A population size bottleneck early in the evolution of the H. sapiens lineage, perhaps at its origin some 2 MYA, has significant explanatory power in resolving some of the contradictions between different sources of data addressing past human population size. This bottleneck is well supported paleontologically, but what about genetically? The long-term inbreeding effective population size of humans is on the order of 104–105 taken over the last 1–2 Myr. The limited amount of genetic variation reflected by this small Ne implies that ancient population size changes, predating 1 Myr, will be difficult to detect using genetic methods. Methods of appraising population bottlenecks from genetic evidence rely on the effects of such bottlenecks on the rate of genetic drift. But if the bottleneck is very ancient, if genetic drift has been powerful at times since the bottleneck, or if there has been selection, then any of the processes that erode genetic diversity may have erased any evidence for a population bottleneck.
Minimally, we may expect to place a lower time limit on the possibility of severe population bottlenecks, such as those that might be associated with a speciation or a period of rapid adaptation. Based on autosomal evidence from several gene systems, we may rule out such a bottleneck at times more recent than 1.5 Myr (this date, the time when significant expansion of the human range out of Africa first began, can be estimated from the autosomal data presented in table 1 ). No date more recent than this is compatible with known neutral nuclear variation. However, nonmolecular sources of information must become more important as we consider demographic events far in the past.
Therefore, considered together, nuclear data allow bottlenecks within a narrow range around 2 MYA, a range of possibilities that is fully compatible with the fossil and archaeological records. We may try to refute the hypothesis of a bottleneck at this time from other genetic data by using the expectation that if there were no bottleneck early in our history, we should expect some ancient variation, older than 2 Myr, to remain at neutral loci in the human population today. This should be true even if forces that erode diversity have been powerful since the bottleneck, as would be reflected by a small long-term average Ne. For example, if Ne has been equal to its lower bound, 104, then the expected coalescence time of a neutral locus should be exponentially distributed with a mean of 4 x 104 generations (Hudson 1990 ). Assuming 23-year generations as before, the mean coalescence time will be about 920,000 years. We can expect from this distribution that 11% of loci will have coalescence times greater than 2 Myr. Likewise, for an Ne of 2 x 104, 34% of loci will coalesce earlier than 2 Myr without a bottleneck. Since a bottleneck would be expected to truncate this distribution at or around 2 Myr, the presence of diversity older than 2 Myr can be reasonably expected if the hypothesis is wrong, and would be a clear disproof that such a bottleneck occurred.
However, while all the genetic systems that have been examined to date are compatible with a population size bottleneck at the origin of H. sapiens, no genetic system for which neutrality has been claimed has been found to have a minimum coalescence time estimate significantly more ancient than 2 Myr. Those analyses based on elements interspersed throughout the genome, such as the human-specific Alu insertions, yield some of the oldest estimates, but all are compatible with the occurrence of a 2-Myr bottleneck. Therefore, although there probably are many factors limiting human genetic diversity, no diversity reasonably interpreted as neutral (and the Alu insertions are the best example of this) has yet been detected that must unquestionably extend from the period before our last speciation.
A second potential for refutation comes from nonneutral systems, such as those under balancing selection, because these address whether the bottleneck sizes noted above are consistent with other data. The most prominent example is the HLA gene system that diverges from neutrality and is likely subject to balancing selection (Klein et al. 1993 ). This system should be expected to retain ancient variation through a population size bottleneck because of the mechanism of balancing selection (Ayala 1995 ). The HLA complex genes have the largest amount of variation studied thus far (Klein 1986 ). As it turns out, retention of a large number of ancestral HLA alleles precludes effective population sizes of much less than 1,000 at any particular point in time during human prehistory (Ayala 1995 ; Ayala and Escalante 1996 ; Takahata and Satta 1998 ). This minimum bottleneck number, 1,000, also seems to be the minimum effective population size compatible with the maintenance of species viability and adaptability (Lande 1995 ). The HLA data, then, do not preclude a speciation bottleneck of minimal population size.
However, this is the only bottleneck not ruled out by the confluence of these data sources. Considerable genetic data are inconsistent with a recent bottleneck in the human lineage, as are data from prehistoric archaeology and paleoanthropology (Jelínek 1982 ; Wolpoff, Wu, and Thorne 1984 ; Eckhardt 1987 ; Kramer 1991 ; Pope 1992 ; Frayer et al. 1993 ; Kennedy 1994 ; Clark 1997 ; Wolpoff and Caspari 1997 ). Information from additional genetic systems will no doubt continue to increase our understanding of the population size of our species at its origin and help further clarify these issues of subsequent population change, but at the moment such studies rest on the horns of a dilemma. If we assume neutrality for the autosomal loci well known at this time, they preclude any recent population size bottleneck for two reasons: (1) they are in equilibrium while mtDNA is not, and (2) they are not consistent with any significant population expansion as must follow such a bottleneck earlier than 10,000 years ago. If we do not assume neutrality, these loci do not give us information about past population size.