---
title: Remote Introgression in Evolutionary Genomics
url: https://www.emergentmind.com/topics/remote-introgression-ri
type: topic
---

# Remote Introgression in Evolutionary Genomics

Searching arXiv for the cited papers and related work on remote introgression to ground the article in the relevant literature.
Remote introgression (RI) denotes the transfer of genetic material between lineages that have been separated by substantial evolutionary time, including introgression from an extinct, genetically divergent “ghost” ancestral population and hybridization between phylogenetically distant taxa whose divergence predates the split of sampled modern groups. In the literature summarized here, RI occupies an intermediate conceptual space between canonical genomic introgression, which typically occurs between closely related taxa through hybridization and backcrossing, and horizontal gene transfer, which often denotes transfer across kingdoms or between very distant lineages. Across these usages, RI is invoked to explain deep gene-tree discordance, tract-level ancestry signals, organelle-to-nucleus molecular fossils such as shared NUMTs, and phylogenetically incongruent loci that are difficult to reconcile with incomplete lineage sorting (ILS) alone [1703.09564] [1810.04796] [2507.07761].

## 1. Definitions and conceptual scope

Hawks defines RI as introgression or low-level gene flow from an extinct ancestral population \(P_2\) that split from the focal population \(P_1\) at time \(T_0\) and later contributed a small fraction \(f\) of alleles to \(P_1\) at time \(T_i\). In that formulation, RI is “remote” because the source population’s divergence predates the split of any sampled modern groups [1703.09564].

A second usage emphasizes phylogenetic depth rather than ghost-population status. In this sense, RI describes the transfer of genetic material between lineages that have been separated by substantial evolutionary time—typically millions of years—so that the donor and recipient taxa are far more diverged than the sister-species pairs that commonly exchange genes. The grass-genome study further defines RI as the transfer of genetic material between phylogenetically distant eukaryotic lineages—here, subfamilies of Poaceae—via mechanisms that resemble hybrid-mediated introgression but occur across deep evolutionary splits \(>80\) Mya in grasses [1810.04796] [2507.07761].

This terminology is explicitly contrasted with other modes of exchange. Canonical genomic introgression or gene flow typically occurs between closely related taxa through hybridization and backcrossing. Horizontal gene transfer usually denotes transfer across kingdoms or between very distant lineages, often via parasitism or asexual mechanisms, and frequently from prokaryotes to eukaryotes. RI is therefore presented as gene exchange that requires recombination-based integration, like introgression, but spans much deeper splits than those generally amenable to classical site-pattern tests [2507.07761].

## 2. Population-genetic and genealogical signatures

In the demographic model used by Hawks, \(P_1\) is a Wright–Fisher population of constant size \(N\), \(P_2\) splits from \(P_1\) at scaled time \(T_0\), symmetric migration occurs at rate \(m\) between \(T_0\) and \(T_i\), and at \(T_i\) a pulse of introgression replaces a fraction \(f\) of \(P_1\)’s alleles. Two alleles sampled from \(P_1\) are traced backward in time across 50,000 independent non-recombining loci of length \(L=50\) kb, with mutation rate \(\mu=3.5\times10^{-8}\)/site/generation and expected pairwise heterozygosity
\[
E[H] = 2 \mu L t_{(\mathrm{gen})}.
\]
Within this framework, introgression and migration suppress the interval-specific coalescence probability
\[
p_i = C_i/E_i,
\]
so that
\[
N_{e,i} \approx 1/(2p_i),
\]
inflating inferred \(N_e\) during intervals in which lineages may reside in different subpopulations [1703.09564].

The best-known consequence in this literature is the RI “wave” in inferred effective population size histories. The leading edge of the \(N_e\) increase coincides with \(T_i\); the trough maps to the interval just before \(T_0\); and the crest occurs just before \(T_0\), smoothed by the mutation process. Its amplitude is monotonic in \(f\) and \(T_0\): \(f\approx1\)–\(2\%\) yields a \(\sim1.2\times\) crest-to-trough \(N_e\) ratio, \(f\approx5\%\) yields \(\sim1.5\times\), and \(f\approx8\)–\(10\%\) yields \(\sim2\)–\(3\times\), comparable to PSMC-inferred waves in humans. A continuous low rate such as \(2Nm=0.2\) produces virtually the same \(N_e\) wave as a single pulse \(f\approx10\%\) at \(T_i\) [1703.09564].

A separate but related RI signature is gene-tree discordance that exceeds ILS expectations. In the Gorilla–Pan–Homo case, roughly \(15\%\) of Gorilla loci are phylogenetically closer to Homo than to Pan, and another \(15\%\) closer to Pan than to Homo. This literature treats the combined \(\sim30\%\) lineage sorting as the signal to be explained by admixture models, ABBA–BABA asymmetry, and shared molecular markers such as NUMTs, rather than by a strict bifurcating tree alone [1808.06307].

## 3. Statistical and computational frameworks

Several frameworks in this literature operationalize RI detection from different data types. In site-pattern analyses, the ABBA–BABA \(D\)-statistic is written as
\[
D = \frac{n_{\mathrm{ABBA}} - n_{\mathrm{BABA}}}{n_{\mathrm{ABBA}} + n_{\mathrm{BABA}}},
\]
with \(n_{\mathrm{ABBA}}\) and \(n_{\mathrm{BABA}}\) counting discordant site patterns. In the gene-tree-topology adaptation implemented by the R package **quaint**, the quartet statistic is
\[
D_q(P_1,P_2,P_3,O)=\frac{n_{\mathrm{ABBA}}-n_{\mathrm{BABA}}}{n_{\mathrm{ABBA}}+n_{\mathrm{BABA}}},
\]
and significance is assessed by
\[
\chi^2 = \frac{(n_{\mathrm{ABBA}} - n_{\mathrm{BABA}})^2}{n_{\mathrm{ABBA}} + n_{\mathrm{BABA}}},
\]
with \(1\) degree of freedom. The package enumerates quartets from a species tree, prunes each gene tree to the quartet, classifies each topology as concordant, ABBA discordant, or BABA discordant, discards quartets with too few gene trees, and summarizes evidence across quartets by the mean of non-zero \(D_q\) values for a candidate taxon pair [2606.22561].

For genome scans in a secondary-contact model, Geneva et al. define
\[
\overline d_{12}=\frac{1}{n_1 n_2}\sum_{i=1}^{n_1}\sum_{j=1}^{n_2} d_{ij},
\qquad
\min d_{12}=\min d_{ij},
\]
and then
\[
G_{\min}=\frac{\min d_{12}}{\overline d_{12}}.
\]
Recent introgression drives \(G_{\min}\ll1\), whereas under isolation \(\min d_{12}\simeq\overline d_{12}\) and \(G_{\min}\to1\). The study reports that \(G_{\min}\) has both greater sensitivity and specificity for detecting recent introgression than \(F_{ST}\), and recommends confirmation by neighbor-joining trees or local genealogical reconstruction [1403.1552].

PhyloNet-HMM embeds phylogenetic networks inside a hidden Markov model so that introgression, recombination, and ILS are modeled jointly. Its likelihood takes the standard HMM form
\[
P[O_1\ldots O_T, x_1\ldots x_T \mid \lambda]
= \pi_{x_1}\prod_{t=2}^{T} a_{x_{t-1},x_t}\prod_{t=1}^{T} b_{x_t}(O_t),
\]
with parental-tree switching corresponding to introgression breakpoints and within-parental-tree switching corresponding to ILS-driven changes among genealogical states. In the Mus musculus domesticus chromosome 7 application, the method assigned about \(12\%\) of all sites within chromosome 7 to introgressive origin, covering about \(18\) Mbp and over \(300\) genes, while detecting no introgression in two negative control data sets [1310.7989].

At a more abstract phylogenetic level, overlaid species forests represent introgression histories by mapping a rooted gene tree \(G\) onto a forest \(F=\{T_1,\dots,T_k\}\). An OSF is a map
\[
\psi : V_G \longrightarrow \bigcup_{i=1}^k V_i
\]
satisfying conditions on leaf labels, ancestry, and the presence of a genuine descendant gene-tree leaf below each mapped vertex. Contact arcs
\[
C^*(\psi)=\{(\psi(u),\psi(v)):(u,v)\in E_G,\ \psi(u)\ \text{and}\ \psi(v)\ \text{lie in different}\ T_i\}
\]
are interpreted as RI events, and the minimum number required to explain a forest triple is
\[
t(\mathcal{F})=\min\{|C(\psi)|:\psi\ \text{is an OSF for}\ \mathcal{F}\}.
\]
The OSF-Builder algorithm is guaranteed to produce a strict OSF with \(|C(\psi)|=t(\mathcal{F})\), reducing RI counting to the Fitch–Hartigan parsimony score on the gene tree [2006.14275].

Large-scale phylogeny-based detection in plants is represented by **RIFinder**, which clusters proteins into Homology Groups, infers and preprocesses maximum-likelihood gene trees, partitions large trees into smaller “ortholog-group-like” subtrees, scores topological incongruence, and then applies a modified branch-length test to exclude ILS. Candidate RI leaves are retained only when bootstrap support exceeds \(70\%\), and focal loci can be subjected to a likelihood ratio test
\[
LRT = 2(\ln L_{RI} - \ln L_{null}),
\]
with significance assessed by chi-square with \(df=1\) [2507.07761].

## 4. Hominin and great-ape interpretations

Nygren’s Gorilla-introgression model treats RI as a deep hybridization event from the Gorilla lineage into the Pan–Homo ancestor around \(6\) Ma and uses three main lines of argument: lineage sorting across roughly \(30\%\) of the Gorilla genome, a shared chromosome 5 NUMT dated to approximately \(6.0\) Ma \((\pm0.5\) Ma), and coalescent-based divergence and admixture estimates under an Isolation-with-Migration model [1808.06307].

In the genomic argument, roughly \(15\%\) of Gorilla loci are closer to Homo than to Pan and another \(15\%\) closer to Pan than to Homo. The ABBA–BABA statistic is used to quantify asymmetry, with reported \(|D|\approx0.05\)–\(0.08\) and \(Z>3\) by block-jackknife, and an \(f_4\)-ratio calibration yielding \(f_{GF}\simeq0.30\). In the same synthesis, a \(\sim8.5\) kb NUMT shared by Gorilla, Pan and Homo is dated by
\[
T_{ins}= d/(2\mu),
\]
using \(\mu \sim 2.5\times10^{-8}\) substitutions/site/year and \(d\approx0.30\)–\(0.32\) substitutions/site, giving \(T_{ins}\approx6.0\) Ma. Coalescent-based simulations are reported to place the probability of this shared insertion arising through ILS alone at \(<10^{-6}\) [1808.06307].

The computational implementation described for this model uses three-population coalescent simulations in **msprime** with \(N_e(\mathrm{Gorilla})=50{,}000\), \(N_e(\mathrm{Pan\text{-}Homo\ ancestor})=40{,}000\), \(T_{\mathrm{gorilla\text{-}PanHomo\ split}}\simeq11\) Ma, \(T_{\mathrm{gf\ pulse}}\simeq5.8\) Ma, and \(f_{GF}=0.30\). It further reports highly asymmetric migration rates, with \(m_{G\to PH}=1.25\times10^{-5}\) per generation and \(m_{PH\to G}\approx0\). Under ILS alone, long tracts \(>100\) kb are described as extremely unlikely \((P<10^{-9})\), whereas a pulse model predicts an exponential decay with mean \(\bar L\approx50\) kb [1808.06307].

Nygren also connects the RI hypothesis to morphological correlations. Within *Paranthropus aethiopicus* and *P. boisei*, quantitative measures of crest height and vault thickness are reported to correlate \((r\approx0.75)\) with the \(30\%\) genomic segments inferred as Gorilla-derived. Masticatory morphology is mapped to introgressed loci including alleles of **MYH16** and **DSPP**, with association mapping in extant primates indicating that variants at these genes explain \(\approx12\%\) of variance in bite force. In *Homo*, Gorilla-derived contributions are proposed for opposable thumb index through **HOXD13** and **BMP2**, for adducted hallux through haplotypes near the **PITX1** regulatory region showing \(r^2\approx0.6\) with hallux adduction metrics, and for subcutaneous fat distribution through loci including **LEP** and **ADIPOQ** [1808.06307].

These claims are presented in the source as evidence for RI-driven hominin speciation, including the suggestion that Australopithecus and Paranthropus emerged from distinct late Miocene hybrid populations. A cautious reading is that the model is explicitly designed to unify molecular signals, demographic simulations, and morphological correlates under a single RI framework, while using tract lengths and the shared NUMT to argue against ILS as a sufficient explanation [1810.04796].

## 5. Plant evolutionary genomics and crop introgression

In grasses, RI is treated as a widespread macroevolutionary process. Using 122 haploid genomes derived from 78 high-quality Poaceae assemblies plus *Ananas comosus* as outgroup, and analyzing approximately \(5.15\) million protein-coding transcripts grouped into \(237{,}491\) Homology Groups, the RIFinder study identifies \(622\) RI events originating from \(543\) distinct homologous genes. Singleton events number \(286\), doubletons \(64\), and multi-species events \(272\). The subfamily Pooideae exhibits the highest number of introgressed genes, while Bambusoideae contains the lowest; PACMAD\(\to\)BOP transfers outnumber BOP\(\to\)PACMAD transfers with \(P=5.85\times10^{-10}\) by \(t\)-test [2507.07761].

The same study links RI to localized adaptation after transfer. Ka/Ks analysis finds localized acceleration in \(\sim12\%\) of RI genes \((Ka/Ks>1.0,\ P<0.05)\). Hypergeometric enrichment identifies stress-response pathways including wax ester synthase, NB-ARC NLR domain, terpene synthases, lipoxygenases, and aconitase-like domains, using
\[
P = \sum_{k=x}^{n} \frac{\binom{M}{k}\binom{N-M}{n-k}}{\binom{N}{n}}.
\]
A specific example is a \(\sim30\)-Kbp segment on chromosome 6A of *Cleistogenes songorica* comprising five genes \((RIG1\text{–}RIG5)\) introgressed from *Achnatherum splendens*. Gene order and orientation are identical over \(20\) Kbp in the donor, coding-sequence identity in sliding windows is \(98.2\%\pm0.6\%\), SH and AU tests reject non-RI trees \((P<0.01)\), and drought expression changes include \(RIG2/RIG3\) down-regulated \((\log_2 FC\approx-2.3)\) and \(RIG5\) up-regulated \((\log_2 FC\approx+1.8)\) [2507.07761].

RI is also used in a breeding and cytogenetic sense in *Brassica*. Atri et al. transfer mustard-aphid resistance from *Brassica fruticulosa* into *B. juncea* through an artificially synthesized amphiploid bridge species, AD-4 \((AAFF)\), derived from *B. fruticulosa* \((FF)\) and *B. rapa* \((AA)\). The study reports \(533\) BC1S4 lines, \(221\) BC1S5 lines, and a core set of \(45\) BC1S5 lines for cytogenetic and molecular work. Nearly all introgression lines carried the euploid complement \(2n=36\), metaphase I showed predominantly \(18\) bivalents, and pollen-grain viability improved from \(70.8\pm20.2\%\) in BC1S4 to \(78.4\pm13.4\%\) in year 1 and \(89.2\pm5.9\%\) in year 2 of BC1S5 testing. Using \(74\) transferable SSR primer pairs, the average proportions of recipient and donor genome in the substitution lines were \(49.72\%\) and \(35.06\%\), respectively, with a minimum donor-genome proportion of \(27.29\%\) in line Ad3K-280 [1703.07987].

These plant studies show that RI is used both for naturally occurring deep transfers across major clades and for deliberate movement of wild-relative segments into crops. This suggests that the term spans macroevolutionary reticulation and applied introgression schemes, provided the transferred material originates from a donor lineage that is genetically or phylogenetically remote relative to the recipient background.

## 6. Interpretation, limitations, and contested points

A recurring issue in RI research is separation of introgression from ILS. Site-pattern tests, tract-length analyses, branch-length tests, HMM segmentation, and shared rare markers such as NUMTs are all introduced precisely because local gene-tree discordance by itself is not sufficient. In the grass framework, modified branch-length testing is used to reject the null \(E[d(\mathrm{Donor},\mathrm{Acceptor})]\approx E[d(\mathrm{Donor},\mathrm{Sister})]\) expected under pure ILS. In the Hominin–Gorilla framework, the shared NUMT and the frequency and length of Gorilla-like tracts are presented as evidence that ILS alone is inadequate [2507.07761] [1808.06307].

A second issue is demographic misinterpretation. Hawks argues that even small RI fractions strongly elevate inferred ancestral \(N_e\), creating a spurious “population expansion” wave without any actual size change, so that PSMC and related methods that assume panmixia will misinterpret structure or introgression as size fluctuations. In the human case discussed there, the consistent wave in modern and archaic genomes—large \(N_e\sim200\) kyr, trough \(\sim500\)–\(800\) kyr—can be partly or largely explained by introgression from an archaic ghost population that split \(\sim500\)–\(800\) kyr ago; for non-Africans, Neandertal introgression of \(\sim1\)–\(4\%\) contributes to the wave’s amplitude, and for Africans, multiple lines of evidence support \(\sim2\)–\(5\%\) introgression from unknown archaic Africans diverged \(\sim700\) kyr ago [1703.09564].

Methodological limitations are explicit in the source literature. Hawks’ simulations assume constant \(N\), no recombination, a single-pulse introgression model, and no selection. OSF models recover only the minimal count of jumps and not explicit timing or branch-length information. PhyloNet-HMM assumes a known network topology, and its state space expands rapidly with taxon number. In **quaint**, gene-tree inference mistakes can masquerade as discordance, motivating bootstrap filtering and multiple-testing correction. In large phylogenomic scans, removal of long-branch outliers and poor alignments is required to control false positives [1703.09564] [2006.14275] [1310.7989] [2606.22561].

Taken together, the literature presents RI as a general explanatory category for deep reticulation, ancestral structure, and trait-associated gene acquisition across animals and plants. A plausible implication is that RI is not a single method or a single demographic model, but a family of hypotheses about gene exchange across substantial evolutionary distance, evaluated with different statistical signatures depending on whether the data are coalescent histories, site patterns, tract structures, gene-tree topologies, or explicitly reconstructed phylogenetic networks.

Source: https://www.emergentmind.com/topics/remote-introgression-ri