Skip to content

Step 46: Lp(a), the LPA KIV-2 Repeat Count

What This Does

Estimates how many kringle IV type 2 (KIV-2) repeats the LPA gene carries on both chromosomes together, from read depth, with a 95% interval. The report's Lp(a) section puts that count beside the two common LPA risk variants that step 42 reads, rs10455872 and rs3798220, and flags a genetically high Lp(a) when either points that way. Opt-in.

It is a research estimate, not a diagnostic test. The test that counts is a blood Lp(a) measurement.

Why

Lp(a) is a lipoprotein particle whose level is mostly inherited, and a high level is a causal risk factor for atherosclerotic cardiovascular disease and is linked to aortic valve stenosis. The 2022 European Atherosclerosis Society consensus statement advises measuring it at least once in every adult (Kronenberg et al., Eur Heart J 2022, PMID 36036785).

Two parts of LPA predict the level:

  • The KIV-2 repeat count. The KIV-2 repeat is a 5.5 kb unit, repeated a variable number of times. Fewer repeats give a smaller apo(a) and, on average, a higher Lp(a). In a Danish population study the total over both chromosomes ranged from 6 to 99, and mean Lp(a) was 56 mg/dL in the quarter with the fewest repeats against 15 mg/dL in the quarter with the most; the count explained 21% to 27% of the variation in Lp(a) level (Kamstrup et al., JAMA 2009, PMID 19509380).
  • Two common variants. rs10455872 and rs3798220 were each associated with a higher Lp(a) level, a smaller apo(a) and coronary disease, with odds ratios of 1.70 and 1.92; one risk allele at either went with an odds ratio of 1.51, two or more with 2.57 (Clarke et al., NEJM 2009, PMID 20032323). The study was of people of European descent.

A VCF cannot show the repeat count: the six KIV-2 units in GRCh38 are near-identical, so short reads from any copy pile up on them. That pile-up is what this step measures.

Tool

  • samtools bedcov -j for the read depth over the exons
  • bin/lpa_kiv2.py (standard library Python) for the count, its interval and the reading

Docker Image

  • SAMTOOLS_IMAGE for the depth, PYTHON_IMAGE for the count

Pinned in versions.env; Image versions lists the current tags. The step adds no image and no reference data.

Input

  • ${SAMPLE}/aligned/${SAMPLE}_sorted.bam with its .bai (ALIGN_DIR as for the other BAM steps), aligned to GRCh38 with chr contig names.
  • The aligner must report a read that fits several places equally well, with mapping quality 0. minimap2 (step 2) and BWA-MEM do. novoalign by default leaves such reads unaligned: then the step finds almost no MAPQ 0 reads over KIV-2 and reports the count as not assessed rather than a count that comes out low.

Command

./scripts/46-lpa.sh your_name

With run-all.sh: TOOLS=...,lpa. It is opt-in and needs no data, so a run lists step 46 only when TOOLS names it, and it is skipped for a sample with no BAM. In the Nextflow pipeline: lpa in --tools (processes LPA_REGIONS, LPA_DEPTH and LPA, for each sample with a BAM). The two risk variants come from step 42 (traits, on by default).

How the count is made

The method is the one published short-read estimators use: depth over the KIV-2 repeat against depth over single-copy sequence (Zekavat et al., Nat Commun 2018, PMID 29973585; Behera et al., BMC Med Genomics 2024, PMID 39449055). Behera et al. give the KIV-2 region in GRCh38 as chr6:160,611,568-160,646,868, six units, and its reads reach into the last exon of KIV-1 and the first exon of KIV-3, which resemble KIV-2.

The regions are exons of the MANE Select transcript ENST00000316300.10 (Ensembl, checked 2026-10-25), numbered from its first exon. bin/lpa_kiv2.py bed prints them:

Group Exons What they are
KIV-2 block 3 to 16 the six KIV-2 units (exons 4 to 15), plus exon 3 (KIV-1) and exon 16 (KIV-3): seven 160 bp and seven 182 bp exons, 2,394 bp
single-copy 19 to 33 kringle exons from the second exon of KIV-4 to KV, each under 90% identical to any KIV-2 exon

Exons 17 and 18 (about 95% identical to KIV-2) are in neither group, nor are exons 1 and 2 or the protease exons, which resemble plasminogen.

samtools bedcov -j sums the depth over each exon, every mapping quality counted and duplicates, secondary and failed reads left out; a second pass counts reads of MAPQ 1 or more only. With D the mean depth over the single-copy exons and M the mean depth over the KIV-2 block, the block holds N + 2 copies of a 342 bp exon pair (N KIV-2 units plus the KIV-1/KIV-3 pair on each chromosome) spread over seven reference pairs:

N = 14 x M / D - 2      (KIV-2 repeats on both chromosomes)

GRCh38's own six units on each chromosome give M = D and N = 12.

The 95% interval comes from how much the depth varies between exons: the relative standard error is sqrt(cv_KIV2² / 14 + cv_single² / 15), where cv is each group's coefficient of variation of the per-exon depth, applied to N + 2 on the log scale. It covers sampling noise only, not a bias of an aligner or a library.

The class compares that interval with the range of most people: about 70% of the 1000 Genomes samples had 30 to 45 KIV-2 copies with DRAGEN's caller (Behera et al. 2024; mean 37.5 in Europeans, 35.7 in Africans, 44.6 in East Asians).

Class When
low the whole interval is under 30: a low count goes with small apo(a) isoforms and a higher Lp(a)
high the whole interval is over 45: on average a lower Lp(a)
typical otherwise

The count is not made when:

Status When
too little depth the single-copy exons have under 10x
not assessed reads of MAPQ 0 make up under 20% of the depth over the KIV-2 block (the aligner left out the reads that fit several copies), the KIV-2 depth is too low for any repeat count, or the BAM has no chr6

Output

File Contents
lpa/${SAMPLE}_lpa_kiv2.tsv metric, value: status, kiv2_copies, ci95_low, ci95_high, class, kiv2_depth, flank_depth, mapq0_share, message, advice
lpa/${SAMPLE}_lpa_kiv2_depth.tsv samtools bedcov -j over the regions, every mapping quality
lpa/${SAMPLE}_lpa_kiv2_depth_q1.tsv the same, reads of MAPQ 1 or more

Both reports have an Lp(a) section: the count with its interval and class, the two risk variants from step 42 with their risk allele counts (rs10455872 G, rs3798220 C), a flag when there is at least one risk allele or the count is low, and the advice to measure Lp(a) in blood. Without step 42's table the two variants are shown as not read.

Runtime

Seconds: bedcov reads only the 29 exons through the BAM index.

Notes

  • Research estimate from read depth, not a diagnostic test, and not a measurement of Lp(a). Lp(a) also depends on variants this step does not read.
  • The count is the total over both chromosomes. Which chromosome carries the shorter allele matters for the level, and depth cannot tell.
  • Exome data can work in principle, since the regions are exons, but the step has only been run on whole-genome reads.
  • On the e2e fixture, a slice of HG002 with no LPA, the step reports the count as not assessed or as too little depth. The e2e case lpa-1 runs it on HG002's real reads over LPA and checks the arithmetic on a copy with half the KIV-2 reads.