; *p < 0.05; **p < 0.01; ***p < 0.001. These analyses allowed for quantitative comparisons of eyelid-based facial metrics across experimental conditions.

Results

We employed the sheltering tube method to focus on the facial regions of rats for behavioral assessment of pain and discomfort after chlorine exposure.^9,10 The sheltering tube allowed orientation of the rats, so their heads were positioned outward for better visualization and video recording of the eye (Figure 1). The animals were previously acclimatized to this sheltering tube to avoid additional stress and to create a stable condition. As reported before for mice, we recorded 5-min videos at 30 frames per second.^9,10 Utilizing machine learning with DeepLabCut, we labeled the inferior and superior margins of the eyelid. The medial and lateral canthus were also identified and measured as shown in Figure 1. We focused on the right eye, and the images of the left eye and the partially acquired images were discarded. We chose the right eye for consistency across all recordings and to minimize variability. Selecting one side ensures uniformity in measurement and analysis, as eyelid movement can differ slightly between eyes. Additionally, the camera setup was optimized for the right eye during behavioral recordings, which provided the most stable and unobstructed view. Images with incomplete labeling due to incorrect head positions or fuzzy images resulting from rapid head movements were also excluded, as indicated in our studies with mice OT. Using these streamlined exclusion methods, we established clear data inclusion criteria and measured eyelid distance and palpebral fissure width using DeepLabCut. The included images provided precise labeling of the superior and inferior eyelid margins and the medial and lateral canthi of the eyes.^9,10 These criteria were used for all the experiments.

OPEN IN VIEWER

Figure 1.

Schematic illustration of the method of rat orbital tightening evaluation using machine learning with DeepLabCut.

To assess whether orbital tightening (OT) occurs upon inhalation of a noxious chemical such as chlorine, we captured images of rats exposed to chlorine within 30 min of exposure (Figure 2). The animals did not show OT before chlorine exposure. Measurement of OT was started 30 min after chlorine (600 ppm) exposure for 45 min. Figure 2(c) shows the frequency distribution of eyelid distance before and 30 min after chlorine exposure, displaying a leftward shift after the stimulation. The average eyelid distance was significantly reduced after chlorine exposure (Figure 2(d)). Similarly, palpebral fissure width was significantly reduced (Figure 2(e) and (f)). This significant reduction in eyelid distance and palpebral fissure width indicates significant orbital tightening in response to pain induced by chlorine inhalation via activation of the sensory neuron nociceptors (Figure 2(d)).

OPEN IN VIEWER

Figure 2.

Quantification of orbital tightening induced by chlorine exposure in the orofacial region of rats. Rats were acclimated to the shelter tube and images of superior and inferior eyelid margins, and the positions of medial and lateral canthus were taken at baseline (a) in unexposed (naïve) rats. Similarly, videos of the eye positions of chlorine-exposed rats were recorded 30 min after exposure (b). Frequency distribution of eyelid distance for images in 5 min for a rat at baseline (blue) and a rat 30 min after chlorine exposure is shown (c). The bin width for the histogram is 0.1 mm. Average eyelid distance at baseline and after chlorine exposure was plotted as a bar graph (n. = 5; d). Similarly, histograms of palpebral fissure width (EF) were plotted. Data represent individual observations and mean ± SEM. *p < 0.05, **p < 0.01.

Pain behavioral responses to noxious chemicals can be short-lived or long-lived. We determined the duration of orbital tightening after chlorine exposure by capturing video images of a different set of rats before chlorine exposure and imaged them at 2, 6, and 24 h after chlorine exposure, as shown in Figure 3. As shown in Figure 2, the eyelid distance remained significantly decreased 2 h after chlorine exposure, displaying a leftward shift after the stimulation. The leftward shift in the scatter plot and the quantitative eyelid distance remained reduced at 6 and 24 h after exposure, indicating that the orbital tightening, which revealed pain and discomfort, persisted for prolonged periods after inhalation of chlorine gas (Figure 3(a) and (b)). Chlorine exposure caused a leftward shift in the scatter plot of palpebral fissure width measurements up to 2 and 6 h after exposure (Figure 3(c)). Interestingly, the palpebral fissure width became nonsignificant and close to normal at 24 h after exposure (Figure 3(c) and (d)).

OPEN IN VIEWER

Figure 3.

Prolonged orbital tightening (OT) induced by chlorine exposure in the orofacial region of rats. Rats were imaged at baseline and then exposed to chlorine and subjected to imaging for OT measurements at different time intervals after exposure. Videos were made of the different eye positions of chlorine-exposed rats at 2, 6, and 24 h after chlorine exposure. Frequency distribution of eyelid distance (A) and palpebral fissure width (C) for images in 2 min for a rat at baseline and at different time intervals is shown. Average eyelid distance and palpebral fissure width at baseline and after chlorine exposure were plotted as bar graph (n = 5; B and D). Data represent individual observations (A&C) and mean ± SEM (B&D). *p < 0.05, **p < 0.01, ****p < 0.001.

Buprenorphine is recommended ¹⁸ as an effective opioid analgesic for rodents. However, it is unknown if chlorine-induced distress or pain can be effectively reduced by its administration. To evaluate whether we can estimate the reduction in chlorine-induced orbital tightening using the DeepLabCut method, we administered buprenorphine before evaluating its effects (Figure 4a). As shown in Figure 4(c), the frequency distribution of eyelid distance after chlorine exposure in vehicle-treated rats displayed a leftward shift after the stimulation, and the quantitative eyelid distance remained reduced after exposure. Similarly, chlorine exposure caused a leftward shift in the scatter plot of palpebral fissure width measurements in the vehicle-treated animals. Buprenorphine treated animals demonstrated a significant increase in the quantitative eyelid distance and the palpebral fissure width in chlorine-exposed animals at 2 h post-exposure (Figure 4(b–d)). The rats were also monitored for their heart rate, breath rate, and tissue oxygenation. Chlorine exposure at these concentrations and for this duration significantly reduces heart rate, breathing rate, and pulse oxygenation (Figure 4(e)). Buprenorphine slightly decreased the already reduced heart rate, but this change was not significant. Similarly, the breath rate was minimally reduced beyond what chlorine exposure itself caused. Buprenorphine treatment also did not further decrease or improve tissue oxygenation at the given dose after chlorine exposure (Figure 4(e)).

OPEN IN VIEWER

Figure 4.

Buprenorphine reverses orbital tightening (OT) induced by chlorine exposure. Rats were imaged at baseline and then exposed to chlorine. Vehicle (saline) or buprenorphine (0.03 mg/kg, IP) was administered 30 min after exposure (a), and rats were subjected to imaging for OT measurements 2 h after exposure (b). Frequency distribution of eyelid distance (c) and palpebral fissure width (d) for images in 5 min for a rat at baseline and 2 h after exposure with or without buprenorphine treatment is shown. Average eyelid distance and palpebral fissure width at baseline and after chlorine exposure were plotted as bar graph (n = 5; b and d). Data represent individual observations and mean ± SEM, *p < 0.05, **p < 0.01, ***p < 0.001.

The pain and discomfort caused by chlorine exposure could also be attributed to the effect on the eyes. We tried to assess the extent to which OT is caused by ocular damage. The skin around the eye was carefully sealed off by a thin layer of petroleum jelly before chlorine exposure (Figure 5(a)). This eye protection was wiped off before recording the OT measurement video. Eye protection had a limited effect on chlorine-induced OT (Figure 5(b) and (c)), suggesting that the origin of OT/discomfort was mainly in the lung periphery/chest wall and systemic following chlorine inhalation.

OPEN IN VIEWER

Figure 5.

Contribution of eye injury to orbital tightening (OT) induced by chlorine exposure. Rats were imaged at baseline and then exposed to chlorine. Petroleum jelly was applied around the eye with a cotton swab in a group of rats before chlorine exposure (a). The petroleum jelly was gently removed, and the rats were imaged for OT measurements 2 h after exposure (b). Frequency distribution of eyelid distance (c) and palpebral fissure width (d) for images in 5 min for a rat at baseline and 2 h after exposure with or without petroleum jelly application to protect the eye is shown. Average eyelid distance and palpebral fissure width at baseline and after chlorine exposure were plotted as a bar graph (n = 4; b and d). Data represent individual observations and mean ± SEM, *p < 0.05, **p < 0.01.

Discussion

Chlorine is a highly reactive toxic gas produced in large quantities in industry and poses a significant risk of accidental or deliberate spill. Exposed victims demonstrate coughing, dyspnea, sore throat, and chest pain.^19–22 Humans exposed to chlorine also report severe pain in their exposed tissues, airways, and stomach.^23–26 In animal models of chlorine exposure, reduced locomotion and physical inactivity, possibly due to pain and discomfort, were reported previously. ⁸ In this study, we employed a machine learning-based approach using DeepLabCut to quantify eyelid distance and palpebral fissure width in naïve and chlorine-exposed rats. Our findings demonstrate that rats exposed to chlorine gas exhibited measurable, time-dependent reductions in both eyelid distance and palpebral fissure width, indicating pain-related changes in facial expression. These results support the utility of automated facial feature tracking as a sensitive method for detecting pain and physical discomfort in chemically exposed rodent models.

Inhalation of chlorine and other toxic gases induces sensations of burning and pain by activating chemoreceptors in somatosensory afferent nerves that innervate the respiratory system. ²⁶ Specialized somatosensory afferent nerves, known as nociceptors, detect painful stimuli and release chemicals. This chemical nociception mainly involves the activation of Aδ- and C-afferent fibers to produce sharp pain or slow-burning pain. ²⁷ Noxious gases inhaled into the respiratory system also act on vagal afferent nerve fibers, resulting in autonomic responses that cause discomfort in animals.^3,4 Chlorine and other reactive chemicals/oxidative agents are known to activate cation channels on these nerve fibers. The transient receptor potential (TRP) family of cation channels serve as the receptors for such chemicals and excite the sensory nerve fibers to elicit pain and inflammation. ²⁸ Halogens, which are potent oxidants, are known to activate the transient receptor potential ankyrin 1(TRPA1) channels that activate the DRG/TG to cause pain and pneumonitis.^29,30 The peripheral ends of the nerve fibers secrete neuropeptides like Substance P and calcitonin gene-related peptide (CGRP) that are proinflammatory and initiate the neurogenic inflammatory response, including immune cell infiltration, mast cell degranulation, and other local inflammatory responses. ²⁶ With low concentration chlorine (1 ppm) inhalation, involvement of neuropeptides was, however, not demonstrated. ³¹

Acute chlorine gas exposure primarily affects the respiratory system, especially in rodent models, where the skin is protected by thick fur. Although richly innervated, the lungs themselves have very few pain receptors, and the chest pain accompanying chlorine exposure could arise from surrounding structures, such as the chest wall, which are supplied by the main afferents of the Vagal nerve and somatosensory neurons. Lung innervating neurons originate primarily in the Vagal ganglion with very little contribution from the dorsal root ganglion. ³² The noxious sensation or the irritant response from chlorine can be sensed by the alveolar irritant receptors and the pulmonary stretch receptors served by the C fibers. However, noxious stimulation could affect other organs after chlorine exposure, as chlorine-derived reactive species formed in the pulmonary bed may reach these distant organs through the bloodstream. ¹

Facial grimace scales are used in animals to measure pain, including in rats, where the rat grimace scale (RGS) allows quantification of pain behavior. ³³ Similarly, the mouse grimace scale (MGS) is used. However, it’s tedious and complex, and its automation is challenging. ³⁴ Machine learning approaches provide enhanced accuracy and serve as a better alternative to pain quantification techniques such as the RGS or MGS. Orbital tightening is an important parameter of the grimace scale of pain^10,35–37 in rodent models. We have previously reported the sensitivity of eyelid distance measurement for capsaicin-induced orbital tightening in mice. ¹⁰ However, nothing has been documented regarding noxious chemical exposure and orbital tightening. Chlorine exposure may physically damage the eyes and cause irritation that exacerbates the painful stimuli.^19,38 So, the early pain experienced could be attributed to eye irritation itself. With the prolonged measurements we performed, the eyelid distance remained significantly reduced even at 24 h post-exposure. However, the degree of reduction in palpebral fissure width was less than the eyelid distance at this time, indicating that eyelid distance is a more sensitive indicator of orbital tightening after chlorine exposure. Orbital tightening following chlorine exposure remains significant in animals whose eyes were protected from chlorine, suggesting that chlorine inhalation in the respiratory system significantly contributes to behavioral pain responses. A previous study showed that buprenorphine administration improved locomotion in chlorine-exposed mice, ⁸ and our study provides the first report of such measured quantitative pain relief with buprenorphine after chlorine exposure in rats. Opioids may reduce orbital tightening by engaging μ-opioid receptors in key brainstem regions such as the periaqueductal gray and rostral ventromedial medulla. This activation strengthens descending inhibitory pathways that dampen nociceptive signaling within the trigeminal nucleus caudalis and spinal dorsal horn. At the same time, opioids suppress activity in cortical and subcortical pain-processing areas, including the anterior cingulate cortex and insula, which helps diminish both the sensory and emotional aspects of pain. Additionally, opioid-driven increases in acetylcholine release within trigeminal motor circuits, along with GABAergic disinhibition in brainstem pain-control centers, may further influence facial motor output. Together, these mechanisms likely contribute to the reduction of pain-related facial expressions such as orbital tightening.^39–43

In conclusion, the quantitative orbital tightening for pain assessment with the use of machine learning with DeepLabCut provides a useful tool for investigating immediate and ongoing pain induced by toxic chemical exposures. Future studies can determine whether this approach allows assessment of pain caused by inhalation of other reactive chemicals. It would also be interesting to quantify the severity of pain induced by different noxious chemicals and determine the effects of other analgesics on pain induced by inhalation of these chemicals.