LD across an inversion (mosquitoes)

TL;DR: a 22 Mb inversion reads as one block, from plink2 --r2-phased output through an LDTrack. The same inversion also loads as a structural variant genotyped per mosquito.

Prerequisites

PLINK 2.0 (plink2), labeled alpha for years despite being the version in general use
htslib (bgzip, tabix)
samtools
curl
python3
node, for the JBrowse CLI

Where the data comes from

Ag1000G phase 2 AR1 (Anopheles gambiae 1000 Genomes Consortium 2020), whose terms of use were lifted in March 2022, so nothing here needs registration or a data-access agreement.

the phased haplotypes and their sample list for chromosome arm 2L, which the commands subset to one population at a time: https://ngs.sanger.ac.uk/production/ag1000g/phase2/AR1/haplotypes/main/shapeit/
the sample metadata the population lists come from, CMgam (Cameroon) and GAgam (Gabon): https://ngs.sanger.ac.uk/production/ag1000g/phase2/AR1/samples/samples.meta.txt
the AgamP4 reference and its gene models, which the gene lane reads: https://ngs.sanger.ac.uk/production/ag1000g/phase3/genome/
the 2La tag SNPs, the ~200 positions whose allele says which arrangement a chromosome carries, which each mosquito's karyotype is scored from (Love et al. 2019): https://raw.githubusercontent.com/rrlove/compkaryo/master/compkaryo/targets/2La_targets.txt
the finished CMgam LD table, rehosted so the track blocks on this page load without the build: https://jbrowse.org/demos/popgen/ag1000g_2L_CMgam.vcor.gz
the 2La genotypes per mosquito: https://jbrowse.org/demos/popgen/ag1000g_2La_CMgam.vcf.gz
the karyotype table the sample lane is grouped by: https://jbrowse.org/demos/popgen/ag1000g_2La_CMgam_samples.tsv

The 2La inversion as one LD block

Two facts set up the page:

Inverted and standard arrangements cannot recombine in a heterozygote, so wherever both are present the whole segment stays correlated.
The 2La inversion in Anopheles gambiae spans roughly 22 Mb of chromosome arm 2L, past what can be computed live from a VCF, so this LD is precomputed with PLINK and read through PlinkLDTabixAdapter.

The sections below build that table, load the same inversion genotyped per mosquito, and read the two together.

Precompute the LD with PLINK

The LD is a file that plink2 --r2-phased writes in three steps: thin the variants, correlate them, index the table. keep.CMgam.txt is the population, two tab-separated columns of the same sample id, the family/individual pair plink asks for.

# the display uploads n(n-1)/2 cells, and ~800 SNPs across an arm is already at
# screen resolution, so keep roughly one variant per 50 kb rather than every
# variant the callset has
plink2 --bfile common --allow-extra-chr --keep keep.CMgam.txt --maf 0.2 \
  --chr 2L --write-snplist --out sel
awk -F'_' -v g=50000 '{p=$2+0; if (p >= nxt) {print $0; nxt = p + g}}' \
  sel.snplist > grid.snplist

# --r2-phased is the haplotype-frequency estimate rather than a correlation
# between dosages, which is what the display draws; dprimeabs adds D' beside it
# as a magnitude, which is how the display reads a precomputed cell.
# --ld-window-r2 0 keeps the uncorrelated pairs. On PLINK 1.9 the pair is one
# flag, `--r2 dprime`, and the columns come out at the same offsets.
plink2 --bfile common --allow-extra-chr --keep keep.CMgam.txt \
  --extract grid.snplist \
  --r2-phased cols=chrom,pos,id,dprimeabs \
  --ld-window 999999 --ld-window-kb 1000000 --ld-window-r2 0 \
  --out ag1000g_2L_CMgam

# plink2 writes tabs and comments its own header, which is what `tabix -H`
# returns. `sort-bed` is `sort -k1,1 -k2,2n` under LC_ALL=C with that `#` line
# kept on top, which is what this table wants too: same first two columns.
jbrowse sort-bed < ag1000g_2L_CMgam.vcor |
  bgzip > ag1000g_2L_CMgam.vcor.gz
tabix -s 1 -b 2 -e 2 -f ag1000g_2L_CMgam.vcor.gz

The track over that file is an LDTrack, and the display reads one of its two metric columns:

Config file CLI (add-track-json)

{
  "type": "LDTrack",
  "trackId": "ag1000g_2l_cmgam",
  "name": "Cameroon, both arrangements segregating (r²)",
  "assemblyNames": ["anoGam3"],
  "adapter": {
    "type": "PlinkLDTabixAdapter",
    "uri": "https://jbrowse.org/demos/popgen/ag1000g_2L_CMgam.vcor.gz"
  },
  "displays": [
    {
      "type": "LDTrackDisplay",
      "ldMetric": "r2",
      "useGenomicPositions": true,
      "showLegend": true,
      "height": 340
    }
  ]
}

jbrowse add-track-json '{
  "type": "LDTrack",
  "trackId": "ag1000g_2l_cmgam",
  "name": "Cameroon, both arrangements segregating (r²)",
  "assemblyNames": ["anoGam3"],
  "adapter": {
    "type": "PlinkLDTabixAdapter",
    "uri": "https://jbrowse.org/demos/popgen/ag1000g_2L_CMgam.vcor.gz"
  },
  "displays": [
    {
      "type": "LDTrackDisplay",
      "ldMetric": "r2",
      "useGenomicPositions": true,
      "showLegend": true,
      "height": 340
    }
  ]
}'

The inversion genotyped per mosquito

The same inversion loads as one <INV> record spanning the breakpoints, genotyped across every mosquito. The regular multi-sample variant display draws each genotype at the call's true span, so a carrier's row begins and ends at the breakpoints.

Those genotypes are what the karyotype lanes in the figure below are: cells shaded by allele dosage, each lane sorted into standard, heterozygous and homozygous-inverted blocks. The karyotype column names the three classes by genotype, and so does the legend: 2L+a/2L+a, 2La/2L+a, 2La/2La, the + marking the non-inverted arrangement.

Load each population as a VariantTrack whose adapter carries the samples TSV, with a LinearMultiSampleVariantDisplay that orders (groupBy) and colors (colorBy) its rows by the karyotype column:

Config file CLI (add-track-json)

{
  "type": "VariantTrack",
  "trackId": "ag1000g_2la_karyotype_cmgam",
  "name": "Cameroon, one row per mosquito",
  "assemblyNames": ["anoGam3"],
  "adapter": {
    "type": "VcfTabixAdapter",
    "uri": "https://jbrowse.org/demos/popgen/ag1000g_2La_CMgam.vcf.gz",
    "samplesTsvLocation": {
      "uri": "https://jbrowse.org/demos/popgen/ag1000g_2La_CMgam_samples.tsv"
    }
  },
  "displays": [
    {
      "type": "LinearMultiSampleVariantDisplay",
      "groupBy": "karyotype",
      "colorBy": "karyotype",
      "referenceDrawingMode": "skip"
    }
  ]
}

jbrowse add-track-json '{
  "type": "VariantTrack",
  "trackId": "ag1000g_2la_karyotype_cmgam",
  "name": "Cameroon, one row per mosquito",
  "assemblyNames": ["anoGam3"],
  "adapter": {
    "type": "VcfTabixAdapter",
    "uri": "https://jbrowse.org/demos/popgen/ag1000g_2La_CMgam.vcf.gz",
    "samplesTsvLocation": {
      "uri": "https://jbrowse.org/demos/popgen/ag1000g_2La_CMgam_samples.tsv"
    }
  },
  "displays": [
    {
      "type": "LinearMultiSampleVariantDisplay",
      "groupBy": "karyotype",
      "colorBy": "karyotype",
      "referenceDrawingMode": "skip"
    }
  ]
}'

What each setting does:

groupBy keeps the karyotype classes contiguous, so each class reads as one block.
referenceDrawingMode is on its default, skip, which fills the lane with the reference color and paints alt cells on top: a solid grey field with the carriers' blocks on it.

Rows divide the lane's height between them, so a 300-pixel lane gives each of 297 mosquitoes about a pixel. The display draws a row for every sample in the file, which makes each population its own track.

The karyotype calls

2La is a cytologically defined arrangement whose breakpoints have been cloned and sequenced (Sharakhov et al. 2006), and the call is drawn at that published extent. PCR across the junctions karyotypes single mosquitoes, checked against polytene cytology on field specimens (White et al. 2007).

Each mosquito's karyotype here is scored from those tag SNPs, the in-silico method MalariaGEN ships for the current Ag3 release: the mean number of alternate alleles across the tags, rounded into a genotype. That score comes out trimodal with empty space between the peaks, and the reproduce script prints the histogram and the karyotype breakdown per population.

The block on the karyotype lanes

Four lanes stack in the figure: each population's r² heatmap over its own karyotype lane, one row per mosquito, 297 from Cameroon and 69 from Gabon.

Ag1000G chromosome arm 2L, the same window and settings throughout. Top: the published extents of 2La and of Vgsc, the two loci the blocks below sit on. r² fills the 2La extent in the Cameroon panel, which segregates both arrangements, and is empty over that span in Gabon, which is near-fixed for the standard arrangement. Open this view in JBrowse ↗ (large dataset — this session takes a while to load)

The block's edges land on the published breakpoint coordinates, and on the karyotype lane beneath the heatmap, whose cells are drawn at those same coordinates from a different file. Across the block, markers at opposite ends are about as correlated as neighbouring ones: correlation holds flat with distance over a recombination-suppressed span.

Two more things stand out beyond the 2La block itself:

The second block is Vgsc. At the low-coordinate end of the arm in both panels, it is reddest along the diagonal and pales away below it: the sodium channel whose codon-995 substitutions confer pyrethroid resistance, and which this release was used to survey (Clarkson et al. 2021). Gabon shows that block too.
Gabon's 2La span reads flat. 64 of its 69 mosquitoes recombine freely across that span, and the MAF floor both files carry drops the variants tagging the 5 heterozygotes.

What an LD block depends on

Four things decide how strongly a block reads, and the reproduce script prints the numbers behind each:

Common-variant density. A sweep that went to fixation leaves few common variants to correlate. Compare density at the locus against a neutral window in the same panel.
Whether the panel segregates the feature. Long-range LD inside a candidate span, against an equally distant control, reads as a block only where both arrangements are present.
Background LD. A bottlenecked panel renders red across the whole arm, so read the absolute background alongside the inside/outside ratio.
Number of haplotypes. r² is a correlation between two biallelic markers, so several haplotypes at one locus fragment the block: each carries a different background, and no single pair of markers tags them all. A soft sweep leaves a patchier block than its strength suggests.

Metric and allele-frequency floor

Two metrics read the same block differently:

D' asks whether recombination has been seen between two markers, so it saturates near 1 wherever no recombinant haplotype has turned up. That makes it the read on where recombination stops, and the reproduce script uses it to recover the breakpoints.
r² asks how well one marker predicts the other, which also requires the two to be at similar frequency, so it draws the sharper boundary and reads on whether a marker can stand in for another.

Switch with ldMetric; the script prints both ratios for every panel.

Raising the minor allele frequency filter (minorAlleleFrequencyFilter) thins dense callsets to the common, block-tagging variants. High enough, it reaches the tagging variants themselves and the block fades.

Reproduce it end to end

build_ag1000g_ld.sh does the whole build for you:

downloads the phased haplotypes
runs each check above and prints the result
builds the tabix-indexed .vcor.gz tracks and the per-mosquito karyotype calls
writes a config.json opening on the inversion

curl -fO https://raw.githubusercontent.com/GMOD/jbrowse-components/main/scripts/build_ag1000g_ld.sh
bash build_ag1000g_ld.sh              # writes ./ag1000g_ld_build/jbrowse2
npx --yes serve ag1000g_ld_build/jbrowse2

The same karyotype track in Drosophila

Selection scans (Drosophila DGRP) builds the same one-record karyotype track for an 11 Mb Drosophila inversion.

References

Anopheles gambiae 1000 Genomes Consortium (2020). Genome variation and population structure among 1142 mosquitoes of the African malaria vector species Anopheles gambiae and Anopheles coluzzii
Clarkson et al. (2021). The genetic architecture of target-site resistance to pyrethroid insecticides in the African malaria vectors Anopheles gambiae and Anopheles coluzzii
Love et al. (2019). In silico karyotyping of chromosomally polymorphic malaria mosquitoes in the Anopheles gambiae complex
Sharakhov et al. (2006). Breakpoint structure reveals the unique origin of an interspecific chromosomal inversion (2La) in the Anopheles gambiae complex
White et al. (2007). Molecular karyotyping of the 2La inversion in Anopheles gambiae

Feedback on this tutorial is welcome: contact us.

LD across an inversion (mosquitoes)

Prerequisites#

Where the data comes from#

The 2La inversion as one LD block#

Precompute the LD with PLINK#

The inversion genotyped per mosquito#

The karyotype calls#

The block on the karyotype lanes#

What an LD block depends on#

Metric and allele-frequency floor#

Reproduce it end to end#

The same karyotype track in Drosophila#

See also#

References#