Dispersal of Mycobacterium tuberculosis via theCanadian fur tradeCaitlin S. Pepperella,1,2, Julie M. Grankab,1, David C. Alexanderc, Marcel A. Behrd, Linda Chuie, Janet Gordonf,Jennifer L. Guthriec, Frances B. Jamiesonc, Deanne Langlois-Klasseng, Richard Longg, Dao Nguyend, Wendy Wobeserh,and Marcus W. Feldmanb

aDivision of Infectious Diseases and bDepartment of Biology, Stanford University, Stanford, CA 94305; cOntario Agency for Health Protection and Promotion,Toronto, ON, Canada M9P 3T1; dDepartment of Medicine, McGill University, Montreal, QC, Canada H3G 1A4; eProvincial Laboratory for Public Health,Edmonton, AB, Canada T6G 2J2; fSioux Lookout First Nations Health Authority, Sioux Lookout, ON, Canada P8T 1B8; gDepartment of Medicine, Universityof Alberta, Edmonton, AB, Canada T6G 2J3; and hDivision of Infectious Diseases, Queen’s University, Kingston, ON, Canada K7L 3N6

Edited by Simon A. Levin, Princeton University, Princeton, NJ, and approved March 3, 2011 (received for review November 9, 2010)

Patterns of gene flow can have marked effects on the evolutionof populations. To better understand the migration dynamics ofMycobacterium tuberculosis, we studied genetic data from Euro-pean M. tuberculosis lineages currently circulating in Aboriginaland French Canadian communities. A singleM. tuberculosis lineage,characterized by the DS6Quebec genomic deletion, is at highest fre-quency among Aboriginal populations in Ontario, Saskatchewan,and Alberta; this bacterial lineage is also dominant among tuber-culosis (TB) cases in French Canadians resident in Quebec. Substan-tial contact between these human populations is limited to aspecific historical era (1710–1870), during which individuals fromthese populations met to barter furs. Statistical analyses of extantM. tuberculosis minisatellite data are consistent with Quebec as asource population for M. tuberculosis gene flow into Aboriginalpopulations during the fur trade era. Historical andgenetic analysessuggest that tiny M. tuberculosis populations persisted for ∼100 yamong indigenous populations and subsequently expanded in thelate 19th century after environmental changes favoring the patho-gen. Our study suggests that spread of TB can occur by two asyn-chronous processes: (i) dispersal of M. tuberculosis by minimalnumbers of human migrants, during which small pathogen popu-lations are sustained by ongoing migration and slow disease dy-namics, and (ii) expansion of the M. tuberculosis populationfacilitated by shifts in host ecology. If generalizable, these migra-tion dynamics can help explain the low DNA sequence diversityobserved among isolates of M. tuberculosis and the difficulties inglobal elimination of tuberculosis, as small, widely dispersed path-ogen populations are difficult both to detect and to eradicate.

genetic drift | demography | Mycobacterium tuberculosis genetics |Native Americans | Approximate Bayesian Computation

Migration of human pathogens between host populations is ofclear biomedical interest, because global spread of infectious

diseases is predicated on pathogen dispersal. For obligate humanpathogens, the paths of migration are shaped by social networksamong hosts and, at a broader scale, by human migrations andpatterns of comingling. The shape and magnitude of these pro-cesses may be inferred from genetic data from both extant pop-ulations of humans (1, 2) and microorganisms carried by humans(3). Statistical genetic inferences about pathogen migration havebeen buttressed by a variety of independent data types, such aslinguistic (3), epidemiological (4), and ecological (5).Human tuberculosis (TB) is caused by Mycobacterium tuber-

culosis. By current estimates, M. tuberculosis infects one third ofthe world’s population; a minority of these infections progress todisease and account for 9–10 million new, transmissible cases peryear (6). Genetic data from M. tuberculosis are characterized bylow DNA sequence diversity (7) and significant population sub-division, both at continental (8, 9) and fine geographic scales(10). TB is characterized by variable transmissibility, dependingon environmental conditions, and irregular disease dynamics (11,12). Variation in temporal dynamics of disease is introduced by

the phenomenon of clinical latency, whereby a host may beinfected with M. tuberculosis for decades before reactivating theinfection, becoming ill, and thus able to transmit the infectionto others. These features and other aspects of disease ecologywould be expected to affect migration of M. tuberculosis amonghost populations. Broad outlines of M. tuberculosis migratoryhistory have been described based on the geographic range ofspecific bacterial clades and lineages (9). However, very little isknown about M. tuberculosis migratory patterns and rates, pop-ulation dynamics during migration, or effects of migratory his-tory on M. tuberculosis population genetic diversity.In this paper, we present evidence suggesting that M. tubercu-

losis dispersed into Canadian Aboriginal populations as a result ofcontact with European fur traders in the 18th century. Althoughcontact between the populations involved small numbers ofindividuals, it was extensive enough to result in the developmentof the Métis society of both European and Aboriginal ancestry.(Additional historical details are given in SI Background Mate-rial.) Large-scale TB epidemics were not evident among WesternCanadian Aboriginal populations until the late 19th and 20thcenturies (10, 13–15), suggesting that epidemics may be uncou-pled from the process of M. tuberculosis dispersal.Our analyses of historical, epidemiological, andM. tuberculosis

genetic data suggest that (i)M. tuberculosismay be spread by smallnumbers of human migrants; (ii) M. tuberculosis populations canpersist at low levels over historical time scales; (iii) these smallbacterial populations may be sustained by ongoing migration andpossibly by slow disease dynamics; and (iv) shifts in host ecologyfavoring the pathogen may be accompanied by bacterial pop-ulation expansions, with marked effects on genetics of M.tuberculosis populations.

ResultsResults of M. tuberculosis population screening for lineage-defining polymorphisms (16, 17) are shown in Fig. 1 and Table S1.DS6Quebec is themost frequent lineage among bacteria fromall fourpopulations: the French Canadian population of Quebec (QU),and Aboriginal populations in Ontario (ON), Saskatchewan (SK),and Alberta (AB). Within Saskatchewan, DS6Quebec was found inall of eight intraprovincial regions (fine-scale geographic data areavailable only for this population; see ref. 10 for definitions ofwithin-Saskatchewan regions). Frequencies of otherM. tuberculosislineages were more variable among populations. A minimum

Author contributions: C.S.P., J.M.G., and M.W.F. designed research; C.S.P. and J.M.G.performed research; C.S.P., D.C.A., M.A.B., L.C., J.G., J.L.G., F.B.J., D.L.-K., R.L., D.N., andW.W. contributed new reagents/analytic tools; C.S.P., J.M.G., and M.W.F. analyzed data;and C.S.P., J.M.G., and M.W.F. wrote the paper.

The authors declare no conflict of interest.

This article is a PNAS Direct Submission.1C.S.P. and J.M.G. contributed equally to this work.2To whom correspondence should be addressed. E-mail: pepperc@stanford.edu.

This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10.1073/pnas.1016708108/-/DCSupplemental.

6526–6531 | PNAS | April 19, 2011 | vol. 108 | no. 16 www.pnas.org/cgi/doi/10.1073/pnas.1016708108

spanning tree, based on 12 minisatellite loci from bacteria with theDS6Quebec polymorphism, reveals a distinctly star-like networkstructure (Fig. 2), with little evidence of population differentiation.

Networks based on minisatellite data within other common line-ages (Rd 182, Rd 219, andH37Rv-like) were not star-like (Fig. S1).Lineage-specific patterns also were evident in analyses of pop-

A

B

Q U É B E C

O N T A R I O

M A N I T O B A

SASKATCHEWAN

ALBERTA

COLUMBIA

BRITISH

NORTHWEST TERRITORIES

H u d s o n B a y

N U N A V U T

Fur Trade Transportation NetworksCanoe Route, E-NW

Canoe Route, Hudson Bay accessRailwaySteamboat Route

Regional character of trade, ca. 1920

Remote/ Traditional

M.tuberculosis LineagesDS6Quebec

H37Rv-like

Non Euro-American

Rd 115

Rd 174

Rd 182

Rd 183

Rd 219

16081710

17601820

18851930

New France Fur tradersreachNorthwest

Britishconquest

HBCmerger

Rail (CPR).Buffalodepleted.

Bushplanes

M.tb dispersal

M.tb expansion

Fig. 1. Fur trade geography, regional frequencies ofM. tuberculosis lineages, and historical timeline. (A) Map of Canada from Natural Resources Canada (49).The main fur trade canoe route from the St. Lawrence River to the Beaufort Sea (Montreal route) is shown in light blue. Canoe routes from Hudson’s Bay tothe interior are shown in red. Geography of canoe routes is based on a map by A. Ray in ref. 24 and was checked against a map of fur trade posts (50).Geography of railways, steamship lines, and areas classified as remote/traditional on the basis of archival evidence (all ca. 1920) are from figure 38 in ref. 21and ref. 24. Proportional lineage frequencies of M. tuberculosis isolates are shown as pie charts in the corresponding province (see also Table S1). Nomen-clature of M. tuberculosis lineages is from refs. 16 and 17. Pie charts are the same size for clarity, although total sample sizes differed among populations.Lineage frequencies were unavailable for the MB population. (B) Events shown in the timeline are the founding of New France (Quebec) in 1608; incursion offur traders to the Northwest around 1710; British conquest of New France (and end of migration from France to Quebec) in 1760; merger of North WestCompany and Hudson’s Bay Company (HBC) in 1820 (with subsequent abandonment of the Montreal route in favor of Hudson’s Bay routes); completion ofthe Canadian Pacific Railway (CPR) in 1885, by which time Western buffalo herds were severely depleted; and, finally, widespread use of bush planes to reachremote areas, starting in the 1930s. Gray boxes indicate the estimated timing of processes in M. tuberculosis demographic history: dispersal of M. tuberculosisto indigenous populations (light gray), and expansion of M. tuberculosis populations as a result of shifts in host ecology (dark gray).

Table 1. Within- and between-lineage AMOVA based on 12 minisatellite loci forM. tuberculosisfrom Quebec, Ontario, Saskatchewan, and Alberta

Bacterial lineage Source of variation D.f. SSVariance

component % Variation FST

DS6Quebec Among populations 3 16.25 0.04 3.81 0.04Within populations 583 521.26 0.89 96.19

H37Rv-like Among populations 3 181.68 0.89 51.56 0.52Within populations 306 256.44 0.84 48.44

Rd 182 Among populations 3 8.26 0.33 22.26 0.22Within populations 59 67.95 1.15 77.74

Rd 219 Among populations 3 14.28 0.25 20.97 0.21Within populations 65 60.68 0.93 79.03

All lineages Among populations 3 75.54 0.10 7.95 0.08Within populations 1,040 1227.16 1.18 92.05

D.f., degrees of freedom; FST, fraction of variation among populations; SS, sum of squares.

Pepperell et al. PNAS | April 19, 2011 | vol. 108 | no. 16 | 6527

EVOLU

TION

ulation genetic structure (Analysis of Molecular Variance,AMOVA; Table 1). Overall, a modest level of differentiation wasobserved among populations [fraction of minisatellite variationamong populations (FST) = 0.08]. However, differentiation variedsubstantially between lineages, with the least variation seen with-in the DS6Quebec lineage (FST = 0.04) and the greatest withinthe H37Rv-like lineage (FST = 0.52). These results suggest that(i) the DS6Quebec lineage was introduced into these populations inthe relatively remote past, resulting in a similarly high frequencyacross populations; (ii)M. tuberculosis belonging to the DS6Quebec

lineage were, or are, exchanged freely among study populations;(iii) theDS6Quebec lineage underwent rapid population expansion,reflected in a star-like network topology, possibly coincident withits introduction into new host populations; and (iv) there has beentemporal and/or spatial variation in patterns of bacterial migra-

tion among human populations, reflected in variable differentia-tion within bacterial lineages.Although the Quebec population is now geographically and

socially isolated from Aboriginal populations in Ontario, Sas-katchewan, and Alberta, there is a history of comingling in thecontext of the Canadian fur trade (Fig. 1). Table 2 outlines theexpected history of the DS6Quebec lineage, assuming it was in-troduced to Canada by a French immigrant to Quebec and dis-persed to Aboriginal populations via fur trade transportation andsocial networks. Estimates derived from M. tuberculosis minis-atellite data of the time to most recent common ancestor(TMRCA) (18) and divergence time between bacterial pop-ulations (TD) (19) suggest that genetic exchange between QU andWestern Aboriginal M. tuberculosis populations occurred over∼100 y; timing of this exchange is consistent with historical humanmigrations connected with the trade in furs. (Additional estimatesare given in SI Results.)Levels of overall genetic diversity of M. tuberculosis restriction

fragment length polymorphism (RFLP) (20) haplotypes from fourpopulations are shown in rarefaction curves in Fig. 3A. Publisheddata were available from Manitoba (MB) and are included alongwith data from QU, SK, and AB; data from ON are not includedbecause of the small sample size (SI Results). The number of dis-tinct haplotypes is highest in QU, consistent with its being a sourcepopulation for pathogen migration. Levels of diversity are lowestin SK and MB, with intermediate diversity in AB. Diversity thusdoes not decrease in an east-to-west pattern, as we might expectwith a serial founder effect (1) originating in Quebec.Epidemics of TB among Canadian indigenous populations

occuring post-European contact have been associated with dra-matic social, economic, and environmental changes that char-acterized the industrial era (late 19th century and beyond; SIBackground Material). There is evidence of regional differences inthe pace of these changes: Published analyses of archival fur tradedocuments indicate that some indigenous populations remainedremote from the cash-based industrial economy as late as the1920s (21). Regions within Saskatchewan and Manitoba fall intothis historically remote/traditional category, whereas Albertadoes not contain any such regions (Fig. 1A). Later development oftransportation networks (starting with the widespread use of bushplanes in the 1930s) allowed commercial development of pre-viously remote regions (21).Given associations between industrialization, attendant shifts

in host ecology, and TB epidemic expansion among indigenouspopulations (SI Background Material), we hypothesized thatM. tuberculosis diversity in Saskatchewan and Manitoba was lowrelative to Alberta because of the more recent expansion of the

QU

ON

SK

AB

1 repeat

Fig. 2. Minimum spanning tree of minisatellite haplotypes (12 loci) fromM. tuberculosis within the DS6Quebec lineage, in QU, ON, SK, and AB pop-ulations. The size of nodes is proportional to the frequency of TB cases as-sociated with that M. tuberculosis haplotype. Scale is indicated by the line atthe bottom of the figure, which represents a difference of one minisatelliterepeat. Minisatellites were not available from the MB population.

Table 2. Genetic timing estimates for DS6Quebec M. tuberculosis lineage in calendar years

Parameter Historical correlate Point estimate(s)* Confidence interval† Historical prediction‡

TMRCA (QU) French migration to Quebec 1740 1709–1771 1608–1760TMRCA (SK) Fur trade expansion west 1797 1777–1815 1730–1870TMRCA (AB) 1779 1751–1804 1750–1870TD (QU/SK) Separation of populations 1789, 1884, 1910 1788–1979 1870TD (QU/AB) 1779, 1901, 1885 1805–1966 1870

*Three point estimates are shown for TD. The first is the earliest bound for divergence, where a variance of 0 is assumed for minis-atellite repeat number in the ancestral population. For the second estimate, we assume that variance in the ancestral population isequal to the variance of haplotypes found presently in the QU population and shared with the founded population (SK or AB). Thethird point estimate is based on an assumption that variance in the ancestral population was equal to the variance found presently inhaplotypes in the founded population (SK or AB) that are also found in QU.†Results presented for TMRCA assume a star-like genealogy. TMRCA confidence intervals are larger if constant population size orexponential growth are assumed (SI Results).‡Estimate of timing of M. tuberculosis population events, based on historical (nongenetic) data. References for date ranges are Char-bonneau (51) for timing of migration to Quebec from France and Innis (24) for other dates. The earlier bounds for fur trade expansioninto Saskatchewan and Alberta are based on the time of establishment of fur trade posts in these regions. Note that this timing is laterthan expansion into the Northwest as a whole (ca. 1710), which includes Western Ontario and Manitoba. Timing of separation ofpopulations is based on analyses of Innis (24) indicating that the Hudson’s Bay Company shifted to less labor-intensive methods ofextracting furs (and thus had no need of French Canadian voyageurs), starting about this time. Analyses of interprovincial migration fromthe 1870s to the early 20th century are also consistent with minimal migration of French Canadians to the Western provinces (25).

6528 | www.pnas.org/cgi/doi/10.1073/pnas.1016708108 Pepperell et al.

M. tuberculosis population in remote/traditional areas. To test thishypothesis, minisatellite and RFLP haplotypes from SK weredivided into historically “remote” (RS) and “nonremote” (NRS)regions according to the classification of trading districts outlinedin ref. 21. Consistent with this classification, diversity in NRSregions is similar to that in AB (Alberta contains no remoteregions), with the lowest diversity observed in RS (Fig. 3B and Fig.S2). Permutation procedures identified RS and QU as havinglower and greater diversity, respectively, than would be expectedunder a random distribution of minisatellite haplotypes amongpopulations (Fig. S3).We analyzed DS6Quebec minisatellite data with rejection sam-

pling, an Approximate Bayesian Computation (ABC) method, toassess whether a demographic model allowing an expansion in RS(in which the historical bacterial effective population size, Ne, wassmaller than the current Ne) is more likely than a null model ofconstant size.We estimated the parameterω, the ratio of historicalNe to contemporary Ne (Materials and Methods and SI MaterialsandMethods). Results using thismethod indicate that the posteriorprobability of the expansion model (versus the null constant-sizemodel) is 1; based on simulations under the null model, the P valueof this posterior probability is 0. We estimate ω to be 0.08 (95%credibility interval 0.01–0.29; Fig. S4A). Translating these resultsto absolute numbers (current Ne = 30) generates an estimate ofeffective number of TB cases in RS before 1930 (the time fixed forthe expansion, based on the social history described above) equalto 2.35 individuals (95% credibility interval 0.35–8.80). SI Resultsgives parameter estimations under alternative mutation rateassumptions, all of which reject the constant-size model in favor ofthe expansion model.Based on the size and dynamics of the population involved in

the French (Montreal-based) trade in the West before 1870 andthe prevalence of TB in European cities in the 18th century, weestimated the absolute number of M. tuberculosis infectionstransmitted to Western indigenous populations (SI Results andTable S2). The historical estimate of Nem, the product of pop-ulation size (Ne) and fraction of the indigenous M. tuberculosispopulation replaced by immigrants (m) per generation, is 0.16.In an island model, Nem <1 results in substantial populationdifferentiation caused by genetic drift (22). Estimates of Nemderived from bacterial minisatellite data were higher than thishistorical estimate: For pooled lineages, Nem derived fromFST (AMOVA) is 6.16, and the private alleles method of Bartonand Slatkin (23) generated an estimate of 3.62 (empirical 95%confidence interval 2.47–4.72).

DiscussionThe recent history of human migration to Western Canadainforms the analyses ofM. tuberculosis genetic data presented here.

Most importantly, substantial contact between French Canadianand Western indigenous populations is limited to a specific his-torical period, the fur trade era. The early bound on this period ofcontact is provided by the dates of incursion of fur traders into theWestern interior [∼1710 (24)], and the later bound is provided byhistorical and demographic analyses indicating that westward mi-gration of French Canadians ceased in the latter half of the 19thcentury for economic and other reasons (25).Based on the observed pattern of M. tuberculosis lineage fre-

quencies in this historical context, as well as bacterial populationgenetic diversities consistent with Quebec as a source population,genetic timing estimates, and patterns of genetic differentiationbetween populations, we infer that the DS6Quebec lineage wasdispersed to indigenous populations by French Canadian furtraders about a century before epidemic forms of TB were mani-fest in Western Aboriginal communities.Several features of the human migration associated with trade

in furs are noteworthy. First, the absolute number of humanmigrants was small: By our estimate, 5,419 individuals migratedfrom east to west during 160 y of trade betweenMontreal andWest-ern Canadian indigenous populations (SI Results). These earlymigrations are dwarfed by population movements of the late 19thand early 20th centuries. Facilitated by massive interprovincialand international migration to the Canadian prairies—fromwhich French Canadians were largely absent—the census ofWestern Canada grew from 110,000 in 1871 to 750,000 in 1911(26). Between 1900 and 1917, a total of 1,671,414 foreign-bornindividuals migrated to Western Canada from the United King-dom, the Ukraine, Russia, Germany, Austria, Hungary, Norway,Denmark, China, and elsewhere (27). Despite the enormousnumber of migrants from regions of high TB incidence, who wouldbe expected to introduce a broad range of Euro-American andEast Asian M. tuberculosis lineages (8), bacterial populations inWestern indigenous communities are dominated by the DS6Quebec

lineage, present at a frequency similar to that in the FrenchCanadian population of Quebec (Fig. 1A).The apparent lack ofM. tuberculosis gene flow from 19th century

homesteaders to indigenous populationsmay be explained by socialdistance between populations. International migrants to the prai-ries, particularly those facing language and cultural barriers to as-similation,were socially andgeographically isolated (27). By the late19th century, prairie indigenous populations also were sociallysegregated, at times forcibly (15, 27), from the non-Aboriginalpopulation. This isolation is in contrast to the earlier fur trade era,whichwas characterizedby intermarriageand trading collaborationsbetweenEuropean immigrants andFirstNations groups (ref. 28 andSI Background Material). Although contact between populationsinvolved small numbers of individuals, we speculate that close socialties between the sending and receiving host populations permitted

A B Fig. 3. Genetic diversity of M. tuberculosispopulations. (A) Number of distinct RFLP hap-lotypes as a function of the number of sampledchromosomes obtained using rarefaction(Materials and Methods). Populations includeQU,MB, SK, and AB; ON is not included becauseof its low sample size (SI Results). Every fourthdata point is presented for clarity. (B) RFLPhaplotype diversity (Shannon index), correctingfor sample size by repeatedly sampling the to-tal number of isolates in the smallest samplefrom each population (NRS, n = 123). Boxplotsindicate values obtained over all samples; valuefor the smallest sample (NRS) is indicated bya line. Populations are as in the left panel, ex-cept that SK is split into RS and NRS pop-ulations. Although Manitoba contains bothremote and nonremote regions (Fig. 1), de-tailed geographic data were not available forthe MB sample; diversity shown here is for theentire sample.

Pepperell et al. PNAS | April 19, 2011 | vol. 108 | no. 16 | 6529

EVOLU

TION

migration of M. tuberculosis through the fur trade. Given thattransmission of M. tuberculosis requires sustained, close contact[exemplified by efficient transmission in high-density shared livingenvironments such as prisons (11)], this observation is likely to begeneralizable, with structure of global M. tuberculosis populationsstrongly influenced by the social architecture of host populations.Although by no means a disease-free interval for Native peoples

(there were devastating epidemics of smallpox and other infectiousdiseases), epidemics of TB were not a feature of the fur trade era(1710–1870). TB epidemics among Western Canadian indigenouspopulations occurred later, starting in the late 1800s (10, 13–15).Expansion of rail and steamship networks into Western Canada inthe late 19th century permitted agricultural development, industrial-scale extraction of natural resources, development of governmentinstitutions, and mass immigration. TB epidemics were amonga chain of sequelae for Aboriginal populations that includeddisplacement, loss of traditional food sources, crowding, andinstitutionalization (ref. 29 and SI Background Material).Indigenous populations in some regions (e.g., Northern Sas-

katchewan and Manitoba; the shaded areas of Fig. 1A) remainedremote from the evolving industrial economy into the 1920s (21).We find that the genetic diversity ofM. tuberculosis in lowest in RSand is highest in QU (Fig. 3); we did not have detailed geographicdata that would allow us to classify M. tuberculosis strains fromManitoba according to the scheme we used in Saskatchewan (i.e.,NRS vs. RS). We note that molecular epidemiological studies ofother historically remote/traditional regions are consistent withlow genetic diversity of the M. tuberculosis population (30, 31).Differences in genetic diversity may result from two distinct

phenomena. First, the process ofM. tuberculosismigration is likelyto involve serial founder events, so that diversity is highest in thesource population (QU) and lower in the subsequently foundedpopulations (ON, RS, NRS, AB, and MB). Another explanation isthat QU, the oldest M. tuberculosis population (founded by 17thand 18th century migrants from France; Table 2), may be closer toan equilibrium distribution of haplotype frequencies; the youngestpopulations stillmay show evidence of a “founder flush” (32). As anhistorically remote/traditional population, RS was probably shiel-ded from at least some of the ecological antecedents of epidemicTB until the regional incursion of air travel networks in the 1930s(21) and thus had the most recent expansion in its pathogen pop-ulation. Demographic modeling with ABC demonstrated a veryhigh probability of expansion in the RSM. tuberculosis populationsince 1930; the magnitude of the bacterial expansion was ∼13-fold.Although our historical estimates of M. tuberculosis migration

rates were low (direct Nem = 0.16), the lack of genetic differentia-tion among populations (DS6Quebec FST = 0.04) and estimates ofNem from the genetic data (indirect Nem = 3.62) would suggest ahigh rate ofM. tuberculosismigration. Slatkin (33, 34) has observedthat indirect (genetic) estimates of gene flow may be higher thandirect measurements of population dispersal, especially in pop-ulations characterized by colonization-extinction-recolonizationdynamics, which can limit population differentiation even whenrates of migration are low. Taken together, our results suggestthat absolute numbers ofM. tuberculosismigration events were low(Nem<1, an order-of-magnitude estimate based on historical data),and patterns of genetic differentiation have been affected by un-stable dynamics of small, preindustrialM. tuberculosis populations.We have delineated two asynchronous processes involved in the

spread of TB from European immigrants to native Canadians (Fig.1B). The first is dispersal of the etiologic agent, M. tuberculosis,populations of which were sustained at very low levels (Ne ≅ 2) for∼100 y by small numbers of human migrants who had intimate,sustained contact with susceptible hosts. In addition to sustainedmigration, variable transmission dynamics of TB may have cush-ioned small bacterial populations against extinction (35). The sec-ondprocess isexpansionof thebacterial population, followinga shiftin host ecology favoring the pathogen.We find evidence of bacterialpopulation expansions in the DS6Quebec M. tuberculosis haplotypenetwork (Fig. 2), patterns of genetic diversity (Fig. 3), and co-alescent-based demographic analysis of bacterial minisatellite data.

This work is a study of a specific historical phenomenon, and itis unknown whether the observed patterns of M. tuberculosis mi-gration are applicable to other settings. Asynchronicity of patho-gen migration and epidemiological phenomena is likely to rendercontrol of TB more difficult and could help explain global per-sistence of the pathogen despite extensive efforts at eradication.Because M. tuberculosis is an obligate human pathogen that re-quires specific environmental conditions and sustained contact fortransmission, barriers to geographic spread of M. tuberculosis arelargely social and therefore somewhat fluid. Perseverance of tinybacterial populations would permit gradual accrual of a largegeographic range despite significant population subdivision. Asa result, by the time the spread of TB is obvious, M. tuberculosispopulations already may be well established.

Materials and MethodsPopulation Descriptions. Clinical isolates of M. tuberculosis in this study arefrom five Canadian populations: the French Canadian population of Quebec(QU), and Aboriginal populations in Ontario (ON), Manitoba (MB), Saskatch-ewan (SK), andAlberta (AB). TheQU sample (n = 297) is described in ref. 16. TheON sample (n = 45) derives from TB cases that occurred between 1997–2009 inFirst Nations communities in a single region of the province (total population∼25,000). MB data (n = 163) are from a published study (36): The total numberof TB cases, number of different bacterial RFLP haplotypes, number of sin-gletons, and the configuration distribution of the five most common M. tu-berculosis haplotypes from First Nations reserve communities in Manitoba(1992–1999) are reported. We made the conservative assumption that theremaining (unreported) haplotypes were doubletons. The SK sample (n = 444)is described in ref. 10. AB samples (n = 283) are from TB cases among FirstNations individuals in the province, excluding urban areas, from 1990–2008.

Genotyping Methods. Isolates of M. tuberculosis were genotyped by RFLP,based on the number and location of IS6110 elements (20). The number oftandem repeats at 12 minisatellite loci also was determined for each isolateusing a standard methodology (37). M. tuberculosis lineage typing, based ongenomic deletions (16, 17), was done with real-time PCR (details are given inSI Materials and Methods and Dataset S1).

Mutation Rate Estimate. We estimated a mutation rate (μ) per transmissiongeneration (10) for minisatellite loci based on simulations using empiricalestimates of intrahost M. tuberculosis population dynamics and in vitro esti-mates of mutations per cell doubling for bacterial minisatellite loci. Givena known per annum RFLP mutation rate (38), we made additional estimatesby examining the number of RFLP types per minisatellite haplotype andcomparing estimates of θ (θ = 2 Neμ, product of effective population size, Ne,and mutation rate, μ) from RFLP and minisatellite data. All estimates were ofa similar order of magnitude and were consistent with published estimates(39); we used 0.001 mutations per locus (SI Materials and Methods, Table S3).

Statistical Calculations. Genetic differentiation. AMOVA, implemented in Arle-quin version 3.5 (40), was used to calculate FST values from M. tuberculosisminisatellite haplotype data.Network analysis. BioNumerics 5.0 (Applied Maths) was used to generateminimum spanning trees ofM. tuberculosisminisatellite haplotypes. This pro-gram implements the Prim–Jarnik algorithm; the BURST priority rule maxi-mizing single- and double-locus variants, was used during network searches.Genetic timing estimates. We estimated TMRCA of the DS6Quebec lineage ineach population using minisatellite genotypes and the method of Ytime(18). We selected as the root the haplotype at the center of the DS6Quebec

network (Fig. 2), present at high frequency in QU, SK, and AB populations.We obtained bootstrap confidence intervals with Ytime assuming a star-likegenealogy and neutrality (SI Materials and Methods). Divergence times forDS6Quebec isolates between pairs of populations were estimated from min-isatellite genotypes by the TD estimator, which is robust to population sizechanges and weak gene flow (SI Materials and Methods) (19). This procedurerequires an estimate of the average variance in repeat number in the an-cestral population at the time of divergence (V0); we estimated TD usingthree different values of V0 for each pair of populations (Results and SIMaterials and Methods). Alternative mutation rate and demographicassumptions have little effect on our main conclusions (SI Results).Population genetic diversity. We assessed diversity of minisatellite and RFLPhaplotypes in each population, for all lineages, by calculating the number ofdistinct haplotypes using rarefaction (41), calculating haplotype diversitywith resampling to the lowest sample size to correct for unequal sample

6530 | www.pnas.org/cgi/doi/10.1073/pnas.1016708108 Pepperell et al.

sizes (42), and implementing a permutation procedure to assess whetherobserved diversities of minisatellite haplotypes could be explained bya random partitioning of isolates from all populations (details are given in SIMaterials and Methods).Migration. Under a simple island model, assuming an equilibrium populationwhere each “island” (QU, SK, and AB) has equal size Ne, we applied Slatkin’sprivate alleles method with minisatellite haplotype data from all lineages toestimate Nem, where m is the probability that an individual is a migrant ineach generation (33, 43). Nem estimates were converted to FST values, andvice versa, using the standard expectation for a haploid population FST = 1/(1+2Nm) (43) (SI Materials and Methods).

Demographic Modeling.Motivatedbytheresultsof thediversity calculations,weused minisatellite genotypes and rejection sampling, an ABC procedure, toestimate the factor (ω) by which the size of the DS6Quebec lineage in RemoteSaskatchewan (RS) before its likely expansion is related to its current size.

Assuming a constant population size after the expansion, we estimated thecurrent effective population size of RS as 544 (multiplying theharmonicmeanoffluctuating DS6Quebec TB case counts in RS from 1986–2004 by 18, the number ofyears over which exhaustive samples were obtained). This epidemiological esti-mate of Ne was interpreted in the standard population genetic manner (dis-cussed further in SI Materials and Methods). One million coalescent simulationsof minisatellite haplotypes under a population-expansion model were per-

formed using SIMCOAL 2.0 (44, 45). The time of expansion was fixed at 54 gen-erations, with ω drawn from a prior Uniform distribution on [0.01, 1]. Thesimulated data were summarized with four statistics (number of distinct andsingleton haplotypes, haplotype diversity, and mean variance in locus repeatsizes), which produced reliable estimates in our method validation (SI Materialsand Methods) and have been used in previous studies (2, 46–48). Values of ωresulting in simulations with summary statistics within a threshold Euclideandistance from those observed in RS were retained to produce a posterior distri-bution. We also estimated the posterior probabilities of the population expan-sionandnull constant-sizemodelsby theacceptanceprobabilitiesofeachmodel.

We tested the method’s performance using simulated data with knownexpansion sizes. The method is most powerful for pronounced expansions(ω ∼ 0.1); with high accuracy, it both can estimate ω and can distinguishbetween the population-expansion and constant-size models (SI Materialsand Methods and Fig. S5).

ACKNOWLEDGMENTS. We thank V. Hoeppner for providing data andbacterial strains, T. MacMillan and A.D. Popescu for data collection;A. Avelar, T. Van, E. Heinemeyer, and C.-Y. Wu for technical assistance; andG. Dolganov for help with real-time PCR assays. This work was supported byNational Institutes of Health Grants 5K08AI67458-2 (to C.S.P.) and GM28016(to M.W.F.). J.M.G. is the recipient of a National Science Foundation GraduateResearch Fellowship.

1. Ramachandran S, et al. (2005) Support from the relationship of genetic andgeographic distance in human populations for a serial founder effect originating inAfrica. Proc Natl Acad Sci USA 102:15942–15947.

2. Ray N, et al. (2010) A statistical evaluation of models for the initial settlement of theamerican continent emphasizes the importance of gene flow with Asia. Mol Biol Evol27:337–345.

3. Moodley Y, et al. (2009) The peopling of the Pacific from a bacterial perspective.Science 323:527–530.

4. Viboud C, et al. (2006) Synchrony, waves, and spatial hierarchies in the spread ofinfluenza. Science 312:447–451.

5. Biek R, Henderson JC, Waller LA, Rupprecht CE, Real LA (2007) A high-resolutiongenetic signature of demographic and spatial expansion in epizootic rabies virus. ProcNatl Acad Sci USA 104:7993–7998.

6. WHO (2009) Global tuberculosis control: A short update to the 2009 report. GlobalTuberculosis Control (World Health Organization, Geneva).

7. Achtman M (2008) Evolution, population structure, and phylogeography ofgenetically monomorphic bacterial pathogens. Annu Rev Microbiol 62:53–70.

8. Hirsh AE, Tsolaki AG, DeRiemer K, Feldman MW, Small PM (2004) Stable associationbetween strains of Mycobacterium tuberculosis and their human host populations.Proc Natl Acad Sci USA 101:4871–4876.

9. Hershberg R, et al. (2008) High functional diversity in Mycobacterium tuberculosisdriven by genetic drift and human demography. PLoS Biol 6:e311.

10. Pepperell C, etal. (2010) Bacterialgenetic signaturesofhumansocialphenomenaamongM. tuberculosis from an Aboriginal Canadian population.Mol Biol Evol 27:427–440.

11. Iseman MD (2000) A Clinician’s Guide to Tuberculosis (Lippincott Williams andWilkins, Philadelphia).

12. Aparicio JP, Castillo-Chavez C (2009) Mathematical modelling of tuberculosisepidemics. Math Biosci Eng 6:209–237.

13. Daschuk JW, Hackett P, MacNeil S (2006) Treaties and tuberculosis: First Nationspeople in late 19th-century western Canada, a political and economic transformation.Can Bull Med Hist 23:307–330.

14. Ferguson R (1928) Tuberculosis among the Indians of the Great Canadian Plains(Adlard and Sons Ltd., London).

15. Lux M (2001) Medicine that Walks: Disease, Medicine, and Canadian Plains NativePeople, 1880-1940 (Univ of Toronto Press, Toronto).

16. Nguyen D, et al. (2004) Genomic characterization of an endemic Mycobacteriumtuberculosis strain: Evolutionary and epidemiologic implications. J Clin Microbiol 42:2573–2580.

17. Gagneux S, et al. (2006) Variable host-pathogen compatibility in Mycobacteriumtuberculosis. Proc Natl Acad Sci USA 103:2869–2873.

18. Behar DM, et al. (2003) Multiple origins of Ashkenazi Levites: Y chromosome evidencefor both Near Eastern and European ancestries. Am J Hum Genet 73:768–779.

19. Zhivotovsky LA (2001) Estimating divergence time with the use of microsatellite geneticdistances: Impacts of population growth and gene flow. Mol Biol Evol 18:700–709.

20. van Embden JD, et al. (1993) Strain identification of Mycobacterium tuberculosis byDNA fingerprinting: Recommendations for a standardized methodology. J ClinMicrobiol 31:406–409.

21. RayAJ (1990)TheCanadianFur Trade in the IndustrialAge (UnivofTorontoPress, Toronto).22. Wright S (1931) Evolution in Mendelian populations. Genetics 16:97–159.23. Barton NH, Slatkin M (1986) A quasi-equilibrium theory of the distribution of rare

alleles in a subdivided population. Heredity 56:409–415.24. Innis HA (1930) The Fur Trade in Canada (Univ of Toronto Press, Toronto) Ed 1970

reprinted 1999 with a new introductory essay.25. Green A, MacKinnon M, Minns C (2005) Conspicuous by their absence: French

Canadians and the Settlement of the Canadian West. J Econ Hist 65:822–849.26. Urquhart MC, Buckley KAH eds (1965) Historical Statistics of Canada (Macmillan,

Toronto).

27. Waiser B (2005) Saskatchewan: A New History (Fifth House, Calgary,AB, Canada).28. Francis D (1985) Traders and Indians. The Prairie West Historical Readings, ed

Francis RD (Pica Pica, Edmonton, AB, Canada), pp 58–70.29. Friesen G (1984) The Canadian Prairies: A History (Univ of Toronto Press, Toronto).30. Kunimoto D, Chedore P, Allen R, Kasatiya S (2001) Investigation of tuberculosis

transmission in Canadian Arctic Inuit communities using DNA fingerprinting. Int JTuberc Lung Dis 5:642–647.

31. Nguyen D, et al. (2003) Tuberculosis in the Inuit community of Quebec, Canada. Am JRespir Crit Care Med 168:1353–1357.

32. Templeton A (2006) Population Genetics and Microevolutionary Theory (John Wiley &Sons, Hoboken, NJ).

33. Slatkin M (1985) Gene flow in natural populations. Annu Rev Ecol Syst 16:393–430.34. Slatkin M (1987) Gene flow and the geographic structure of natural populations.

Science 236:787–792.35. Blaser MJ, Kirschner D (2007) The equilibria that allow bacterial persistence in human

hosts. Nature 449:843–849.36. Blackwood KS, Al-Azem A, Elliott LJ, Hershfield ES, Kabani AM (2003) Conventional

and molecular epidemiology of tuberculosis in Manitoba. BMC Infect Dis, 10.1186/1471-2334-3-18.

37. Supply P, et al. (2006) Proposal for standardization of optimized mycobacterialinterspersed repetitive unit-variable-number tandem repeat typing ofMycobacteriumtuberculosis. J Clin Microbiol 44:4498–4510.

38. Tanaka MM, Rosenberg NA (2001) Optimal estimation of transposition rates ofinsertion sequences for molecular epidemiology. Stat Med 20:2409–2420.

39. Reyes JF, Tanaka MM (2010) Mutation rates of spoligotypes and variable numbers oftandem repeat loci in Mycobacterium tuberculosis. Infect Genet Evol 10:1046–1051.

40. Excoffier L, Lischer HEL (2010) Arlequin suite ver 3.5: A new series of programs toperform population genetics analyses under Linux and Windows. Molecular EcologyResources, in press.

41. Kalinowski ST (2004) Counting alleles with rarefaction: Private alleles and hierarchicalsampling designs. Conserv Genet 5:539–543.

42. Shannon CE (1948) A mathematical theory of communication. Bell Syst Tech J 27:379–423, 623–656.

43. Slatkin M, Barton NH (1989) A comparison of three indirect methods for estimatingaverage levels of gene flow. Evolution 43:1349–1368.

44. Excoffier L, Novembre J, Schneider S (2000) SIMCOAL: A general coalescent programfor the simulation of molecular data in interconnected populations with arbitrarydemography. J Hered 91:506–509.

45. Laval G, Excoffier L (2004) SIMCOAL 2.0: A program to simulate genomic diversityover large recombining regions in a subdivided population with a complex history.Bioinformatics 20:2485–2487.

46. Estoup A, Wilson IJ, Sullivan C, Cornuet JM, Moritz C (2001) Inferring populationhistory from microsatellite and enzyme data in serially introduced cane toads, Bufomarinus. Genetics 159:1671–1687.

47. Beaumont MA, Zhang W, Balding DJ (2002) Approximate Bayesian computation inpopulation genetics. Genetics 162:2025–2035.

48. Estoup A, Beaumont M, Sennedot F, Moritz C, Cornuet J-M (2004) Genetic analysis ofcomplex demographic scenarios: Spatially expanding populations of the cane toad,Bufo marinus. Evolution 58:2021–2036.

49. NaturalResourcesCanada (2007) Lakes Rivers and Names of Canada. Available athttp://atlas.nrcan.gc.ca/site/english/maps/reference/outlinecanada/canada06. AccessedSeptember 29, 2010.

50. NaturalResourcesCanada (1974) Posts of the Canadian Fur Trade. Available at: http://atlas.nrcan.gc.ca/site/english/maps/archives/4thedition/historical/079_80. Accessed Sep-tember 29, 2010.

51. Charbonneau H, et al. (1993) The first French Canadians. Pioneers in the St. LawrenceValley (Univ of Delaware Press, Newark, NJ).

Pepperell et al. PNAS | April 19, 2011 | vol. 108 | no. 16 | 6531

EVOLU

TION