Footprint evidence of early hominin locomotor diversity at Laetoli, Tanzania

Wild bear behavioural data

Wild black bear behaviour was quantified using video data recorded by B.K. over the course of several years at his ongoing field site in Lyme, New Hampshire, USA. Video data captured bears of different ages (cubs, adolescents and adults). Bears were present on screen for a total of 50 h 55 min 18 s. For each terrestrial bipedal incident, the length of the event, the approximate age of the bear and the number of steps were recorded. Additionally, steps were evaluated on whether they were completed independently, or the individuals used other environmental objects for balance.

Comparative kinematic data

Comparative kinematic data were collected on three species: U. americanus, P. troglodytes, and Homo sapiens. For bears and chimpanzees, the sample size included all available individuals housed at each location. For the human sample size information see below. Randomization was not relevant to our study as we were interested in measuring footprint characteristics from whole sample populations, as opposed to comparisons within those populations. Blinding was not relevant to the data collected on the non-human comparative species (for example, bears and chimpanzees) nor to the data collection on fossilized footprints. The human participants were unaware of the site A tracks at Laetoli and therefore had no knowledge of how the data obtained from their footprints would be used in this study.

Ursus americanus

Data were collected on four juvenile semi-wild U. americanus (n = 3 male, 1 female), whose feet were within 10% of the length (average foot length = 145.7 mm) of the recorded footprints of Laetoli site A (average foot length = 161.7 mm). These orphaned, approximately 20-kg bears were located at the Kilham Bear Center (Lyme, NH), awaiting reintroduction to the wild. This study examined the bears between the ages of 5–8 months old. Our protocol was reviewed and approved by the Institutional Animal Care and Use Committee (IACUC) of Dartmouth College. The bears were enticed to independently walk bipedally through a constructed mud trackway for either an applesauce or maple syrup reward (Extended Data Fig. 3). Measurements were collected on the footprints, including foot length, heel width, forefoot width, step length and stride width using the definitions from Tuttle4. For a subset of footprints (n = 5), the width of the impression for the 1st, 2nd, 4th and 5th digits were measured. The presence or absence of claw impressions was also documented.

Pan troglodytes

Data for extant chimpanzees were extracted from three sources to collect all the relevant gait metrics. Two published datasets examined the same two subadult individuals housed at Stony Brook University. The third set were recorded on semi-wild individuals (n = 46), using a plantar pressure mat at the Ngamba Island Chimpanzee Sanctuary (Entebbe, Uganda). While this third data set increases sample size and captures intraspecific variation, we recognize that plantar pressure data do not always align perfectly with footprints made in a deformable substrate20.

Stride width data and step length comparisons

Chimpanzee stride width data were taken from Thompson et al.28 on two subadult male chimpanzees (7.0 ± 0.1 years of age; 34.8 ± 1.2 kg) and were supplemented with step length data for the same steps. Three-dimensional kinematic methods and step width calculation have been described previously28. Step length was calculated as the distance between left and right calcaneus markers in the sagittal plane during consecutive hind limb midstance periods. Chimpanzee step lengths are typically asymmetric, so step length was averaged over the two consecutive steps which defined the stride.

Forefoot width, heel width and foot length comparisons

Footprint dimensions and stride length data were recorded on the same two subadult male chimpanzees as above, though at a slightly younger age (6.5 and 6.9 years of age, 30.7 and 27.8 kg, respectively). The experimental design is described in detail elsewhere42. In brief, chimpanzees traversed a runway, at the centre of which was a pressure mat (RSScan International) and a container of hydrated sediment in which the chimpanzees could produce footprints. This sediment was taken directly from a layer that preserves 1.5 Ma hominin footprints near Ileret, Kenya50. Laterally positioned video cameras were used to record the chimpanzees as they walked along this trackway and produced footprints. Two digitization softwares, MaxTRAQ Lite+ (v. (Innovisions Systems) and ImageJ v.1.4751, were used to quantify various aspects of their gaits, including stride length. Tape measures and digital callipers were used to directly measure the external dimensions of each chimpanzee’s feet. Scaled photographs were taken of the footprints produced in each trial, and these were later measured using ImageJ software.

Forefoot width, width of digits 1 and 2, divergence ratio, and foot length comparisons

Data were collected by E.J.M. at the Ngamba Island Chimpanzee sanctuary (Entebbe, Uganda) managed by the Chimpanzee Sanctuary and Wildlife Conservation Trust (CSWCT) using procedures approved by the Dartmouth College IACUC. The Tekscan plantar pressure mat (PPM) was positioned within a walkway connecting the overnight enclosure to the open forest habitat, near a gate and underneath a solid cross section to help prevent individuals from jumping over the mat using the ceiling bars. This location was determined using the expertise of the sanctuary keepers. The animals were first introduced to a mat shell that was lacking the internal sensors, to habituate them to the novel stimulus. All subsequent data were collected using both the empty and real PPMs, positioned such that they covered the entire width of the walkway to force individuals to walk across one of the two mats. Both mats were covered with thin green sacks to help disguise them from the chimpanzees and facilitate faster removal if necessary. It was determined that the southeast facing direction was the preferred path for the chimpanzees and the sensor-containing PPM was always positioned there from the second collection onwards. Data were collected twice a day; once in the morning (between 06:45 and 08:00) when the chimpanzees were headed to the forest for their first feeding, and once at 18:00, when the chimpanzees were headed into the overnight enclosure to sleep and receive their last feeding. Data were collected on 46 adult chimpanzees (18 male, 28 female, ages 12–36 years). A subset of 54 dynamic pressure records was analysed. Using the associated Tekscan PPM software, Footmat Research (v. 7.10), the pressure recordings were analysed to determine foot length, forefoot width, the width digits 1 and 2, and the linear distance between the centre of digits 1 and 2. The divergence ratio was calculated by dividing the distance between digits 1 and 2 by the individual’s foot length.

Homo sapiens

Data were extracted from previous studies of two modern human populations in order to collect all the relevant foot, footprint and gait metrics.

Stride width and step length comparisons

Data were taken on 654 participants, recruited through the Living Laboratory at the Boston Museum of Science19. Sample size was determined by museum visitor traffic and willingness to participate in a scientific study. In brief, this dataset included 73 children between the ages of 2 and 7 years old (29 female and 44 male) and 581 individuals (366 female and 215 male) between the ages of 8 and 80 years. A pressure-sensitive gait carpet (6.1 m long × 0.89 m wide) with a spatial resolution of 1.27 cm and collecting data at 120 Hz (GAITRite) was used to collect stride length and stride width. For a subset of 33 adults, additional data were collected using a Tekscan PPM and analysed with FootMat Research (v. 7.10) to calculate foot length, the width of digits 1 and 2, and the linear distance between the centre of digits 1 and 2. These measurements were used to calculate a divergence ratio as described above in ‘P. troglodytes’.

Forefoot width, heel width and foot length comparisons

Footprint dimensions and stride length data for 29 Daasanach adults (15 male and 14 female, ages 18–47) and 12 children (10 male and 2 female, ages 4–15), who live near the town of Ileret, Kenya and grew up either habitually unshod or minimally shod were taken from Hatala et al.20,21. Details of the experimental protocol largely mirrored the procedures described above. In brief, subjects generated footprints while walking through a rehydrated sample of the same sediments that preserve 1.5 Ma hominin tracks near Ileret. Video cameras were used to record subjects as they produced footprints, and two digitization software packages (MaxTRAQ Lite+ v. and ImageJ v.1.47) were used to measure stride lengths and other kinematic variables. The external dimensions of subjects’ feet were directly measured with tape measures and digital callipers. Scaled photographs of the footprints produced in each trial were measured using ImageJ.

Human cross-stepping footprint experiments

Experiments were carried out by K.G.H. and E.M.W.-H. to investigate whether and how cross-stepping kinematics influence the perimeter dimensions and internal topologies of an individual’s footprints. We could thereby evaluate whether the size and shape of the Laetoli site A tracks could have been generated by a hominin with feet similar to those who left tracks at sites G and S, but while cross-stepping. Detailed methods are provided in Supplementary Information. In brief, ten adult subjects (including six female, three male, and one non-binary between 19 and 52 years old) each completed ten trials in which they produced tracks in sedimentary conditions meant to mimic those at Laetoli37,52. Sample size was determined by availability and willingness to participate in the study. Five trials were completed with a normal, self-selected walking gait and another five were completed with a cross-stepping gait, as inferred for the Laetoli site A trackmaker. In each trial, a focal footprint was selected, measured in situ, and photographed (25–30 photos per footprint). The lengths and widths of the steps bracketing the track were also measured. Photographs were used to generate 3D models of the tracks using Agisoft Metashape software (v.1.7.3), and average normal and cross-stepping tracks were generated for each subject using DigTrace Pro (v.1.8.1)53. Lengths and widths of these averaged tracks were measured using Geomagic Wrap (v. 2021.0.0) (3D Systems). Regional depths were measured and evaluated using the same methods described below (‘Comparative analyses of Laetoli footprint shapes’). Within-subject comparisons enabled us to understand how cross-stepping influenced the dimensions of the perimeter and the internal topology of a subject’s footprints.

Fossil footprint data and analysis

Comparative metrics were quantified from a set of modern human footprints from the Late Pleistocene at Engare Sero, Tanzania. These footprints are an important comparison with the Laetoli footprints, as they were generated in a similar circumstance (footprints in volcanic ash) and represent an early population of unshod modern humans.

A scaled 3D orthophoto of the Engare Sero site was created via photogrammetry by B.Z. and C.L.-P. to visualize the distribution of footprint trackways across the entire site using Agisoft Photoscan (now Agisoft Metashape v. 1.4.4). The model was created from hundreds of photos originally taken by the Smithsonian 3D Digitization Program in 2010. Measurements that were defined in Tuttle4, were taken from the fossil tracks using the software, ImageJ (v. 1.49), and included foot length, forefoot width, heel width and step width of each footprint at Engare Sero. In some cases, partial footprints were included for measurement, as long as they included the requisite landmarks for those measurements. Overall, data were collected from 151 footprints at the Engare Sero site. Of the 151 footprints, 61 footprints were considered partial footprints and 90 footprints were considered complete footprints. From these, 67 step length and stride width measurements and 105 heel width and ball width measurements were included in our analyses.

All measurements for Laetoli trackways G and S were obtained from published sources4,32,42. Box and whisker plots and bivariate graphs (using ggplot254) were generated using R (v. 3.6.1), while the table and pie chart were generated using Microsoft Excel (v. 2102).

Comparative analyses of Laetoli footprint shapes

Comparative analyses followed methods similar to resampling analyses published previously42. In brief, the human comparative sample included 245 footprints produced by 29 adult and 12 juvenile habitually unshod Daasanach individuals traveling at walking speeds. The chimpanzee comparative sample included 45 footprints produced by two individuals walking bipedally. Laetoli samples included only the best-preserved tracks from each site, leaving samples of five footprints from site G that were described by their original excavators as free from taphonomic damage that would obscure track topology (G1-25, G1-27, G1-33, G1-34 and G1-35), and two from site S (L8-S1-2 and L8-S1-4). For site A, we included tracks A2 and A3, as these were the only two for which we were relatively confident in identifying regions of interest across the entire track. Larger sample sizes would be desirable, but we did not want to sacrifice data quality for quantity by including tracks that were overprinted or that did not appear to represent complete foot anatomy. We did not rely on parametric statistical tests for which larger sample sizes would be a necessity, and instead used an analytical approach that could handle smaller sets of observations (see below).

For each experimental and fossil footprint, 3D models were constructed using photogrammetry, through a variety of methods described here for Laetoli site A and elsewhere by the authors for other samples20,21,35,42. Using Geomagic Wrap (v. 2021.0.0) (3D Systems), a best-fit plane was fit to the undisturbed substrate surrounding each track, and this was fixed to the xy plane in world coordinate space. In this orientation, depths of the footprint were measured in the regions of the medial and lateral heel, medial and lateral midfoot, and all five metatarsal heads and toes. Raw depth measurements were normalized, within each footprint, to a scale of 0 to 1 in order to compare the topologies of footprints that may vary in depth. However, a Wilcoxon signed-rank test showed that, overall, human and Laetoli track samples did not differ significantly in their depths (P = 0.08). Within-subject means of the 14 normalized depth measurements were calculated, and a between-subject covariance matrix was created using the subject averages for normalized depths at each of the 14 measured regions. An overall ‘human mean footprint’ was also computed by averaging the within-subject mean normalized depths, and this represented a measure of central tendency as described below.

To represent the range of observed variation in human footprint topography, for 1,000 iterations we randomly sampled a human subject and drew a sample of two of their footprints. We then averaged the normalized depths of those two footprints and computed the Mahalanobis distance (using the between-subject covariance matrix) between this track and the mean of all other subjects’ footprints. Also, for 1,000 iterations we selected a random chimpanzee subject, drew a random sample of two of their footprints and computed the Mahalanobis distance between the average of those tracks and the overall mean human footprint. For the Laetoli tracks, site A and site S samples only included two tracks, so these were simply averaged and the Mahalanobis distance was calculated between each averaged track and the mean human footprint. For Laetoli site G, all possible two-track combinations (ten) were drawn from the sample described above, and the Mahalanobis distance was calculated between the averaged track from each combination and the human mean. In all cases, we calculated multivariate distances using the human between-subject covariance matrix (that is, treating the chimpanzee and fossil tracks as if they came from different human subjects). All analyses described above, and the histogramdisplaying multivariate distances (Fig. 2), were generated using R (v. 3.6.1), with custom scripts and functions from the dplyr55, ggplot254 and reshape256 packages.


While casts of the site A bipedal footprints existed at one point, all our attempts to locate them (see Acknowledgements for a complete list) were unsuccessful. Prior to our fieldwork at Laetoli in 2019, we modelled the original trackway photogrammetrically using extant photography from the site. Original photography of Laetoli site A was taken by J.R. We obtained his photographs of trackway A through Science Photo Library. All photographs were taken with a Nikon F2 on 35-mm Kodachrome slide film. Digital scans from these slides were used to produce a 3D model of the Laetoli A footprints. Unfortunately, since the images were taken in 1977, they were not recorded with modern photogrammetry processing in mind. Several features of the digitized images limit successful and accurate construction of a 3D model. First, there are only four images of the footprints. One of these images has noticeably different exposure settings that caused significant alignment problems during processing, and thus was excluded. All images were shot at oblique angles, from a relatively narrow range of camera positions. A yellow string defining the site grid lies over one of the footprints, obscuring part of it, and casting a shadow. The images were all taken relatively early in the day, so there are shadows within each footprint that create strong contrasts. The slides were digitized at 4,000 dpi, but they were not scanned with specialized equipment to guarantee geometric accuracy, and this potentially introduced more sources of distortion.

However, despite the limitations of the images, it was possible to extract 3D data for the Laetoli A footprints. All processing was done in Agisoft Photoscan Pro (v. 1.7.1). The standard processing steps (align photos, build dense cloud, build mesh, build texture) were run to produce a 3D model, though the process had to be done iteratively to remove noise, ensure accurate alignment of the photos, scale the model appropriately using published measurements, add manual tie points, and refine the model. A DEM (digital elevation model) and orthophotograph were exported for further visualization and analysis in ArcGIS (v. 10.6.1). The 3D model was also exported to Autodesk Meshmixer (v. 3.5.474) to create a ‘watertight’ 3D volume that could be 3D printed for further visualization (1977 model is hosted on Morphosource, ID: 000390119). Photogrammetric reconstruction was validated using published measurements of the footprints. It is important to note however, that there were no published measurements for the depths of the footprints and that the internal anatomy of this reconstruction is potentially misleading because of the incomplete excavation of the footprints2,8,12.

A second, more accurate reconstruction was done using photogrammetry from the re-excavated site A bipedal trackway using 57 images taken in June 2019. The images were captured in a systematic manner using a Nikon D7000 camera and Nikon DX AF-S Nikkor 18–105 mm lens. All photos were taken by hand, from an eye level, while walking a series of transects, across the area of interest. Spacing between shots was kept low to ensure a minimum of approximately 65% overlap between adjacent images. All processing was done using Agisoft Photoscan Pro/Metashape Pro (v. 1.7.1). Standard processing steps (for example, as described15,51) were taken to create a 3D model of the A trail. This included photo alignment, manual editing of the sparse cloud to remove points with high ‘reprojection uncertainty’, building a dense cloud, building a mesh, refining the mesh, then building a texture. During processing, images were checked for sharpness using the ‘image quality’ tool and any images with significantly lower quality were removed. The model was scaled to the real-world using scale bars placed across the region of interest. Finally, an orthophotograph as well as a DEM (digital elevation model) were exported as geotiffs into ArcGIS in an arbitrary local coordinate system for further analysis (2019 model hosted on Morphosource, ID: 000390114).

To generate contour maps, two approaches were used. First, starting with the raw stereolithography scans (.stl file format), Ultimaker Cura software (v.4.8.0) was used to rotate the raw scans and align them with x and y axes. This was a manual process. These rotations were exported to binary-format .stl files. The rotated files were then run through an R script using R version 4.0.3. The R script uses the tidyverse and rgl libraries to load the .stl files into R-friendly dataframes and plot them as contours using ggplot’s geom_contour function. The script is available through GitHub.

Using a second approach, the .stl files were brought into Cloud Compare (v. 2.11.3) to check model orientation. If necessary, models were reoriented to allow the local ground surface to be level using the “level” tool, and then the files were exported. The correctly oriented model was imported into SAGA GIS using the import stereo lithograph file (STL) tool.  This tool converts the .stl directly to a DEM raster. The rasters were checked in SAGA and a hillshade generated with the analytic hillshading function using the standard sun position setting of 315° azimuth and 45° height. Both the DEM and hillshade were then exported as geotiffs. These geotiffs were imported into ArcGIS for visualization. The DEM was colored using a red-blue colour ramp to indicate relative depth and this was layered onto the hillshade raster using the NAGI fusion method57 (Extended Data Fig. 8). Cloud compare was used to quantify erosive alterations to the site A footprints from 1977 to 2019.

3-D surface scanning

Three-dimensional surface scans of Laetoli A and plaster casts of bear prints were collected using a Creaform Go!Scan 50.

Reporting summary

Further information on research design is available in the Nature Research Reporting Summary linked to this paper.

Leave a Reply

Your email address will not be published. Required fields are marked *