We report here the development, reliability, accuracy, analgesic sensitivity, and utility of the RGS, a method to quantify spontaneous pain in the laboratory rat. In addition, we have developed an automated system--the RFF software package--that can successfully extract scorable image files from digital video, previously the most labor-intensive step in the application of the RGS (or MGS). The RFF can be obtained directly from one of the authors (ODK at firstname.lastname@example.org) upon request by interested individuals, at no charge for academic users and via licensing agreement for corporate users.
RGS vs. MGS
As predicted by the evolutionary conservation of facial expressions of emotions , including pain , the "pain face" of the rat was found to be broadly similar to that of the mouse, with three of the RGS action units essentially unchanged from the MGS. A major exception is the nose and cheek, whereby pain in the mouse results in bulging, but in the rat bulging occurs naturally and this characteristic actually diminishes when the rat is in pain.
Observed accuracy rates for the RGS are lower than those observed in the MGS using similar high-definition video (97% by S.G.S.) , but still far above chance levels. We note that when testing the accuracy of the MGS, another well-validated pain-related behavior (abdominal constrictions) was used to verify the existence of spontaneous pain in each subject, whereas here, using CFA, we were forced to simply assume its existence. This fact likely accounts for the lower accuracy values obtained, since rats in some of the pain photographs may not have been, at that precise moment, actually in pain, which would artificially inflate the miss rate.
Inter-rater reliability of the RGS was very high, as high as on the MGS. We note, however, that this was only tested in five individuals in one laboratory. We encourage others to use the method so that true reliability and accuracy rates can be ascertained.
Time course of inflammatory pain
Peak RGS scores were observed at 6 h post-CFA, 3 h post-kaolin/carrageenan, and 4 h post-laparotomy. It is tempting to conclude that this represents the peak of spontaneous pain in these assays, as opposed to allodynia. There are, of course, very few extant studies where spontaneously emitted behaviors have been recorded in these assays, and even then it's not clear that what is being measured is spontaneous pain (as opposed to mechanical allodynia), or even pain at all [1, 4, 5]. In an early study using intraplantar 100% CFA in the rat , a number of behavioral characteristics including food intake, open field behavior, and core body temperature were altered by CFA, some for over 5 weeks. In contrast, Djouhri and colleagues , using spontaneous foot lifting as a measure of spontaneous pain in Wistar rats, noted that all rats displayed foot lifting 1 day after injection, but less than 20% did by day 2 and none did at 4-7 days post-injection. Using a suite of behaviors (including muscle twitching, back arching, staggering, and abdominal writhing), Roughan and Flecknell  concluded that postoperative pain after laparotomy decreased significantly after 3-5 h post-surgery. Using exploratory activity and conditioned operant responding for sucrose pellets as the measure, in contrast, Martin et al.  observed changes after surgery lasting up to 2-3 days.
The time course of mechanical allodynia and thermal hyperalgesia in these models is better known, albeit dose- and strain-dependent. The first study to use 50% CFA in the rat observed peak thermal hyperalgesia at 4 h post-injection and a return to baseline by 15 days; mechanical allodynia peaked at 2 days post-injection and was resolved by 5 days . The duration of changes in the other two models is much more limited. Thermal hyperalgesia in the kaolin-carrageenan model was found to peak at 8-12 h and resolve by 2 days post-injection . Electrophysiological experiments have shown that primary afferent fibers in the joint are sensitized in the kaolin-carrageenan model 3-6 h post-injection . After laparotomy in the Wistar rat, mechanical allodynia was noted from 2.5-6.5 h post-surgery , although in a recent study (involving in addition to the incision the implantation of a radiotelemetry transmitter) significant allodynia was observed for 9 days .
Overall there is good concordance between the time course of inflammatory pain inferred from the literature and our current data. It is important to note, however, that the disappearance of the facial grimacing may not necessarily represent the disappearance of spontaneous pain, as there are adaptive advantages to inhibiting a "pain face" as soon as possible.
New approaches to algesiometry
The problematic symptoms of chronic pain in humans include spontaneous pain, numbness, dysesthesias, and evoked (mechanical, heat and cold) hypersensitivity. But these are not equally common, or of equal concern. For both neuropathic and non-neuropathic pain, spontaneous or ongoing pain (especially deep pain) is far more prevalent than evoked pain, especially touch- and warmth-evoked pain [24, 25]. Spontaneous pain is also rated as more bothersome, and more highly correlated with global ratings of pain severity . Despite this clinical reality, preclinical studies of pain are strongly weighted towards the study of mechanical and thermal hypersensitivity states, largely for reasons of practicality and inertia .
However, new approaches to measuring pain (and/or the impact of pain) appear to be gaining popularity; these include thermal preference/escape models [e.g., [26, 27]], conditioned place aversion [28, 29], conditioned place preference (to pain inhibition) [30, 31], and ultrasonic vocalization . Compared to these, facial expression coding has the considerable advantage that no subject training or special equipment (other than a video camera) are required. It also provides the advantage of more complete blinding of the experimenter , since during scoring the presence or absence of an inflamed or guarded hind paw is completely obscured. Quantifying pain by facial expression is also the only technique of practical value in veterinary medicine (including laboratory animal welfare), as it can in fact be performed in real time by trained investigators, animal technicians and/or veterinarians.
The major disadvantages to blinded facial expression coding for research purposes are the labor-intensive nature of frame grabbing, a problem now largely solved with RFF software, and the limited duration (< 48 h) of the pain face. This limitation is imposed by the nature of facial grimacing itself, which is also not observed in human chronic pain patients. Thus, the study of real-time spontaneous pain in chronic neuropathic assays awaits the development of a useful dependent measure.