Differential Decomposition Among Pig, Rabbit, and Human Remains

J Forensic Sci, 2018 doi: 10.1111/1556-4029.13784 PAPER Available online at: onlinelibrary.wiley.com ANTHROPOLOGY Angela Dautartas,1 M.A.; Michael W. Kenyhercz,2,3 Ph.D.; Giovanna M. Vidoli,1 Ph.D.; Lee Meadows Jantz,1 Ph.D.; Amy Mundorff,1 Ph.D.; and Dawnie Wolfe Steadman,1 Ph.D. Differential Decomposition Among Pig, Rabbit, and Human Remains*,† ABSTRACT: While nonhuman animal remains are often utilized in forensic research to develop methods to estimate the postmortem inter- val, systematic studies that directly validate animals as proxies for human decomposition are lacking. The current project compared decomposi- tion rates among pigs, rabbits, and humans at the University of Tennessee’s Anthropology Research Facility across three seasonal trials that spanned nearly 2 years. The Total Body Score (TBS) method was applied to quantify decomposition changes and calculate the postmortem interval (PMI) in accumulated degree days (ADD). Decomposition trajectories were analyzed by comparing the estimated and actual ADD for each seasonal trial and by fuzzy cluster analysis. The cluster analysis demonstrated that the rabbits formed one group while pigs and humans, although more similar to each other than either to rabbits, still showed important differences in decomposition patterns. The decomposition trends show that neither nonhuman model captured the pattern, rate, and variability of human decomposition. KEYWORDS: forensic science, forensic anthropology, taphonomy, postmortem interval, decomposition, animal models Much of the published literature on postmortem interval esti- substitutes for humans in decomposition research designed to mation has been based upon nonhuman animal subjects as prox- help estimate the postmortem interval in forensic casework. ies for humans, including dogs (1), cats (2), rats (3–5), elephants Despite the vast number of animal decomposition studies, (6), monkeys (7), guinea pigs (8), rabbits (9–17), and pigs (18– there have been warnings that human and nonhuman species 30). Much of this research has focused on basic ecological stud- vary in significant ways that can affect decomposition rates. For ies designed to understand insect succession and development instance, Vass et al. (33) detected significant biochemical differ- rates or insect behavior in specific ecological niches (e.g., ences between the bones of decomposed human and other mam- 3,9,11,13). For such studies, the type of carrion used was rela- malian species. Within a laboratory setting, Notter et al. (34) tively inconsequential and depended upon what animals were found differences in adipocere formation and composition available locally. A potential problem may arise, however, when between pig and human tissues that they argue is due to the dif- data derived from animal studies are applied to estimate the ferential quantity and distribution of fat between species. The postmortem interval of human remains in forensic contexts (31). only field comparison of species decomposition was conducted Some argue that pigs are appropriate proxies for human subjects by Schoenly et al. (30), who placed two pigs and one human at based on size, skin thickness, hair coverage, and arthropod suc- the Anthropology Research Facility (ARF) in Knoxville, Ten- cession patterns (30), while rabbits/hares have been increasingly nessee for 35 days. They found limited differences in arthropod used for decomposition research outside the United States (14– preference between species and they “confirmed the claim that 17,32). The purpose of this research project was to evaluate pig carcasses (of 23–27-kg starting mass) can substitute for whether common nonhuman animal models are appropriate human corpses in research and training programs, at least for summer-exposed and unconcealed remains in the first 5 weeks postmortem” (30:881). However, their study had small sample sizes, was not replicated in other seasons, and did not document 1 Department of Anthropology, University of Tennessee, 502 Strong Hall, what factors other than insects contributed to decomposition. 1621 Cumberland Ave., Knoxville, TN 37996. This study is based on a project initiated in 2014 that directly 2 Central Identification Laboratory, Defense POW/MIA Accounting Agency, 570 Moffet Street, JBPHH, HI 96853. compared 45 pig, rabbit, and human subjects across three sea- 3 Department of Anatomy, University of Pretoria, Private Bag x323, sonal trials to assess if overall decomposition patterns and rates Arcadia 0007, South Africa. differ among carrion species and quantify any differences in *Presented at the 67th Annual Scientific Meeting of the American Acad- scavenger and insect abundance, diversity and behavior. The cur- emy of Forensic Sciences, February 16–21, 2015, in Orlando, FL, and at the rent research directly informs the scientific community about the 68th Annual Scientific Meeting of the American Academy of Forensic Sciences, February 22–27, 2016, in Las Vegas, NV. applicability of nonhuman animal models in forensic research † Supported by the National Institute of Justice, Office of Investigative and and their probative value in informing casework. The null Forensic Sciences, United States Department of Justice, grant # 2013-DN- hypothesis is that if external environmental factors are held con- BX-K0. stant, nonhuman animals will exhibit the same forensically Received 5 Nov. 2017; and in revised form 5 Jan. 2018, 5 Mar. 2018; important characteristics as humans, such as arthropod diversity accepted 5 Mar. 2018. © 2018 American Academy of Forensic Sciences 1 2 JOURNAL OF FORENSIC SCIENCES and succession, scavenging behavior (with respect to scavenger The Total Body Score (TBS) method evaluates morphological type and frequency) and morphological changes. This study changes and tissue loss on each body segment independently details the overall morphological differences in seasonal decom- (37). The body is divided into three regions (head and neck, position patterns among species while subsequent papers will trunk, limbs), each of which is scored according to a progressive assess whether scavenger activity differs among subject species chart of decomposition characteristics. The scores for each (35) as well as differences in arthropod activity. region are then summed to calculate the Total Body Score. The TBS is then used to calculate the accumulated degree days, or ADD, which is a measure of heat energy that drives biological Methods processes of the insects and bacteria involved in decomposition, The field site for this project was the ARF at the University of using the equation, ADD = 10(0.002*TBS*TBS+1.81) 388.16. Total Tennessee, Knoxville; the first outdoor natural laboratory to study Body Scores and photographs were recorded twice daily (AM human decomposition. Located on approximately three acres of and PM) for each subject. Longitudinal weather data recorded bluff above the Tennessee River, the area is naturally forested by on-site temperature and humidity data loggers provided the with oak, maple, and hickory trees. The study area was in a average daily temperatures to calculate the actual ADD. Follow- newly fenced section of the ARF containing virgin soil where no ing Megyesi et al. (37), only temperatures above 0°C were previous studies of surface decomposition had occurred. Five included in ADD calculations as previous research has shown human (Homo sapiens), five domestic pig (Sus scrofa), and five that microbial and insect activity can be substantially slowed domestic rabbit (Oryctolagus cuniculus) subjects were placed at below this temperature (38). Observations continued daily until the ARF in each of three seasonal trials (spring, summer, and 84 ADD were reached. A total of 84 ADD was selected as the winter) for a total of 45 individuals (Table 1). Human subjects benchmark for data collection as this has been suggested to were enrolled in the study if their weights were between 100 and cover the total active decomposition period (38). A total of 250 pounds and the body exhibited no signs of external trauma. 47,231 daily photographs and 486 TBS observation events were Pigs and rabbits were supplied by local farms and euthanized by analyzed for this component of the study. a veterinarian. All subjects were placed in a 38°F (3°C) morgue Loess smoothed line plots were prepared of averaged TBS for cooler for at least 24 h before placement to equalize body tem- each species over ADD in R (39) using the ggplot2 package perature. The protocols for animal handling and euthanasia were (39,40). These plots show the relationship of TBS with actual approved by the University of Tennessee Institutional Animal ADD, but not the resulting estimated ADD scores. To examine Care and Use Committee (IACUC). Additionally, the University the relationship of actual and estimated ADD scores, decomposi- of Tennessee Institutional Review Board (IRB) reviewed the tion trajectories were analyzed graphically with a novel approach human subjects protocols. in which the difference between the actual ADD (known time For each trial, 15 subjects were placed in random order along since placement) and the average of the estimated ADD (calcu- transects in the same microenvironment within the ARF follow- lated from the TBS), that is, the residuals of the actual and esti- ing a randomized block design conducted in SAS version 9.2 mated ADD, were plotted for each species using loess (36). The subjects were placed a minimum of three meters apart. smoothing with the ggplot2 package in the statistical program R. The first trial took place near the bottom of the slope and each These residual ADD plots show the over- or underestimation of successive trial moved a minimum of five meters up slope from ADD for each species through time. Theoretically, if the rela- the previous transect. Rabbits were placed in wire cages to pro- tionship between estimated and actual ADD were perfect, there tect them from being removed from the study area by scavengers would be straight lines at 0 on the y-axis. Individual trial plots as one of the specific objectives was to examine insect activity and plots of all trials pooled were prepared to examine inter-trial, variation by subject species. In Trials 2 and 3, the cage tops or seasonal, variation and more general, overarching trends. were unlocked to allow scavenger access yet limit the ability for The impact of season and species on estimated ADD was scavengers to remove the subjects. Pig and human remains were examined with a multi-way ANOVA, one in which season and not caged during any of the trials. species were examined independently, and again as an interac- tion of season and species. The multi-way ANOVA with the interaction of season and species was then subjected to an effects TABLE 1––Weight in pounds (and kilograms) and sex data for human sub- jects and weight data for pig and rabbit subjects in all three trials. plot to visualize the relationship of species and season on ADD estimates. The multi-way ANOVA and effects plot were pre- Trial 1 Trial 2 Trial 3 pared in R (40). Sex/Weight Sex/Weight Sex/Weight NCSS version 11 (41) was used to analyze TBS segment Species lbs (kg) lbs (kg) lbs (kg) scores (head and neck, trunk, and limbs) from all three species Human 1 Male/179 (81) Female/123 (56) Female/126 (57) within each of the three trial periods with fuzzy clustering. Human 2 Female/159 (72) Male/147 (67) Female/173 (78) Fuzzy clustering is an analytical method that uses an algorithm Human 3 Male/186 (84) Female/116 (53) Male/188 (85) to compare sets of independent variables across multiple individ- Human 4 Male/167 (76) Male/191 (87) Female/186 (84) uals to determine the group structure of those individuals. Fuzzy Human 5 Male/167 (76) Female/235 (107) Male/161 (73) Pig 1 Male/140 (64) Female/113 (51) Female/103 (47) clustering was selected because it does not require the number Pig 2 Female/133 (60) Female/104 (47) Female/106 (48) of groups to be known before analyzing the data, and it also Pig 3 Female/140 (64) Male/106 (48) Male/126 (57) allows for an observation or individual to belong to more than Pig 4 Male/148 (67) Male/130 (59) Male/117 (53) one group (42). To use this method, a “fuzzifier” or degree of Pig 5 Male/150 (68) Female/88 (40) Male/106 (48) Rabbit 1 Male/8 (4) Male/9 (4) Female/7 (3) overlap must be selected. A fuzzifier can range from 1.01 to 10, Rabbit 2 Male/8 (4) Male/8 (4) Male/8 (4) and the lower the fuzzifier is set, the more discrete the groups. Rabbit 3 Female/6 (3) Female/7 (3) Female/5 (2) Fuzzifiers tested for these models ranged from 2.5 to 1.01, Rabbit 4 Female/7 (3) Female/8 (4) Female/7 (3) which are commonly chosen values (43). These fuzzifier values Rabbit 5 Male/8 (4) Female/8 (4) Male/9 (4) also allow for significant amounts of overlap while not being so  DAUTARTAS ET AL. . DIFFERENTIAL DECOMPOSITION 3 vague as to completely obscure data structure. Minimum and TABLE 2––Seasonal information, temperatures and study duration of the maximum number of groups must also be selected. Here, the three trials. minimum number of groups was 1, to account for the possibility Max Avg that all three species could cluster together. The maximum num- Placement Max High Temp Study ber of clusters was 5. This maximum allows for outliers or Trial Season Date Low (°C) (°C) (°C) Duration unique cases to be identified by clustering into their own groups 1 Spring March 14, 2014 2.6 24.2 16.2 75 days and yet does not have so many possible clusters as to create an 2 Summer June 13, 2014 19.5 28.7 24.2 45 days overly complex model. The optimal model is chosen based on 3 Winter December 1, 2014 0–12.7 17.8 5.1 123 days three model fit statistics; the average silhouette and Dunn’s coef- ficient, which should both be high, and a partition coefficient, the Dc(U), which is a low value in a strong model with a good temperatures too cold for insect activity yet warmed by the end fit to the data (42). The average silhouette is a measurement of of the trial 75 days later. Trial 2 began and ended quickly during the tightness of the cluster; if it is high, then it reflects that the peak insect activity, taking only 16 days to reach 84 ADD, observations assigned to that cluster are grouped close to the although data continued to be collected for 45 days. Trial 3 mean. This indicates that those observations strongly grouped began in the winter and took over 120 days to achieve 84 ADD. together as opposed to a low silhouette, which would suggest that the observations are on the border of belonging to other Decomposition Trends Among Species groups (41). Dunn’s coefficient is a measurement of the overlap or separation of the identified clusters. A high Dunn’s coefficient The residual ADD values plotted against PMI (in days) are will indicate that the clusters are discrete; a low Dunn’s coeffi- shown in Fig. 1 for each trial separately, as well as combined. cient shows that there is overlap between the clusters. The Dc Figure 1A shows the decomposition trends from the spring trial (U) is another partition coefficient similar to the Dunn’s partition (Trial 1) in which temperatures were initially very cold in March coefficient, which aids in determining the optimum number of and then warmed to a maximum temperature of 24°C later in clusters. With this second partition coefficient, a high value indi- the season. Decomposition began in the rabbits earlier and pro- cates significant overlap (or “fuzziness”) between groups, while ceeded at a faster rate than that of either the pigs or humans. a low value means that a strong solution with discrete groups The pigs and humans tracked together well until insect activity has been identified (42). began in earnest and the pigs began to skeletonize faster than Data were separated by trial to eliminate the effect of season- the humans. Due to the initially cool temperatures and lack of ality on model development. Independent variables were seg- insect activity, the human bodies mummified such that the flies ment scores at ADD 100, 300, 400, and 500 for Trials 1 and 3, were largely uninterested in nonfresh tissues for oviposition and segment scores at ADD 100, 300, and 400 for Trial 2. Trial when temperatures warmed. Thus, large maggot masses did not 2 TBS plateaued more quickly than in Trials 1 and 3, and there- form or spread across the human bodies. The pigs did not mum- fore did not have a measurement at ADD 500. Segment scores mify and exhibited body-encompassing maggot masses, thereby were used instead of TBS to get a finer scale analysis that skeletonizing faster. Figure 1A demonstrates that TBS overesti- reflected decomposition pattern by location as well as total mates ADD early in human decomposition (as shown by the amount of decomposition. Analyses were run twice for each positive residual value) and, as time passes, TBS progressively trial; once with all three species, and a second time comparing underestimates ADD. Conversely, ADD for pigs are underesti- humans and pigs only. mated early in decomposition and, through time, progressively Scatter plots for three species comparisons were constructed are overestimated as decomposition accelerates due to insect for each trial. The plots compare the TBS of each individual (on activity. The residual ADD for the rabbits oscillates about the 0 the y-axis) to the actual ADD (shown on the x-axis). The time line, showing an initial overestimation followed by a slight period examined was 100–500 ADD. This time period was underestimation and then a progressive overestimation through selected to mirror the time analyzed by the fuzzy clustering. time and, by the end, underestimated, but not nearly as much as These plots show the progression of decomposition for all spe- the humans. cies across time and aid in visualizing the different patterns Trial 2 began in the middle of June and reached 84 ADD by observed in each species. June 29. Figure 1B shows that pig decomposition occurred more To validate the solutions found with the fuzzy clustering, a ser- rapidly than that of the rabbits and humans. Insect oviposition ies of discriminant analyses were run. Discriminant analysis is a on the pigs was immediate and maggot masses quickly envel- classification method that identifies observations by assigning oped the pigs, promoting faster skeletonization than the other them to predicted groups based on dependent variable data. species. While the human remains were affected by insects, the Unlike fuzzy clustering, discriminant analysis requires that the progression toward skeletonization was slower than that of the number of groups be specified at the start of the calculations. The pigs and the soft tissue of the humans often mummified. How- number of groups for the discriminant analysis was taken from ever, ADD in humans is accurately predicted, as much of their the solutions found during the fuzzy clustering. If the fuzzy clus- trajectory hovers around 0 (i.e., little difference between actual tering solutions work well, then there should be very low classifi- and estimated ADD), although overestimates ADD toward the cation error when evaluated through the discriminant analysis. end of Trial 2. Pigs are progressively overestimated through time in Trial 2 and rabbits are intermediate between pigs and humans. Trial 3, which took place in the winter, shows yet a different Results trend (Fig. 1C). Trial 3 began on December 1 and no insect The results presented here focus on the general decomposition activity occurred for over 100 days. Human decomposition patterns and rates observed among the three species across three occurred at a much faster rate than that of the nonhuman spe- seasons. Table 2 demonstrates that seasonality was captured in cies. The only variable affecting decomposition during this trial the study. Trial 1 took place in the spring and began in was scavenging and the human subjects were clearly 4 JOURNAL OF FORENSIC SCIENCES FIG. 1––Decomposition trends for pig, rabbit, and human subjects in Trial 1 (A), Trial 2 (B), Trial 3 (C), and all trials combined (D). The y-axis represents the averaged residual ADD for each species, thus any positive residual indicates and overestimation of ADD, on average, and vice versa for negative residuals. preferenced, thereby accelerating decomposition. One human regardless of season. The humans and pigs show the greatest subject in particular was scavenged and skeletonized more similarity in their decomposition progression in Trial 1, although quickly than any of the other subjects (35). In general, human they begin to separate at ADD 200 and the lines do not intersect PMI was progressively overestimated while that of pigs and rab- again until ADD 1100, indicating that the patterns are different bits are underestimated through time in Trial 3. Raccoons were between the two species during the most active stages of decom- the primary scavenger during this trial. Game cameras placed position. Trial 3 had the most distinct differences between spe- along the study area captured clear photographs of the scaveng- cies, as shown in Fig. 2C. The average human TBS rapidly ing activity. A detailed analysis of scavenging behavior is dis- increased around ADD 300, at the same time as the pig and rab- cussed by Steadman et al. (35). bit TBS began to plateau. Figure 2D shows the average TBS Figure 1D represents all of the species across all three seasons over ADD for all of the trials combined and shows humans and (trials) and shows that none of the estimated ADD of any spe- pigs following a similar trend until approximately 250 ADD cies had a strong correlation with known ADD, and each species when the pig TBS values begin to plateau relative to humans. shows different trajectories through time. Humans are initially Past 500 ADD, each species follows its own trajectory with esti- overestimated, tend to be underestimated midway through the tri- mated TBS converging by 1250 ADD. als, and then are overestimated by the end of the trials, while A multi-way ANOVA examines the estimated ADD as a func- pigs and rabbits show a progressive underestimation of ADD tion of season and species separately, as well as the interaction through time. The graph clearly shows that the pattern for pigs of season given species. Figure 3 demonstrates that season and and rabbits is vastly different than that for humans. the interaction of season given species are significant Figure 2A–D shows the raw ADD values plotted against TBS. (p < 0.001), but species alone is not (p = 0.2541). This helps Figure 2A is the data from Trial 1, Fig. 2B is Trial 2, Fig. 2C is explain why each trial showed different trends in decomposition Trial 3, and Fig. 2D is all three trials averaged. The rabbits whereby, for example, ADD for humans was overestimated in clearly have a different trajectory than either the humans or pigs, winter but underestimated in spring and summer.  DAUTARTAS ET AL. . DIFFERENTIAL DECOMPOSITION 5 FIG. 2––Decomposition trends for pig, rabbit, and human subjects in Trial 1 (A), Trial 2 (B), Trial 3 (C), and all trials combined (D). ADD is depicted on the x-axis, and average TBS is on the y-axis. model when all three species were analyzed. In Trial 2, a two- Fuzzy Cluster Analysis cluster solution was also the best model when pigs and humans A two-cluster solution with a fuzzifier of 2.0 was found to be were examined without the rabbits. A five-cluster solution was the strongest model for the three species comparison, with an the best model for analyzing pigs and humans for Trial 3. Model average silhouette of 0.59 and a Dunn’s index of 0.71 (Table 3). fit statistics for the three species comparisons in all trials are The Dc(U) was the lowest of the possible solutions at 0.16. In summarized in Table 7, and probability of belonging to each this model, the pigs and humans formed one cluster, while the cluster is shown in Table 8. Model fit statistics for comparisons rabbits formed the second cluster. Probabilities of group mem- of humans and pigs only are summarized for all three trials in bership are presented in Table 4. The probabilities show that the Table 9, and the probabilities of group membership for the two rabbits strongly cluster into their own group, with an average species comparison are presented in Table 10. probability of belonging of approximately 85%. They have a The results of the discriminant analysis show that solutions very low chance of belonging to the cluster formed by the pigs identified by the fuzzy clustering performed well. For each of and humans, which demonstrates that the pigs and humans are the three species solutions, where two clusters were identified, more similar to each other than either species is to the rabbits. there was a perfect (100% correct) classification rate. The rabbits When the rabbits were removed from the analysis, a five-clus- were consistently classified into their own group, and pigs and ter solution with a fuzzifier of 1.5 produced the best model humans grouped together. The results of the three species analy- (Table 5). The average silhouette was lower, at 0.31, but the ses are presented in Tables 11, 12, and 13. Dunn’s index was still high at 0.60, and the Dc(U) was 0.32. When the pigs and the humans were analyzed separately from This shows that the structure of the data is more complicated. the rabbits using the proposed number of groups determined by Probabilities of group membership are shown in Table 6. the fuzzy clustering, the majority of the observations were classi- Trials 2 and 3 followed similar patterns to that of Trial 1. For fied correctly. Table 14 shows the results for Trial 1, where all both trials, a two-cluster solution was found to be the strongest five humans were correctly classified as human, and four of the 6 JOURNAL OF FORENSIC SCIENCES TABLE 5––Model fit statistics, pigs and humans only, Trial 1. Number of Average Clusters Silhouette Dunn’s Coefficient Dc(U) 1 1.000000 1.0000 Not calculable 2 0.062228 0.5011 0.9380 3 0.137499 0.4418 0.6972 4 0.162959 0.5207 0.4373 5 0.311313 0.6031 0.3234 Best fit model in bold. The lower average silhouette in the best fit model demonstrates that fewer of the observations were close to the group mean. However, there was still clear separation between the clusters. TABLE 6––Probabilities of group (cluster) membership for pigs and humans only, Trial 1. Prob Prob Prob Prob Prob Individual Cluster in 1 in 2 in 3 in 4 in 5 Human 1 3 0.1373 0.1316 0.3790 0.1728 0.1792 Human 2 1 0.9750 0.0048 0.0056 0.0065 0.0081 Human 3 1 0.2116 0.1923 0.2029 0.1884 0.2048 Human 4 2 0.0037 0.9840 0.0040 0.0041 0.0042 Human 5 3 0.0701 0.0651 0.6646 0.0893 0.1109 FIG. 3––Effects plot of the interaction of season and species on estimated Pig 1 3 0.1231 0.1128 0.2594 0.2554 0.2494 ADD. Pig 2 4 0.0027 0.0022 0.0031 0.9876 0.0044 Pig 3 4 0.0027 0.0022 0.0031 0.9876 0.0044 TABLE 3––Model fit statistics for three species comparison (pig, rabbit and Pig 4 5 0.0624 0.0449 0.0701 0.0847 0.7379 human) for Trial 1. Pig 5 5 0.0789 0.0461 0.0767 0.0857 0.7125 Analyzing the pigs and humans without the rabbits resulted in a five-clus- Number of Clusters Average Silhouette Dunn’s Coefficient Dc(U) ter solution, without a high degree of overlap between groups. The probabil- ity that each observation belongs in a specific cluster is presented in the table. 1 1.000000 1.0000 N/A 2 0.595919 0.7130 0.1644 3 0.238387 0.4793 0.5090 4 0.062987 0.3946 0.5678 TABLE 7––Summary of model fit statistics for three species comparisons, all 5 0.129267 0.4094 0.4872 trials. Best fit model is in bold. The average silhouette indicates that most of the Number of Average Dunn’s observations cluster close to the group mean, and the two partition coeffi- clusters Silhouette Coefficient Dc(U) cients demonstrate that there is not a high degree of overlap between the groups. Trial 1 2 0.595919 0.7130 0.1644 Trial 2 2 0.230658 0.5001 0.9757 Trial 3 2 0.532449 0.6850 0.2034 TABLE 4––Probabilities of group (cluster) membership for three species, Trials 1 and 3 produced solutions that had observations clustered close to Trial 1. the group means and had clear distinctions between each cluster. Trial 2 pro- duced a solution that had fewer observations close to the mean and had a Probability of Probability of greater degree of potential overlap between the clusters, indicating more sim- Individual Cluster Membership, Cluster 1 Membership, Cluster 2 ilarity in decomposition patterns during the summer months. Human 1 1 0.5824 0.4176 Human 2 1 0.7778 0.2222 Human 3 1 0.7817 0.2183 TABLE 8––Average probabilities of group membership for three species Human 4 1 0.8117 0.1883 comparisons, all trials. Human 5 1 0.6615 0.3385 Pig 1 1 0.8308 0.1692 Avg. Avg. Avg. Avg. Pig 2 1 0.8862 0.1138 Human Avg. Pig Rabbit Human Avg. Pig Rabbit Pig 3 1 0.8862 0.1138 Cluster 1 Cluster 1 Cluster 1 Cluster 2 Cluster 2 Cluster 2 Pig 4 1 0.8731 0.1269 Pig 5 1 0.8527 0.1473 Trial 1 0.72302 0.8658 0.14256 0.27698 0.1342 0.85744 Rabbit 1 2 0.1546 0.8454 Trial 2 0.49532 0.50668 0.49694 0.50468 0.49332 0.50306 Rabbit 2 2 0.1362 0.8638 Trial 3 0.75594 0.8138 0.19484 0.24406 0.1862 0.80516 Rabbit 3 2 0.1259 0.8741 Rabbit 4 2 0.1626 0.8374 Trials 1 and 3 consistently had high likelihoods of the observations being Rabbit 5 2 0.1335 0.8665 consistently classified into their assigned group. Trial 2 showed more poten- tial for observations to be classified into more than one cluster. Trial 2 had The probabilities indicate the likelihood of each observation being rapid decomposition across all species, leading to more similar TBS values assigned to each group. The majority of the observations have a strong prob- in all species, and therefore less distance between clusters. ability of membership in the cluster to which they have been placed. five pigs were correctly identified. One pig was misclassified as For Trial 2, the optimum solution proposed was two clusters, a human. This means that the humans were 100% correctly clas- and the discriminant analysis results, as presented in Table 15, sified, and the pigs were classified correctly 80% of the time. showed all five humans correctly classified, and again, four of  DAUTARTAS ET AL. . DIFFERENTIAL DECOMPOSITION 7 TABLE 9––Model fit statistics for pigs and humans only, all trials. TABLE 11––Discriminant analysis results for all species in Trial 1 showing when two groups are specified (as determined by the fuzzy clustering analy- Number of Dunn’s sis) then all the pigs and humans classify into one group while the rabbits Clusters Average Silhouette Coefficient Dc(U) are grouped separately. Trial 1 5 0.311313 0.6031 0.3234 Predicted Trial 2 2 0.264526 0.5569 0.6097 Trial 3 5 0.278720 0.6322 0.3345 Actual Human/Pig Rabbit Total % Correct In all three trials, fewer observations were grouped close to the mean, but Human/Pig 10 0 10 100 there was a clear separation between clusters. Rabbit 0 5 5 100 Total 10 5 15 the five pigs correctly classified, again corresponding to 100% This table corresponds to the cluster analysis described in Tables 3 and 4. This shows that all of the observations were correctly classified. correct classification for the humans and 80% correct classifica- tion for the pigs. Table 16 presents the results of the discrimi- nant analysis for Trial 3. Here, the optimum number of clusters TABLE 12––Discriminant analysis validation of fuzzy clustering for all spe- proposed was five; when five groups were specified, all of the cies in Trial 2. pigs were correctly classified into one of two pig groups, and four of the five humans were correctly assigned to human Predicted groups. One human was misclassified as a pig. Here, the humans Actual Human and Pig Rabbit Total % Correct were classified correctly 80% of the time, and the pigs had a 100% correct classification rate. Human/Pig 10 0 10 100% Figure 4A–C shows the trajectory of decomposition over the Rabbit 0 5 5 100% Total 10 5 15 time period between 100 and 500 ADD. Figure 4A focuses on Trial 1 and shows the rabbits have a unique pattern, while the When two groups are specified all the humans and pigs classify together humans and pigs are much more similar. Humans have a greater and all of the rabbits are grouped separately. This table corresponds to the fuzzy clustering analysis described in Tables 7 and 8. Again, all observations degree of variability in TBS, which is shown by their wider were correctly classified. range. Figure 4B demonstrates that decomposition happens much more rapidly in warmer months and that seasonality is the driv- ing force behind the decomposition patterns in Trial 2. This can TABLE 13––Discriminant analysis validation of fuzzy clustering for all spe- be seen by the fact that all three species have much more similar cies for Trial 3. progressions than in either Trial 1 or Trial 3, and the Total Body Scores are higher. Figure 4C, based on Trial 3, shows the great- Predicted est separation of the three species, and again demonstrates that Actual Human/Pig Rabbit Total % Correct the humans are the most variable. It also shows that the Total Body Score of the rabbits remains much lower throughout the Human/Pig 10 0 10 100% analysis time period, reflecting a slower decomposition progres- Rabbit 0 5 5 100% Total 10 5 15 sion in rabbits compared to pigs or humans. External morpho- logical indicators of decomposition were also much more When two groups are specified, all of the pigs and humans fall into one difficult to visually observe in the rabbits. The fur largely hid group and the rabbits are classified into a separate group. All observations color changes and skin slippage, which also contributed to the were correctly classified. lower TBS. and is common in the temperate environment of Tennessee. Mummification inhibits or significantly delays insect activity on Discussion the body, thereby slowing the rate of decomposition. The fact The results show that pigs and rabbits demonstrated different that neither animal proxy experienced mummification reduces decomposition patterns and trajectories from humans in each sea- their modeling potential for human decomposition in certain son. Rabbits decomposed more quickly in the spring and pigs environments. more rapidly in the summer due to insect activity in Trials 1 and Fuzzy clustering of the TBS similarly shows decomposition 2, respectively. In contrast, humans were much more variable in variation among species. If the animal models are a sufficient their decomposition rates and patterns. Humans were more likely proxy for human remains, then the animal subjects should show to be scavenged (35) and to mummify in each trial. Mummifica- a similar TBS as the humans at the same number of ADD. This tion did not occur in either pigs or rabbits. Human mummifica- would then lead them to be assigned to the same group or clus- tion transpires in a number of different arid microenvironments ter. When all three species were analyzed together, the pigs and TABLE 10––Average probabilities of group membership for pigs and humans only, all trials. This again demonstrates the likelihood that the observations would be consistently classified into the cluster to which they were assigned. Avg. Human Avg. Pig Avg. Human Avg. Pig Avg. Human Avg. Pig Avg. Human Avg. Pig Avg. Human Avg. Pig Cluster 1 Cluster 1 Cluster 2 Cluster 2 Cluster 3 Cluster 3 Cluster 4 Cluster 4 Cluster 5 Cluster 5 Trial 1 0.27954 0.05396 0.27556 0.04164 0.25122 0.08248 0.09222 0.47420 0.10144 0.34172 Trial 2 0.60054 0.39946 0.37692 0.62308 N/A N/A N/A N/A N/A N/A Trial 3 0.28262 0.06474 0.2458 0.03332 0.29918 0.09808 0.09376 0.3017 0.07862 0.50214 8 JOURNAL OF FORENSIC SCIENCES TABLE 14––Discriminant analysis validation results for pigs and humans rely upon visual cues of decomposition. While indicators such as from Trial 1 showing nine individuals correctly classified by species, and color changes, skin slippage, and bloat were easily visible on the one pig misclassified as a human. pigs, the thick fur and small body size of the rabbits often Predicted obscured these indicators, making it more difficult to determine a consistent score. For instance, Fig. 5 shows Rabbit 4 4 days Actual Human Pig Total % Correct after placement in Trial 2 and demonstrates significant maggot Human 5 0 6 100% activity in the lower half of the body and skeletal elements are Pig 1 4 4 80% visible. However, less than 24 h earlier, this same rabbit showed Total 6 4 10 no external signs of decomposition. There were no externally observed maggot masses, including the face or anal region, such This table corresponds to the fuzzy clustering analysis in Tables 5 and 6. that maggot activity was not scored until significant internal con- sumption had occurred. Thus, the mode and timing of insect TABLE 15––Discriminant analysis validation of fuzzy clustering for pigs activity were less visible among rabbits than the pigs or humans. and humans from Trial 2; when two groups are specified for the pigs and Figure 6, in contrast, shows Human 2 and Pig 5 4 days after humans, all humans and four of the pigs are correctly classified. One pig is placement in Trial 2 (ADD = 122.5) where the maggot activity misclassified as a human. is clearly visible in the faces of both individuals and has not led to bone exposure as it had in the rabbit. Use of consistent meth- Predicted ods of evaluation is needed in order for research data to be com- Actual Human Pig Total % Correct parable to other studies and must be accurately applied to all specimens. Keough et al. (47) have proposed a modification of Human 5 0 5 100% Pig 1 4 5 80% the TBS system for pigs in South African environments, but cur- Total 6 4 10 rently, there is no such modification available for rabbits. It should also be noted that the differences in decomposition rate observed in rabbits cannot be simply attributed to the fact that the rabbits were caged while the pig and humans were not. The TABLE 16––Discriminant analysis validation of fuzzy clustering for pigs cages were open to the air and elements and only served to pre- and humans in Trial 3. vent the rabbits from being removed from the study area. The study did not seek to validate the TBS method as a tool Predicted to help estimate the time since death, although it is notable that the estimated ADD did not conform well to actual ADD for any Actual Human Pig Total % Correct of the three species in the trials. The closest correlation occurred Human 4 1 5 80% among the humans in the summer trial, which was the shortest Pig 0 5 5 100% of all of the trials and therefore provided less time for deviation. Total 4 6 10 Notably, TBS does not take into account scavenging, which was Four of the five humans correctly classified into human groups, while one a significant factor affecting human decomposition in this study human was misclassified as a pig. All five pigs were correctly classified as as scavenger activity often lead to early bone exposure and thus pigs. artificially high TBS (35). Even in the absence of significant scavenging in Trials 1 and 2, however, decomposition among the human subjects was overall more variable than the other spe- humans consistently grouped together in one cluster, while the cies, which resulted in both over- and underestimation of ADD. majority of the rabbits formed their own group. This reflects that Possible reasons for greater variability in decomposition among the decomposition pattern of pigs and humans is much more human subjects may relate to diet and microbial communities comparable than either species is to rabbits. When only the pigs and body size. Farm-fed animals typically used for decomposi- and humans were compared, the pigs formed one cluster, with tion research are given homogenous diets, share comparable one of the humans included with the pigs, and some humans in pathogen exposures, and are of similar body weights. In a study a separate group. The misclassification of the human reflects the of bacterial communities among pigs and humans, Parkinson higher degree of decomposition variability observed among the (48) found that the high variability of microflora in humans humans as opposed to the consistency within the pig subjects. likely explained the greater diversity of bacterial community When validated with the discriminant analysis, the fuzzy cluster- changes in soil. While there were similar soil bacterial changes ing solutions performed well, with all of the three species com- between species, pig decomposition rates were considered more parisons having perfect classification results. Misclassifications stable than that of humans. Parkinson (48) noted that such for just the pigs and humans were infrequent, again demonstrat- microflora variability would cause challenges when producing ing that the patterns identified in the data by the fuzzy clustering PMI techniques. More research is required in the area of individ- were accurate. ual microflora before making a conclusion, however. In contrast Another factor to consider concerning nonhuman animal prox- to Parkinson (48), Metcalf et al. (49) found that soil microbe ies for human decomposition is the inability to effectively apply communities were not distinctly different based on carrion type. the TBS system to the nonhuman species, which will affect the To investigate what is driving the intraspecies variation, it would estimated ADD results. This was particularly true for the rabbits be useful to focus microbial analyses on the internal body envi- during early decomposition. Although TBS was designed to be ronment as opposed to skin or soil microbe communities. applied to humans, multiple studies have applied TBS to pigs Numerous studies have also found that body weight and com- (18,19,44–46) and rabbits (15,16) in single-species studies only position, within and among carrion species, can cause significant and the limitations of such applications were not discussed or variation in decomposition rates and byproducts in the soil (44). were unknown. However, the stages listed in the TBS system For instance, Sutherland et al. (18) found that small pigs  DAUTARTAS ET AL. . DIFFERENTIAL DECOMPOSITION 9 FIG. 4––Scatter plot of Total Body Score across actual ADD for pigs, rabbits, and humans during Trial 1 (A), Trial 2 (B), and Trial 3 (C) showing clustering of the three species. FIG. 5––Decomposition of Rabbit 4 in Trial 2 on Day 4 (left) where maggots have exposed bone on the hind limbs but 24 h earlier no observable egg or maggot masses were visible through the fur (left). decompose nearly three times faster than larger pigs, while the pattern reversed. Although Simmons et al. (32) note that Matuszewski et al. (50) found that larger pig carcasses had body mass differences are only a factor when insects are present, higher TBS than smaller pigs for the first 100 ADD, after which the majority of these studies have shown that intraspecific body 10 JOURNAL OF FORENSIC SCIENCES FIG. 6––Decomposition of Human 2 and Pig 5 on Day 4 of Trial 2 where skin color changes are visible and maggot activity has not led to bone exposure. mass is a significant factor in decomposition. Therefore, it is faster than humans when insects are present and scavenging was likely that smaller-bodied animals than humans are unlikely to more extensive in the winter. In sum, our study indicates that accurately model human decomposition rates. Conversely, it is human data are best for determining human patterns of decom- possible that pigs and humans sometimes had similar TBS position in forensic cases and should form the basis for research because they were of a more comparable body mass. It is impor- seeking to develop forensic-related PMI methods. tant to note that more similar does not mean identical; examining Fig. 2 clearly shows that the progression of decomposition as measured by comparison of TBS to ADD was not the same for References humans and pigs. Although the lines overlapped at points, they 1. Reed J. A study of dog carcass communities in Tennessee, with special did not remain consistently together, nor did they deviate in such reference to the insects. American Midland Nature 1958;59(1):213–45. a way that a simple correction factor would render the difference 2. Early M, Goff ML. Arthropod succession patterns in exposed carrion on the island of O’ahu, Hawaiian Islands. USA. J Med Entomol 1986;23 negligible. When predicting or estimating PMI, minimal error is (5):520–31. required. This research suggests that using a nonhuman animal 3. Keiper JB, Chapman EG, Foote BA. Midge larvae (Diptera: Chironomi- proxy will introduce a higher error rate. dae) as indicators of postmortem submersion interval of carcasses in a woodland stream: a preliminary report. J Forensic Sci 1997;42(6):1074– 9. Conclusions 4. Tomberlin JK, Adler PH. Seasonal colonization and decomposition of rat carrion in water and on land in an open field in South Carolina. J Med The use of nonhuman animal proxies for human remains in Entomol 1998;35(5):704–9. decomposition studies has been based primarily on the assump- 5. Carter DO, Yellowlees D, Tibbett M. Temperature affects microbial tion that most carrion, especially those that are similar in size decomposition of cadavers (Rattus rattus) in contrasting soils. Appl Soil Ecol 2008;40(1):129–37. and integument, will undergo the same decomposition process at 6. Coe M. The decomposition of elephant carcasses in the Tsavo (East) an analogous rate as humans, or with negligible error. However, National Park. Kenya. J Arid Environ 1978;1(1):71–86. Moreau et al. assert that “Until more peer-reviewed studies are 7. Lee HL, Marzuki T. Preliminary observation of arthropods in carrion and published that directly test the pig-as-surrogate claim, it appears its application to forensic entomology in Malaysia. Trop Biomed that the use of medium-sized pig carcasses [22–27 kg] relies 1993;10(1):5–8. 8. Bornemissza GF. An analysis of arthropod succession in carrion and the more on logic and practicality than experimental evidence” effect of its decomposition of the soil fauna. Aust J Zool 1957;5(1):1– (51:215). The purpose of this project was to provide experimen- 12. tal evidence to determine if nonhuman animal data are relevant 9. Al-Mesbah H, Moffatt C, El-Azazy OME, Majeed QAH. The decompo- and reliable for human forensic contexts and define the condi- sition of rabbit carcasses and associated necrophagous Diptera in Kuwait. Forensic Sci Int 2012;217(1–3):27–31. tions in which entomologists or anthropologists can apply the 10. Denno RF, Cothran WR. Competitive interactions and ecological strate- results of nonhuman animal studies to PMI estimation for a gies of Sarcophagid and Calliphorid flies inhabiting rabbit carrion. Ann specific human case. Entomol Soc Am 1976;69(1):109–13. Locally available nonhuman animals are fully appropriate for 11. Tantawi TI, El-Kady EM, Greenberg B, El-Ghaffar HA. Arthropod suc- ecological studies designed to investigate arthropod variation cession on exposed rabbit carrion in Alexandria. Egypt. J Med Entomol 1996;33(4):566–80. within a particular environment or to address questions concern- 12. Tessmer JW, Meeks CL. Dispersal and distribution of Calliphordae ing pest control, soil sciences, wildlife management, forestry, (Diptera) immature from animal carcasses in southern Louisiana. Ento- and natural scavenging behavior. The results of this study mol Soc Am 1996;33(4):665–9. demonstrate, however, that human remains behave less pre- 13. Bourel B, Martin-Bouyer L, Hedouin V, Cailliez JC, Derout D, Gosset D. Necrophilous insect succession on rabbit carrion in sand dune habitats dictably than those of pigs or rabbits such that the nonhuman in northern France. J Med Entomol 1999;36(4):420–5. models could not replicate the impacts of differential insect 14. Adlam RE, Simmons T. The effect of repeated physical disturbance on activity, scavenging or physical state changes (e.g., mummifica- soft tissue decomposition – are taphonomic studies an accurate reflection tion) exhibited by the human subjects. While this paper seeks to of decomposition. J Forensic Sci 2007;52(5):1007–14. document the decomposition pattern differences among carrion 15. Bachmann J, Simmons T. The influence of preburial insect access on the decomposition rate. J Forensic Sci 2010;55(4):893–900. species, subsequent papers will focus on different variables 16. Simmons T, Cross T, Adlam RE, Moffat C. The influence of insects on responsible for those pattern variations. Seasonality also plays a decomposition rates in buried and surface remains. J Forensic Sci role in differential decomposition as pigs and rabbits decompose 2010;55(4):889–92.  DAUTARTAS ET AL. . DIFFERENTIAL DECOMPOSITION 11 17. Widya M, Moffatt C, Simmons T. The formation of early stage adipo- 36. SAS Institute. The SAS system for windows. Release 9.2. Cary, NC: cere in submerged remains: a preliminary experimental study. J Forensic SAS Institute, 2011. Sci 2012;57(2):328–33. 37. Megyesi MS, Nawrocki SP, Haskell NH. Using accumulated degree days 18. Sutherland A, Myburgh J, Steyn M, Becker PJ. The effect of body size to estimate the postmortem interval from decomposed human remains. J on the rate of decomposition in a temperate region of South Africa. Forensic Sci 2005;50(3):618–26. Forensic Sci Int 2013;231(1):257–62. 38. Vass AA. The elusive universal postmortem interval formula. Forensic 19. Cross P, Simmons T. The influence of penetrative trauma on the rate of Sci Int 2011;204(1–3):34–40. decomposition. J Forensic Sci 2010;55(2):295–301. 39. R Core Team. R: a language and environment for statistical computing. 20. Payne JA. A summer carrion study of the baby pig Sus scrofa. Ecology Vienna, Austria: R Foundation for Statistical Computing, 2016; https:// 1965;46:592–602. www.r-project.org (accessed March 6, 2018). 21. Goff ML, Omori AL, Gunatilake K. Estimation of postmortem interval 40. Wickham H. ggplot2: elegant graphics for data analysis. New York, NY: by arthropod succession. Three case studies from the Hawaiian Islands. Springer-Verlag, 2009. Am J Forensic Med Pathol 1988;9(3):220–5. 41. NCSS. NCSS 11 Statistical Software. Kaysville Utah: NCSS, LLC, 22. Shean BS, Messinger L, Papworht M. Observations of differential 2016; https://www.ncss.com/software/ncss/(accessed March 6, 2018). decomposition on sun exposed v. shaded pig carrion in coastal Washing- 42. Seaver B, Traintis K, Reeves C. The identification of influential subsets ton State. J Forensic Sci 1993;38(4):938–49. in regression using a fuzzy clustering strategy. Technometrics 1999;41 23. Anderson GS, VanLaerhoven SL. Initial studies on insect succession on car- (4):340–51. rion in southwestern British Columbia. J Forensic Sci 1996;41(4):617–25. 43. Huang M, Xia Z, Wang H, Zeng Q, Wang Q. The range of the value for 24. Horenstein MB, Xavier Linhares A, Rosso de Ferradas B, Garcia D. the fuzzifier of the fuzzy c-means algorithm. Pattern Recogn Lett Decomposition and Dipteran succession in pig carrion in central Argen- 2012;33(16):2280–4. tina: ecological aspects and their importance in forensic science. Med 44. Spicka A, Johnson R, Bushing J, Higley LG, Carter DO. Carcass mass Vet Entomol 2010;24(1):16–25. can influence rate of decomposition and release of ninhydrin-reactive 25. Richards EN, Goff ML. Arthropod succession on exposed carrion in nitrogen into gravesoil. Forensic Sci Int 2011;209(1–3):80–5. three contrasting tropical habitats on Hawaii Island. Hawaii. J Med Ento- 45. Meyer J, Anderson B, Carter DO. Seasonal variation of carcass decom- mol 1997;34:328–39. position and gravesoil chemistry in a cold (Dfa) climate. J Forensic Sci 26. Carvalho LML, Linhares AX. Seasonality of insect succession and pig 2013;58(5):1175–82. carcass decomposition in a natural forest area in southeastern Brazil. J 46. Myburgh J, L’Abbe EN, Steyn M, Becker PJ. Estimating the postmortem Forensic Sci 2001;46(3):604–8. interval (PMI) using accumulated degree-days (ADD) in a temperate 27. Shalaby O, DeCarvalho LK, Goff ML. Comparison of patterns of region of South Africa. Forensic Sci Int 2013;229(1):165e1–6. decomposition in a hanging carcass and a carcass in contact with soil in 47. Keough N, Myburgh J, Steyn M. Scoring of decomposition: a proposed xerophytic habitat on the Island of Oahu. Hawaii. J Forensic Sci 2000;45 amendment to the method when using a pig model for human studies. J (6):1267–73. Forensic Sci 2017;62(4):986–93. 28. Centeno N, Maldonado M, Oliva A. Seasonal patterns of arthropods 48. Parkinson R. Bacterial communities associated with human decomposi- occurring on sheltered and unsheltered pig carcasses in Buenos Aires tion [dissertation]. Wellington, New Zealand: Victoria University of Province (Argentina). Forensic Sci Int 2002;126(1):63–70. Wellington, 2009. 29. Grassberger M, Frank C. Initial study of arthropod succession on pig carrion 49. Metcalf JL, Xu ZZ, Weiss S, Lax S, van Treuen W, Hyde ER, et al. in a central European urban habitat. J Med Entomol 2004;41(3):1210–4. Microbial community assembly and metabolic function during mam- 30. Schoenly KG, Haskell NH, Hall RD, Gbur JR. Comparative performance malian corpse decomposition. Science 2016;351(6269):158–62. and complementarity of four sampling methods and arthropod preference 50. Matuszewski S, Konwerski S, Fraz tczak K, Szafałowicz M. Effect of tests from human and porcine remains at the Forensic Anthropology body mass and clothing on decomposition of pig carcasses. Int J Legal Center in Knoxville. Tennessee. J Med Entomol 2007;44(5):881–94. Med 2014;128(6):1039–48. 31. Schoenly KG, Haskell NH, Hall R. Testing reliability of animal models 51. Moreau G, Michaud J-P, Schoenly KG. Experimental design, inferential in research and training programs in forensic entomology. Part II. Wash- statistics, and computer modeling. In: Tomberlin JK, Benbow ME, edi- ington, DC: National Institute of Justice, 2002. tors. Forensic entomology: international dimensions and frontiers. Boca 32. Simmons T, Adlam RE, Moffatt C. Debugging decomposition data-com- Raton, FL: CRC Press, 2015;205–30. parative taphonomy studies and the influence of insects and carcass size on decomposition rate. J Forensic Sci 2010;55(1):8–13. Additional information and reprint requests: 33. Vass AA, Smith RR, Thompson CV, Burnett MN, Dulgerian N, Ecken- Angela Dautartas, M.A. rode BA. Odor analysis of decomposing buried human remains. J Foren- Department of Anthropology sic Sci 2008;53(2):384–91. University of Tennessee 34. Notter SJ, Stuart BH, Rowe R, Langlois N. The initial changes of fat 502 Strong Hall, 1621 Cumberland Avenue deposits during the decomposition of human and pig remains. J Forensic Knoxville Sci 2009;54(1):195–201. TN 37996 35. Steadman DW, Dautartas AM, Kenyhercz MW, Vidoli GM, Jantz E-mail: [email protected] LM, Mundorff AZ. Differential scavenging activity among pig, rabbit and human subjects. J Forensic Sci. In press.