Supporting InformationTheis et al. 10.1073/pnas.1306477110SI MethodsAnesthesia Protocols. Spotted hyenas were anesthetized withTelazol (6.5 mg/kg) delivered from a CO2-powered darting rifle,typically while the hyena was alone and lying down. Striped hy-enas were anesthetized with Telazol (2.5 mg/kg) or with a com-bination of ketamine HCl (3.6 mg/kg) and medetomidine HCl(0.06 mg/kg) delivered from a blowpipe or darting rifle after thehyena had been caught in a soft-catch foot-hold or cage trap (1).

Scanning Electron Microscopy. Paste samples were fixed for 1 h in4% (vol/vol) glutaraldehyde buffered with 0.1 M sodium phos-phate at pH 7.4. They then were rinsed and fixed in 1% (wt/vol)osmium tetroxide for 1 h before being rinsed again and dehy-drated in an ethanol series (25%, 50%, 75%, 95%; vol/vol) for15 min at each concentration and for three 15-min changes inpure ethanol. The samples were dried in a Balzers Model 010critical point dryer using liquid CO2 as the transitional fluid andwere mounted on aluminum stubs using Quick Cure-5 epoxy.They were then coated with ∼10 nm osmium and were examinedin a JEOL JSM-7500F (cold field emission electron emitter)scanning electron microscope. Digital images were acquired us-ing JEOL PC-SEM Version 2.0.0.8.

PCR Program and Amplicon Clean-Up. The PCR program consistedof an initial dissociation step of 94 °C for 3 min, followed by 30cycles of 94 °C for 30 s (denaturing), 60 °C for 45 s (annealing),and 72 °C for 1 min (extending) and a final extension step of 72 °Cfor 2 min. During amplification, unique molecular barcodes wereincorporated into each sample’s amplicons, enabling multiplexsequencing of samples (2). To minimize PCR bias (3), eachsample was amplified in triplicate and then pooled. Nontemplatecontrols were run for each barcoded primer pair. Ampliconswere purified using solid-phase reversible immobilization (AM-Pure XP beads) and were quantified using an Invitrogen high-sensitivity Quant-iT dsDNA assay kit, and amplicons from allsamples were pooled at equimolar concentrations.

Sequence Processing. We processed 454 run files using mothursoftware (v. 1.27.0) (4). Sequences were discarded if they (i)contained any ambiguous base calls, (ii) did not have perfectbarcode and primer matches, (iii) had a homopolymer run greaterthan eight bases, or (iv) had an average quality score of less than35 using a sliding window of 50 bases. The remaining sequenceswere aligned using the SILVA-based bacterial reference align-ment in mothur, and those sequences that did not begin exactly atour target site on the bacterial 16S rRNA gene were discarded, aswere sequences that fell in the bottom 6% of total sequencesbased on read length (<233 bases). Last, sequences were dis-carded if they were deemed chimeric by mothur’s uchime tool or iftheir origin was determined to be a mitochondrion or chloroplastby mothur’s classifier tool. The remaining 583,551 sequences werebinned into 2,274 operational taxonomic units (OTUs; 1,284 sin-gletons), based on a 97% sequence similarity, using the averageneighbor clustering algorithm. Iterative subsampling was per-formed as described in Methods in the main text.

OTU Representative Sequences. Representative sequences of the865 OTUs have been deposited in GenBank (accession nos.KC705471–KC706325), except for 10 that were flagged as po-tentially chimeric by GenBank using the BLAST algorithm. The10 OTUs represented by these sequences did not affect any

analyses. All 865 representative sequences and their associatedmetadata are provided as Dataset S2.

Abundances of OTUs in Pastes.Given that many of the OTUs in thisstudy could not be assigned confidently to particular knowngenera (Table S2) and that among those which could be confi-dently assigned the 16S rRNA gene copy number was largelyunknown or variable, it was not possible to adjust OTU data toreflect potential variation in 16S rRNA gene copy number amongcommunity members (5, 6). Therefore, although our OTU dataaccurately reflect patterns in bacterial community structure amongthe pastes of sampled hyenas, they do not necessarily provide de-finitive information about the absolute or rank abundances ofspecific bacteria in paste communities.

Protocols for the Extraction and Detection of Volatile Fatty Acidsfrom Paste. All extractions were performed at room tempera-ture. Formic acid (85% wt/vol) was added to methyl tert-butylether (MTBE) at a final concentration of 0.2% (vol/vol) to reducethe pH, thereby preventing protonation of the volatile fatty acids(VFAs) and improving recovery. We used pivalic acid as theinternal standard at a final concentration of 0.15 mM in MTBE.To begin, a paste aliquot was transferred to a 2-mL glass vial fittedwith a Teflon-lined screw cap. Next, 500 μL of the extractionsolvent mixture (MTBE + formic acid + pivalic acid) and 500 μLof Milli-Q water were added to the paste. This mixture wasvortexed until the paste dissolved in the solvent (∼45 s) and wascentrifuged for 5 min at 835 × g to enhance the transfer of theVFAs into the extraction solvent and to aid separation of thehydrophobic and hydrophilic layers. Next, both the aqueous(bottom) layer and the solvent (top) layer were transferred intoa second vial, leaving the residual lipid (middle) layer behind.Then 500 μL of the extraction solvent mixture was added to thesecondary vial, and the process of blending, centrifuging, andtransferring the aqueous and solvent layers was repeated, asspecified above. Another 500 μL of solvent mixture was added toa fresh third vial, and the extraction procedure was repeated oncemore. After this step, we transferred 1,500 μL of the solvent mix-ture and 500 μL of the aqueous layer to a final fourth vial, whichwas stored overnight at −80 °C. The aqueous layer froze, facili-tating removal of the supernatant solvent layer from each sample.In preparation for the GC/MS analysis, 500 μL of the super-

natant layer was transferred from the extraction vial into a fresh2-mL glass vial fitted with a screw cap and ethylene tetrafluoro-ethylene/silicone septum. An additional 5 μL of 85% (wt/vol) formicacid was added to the GC vial to ensure deprotonation of the VFAsand improve detection. Standards were prepared by dissolvingacetic, propanoic, isobutanoic, butanoic, isopentanoic, penta-noic, isohexanoic, hexanoic, and heptanoic acids in the solventmixture to yield final concentrations of 1 mM, 2 mM, 6 mM, and8 mM.GC/MS analysis was conducted on an Agilent 6890 gas chro-

matograph coupled to an Agilent 5793 mass spectrometer and anAgilent 7683 auto-injector equipped with a DB-WAX column(30 m × 250 μm × 0.25 μm film thickness). Helium carrier gas wasused with a flow rate of 1.2 mL/min. Splitless injections of 1-μLsamples were made into an inlet held at 240 °C; the inlet waspurged after 0.5 min at a flow rate of 50 mL/min. The GC ovenramp (18.25 min overall time) used was 40 °C (hold 1 min); rampat 40 °C/min to 100 °C (hold 0 min); ramp at 10 °C/min to 220 °C(hold 0 min); ramp at 40 °C/min to 250 °C (hold 3 min). Param-eters used for the mass spectrometer included a solvent delay of

Theis et al. www.pnas.org/cgi/content/short/1306477110 1 of 9

3 min, and data were acquired in scan mode with a mass range of30–400 m/z.Although paste samples were collected over a 15-y period

(1994–2009), they were assayed simultaneously in 2012. There-fore, to ensure that storage artifacts were not driving any of theanalyses in this study, we evaluated potential correlations be-tween the relative abundances of individual volatiles in paste andstorage time (i.e., time in months since sample collection). Nosignificant correlations were found (Table S3).

Graphical and Statistical Data Analyses. Variation in the OTU andVFA profile similarities was visualized via 2D, nonmetric mul-tidimensional scaling (nMDS) plots and was statistically evalu-ated using nonparametric MANOVA (NPMANOVA) when thesample sizes of compared treatment groups were balanced andusing analyses of similarity (ANOSIM) when they were not (7–10). For these permutation analyses, the application of Bonfer-roni corrections to planned comparisons is overly conservativeand substantially increases the likelihood of type II statistical er-rors (11, 12). When differences in profile structure were found,

similarity percentage (SIMPER) analyses were performed toelucidate the contributions of specific OTUs or VFAs to thosedifferences (10). These analyses have been provided as DatasetS1. To maximize interpretability, SIMPER analyses of OTUprofiles were performed using untransformed abundance data.Covariance of the OTU and VFA profiles of pastes was evaluatedthrough Mantel tests using the Bray–Curtis similarity index (7).Permutational tests of multivariate dispersions (PERMDISP)were used to evaluate whether treatment groups differed in thedegree to which their pastes varied with respect to OTU or VFAprofiles (13). These analyses were done using the Bray–Curtissimilarity index and permutation of the least-absolute-deviationsfrom spatial medians. All permutation tests (i.e., NPMANOVA,ANOSIM, Mantel test, PERMDISP) were conducted using 9,999permutations. Throughout, univariate parameters are presentedas means ± SD. Statistical and graphical analyses were completedusing PAST software (v. 2.17) (14), except for PERMDISP anal-yses, which were completed using PERMDISP2 (13), and heatmaps, which were created using Matrix2png (15).

1. Wagner AP, Creel S, Frank LG, Kalinowski ST (2007) Patterns of relatedness andparentage in an asocial, polyandrous striped hyena population. Mol Ecol 16(20):4356–4369.

2. Binladen J, et al. (2007) The use of coded PCR primers enables high-throughputsequencing of multiple homolog amplification products by 454 parallel sequencing.PLoS ONE 2(2):e197.

3. Kanagawa T (2003) Bias and artifacts in multitemplate polymerase chain reactions(PCR). J Biosci Bioeng 96(4):317–323.

4. Schloss PD, et al. (2009) Introducing mothur: Open-source, platform-independent,community-supported software for describing and comparing microbial communities.Appl Environ Microbiol 75(23):7537–7541.

5. Rastogi R, Wu M, Dasgupta I, Fox GE (2009) Visualization of ribosomal RNA operoncopy number distribution. BMC Microbiol 9(1):208.

6. Lee ZM, Bussema C, 3rd, Schmidt TM (2009) rrnDB: Documenting the number of rRNAand tRNA genes in bacteria and archaea. Nucleic Acids Res 37(Database issue):D489–D493.

7. McCune B, Grace JB (2002) Analysis of Ecological Communities (MjM Software Design,Gleneden Beach, Oregon).

8. Ramette A (2007) Multivariate analyses in microbial ecology. FEMS Microbiol Ecol62(2):142–160.

9. Anderson M (2001) A new method for non-parametric multivariate analysis ofvariance. Austral Ecol 26:32–46.

10. Clarke KR (1993) Non-parametric multivariate analysis of changes in communitystructure. Aust J Ecol 18:117–143.

11. Clarke KR, Warwick RM (2001) Change in Marine Communities: An Approach toStatistical Analysis and Interpretation (PRIMER-E Ltd., Plymouth, UK), 2nd Ed.

12. Hammer O (2011) PAST: PAleontological STatistics Manual (Natural History Museum,Oslo) 2.07 Ed.

13. Anderson MJ (2006) Distance-based tests for homogeneity of multivariatedispersions. Biometrics 62(1):245–253.

14. Hammer O, Harper DAT, Ryan PD (2001) PAST: PAleontological STatistics softwarepackage for education and data analysis. Palaeontol Electronica 4(1):1–9.

15. Pavlidis P, Noble WS (2003) Matrix2png: A utility for visualizing matrix data.Bioinformatics 19(2):295–296.

Theis et al. www.pnas.org/cgi/content/short/1306477110 2 of 9

Fig. S1. Map of Kenya showing the locations of the sampled hyena populations. We collected pastes directly from the subcaudal scent pouches of spottedhyenas in the Masai Mara National Reserve (MMNR), Kenya. The Reserve consists of rolling grassland, scattered brushland, and narrow stretches of riparianforest. We sampled 19 adult spotted hyenas (nine males/10 females) from the general MMNR population, representing at least 10 distinct clans. We addi-tionally sampled 21 adult spotted hyenas (seven immigrant males/seven lactating females/seven pregnant females) that were members of the same socialgroup, the Talek clan, within the MMNR. We collected pastes directly from the scent pouches of striped hyenas in the Laikipia and Shompole regions of Kenya.The northern portion of the Laikipia District includes the Loisaba, Mpala, and Kisima livestock ranches. It is characterized by semiarid grassland with scatteredshrubland. In this area, we collected pastes from 20 striped hyenas (eight males/12 females), of which seven were juveniles. We also sampled striped hyenasfrom two properties in the Shompole region of Kenya (Shompole and Olkirimatian group ranches). This area consists of semiarid plains and riverine forest. Inthis area, we collected pastes from 13 striped hyenas (six males/seven females), of which four were juveniles.

Theis et al. www.pnas.org/cgi/content/short/1306477110 3 of 9

Fig. S2. Scanning electron micrographs (SEMs) showing that the scent glands of spotted and striped hyenas are inhabited by symbiotic (i.e., resident) mi-crobes. SEMs reveal bacteria-sized cocci and bacilli in the pastes of (A) male spotted, (B) female spotted, (C) male striped, and (D) female striped hyenas. Arrowshighlight examples of apparent binary fission events, which are indicative of actively growing and therefore metabolically active microbes. (Magnification:5,000×.) Detailed SEM methodology is provided in SI Methods.

Theis et al. www.pnas.org/cgi/content/short/1306477110 4 of 9

Fig. S3. Differences in the composition and structure of bacterial communities in the pastes of spotted and striped hyenas. nMDS plots showing differences inthe (A) composition (Jaccard similarity index) and (B) structure (Bray–Curtis similarity index) of the paste bacterial communities of spotted hyenas in the MMNR(n = 19) and Shompole (n = 2) and of striped hyenas in Shompole (n = 13). Light blue shading of striped hyena paste samples indicates that the samples wereobtained from juveniles. For these particular plots, each sample’s dataset was iteratively subsampled 15 times to a uniform depth of 1,272 sequences, and meanabundance values for the sample’s OTUs were calculated. Data were log-transformed before analyses.

Theis et al. www.pnas.org/cgi/content/short/1306477110 5 of 9

Fig. S4. Correlations among specific bacteria (OTUs) and VFAs in hyena pastes. Heat maps illustrating the strength and direction of linear correlationsbetween prominent OTUs (i.e., top 15 based on average sequence abundance) and VFAs in pastes among (A) spotted hyenas in the general MMNR population,(B) spotted hyenas in the Talek clan, and (C) striped hyenas in Laikipia and Shompole.

Theis et al. www.pnas.org/cgi/content/short/1306477110 6 of 9

Table S1. Variation in the alpha diversities of bacterial communities in the pastes of spotted and striped hyenas

Sample groupSample

sizeGood’s

coverageNumber of

OTUsChao1

richnessSimpson index

(1 – D)

A. Alpha diversities among general populations of adult spotted and striped hyenas*

Spotted hyena (MMNR) 19 0.98 ± 0.01 61.3 ± 29.0 117.0 ± 79.5 0.57 ± 0.27Striped hyena (Laikipia) 13 0.99 ± 0.00 30.1 ± 3.8 63.7 ± 37.4 0.76 ± 0.07Striped hyena (Shompole) 9 0.99 ± 0.00 38.6 ± 4.5 73.1 ± 20.1 0.70 ± 0.05

B. Alpha diversities among immigrant male, lactating and pregnant female spotted hyenas in Talek†

Immigrant male 7 0.98 ± 0.01 81.1 ± 16.6 150.2 ± 55.2 0.84 ± 0.07Lactating female 7 0.98 ± 0.01 60.3 ± 13.1 112.6 ± 41.1 0.68 ± 0.23Pregnant female 7 0.99 ± 0.01 46.7 ± 26.8 89.3 ± 70.7 0.42 ± 0.22

C,D. Rank abundance plots of (C) MMNR, Laikipia, Shompole and (D) Talek adult hyenas‡

1

10

100

1000

1 6 11 16 21 26

Sequ

ence

s re

cove

red

(log

scal

e)

OTU rank

1

10

100

1000

1 6 11 16 21 26

Sequ

ence

s re

cove

red

(log

scal

e)

OTU rank

*Comparison of the richness (Chao1 richness estimator) and evenness (Simpson evenness index) of paste bacterial communities among adult spottedand striped hyenas. Good’s coverage is provided as an indicator of the appreciable depth of sampling in this study.†Comparison of the richness and evenness of paste bacterial communities among immigrant male, lactating female and pregnant female spottedhyenas in the Talek clan.‡Rank abundance plots further illustrating degrees of bacterial community evenness in the pastes of adult spotted and striped hyenas.§Data are presented as means ± SD.

Theis et al. www.pnas.org/cgi/content/short/1306477110 7 of 9

Table S2. Phylum- and genus-level classification of 16S rRNA gene sequences obtained from the pastes of adult spotted andstriped hyenas

phylumBacterial

Percentageof sequences in

phylum (# of OTUs)

Percentage of those sequences

assigned to a genus Genera to which sequences were assigned

A. Spotted hyenas in the MMNR (N = 19)Firmicutes 95.59 ± 2.10 (256) 76.09 ± 17.09 Anaerococcus, Bacillus, Blautia, Butyricicoccus, Clostridium

IV, Clostridium sensu stricto, Clostridium XI, Coprobacillus, Dehalobacter, Dialister, Erysipelotrichaceae incertae sedis, Eubacterium, Facklamia, Fastidiosipila, Finegoldia, Lachnospiracea incertae sedis, Lactobacillus, Moryella, Murdochiella, Negativicoccus, Peptoniphilus, Solobacterium, Staphylococcus, Streptococcus, Tissierella

Actinobacteria 3.73 ± 2.08 (30) 99.15 ± 2.31 Actinoplanes, Arthrobacter, Atopobium,Auritidibacter, Corynebacterium, Curtobacterium, Dermabacter, Microbacterium, Propionibacterium, Slackia, Solirubrobacter, Terrabacter

Bacteroidetes 0.31 ± 0.49 (19) 86.59 ± 16.05 Anaerorhabdus, Bacteroides, Hallella, Porphyromonas

Fusobacteria 0.18 ± 0.30 (5) 100 ± 0.00 Cetobacterium, FusobacteriumProteobacteria 0.17 ± 0.20 (32) 81.46 ± 30.91 Altererythrobacter, Aquabacterium, Bdellovibrio,

Brevundimonas, Buttiauxella, Escherichia, Hyphomicrobium, Massilia, Methylobacterium, Methylosinus, Microvirga, Moraxella, Neisseria, Oligotropha, Pseudomonas, Ralstonia, Ramlibacter, Raoultella, Rubrivivax, Sphingomonas, Sutterella, Tepidamorphus

Tenericutes 0.01 ± 0.06 (1) 100 AnaeroplasmaAcidobacteria 0.01 ± 0.02 (2) 100 Gp3, Gp7Chloroflexi < 0.01 ± 0.01 (1) 0 n/aunclassified < 0.01 ± 0.01 (1) n/a n/a

B. Striped hyenas in Laikipia (N = 13)Firmicutes 99.96 ± 0.07 (71) 0.75 ± 0.42 Dialister, Erysipelotrichaceae incertae sedis, Fastidiosipila,

Peptoniphilus, StaphylococcusBacteroidetes 0.04 ± 0.07 (1) 100 Porphyromonas

C. Striped hyenas in Shompole (N = 9)Firmicutes 98.83 ± 1.81 (85) 0.67 ± 0.46 Anaerosphaera, Clostridium sensu stricto, Fastidiosipila, Bacteroidetes 1.16 ± 1.79 (9) 100 ± 0.00 Peptoniphilus, PorphyromonasProteobacteria 0.01 ± 0.03 (2) 50 Devosia

Taxonomic classification of OTUs and their respective sequences were determined using Ribosomal Database Project’s Classifier (training set 9;bootstrap cutoff of 50%) (1, 2). Data are presented as means ± SD.

1. Wang Q, Garrity GM, Tiedje JM, Cole JR (2007) Naive Bayesian classifier for rapid assignment of rRNA sequences into the new bacterial taxonomy. Appl Environ Microbiol 73(16):5261–5267.

2. Claesson MJ, et al. (2009) Comparative analysis of pyrosequencing and a phylogenetic microarray for exploring microbial community structures in the human distal intestine. PLoS ONE4(8):e6669.

Theis et al. www.pnas.org/cgi/content/short/1306477110 8 of 9

Table S3. Consideration of potential correlations between the percent abundances of individual VFAs in hyena paste and timein storage before chemical analysis

Analysis (N) A Pr IB B IP Pe Ihex Hex Hep

MMNR(19)

0.2053(0.3992)

0.2386(0.3253)

0.3000(0.2121)

0.3053(0.2038)

0.2035(0.4034)

-0.1158(0.6369)

0.2000(0.4117)

-0.1421(0.5617)

0.0175(0.9432)

Talek(21)

-0.1675(0.4679)

0.0351(0.8801)

-0.1533(0.5072)

0.2455(0.2835)

-0.1052(0.6500)

0.0766(0.7413)

0.3507(0.1191)

0.1273(0.5825)

-0.0104(0.9644)

Talek males(7)

-0.1786(0.7131)

0.2143(0.5948)

-0.2857(0.4976)

0.1429(0.7131)

-0.5357(0.2357)

0.4643(0.3024)

0.5000(0.2667)

0.1429(0.7131)

-0.1429(0.7131)

Talek lactating(7)

-0.0714(0.8397)

-0.1071(0.7825)

-0.0714(0.8397)

0.3214(0.4444)

0.0357(0.9064)

-0.5714(0.1667)

0.2500(0.5948)

0.2500(0.5948)

0.1429(0.7131)

Talek pregnant(7)

-0.5357(0.2357)

-0.1786(0.7131)

-0.5714(0.1667)

0.2143(0.5948)

0.2500(0.5948)

0.3929(0.3956)

0.6429(0.1095)

0.4286(0.3024)

0.1429(0.7131)

Laikipia (20)

-0.0655(0.7838)

0.2485(0.2908)

0.2417(0.3046)

0.5286(0.0166)

-0.3923(0.0871)

0.4345(0.0556)

0.2538(0.2803)

0.5949(0.0057)

0.1747(0.4613)

Laikipia adults(13)

0.3559(0.2327)

0.5490(0.0520)

0.1710(0.5764)

0.4966(0.0843)

-0.5131(0.0729)

0.4993(0.0824)

0.2428(0.4242)

0.3724(0.2102)

-0.1628(0.5952)

Shompole (13)

-0.2528(0.4048)

0.4154(0.1581)

0.1978(0.5171)

0.4396(0.1329)

0.1099(0.7208)

0.2198(0.4706)

0.41210.1618)

0.5769(0.0390)

0.1374(0.6545)

Shompoleadults(9)

0.0667(0.8432)

0.7280(0.0321)

0.3833(0.3125)

0.4167(0.2499)

-0.3000(0.4101)

0.3667(0.3363)

0.5667(0.1206)

0.6833(0.0503)

-0.1833(0.6436)

Spearman rank correlation coefficients (r) and the results of correlation tests are provided. P values are in parentheses. If Bonferroni correctionsare modestly applied to each analysis separately (α = 0.05/9 = 0.0056), no significant correlations are found. A, acetic acid; B, butanoic acid; Hep,heptanoic acid; Hex, hexanoic acid; IB, isobutanoic acid; Ihex, isohexanoic acid; IP, isopentanoic acid; Pe, pentanoic acid; and Pr, propanoic acid.

Other Supporting Information Files

Dataset S1 (XLSX)Dataset S2 (XLSX)

Theis et al. www.pnas.org/cgi/content/short/1306477110 9 of 9