Ancient Rome was the capital of an empire of roughly 70 million people, and in 2019 a team led by Margaret Antonio and Jonathan Pritchard sequenced 127 genomes from 29 sites in and around the city to find out what that actually looked like in DNA. Their headline finding is well known by now: during the Imperial period, only 2 of 48 sequenced individuals show a western European genetic profile, while the rest shift heavily toward the Near East. What gets lost in the retelling is how total that shift was. Reconstructing the same population in Global25 space, the Imperial-era Romans that Antonio et al. did not flag as anything unusual come out looking almost as far from the pre-Imperial baseline as the ones they explicitly labelled outliers. Rome's population did not pick up a foreign fringe. Between the Republic and the Empire, its centre of gravity moved.
A city that stopped looking like its own Iron Age
Antonio et al.'s central claim is simple to state and easy to under-read: Rome's Imperial population received net immigration from the Near East, then a further shift toward Europe as the Empire declined. The 2 of 48 statistic is the number most often repeated, because it is dramatic and precise. But a statistic about a minority of individuals invites the assumption that the other 46 stayed put, genetically speaking, and that is not what the data show. This article rebuilds the same population in Global25 coordinates, at both the aggregate and the individual level, and the aggregate result reproduces the published finding almost exactly. The individual-level result does something the published summary does not foreground: it shows that ordinary, unflagged Imperial genomes already sit closer to the Near East and North Africa than to Republican Rome, and that the outlier label the original paper attached to a subset of individuals is a matter of degree on a shared gradient, not a separate population mixed into a stable local one.
Data and method
This reconstruction uses 34 Imperial-period individuals (Italy_Imperial, AD 87 to 300, 45 to 98 per cent SNP coverage), 7 pre-Imperial Republic individuals (Italy_IA_Republic, 650 to 400 BCE), 9 Medieval individuals (Italy_Medieval and Italy_Medieval_EarlyModern, AD 855 to 1250), and 14 individuals the original study or its later re-analyses flagged as outliers by direction: oLevant, oAfrica, oCentralEuropean, oAnatoliaCaucasus, and two labelled o1/o2 with no direction given. All are drawn from the realigned Global25 ancient-samples file, merged first-occurrence-wins with the standard priority order used throughout this site. Four sources anchor every model below: the Republic consensus itself (the mainstream pre-Imperial average, used as the local baseline), Lebanon_ERoman (a Roman-era Levantine average), Algeria_NumidoRoman_Berber (a Roman-era North African average), and France_Metz_GalloRoman (a Roman-era Gallic average). Distances are scaled Euclidean distances multiplied by one thousand, and ancestry proportions come from non negative least squares with a sum to one constraint, the conventions used throughout this site.
Two honest notes before any number is used. Most of these are single ancient genomes at 25 to 98 per cent coverage, and individual-level noise is real; the placebo test below exists specifically to show how much of the apparent signal survives that noise. And the four sources used are broad regional proxies from the Roman period itself, not the specific communities any given migrant actually came from; a Levantine-shifted individual could descend from Syria, Judaea, Cyprus or the Aegean coast, and this model cannot separate those.
Where the aggregate lines up with the published paper
Modelled as a four-source mixture, the Republic consensus comes out at 100.0 per cent local, exactly as a placebo should. The Imperial consensus (n=34, spanning every mainstream individual with no outlier flag) comes out at 19.9 per cent local, 29.1 per cent Levant, 48.7 per cent North Africa, and 2.3 per cent Europe. That 2.3 per cent figure is the aggregate cross-validation this site always looks for: Antonio et al. reported that only 2 of 48 Imperial individuals showed a western European affinity, and an independent Global25 NNLS fit on a different subset of the same population lands on almost exactly that same near-zero European share. The Medieval consensus shifts to 27.6 per cent local, 19.0 per cent Levant, 38.3 per cent North Africa and 15.0 per cent Europe, tracking the published finding of a subsequent swing toward Europe as the Empire's eastern connections weakened. Modern Lazio sits close to the Medieval profile, at 35.4 per cent local, 18.2 per cent Levant, 29.2 per cent North Africa and 17.1 per cent Europe: closer to the pre-Imperial baseline than the Imperial population ever was, but still a long way from the 100 per cent of the Republic.
Four-source NNLS model run identically on the Republic consensus, the Imperial consensus, the Medieval consensus, and modern Lazio. The Imperial bar's 2.3 per cent Europe share is the number to compare against Antonio et al.'s 2 of 48 individuals.
The outliers, and what they are actually mixed with
Antonio et al. flagged 14 of their Imperial and late-Republican individuals as directional outliers: oLevant, oAfrica, oCentralEuropean, and oAnatoliaCaucasus. Testing each one against the four sources individually confirms most of these labels cleanly. R42 (AD 87), R1547 and R1550 (AD 136), and R70 (AD 200), all labelled oLevant, each fit best against the Levantine source, with the model assigning them 85 to 96 per cent Levantine ancestry and a clearly lower residual than any competing source. R132 (AD 280, oAfrica) fits best against North Africa at 90.5 per cent. R116 (AD 100, oCentralEuropean) is the one individual in this entire dataset who fits Europe convincingly, at 75.8 per cent, with the lowest residual of any candidate by a wide margin.
Two labels do not survive the check. R67 and R68, both dated to AD 200 and labelled oAnatoliaCaucasus, actually fit the Levantine source better than the Armenia Iron Age proxy used here for Caucasus ancestry, at 67 to 70 per cent Levant against 29 to 30 per cent Africa and essentially nothing assigned to the Caucasus proxy. This is a case where this site's own method finds something narrower than the original PCA-based label rather than confirming it, and it is reported that way rather than forced into agreement. R11118 and R11121, both labelled oEurope, are stranger still: R11118 comes back 100 per cent local in the four-source model, with no source able to improve the fit, and R11121 splits mostly between Africa and Republic with only a modest, low-confidence Europe component. Whatever pulled these two into the original study's European PCA region, it does not register as a clean single-source foreign pull in this Global25 reconstruction.
The placebo test: mainstream individuals are not "clean" either
The methodological standard this site holds itself to is to test whether a zero, or in this case a clean local reading, actually means what it looks like it means. Three Imperial individuals that Antonio et al. did not flag as anything unusual, R75 (AD 100), R43 (AD 200) and R66 (AD 200), were run through the identical four-source model as a placebo. If "mainstream" meant "genetically similar to Republican Rome," these three should come back close to 100 per cent local, the way the Republic consensus itself does. They do not. R75 comes back 3.0 per cent local, 96.5 per cent everything else, mostly Africa and Levant. R43 comes back 21.5 per cent local. R66 comes back 0.0 per cent local, entirely Levant and Africa. Only one control, R435 from the pre-Imperial Republic itself, returns a majority-local reading, at 66.8 per cent, and even that carries a 33.2 per cent Europe component this article cannot fully explain from a single 71 per cent coverage genome.
Republic-local share for a mix of labelled outliers and unflagged mainstream individuals, same four-source model throughout. Names in muted grey were flagged as outliers by Antonio et al.; names in bold were not. The two groups interleave rather than separate.
The honest reading is that "mainstream" and "outlier" in the original PCA-based classification describe a continuum along one axis of genetic distance from the pre-Imperial baseline, not two distinct populations. The individuals the original study flagged are simply the ones furthest out on that axis, or furthest out in a specific direction. Almost the entire sampled Imperial population, flagged and unflagged alike, already sits closer to the Mediterranean-wide profile than to Republican Rome. That is a stronger and less comfortable version of the published finding, not a contradiction of it.
A 2025 postscript: the shift may have started before the Empire existed
A study published in Genome Biology and Evolution in 2025 sequenced a new individual, BSP71, from Villa Falgari near Tarquinia, dated to the last two centuries BCE, squarely within the Republic rather than the Empire. BSP71 already carries a strong Eastern Mediterranean genetic signature, and the same team's re-analysis of two previously published Republican-era Tarquinia individuals, labelled Italy_IA_Republic_oEasternMediterranean and carrying the individual IDs R437 and R850, the same two individuals this article's own Republic-period outlier check flags as the region's most Eastern-shifted, found the same pattern independently. Their qpAdm modelling put roughly 50 per cent Eastern Mediterranean ancestry in R437 and R850 versus roughly 73 to 86 per cent in Imperial-period individuals and effectively 100 per cent in BSP71 and a second oLevant Tarquinia individual, a gradient that tracks Rome's own military and political expansion into the eastern Mediterranean between 168 and 64 BCE rather than the founding of the Empire in 27 BCE. Their proposed reading, consistent with what this article's placebo test finds independently, is that these late-Republican "outliers" were not oddities on the margin of a stable population. They were the leading edge of the shift that later defined the entire Imperial gene pool.
What this method cannot say
Four limits, stated before anyone else states them. Most individuals here are single ancient genomes at well under 100 per cent coverage, and the placebo test above shows that noise alone can push a genuinely local genome to a near-zero local reading in this four-source system; a reading of "0 per cent local" for any one individual should be read as "at or below this system's floor," not as a precise number. The four sources used, Roman-era Levant, Roman-era North Africa, Roman-era Gaul and the Republic consensus, are broad regional proxies, not the actual home communities of any specific migrant, and a Global25 coordinate carries no information on which of many possible Eastern Mediterranean regions a Levantine-shifted individual's ancestors actually came from. Two of the four originally labelled oAnatoliaCaucasus or oEurope individuals did not reproduce their published direction in this reconstruction, most likely because the specific Caucasus and Central European proxies chosen here are imperfect stand-ins for whatever regional signal the original PCA captured; this is a limit of the proxies chosen, not necessarily of the original classification. And 34 Imperial individuals from Rome and its immediate surroundings cannot stand in for the demographic history of an empire of 70 million people; they describe the city Antonio et al. sampled, not every province it ruled.
Myth and measurement
| The claim | What the evidence shows |
|---|---|
| The famous "only 2 of 48" statistic means the other 46 Imperial Romans looked genetically similar to their Republican ancestors. | The 46 non-outlier individuals average 19.9 per cent local ancestry as a group, and three tested individually (R75, R43, R66) return 0 to 21.5 per cent local on their own. The 2 of 48 figure describes western European affinity specifically, not similarity to Republican Rome. |
| Antonio et al.'s "outlier" individuals were a small foreign minority mixed into an otherwise stable Roman population. | Placebo testing shows unflagged mainstream individuals sit on the same Mediterranean-shifted gradient as the labelled outliers, just somewhat less far along it. Outlier and mainstream is a matter of degree on one axis, not two separate populations. |
| The Near Eastern shift in Rome's population was a consequence of the Empire's founding in 27 BCE. | A 2025 study of a newly sequenced late-Republican individual (BSP71, Villa Falgari) and a re-analysis of two previously published Republican outliers (R437, R850) found the same Eastern Mediterranean signature 150 to 200 years before the Empire existed, tracking Rome's Republican-era conquests in Greece, Anatolia and Syria instead. |
| Every individual labelled an outlier by the original PCA pulls toward the region its label names. | Two of the individuals labelled oAnatoliaCaucasus (R67, R68) fit a Levantine source better than a Caucasus proxy in this reconstruction, and one labelled oEurope (R11118) returns no detectable foreign signal at all in a four-source Global25 model. |
References
- Science 2019 Antonio ML, Gao Z, Moots HM, Lucci M, Candilio F, Sawyer S, Oberreiter V, Calderon D, Devitofranceschi K, Aikens RC, et al. Ancient Rome: a genetic crossroads of Europe and the Mediterranean. Science 366:708-714. https://doi.org/10.1126/science.aay6826
- GBE 2025 Ravasini F, Conati Barbaro C, Scheib CL, Tambets K, Metspalu M, Cruciani F, Trombetta B, D'Atanasio E. A new perspective on the arrival of the Eastern Mediterranean genetic influx in Central Italy before the onset of the Roman Empire. Genome Biology and Evolution 17(8):evaf149. https://doi.org/10.1093/gbe/evaf149
- Nat Ecol Evol 2023 Moots HM, Antonio M, Sawyer S, Spence JP, Oberreiter V, Weiss CL, Lucci M, Cherifi YMS, La Pastina F, Genchi F, et al. A genetic history of continuity and mobility in the Iron Age central Mediterranean. Nature Ecology and Evolution 7:1515-1524. https://doi.org/10.1038/s41559-023-02143-4
The coordinates
Global25 scaled coordinates for the period consensus points, the four sources, and the four most discussed individuals used above, for anyone who wants to reproduce the models in Vahaduo.
Italy_IA_Republic_consensus_n7,0.127319,0.152764,0.0383584,-0.0137967,0.046778,-0.00501986,0.000872857,0.000593571,0.0232866,0.040873,-0.000185714,0.010555,-0.019772,-0.0103414,-0.00126029,-0.002254,0.00428414,0.00159271,0.00657214,-0.00191143,-0.000392143,0.00293243,-0.00382057,-0.00351157,-0.00196729
Italy_Imperial_mainstream_consensus_n34,0.109017,0.14965,-0.0210245,-0.0552151,0.00575311,-0.0196928,-0.000261139,-0.00555747,-0.00244858,0.0210381,0.00353197,0.00278497,-0.00586381,-0.000806611,-0.00941361,-0.00172736,0.0047445,0.00112614,0.00214033,-0.00415125,-0.0026655,0.00308439,-0.0013625,0.000592472,2e-05
Italy_Medieval_consensus_n9,0.109776,0.1468,-0.00100578,-0.0389753,0.0180546,-0.0134798,-0.000861667,-0.00241011,0.00640844,0.0216659,0.00128089,0.00288078,-0.00936544,-0.001942,-0.00778133,-0.00195944,0.00128922,0.000492667,0.000335111,-0.00373789,0.000457667,0.00272044,-0.00160233,-0.000883667,-0.000944556
Italian_Lazio_modern,0.11314,0.152736,0.006788,-0.034303,0.021173,-0.010933,0.003243,-0.000277,0.006422,0.024237,0.00052,0.002548,-0.009574,-0.00523,-0.006732,0.002943,0.001043,0.000456,0.00269,-0.004227,-0.001922,0.004229,0.000567,0.000723,-0.000575
Lebanon_ERoman.SG,0.0838497,0.146575,-0.05506,-0.092055,-0.016516,-0.0304923,-0.00235,-0.0073843,0.0068173,0.0093547,0.010555,-0.008942,0.0155597,-0.0015137,-0.011446,0.0152037,0.010909,0.0031247,-0.000461,0.0019593,0.0011647,0.0046163,-0.0064497,0.0060247,-0.0050693
Algeria_NumidoRoman_Berber.SG,0.0916275,0.150298,-0.024324,-0.0521645,0.008155,-0.021335,0.0034075,-0.0003465,0.006954,0.021048,0.0022735,0.003822,-0.0039395,0.0001375,-0.002375,-0.0028505,0.005411,-0.002154,-6.3e-05,-0.005565,-0.005116,-0.005626,0.00228,0.001145,0.0044305
France_Metz_GalloRoman.SG,0.129303,0.14258,0.0538528,0.024548,0.0424694,0.0064702,0.0051702,0.0047074,0.0113306,0.0114806,-0.0056512,0.0076732,-0.0110306,-0.0083122,0.0068944,0.0041632,0.000313,0.001191,0.0008044,0.0033018,0.0054902,0.003141,-0.0025144,0.0043862,-0.003401
Italy_Imperial.SG_R42_oLevant__AD_87__Cov_98.16,0.080814,0.14319,-0.061471,-0.093024,-0.01508,-0.027331,-0.00658,-0.00923,0.009817,0.00893,0.007145,-0.004496,0.014717,0.004542,-0.008686,0.003978,-0.002868,0.002787,0.004148,0.007629,-0.001497,0.011005,0.000493,-0.001205,-0.003832
Italy_Imperial.SG_R132_oAfrica__AD_280__Cov_83.33,0.023903,0.140143,-0.011314,-0.072675,0.02462,-0.03263,-0.016921,0.003923,0.049495,0.031527,0.004384,0.001499,0.005352,-0.008257,0.000543,-0.010077,0.00326,0,-0.020237,-0.002126,-0.010731,-0.018795,0.011462,-0.008435,0.002754
Italy_Imperial.SG_R116_oCentralEuropean__AD_100__Cov_71.66,0.127482,0.138112,0.052043,0.018088,0.042469,0.001673,0.003525,0.007846,0.01943,0.020775,0.001461,0.007493,-0.013528,-0.010322,0.008958,0.00053,0.000391,-0.002914,-0.000251,-0.002501,0.00262,0.001978,-0.006162,-0.001566,0.00479
Italy_Imperial.SG_R11118_oEurope__AD_200__Cov_80.45,0.127482,0.170609,0.037712,-0.042959,0.072321,-0.029005,-0.007755,0.003923,0.059312,0.089842,-0.004872,0.016335,-0.04103,-0.016377,-0.015336,0.001856,0.02021,-0.006081,0.00176,0.004127,-0.004866,-0.002226,-0.008874,-0.026992,0.005389