Large haploblocks underlie rapid adaptation in an invasive weed

Adaptation is the central feature and leading explanation for the evolutionary diversification of life. Adaptation is also notoriously difficult to study in nature, owing to its complexity and logistically prohibitive timescale. We leverage extensive contemporary and historical collections of Ambrosia artemisiifolia—an aggressively invasive weed and primary cause of pollen-induced hayfever—to track the phenotypic and genetic causes of recent local adaptation across its native and invasive ranges in North America and Europe, respectively. Large haploblocks— indicative of chromosomal inversions—contain a disproportionate share (26%) of genomic regions conferring parallel adaptation to local climates between ranges, are associated with rapidly adapting traits, and exhibit dramatic frequency shifts over space and time. These results highlight the importance of large-effect standing variants in rapid adaptation, which have been critical to A. artemisiifolia’s global spread across vast climatic gradients.


202
. CC-BY-NC-ND 4.0 International license available under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (which this version posted July 5, 2022.

258
. CC-BY-NC-ND 4.0 International license available under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (which this version posted July 5, 2022. ; https://doi.org/10.1101/2022.03.02.482376 doi: bioRxiv preprint of Helianthus species 26 . We used a similar approach to identify genomic signatures of putative 261 inversions (haploblocks) contributing to local adaptation in A. artemisiifolia. Briefly, we identified 262 genomic regions in which population structure was divergent and fell into three clusters, 263 putatively representing the heterozygous and two homozygous genotypic classes of an 264 inversion. Further, we looked for pronounced shifts in population structure (  . CC-BY-NC-ND 4.0 International license available under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (which this version posted July 5, 2022. ; https://doi.org/10.1101/2022.03.02.482376 doi: bioRxiv preprint and Europe, based on 10,000 unlinked SNPs that were randomly selected from outside 294 haploblocks and genes. The estimated slopes for haploblocks HB27a, HB27b and HB31 fell into 295 the 5% tail of the distribution for at least one of the ranges (fig. S12). However, this approach 296 did not examine temporal changes nor the combined signatures of selection over space and 297 time. To do so, we ran generalized linear models comparing haplotype frequency with latitude, 298 time (date of specimen in years) and range (North America vs. Europe; fig. 3A; fig. S13; table 299 S12-S17). All of the haploblocks showed significant changes over time, either in their average 300 frequency in one or both ranges, or in their relationship with latitude within each range, a pattern 301 that is consistent with local selection on the haploblocks. All seven haploblocks showed 302 significant associations with time or with latitude in at least one range, indicative of climate 303 adaptation. These patterns were robust to time being coded as discrete ( on the haploblocks, we tested whether estimates of selection inferred from contemporary spatial 315 data were associated with temporal changes in haploblock frequencies between historical and 316 contemporary European populations. We used spatial variation in contemporary haploblock 317 frequencies to estimate the relative strength of local selection on these haploblocks (see 318 supplementary text S2). We specifically compared estimates of the maximum slope of latitudinal 319 clines for each putative inversion's frequency to simple population-genetic models for clines at 320 equilibrium between local selection and gene flow. In these models, cline slopes are 321 proportional to √ / , where s represents the strength of local selection for a given inversion and 322 is the average dispersal distance of individuals in the range 46 (supplementary text S2). While 323 our estimates of selection are, therefore, scaled by the dispersal rate, dispersal should equally 324 affect all inversions within a given range, allowing us to infer the relative strength of spatially 325 varying selection for each putative inversion. We found that estimates of the relative strengths of 326 . CC-BY-NC-ND 4.0 International license available under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (which this version posted July 5, 2022. ; https://doi.org/10.1101/2022.03.02.482376 doi: bioRxiv preprint changes in the haploblock cline slopes between historic and modern time points within Europe 329 were significantly correlated with our estimates of the relative strength of spatially varying 330 selection of the haploblocks across the European range (r = 0.86, p = 0.01 fig. S14). Such a 331 pattern is consistent with a scenario in which historical European populations were not initially 332 locally adapted (haploblock frequencies were initially far from local optima) and where the 333 haploblocks subject to relatively strong local selection exhibited the greatest temporal changes 334 in local frequency over the ensuing century. The same pattern was not observed in the North 335 American native range, whose historic populations are likely to have been consistently closer to 336 the local optima across the timescale of our analysis (r = 0.61, p = 0.15). Cline slopes for 337 latitude were also shallower in historic relative to modern European populations (t6.58 = -2.39, p 338 = 0.05; mean absolute slope: historic EU = 0.05; modern EU = 0.15), but not so in North 339 America (t11.9 = 0.24, p = 0.81), consistent with initial maladaptation in Europe, followed by 340 adaptation to local climates. 341

342
The HB448 haplotype frequency dramatically increased in the European invasive populations 343 over time, consistent with strong range-wide selection in Europe (table S12). The haploblock 344 also showed a significant range effect, with a significantly higher frequency in modern Europe 345 compared to North America (historic and modern) and historic Europe (table S14) was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (which this version posted July 5, 2022.  was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (which this version posted July 5, 2022. ;

384
Examples of significant associations between haploblock alleles and phenotypes (boxes denote mean 385 and SEM). C. HB27b allele frequency in modern A. artemisiifolia populations.

386
. CC-BY-NC-ND 4.0 International license available under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (which this version posted July 5, 2022. we have shown that A. artemisiifolia was present in regions throughout Europe well before 441 many of these adaptive variants became locally common, suggesting the species' extensive 442 phenotypic plasticity may have facilitated its initial expansion. Strong local selection further 443 improved the match between genotypes and local environments, even appearing to affect 444 reproductive output in herbarium specimens. Many of the selected variants we identified are 445 linked to traits that are key factors in the timing, length and severity of the local pollen season 446 (e.g. days to flowering onset, days to the end of pollen production, and biomass). Consequently, 447 local adaptation has played a central role in shaping the allergy season in Europe and will likely 448 continue to be critical as climate change and continued range expansion further amplify the 449 damaging effects of this hazardous weed 50 . 450 . CC-BY-NC-ND 4.0 International license available under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (which this version posted July 5, 2022. The HiFi assembly yielded a 2.1-Gbp genome, substantially larger than previous genome size 478 estimates of 1.13-1.16Gbp 31,32 . We used GenomeScope 2.0 to estimate the genome size and 479 ploidy using 21mers identified in the reads with Jellyfish 2.3.0 52 . Genomescope estimated the 480 haploid genome size to be 1.04Gb using a diploid model ( fig. S16), a better model fit (95%) than 481 the tetraploid model (91%), which also vastly underestimated the haploid genome size (497 indicated diploidy (fig. S17). Genomescope also identified that the sample was highly 484 . CC-BY-NC-ND 4.0 International license available under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (which this version posted July 5, 2022. ; https://doi.org/10.1101/2022.03.02.482376 doi: bioRxiv preprint assemblies using the eukaryota odb10 dataset (table S1) demonstrated a two-fold reduction of 491 duplicated BUSCO genes (from 233 to 115) with an increase in missing BUSCO genes of only 492 zero to two. 493 494 The HiFi de novo assembly, purged of haplotigs, and Dovetail OmniC library reads were used 495 as input data for HiRise, a software pipeline designed specifically for using proximity ligation 496 data to scaffold genome assemblies 29 . Dovetail OmniC library sequences were aligned to the 497 draft input assembly using bwa 53 . The separations of Dovetail OmniC read pairs mapped within 498 draft scaffolds were analyzed by HiRise to produce a likelihood model for genomic distance 499 between read pairs, and the model was used to identify and break putative misjoins, to score 500 prospective joins, and make joins above a threshold ( fig. S18). The NCBI 54 genome submission 501 portal identified 16 contigs containing bacterial contamination (totalling 7.3 Mb) which were 502 subsequently removed from the final assembly. 503

504
To assess the presence of remnant haplotigs and other assembly artifacts, we mapped Illumina 505 reads used in the reference genome assembly to the haplotig-purged reference genome FASTA 506 file using AdapterRemoval 55 , BWA-MEM 53 and Picard MarkDuplicates 507 (https://broadinstitute.github.io/picard/), and measured average sequencing depth and 508 heterozygosity of the alignment in non-overlapping 1Mb windows across the genome. Window 509 depth was never greater than two times higher or 0.5 times lower than the mean, and 510 furthermore regions of both low depth and low heterozygosity were distributed throughout the 511 genome. The fact that there were no large regions with both low read-depth and low 512 heterozygosity points to the success of the haplotig removal ( fig. S19). Minimap2 was used to 513 align the genome against itself, after filtering for alignments less than 20kb and fewer than 1000 514 matches, to identify homologous blocks that may represent haplotigs. This analysis revealed no 515 large syntenic blocks among scaffolds, further confirming the absence of recent allopolyploidy 516 and the success of the haplotig removal ( fig. S20). 517 518 . CC-BY-NC-ND 4.0 International license available under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (which this version posted July 5, 2022. 20.0, InbreedingCoeff < -0.8). Additionally, SNPs and indels were separately filtered for sites 543 with depth (DP) less than one standard deviation below the mean, and greater than 1.5 544 standard deviations above the mean. Individual genotypes were set to missing if their depth was 545 less than three, then variants with greater than 20% missing across all samples were removed. 546 Samples with greater than 50% missing variants were removed. For the remaining 311 modern 547 samples, contigs with fewer than 100 variants were removed, and the remaining 30 contigs 548 were phased and imputed using Beagle 5.2 60 . 549 550

Genome annotation 551
To obtain RNA transcript sequences for annotation of the genome, after 160 days of growth 552 additional samples of leaf, stem, flower, root, and branch were taken from individual "NSS02/B" 553 . CC-BY-NC-ND 4.0 International license available under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (which this version posted July 5, 2022. ; https://doi.org/10.1101/2022.03.02.482376 doi: bioRxiv preprint the Clontech SMARTer PCR cDNA Synthesis Kit, followed by first-strand synthesis with 559 SMARTScribeTM Reverse Transcriptase. After cDNA amplification, a portion of the product was 560 used directly as a non-size selected SMRTbell library. In parallel, the rest of amplification was 561 first selected using either BluePippin or SageELF, and then used to construct a size-selected 562 SMRTbell library after size fractionation. DNA damage and ends were then repaired, followed 563 by hairpin adaptor ligation. Finally, sequencing primers and polymerase were annealed to 564 SMRTbell templates, and IsoSeq isoform sequencing was performed by Novogene Europe 565 (Cambridge, UK) using a PacBio Sequel II instrument, yielding 97,819,215 HiFi reads. To 566 prepare the raw IsoSeq RNA data for downstream use in the annotation of the genome, we first 567 identified the transcripts in the PacBio single-molecule sequencing data by following the IsoSeq 568 v3 pipeline provided by PacificBiosciences (https://github.com/PacificBiosciences/IsoSeq). 569 Briefly, the pipeline takes PacBio subread files as an input and undergoes steps of consensus 570 generation, demultiplexing of primers, IsoSeq3 refinement, followed by a final clustering of the 571 reads. 572 573 Prior to annotation of the genome, repetitive elements were identified using RepeatModeler2 61 . 574 ProExcluder 62 was then run to remove any protein coding genes from the repeat library. 575 RepeatMasker 63 was used to mask the genome using the finalized repeat library (table S2) was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (which this version posted July 5, 2022. Fisher's exact test, the 'weight01' algorithm, and a p-value < 0.05 to assess significance. was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (which this version posted July 5, 2022. ; https://doi.org/10.1101/2022.03.02.482376 doi: bioRxiv preprint genome, we calculated a standardized measure of frequency change, yt, using estimates of 691 historic, ! , and contemporary, " , allele frequencies according to the equation: 692 where t is the number of generations separating the frequency estimates (equivalent to the 696 number of years due to ragweed's annual lifecycle). As we show in supplementary text S3, the 697 distribution of yt estimates under neutrality are predictable and roughly independent of the initial 698 frequency of each neutral variant once the loci with low-frequency initial minor allele frequencies 699 are filtered out. To further assess if selection was the likely cause of temporal changes of the 700 ELF3 and HB448 variants, we estimated the distribution of yt estimates for computed from 701 10,000 randomly-sampled SNPs that were located outside annotated genes and haploblocks 702 and pruned for linkage disequilibrium using a window size of 50kb, a step size of 5bp and an r 2 703 of 0.5 in PLINK 1.9 72 . Prior to calculation of yt, sampled SNPs were then filtered for a minor 704 allele frequency > 0.2 for HB448 comparisons and MAF > 0.05 for ELF3 comparisons (due to 705 the low historic frequency of ELF3 in historic Berlin populations). We then compared the 706 distributions to the yt values of candidate adaptation loci to test whether candidate regions were 707 more divergent than the putatively neutral distribution. As a point of comparison we repeated 708 this analysis for the ELF3 allele in Quebec. As in Berlin, this allele is at high frequencies, but 709 substantial temporal change was not expected as the populations were predicted to be closer to 710 the equilibrium over the temporal sampling period in the native range. Samples within 200km of 711 Quebec City (ten historic and eleven modern) were pooled at both timepoints. Allele frequency 712 changes of the 10,000 randomly-sampled SNPs and the non-synonymous ELF3 allele were 713 assessed as above. 714 715

Haploblock identification 716
To identify signatures of large, segregating haploblocks across the genome, we performed local 717 windowed principal component analysis with Lostruct 45 . Using SNP data from 311 modern 718 samples, we extracted the first ten multidimensional scaling (MDS) coordinates across each of 719 the 30 imputed scaffolds (scaffolds with > 100 SNPs remaining after filtering) in windows of 100 720 SNPs. These MDS coordinates were then plotted along each scaffold to observe regions of 721 local structure, indicative of segregating haploblocks. We focused on outlier MDS signals that 722 . CC-BY-NC-ND 4.0 International license available under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (which this version posted July 5, 2022. haploblock coordinates) to reduce clustering ambiguity which was heightened by the low 739 coverage historic samples. We also conducted a PCA on 10,000 SNPs randomly-sampled from 740 the 311 modern genomes that were located outside annotated genes and haploblocks, and 741 pruned for linkage disequilibrium using a window size of 50 kb, a step size of 5 bp and an r 2 of 742 0.5 in PLINK 72 . Following this, we used generalized linear models (glm R) to assess how 743 haplotype frequency (binomial response) changed over time and space. A count of each 744 haplotype at a geographic location and year was the binomial response variable and time period 745 (historic or modern), range (North America or Europe), latitude, and all interactions between 746 these three main effects were used as predictors. Non-significant interactions were removed in 747 a stepwise fashion, starting with the highest order. PC1 from the PCA of 10,000 randomly-748 sampled SNPs was included as a covariate to control for the effects of population structure on 749 haplotype frequency. We tested the significance of the effects in our model using the Anova 750 function (Car package R) 76 with type 3 tests. Significant differences among groups for means or 751 slopes were tested with the emmeans package using an FDR correction 79 . To determine if the 752 classification of samples into modern or historic timepoints influenced our results we ran a 753 second set of generalized linear models examining haplotype frequency as a function of 754 collection year, range (North America or Europe), latitude, and all interactions between these 755 three main effects as well as PC1, using the same approach as above. For interactions 756 . CC-BY-NC-ND 4.0 International license available under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (which this version posted July 5, 2022. between these three main effects were used as predictors. All interactions were retained in the 769 model and slopes and their confidence intervals estimated for each group using the function 770 emtrends (emmeans package R 79 ; table S13). PC1 was included as a covariate to control for 771 the effects of population structure on haplotype frequency. We expected the slopes to be 772 shallower in the historic versus the modern European group, but similar across timepoints in 773 North America. To test this, we used a t-test and compared slopes for modern and historic 774 timepoints in each range. We also expected that the magnitude of change in the slope over time 775 would be the greatest in haploblocks showing the largest estimates of selection in Europe (table 776 S15). We estimated the relative strength of selection for the modern European range for each 777 haploblock and tested if the change in slope for each haploblock was correlated with this 778 estimate. We also examined if there was a correlation in the relative strength of selection for 779 modern North American and European haploblocks, which would indicate parallel selection 780 along the cline in each range. 781

782
We compared our slope estimates of the haploblocks to the genome wide distribution in each 783 range using 10,000 randomly selected SNPs outside of genes and haploblocks. We did this to 784 determine if our haploblocks showed stronger latitudinal patterns than the majority of SNPs, in 785 one or both ranges, which may be indicative of spatially varying selection. For the modern 786 samples in each range (North America or Europe), we fit a generalized linear model with latitude 787 as the only predictor. We did this for each null SNP and each haploblock that was statistically 788 associated with latitude. 789 790 . CC-BY-NC-ND 4.0 International license available under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (which this version posted July 5, 2022.

864
. CC-BY-NC-ND 4.0 International license available under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (which this version posted July 5, 2022.

867
. CC-BY-NC-ND 4.0 International license available under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (which this version posted July 5, 2022.

872
. CC-BY-NC-ND 4.0 International license available under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (which this version posted July 5, 2022.

875
. CC-BY-NC-ND 4.0 International license available under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (which this version posted July 5, 2022.

884
. CC-BY-NC-ND 4.0 International license available under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (which this version posted July 5, 2022.

888
. CC-BY-NC-ND 4.0 International license available under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (which this version posted July 5, 2022.

892
. CC-BY-NC-ND 4.0 International license available under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (which this version posted July 5, 2022.

897
. CC-BY-NC-ND 4.0 International license available under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (which this version posted July 5, 2022.

916
. CC-BY-NC-ND 4.0 International license available under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (which this version posted July 5, 2022.

922
. CC-BY-NC-ND 4.0 International license available under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (which this version posted July 5, 2022.

927
. CC-BY-NC-ND 4.0 International license available under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (which this version posted July 5, 2022. ; https://doi.org/10.1101/2022.03.02.482376 doi: bioRxiv preprint Figure S13. Logistic regression models with 95% CI ribbons (see table S12-S17 for model details) of 929 haploblock frequency (allele 1) against latitude for each haploblock across five time bins ranging from 930 most historic (purple) to most modern (green).

931
. CC-BY-NC-ND 4.0 International license available under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (which this version posted July 5, 2022. ;

936
. CC-BY-NC-ND 4.0 International license available under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (which this version posted July 5, 2022.

941
. CC-BY-NC-ND 4.0 International license available under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (which this version posted July 5, 2022.

944
. CC-BY-NC-ND 4.0 International license available under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (which this version posted July 5, 2022.

951
. CC-BY-NC-ND 4.0 International license available under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (which this version posted July 5, 2022.

956
. CC-BY-NC-ND 4.0 International license available under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (which this version posted July 5, 2022.

958
. CC-BY-NC-ND 4.0 International license available under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (which this version posted July 5, 2022.

973
. CC-BY-NC-ND 4.0 International license available under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (which this version posted July 5, 2022.

982
. CC-BY-NC-ND 4.0 International license available under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (which this version posted July 5, 2022. ; https://doi.org/10.1101/2022.03.02.482376 doi: bioRxiv preprint models provided the differences in fitness among genotypes are small (consistent with our 1045 . CC-BY-NC-ND 4.0 International license available under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (which this version posted July 5, 2022. 1065 To infer the strength of selection (S) that would be consistent with observed inversion 1066 frequencies and a given equilibrium value ( &' ), we simulated 10 6 inversion frequencies 1067 consistent with the estimated frequency and its sample size at historical time point (~1883) and 1068 . CC-BY-NC-ND 4.0 International license available under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (which this version posted July 5, 2022. For steep clines-those with a narrow clinal region, where the maximum slope will be 1133 underestimated using the logit function-we have: 1134 . CC-BY-NC-ND 4.0 International license available under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (which this version posted July 5, 2022.

Comparisons of spatially varying selection among haploblocks 1145
All of these estimates rely on the assumption that the system is at equilibrium within each range 1146 and time point, though that assumption may be more valid for some cases then others. To the 1147 extent that it is a reasonable assumption, and if dispersal (gene flow) is consistent across 1148 ranges and times, we can estimate the relative strength of selection in different contexts as: 1149 was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (which this version posted July 5, 2022. where Ne is the effective population size, and n0 is the number of genes sampled in the historic