HLA-B27 in your DNA file: common in Finland, rarely a diagnosis

One version of a very variable immune gene

HLA-B makes a protein that displays short fragments of protein on the surface of cells for the immune system to inspect. If a fragment comes from a virus or a bacterium, the immune system triggers the cell showing it to self-destruct^[1]. The gene comes in hundreds of versions, and HLA-B27 is the name for one family of them^[1]. It is a family, not a single sequence: by 2017 it counted 213 alleles grouped into 160 subtypes, numbered B*27:01, B*27:02 and onward^[4].

The family matters because of one disease. Ankylosing spondylitis is a painful, lasting inflammation that starts mostly in the joints between the pelvis and the base of the spine, with first symptoms typically between the ages of 15 and 30. Over years it can spread up the spine and fuse the vertebrae^[2]. It belongs to a group of related conditions called spondyloarthritis, and HLA-B27 is linked to others in the group too, among them reactive arthritis^[1]. How HLA-B27 raises the risk is not settled; MedlinePlus describes two ideas, unusual display of some fragments and improper folding of the protein^[1].

About one Finn in seven carries it

How common HLA-B27 is depends on where people's ancestors lived. A US national survey in 2009 found it in 6.1 percent of adults, and in 7.5 percent of non-Hispanic white adults^[5]. Finnish researchers put the frequency in Finland at approximately 15 percent, higher than in most other European populations^[3].

Where HLA-B27 is common the disease tends to be common, and in Japan both are rare. The match is loose: Finland and Iceland carry HLA-B27 about as often as Norway and record far less disease, which may say as much about how each study counted cases as about the populations.

Public genome data show the Finnish excess without anyone being typed. gnomAD, the largest open collection of human genome sequences, reports how often each DNA letter occurs in each ancestry group. From its counts for rs4349859, the tag marker described below, we worked out the share of people in each group carrying at least one copy of the tag letter^[7].

The Finnish bar, 795 of 5,302 people, lands on the typed figure of about 15 percent^{[3, 7]}. The other bars are less reliable as estimates of HLA-B27 itself, because a tag is only as good as its link to the subtypes a population carries.

Most carriers never develop ankylosing spondylitis

Seen from the patient side, the link is strong. In a Finnish study of 261 families with ankylosing spondylitis, 93 percent of patients carried HLA-B27. Carriers developed symptoms earlier, at an average age of 26.7 against 35.7 years, and had acute anterior uveitis, an eye inflammation, more often: 50 percent against 16 percent^[9]. Seen from the carrier side the link looks weak, because carriers vastly outnumber patients.

That range rests on the few studies that went looking for disease among carriers. In the best known, from 1984, van der Linden and colleagues examined 2,957 Dutch people aged 45 or over. About 230 of them would have carried HLA-B27, and 3 had ankylosing spondylitis: 1.3 percent of carriers. Among carrier relatives of a carrier patient, the figure was 21 percent^[10].

A rough Finnish cross-check, our own arithmetic and only an estimate: clinically significant ankylosing spondylitis was found in 0.15 percent of Finnish adults aged 30 or over^[6]. If 93 percent of them carried HLA-B27^[9] and 15 percent of all Finns carry it^[3], about 0.9 percent of Finnish carriers had the disease. The sources span decades, so this says only that Finland sits at the low end of the published range. Two copies count for more than one: in the Finnish families, 11 percent of patients carried two copies, where 4 percent would have been expected^[9].

How a DNA file can point to HLA-B27

Laboratories determine HLA-B27 directly, by serology or with a PCR test aimed at the gene^[12]. A consumer DNA file does neither. A chip reads a fixed list of single positions, and a sequencing VCF lists where a genome differs from the reference. HLA-B is too variable for a few letters inside it to name the version, so studies use a tag: a nearby position whose letter is inherited together with HLA-B27.

Two tags dominate the literature. In 2011 Evans and colleagues searched the region for one position to stand in for HLA-B27 and found rs4349859, about 41,000 letters from HLA-B, next to the neighbouring gene MICA^[8]. In 2013 the International Genetics of Ankylosing Spondylitis Consortium used a second marker, rs116488202, to tag HLA-B27 throughout its analysis^[13]. On the forward strand of the reference genome the tag letters are A at rs4349859 and T at rs116488202; each is the rarer letter, and one copy counts as positive^{[12, 14, 15]}.

How well the stand-in works has been measured against typing in several populations: the share of typed carriers the tag found (sensitivity), and where reported the share of non-carriers it left negative (specificity).

In people of European ancestry, then, one tag letter agrees with typing about 98 or 99 times in 100. The misses are not random. Evans and colleagues showed that rs4349859 follows the main European subtypes linked to the disease, B*27:02, B*27:05 and B*27:08, but not B*27:03, the African subtype linked to it, nor the Asian subtypes B*27:04, B*27:06 and B*27:07. It does follow B*27:05 in Asian carriers, and also B*27:09, a European subtype not associated with the disease^[8].

Why a negative tag is not a negative type

The New Zealand study shows the gap plainly. Among 13 Māori patients typed as carriers, both tags missed 3, and high-resolution typing found all three carried B*27:04. Of three Asian patients typed the same way, rs116488202 identified only the one carrying B*27:05^[12]. For those patients a tag reading of nothing was wrong.

Whether a file has the row depends on the file. A chip reports a fixed list of positions, and whether rs4349859 or rs116488202 is on it depends on the chip and its version; we have not checked which consumer chips carry them. A plain VCF lists only differences from the reference, so a person without the tag letter usually has no line there, and a missing line looks the same whether the position was read or never read; a missing VCF record is not a reference call. A gVCF records reference stretches, so it can show the position was read.

The tag itself reads normally in sequencing data, which is not true of every HLA tag; our celiac guide describes one that nearly vanishes. dbSNP counts the HLA-B27 tag letter on 68 of 5,008 chromosomes in the 2015 1000 Genomes release and on 91 of 6,404 in its 30x genomes, about the same share^[18].

When a raw file does have the rows, this is what they mean, on the forward strand as Ensembl gives it. A file written on the opposite strand shows the complementary letters.

What our report shows for HLA-B27

Our report does not read HLA-B27 from any file; no part of it looks at rs4349859 or rs116488202. It reads two other HLA-B versions through their own tag markers: B*15:02 and B*57:01, linked to severe reactions to carbamazepine and abacavir. Those appear in the medication section and say nothing about HLA-B27.

Deep Read types HLA genes from aligned reads, so it could in principle type HLA-B. We tried. Deep Read reads a slice of the genome file around the genes it needs, and we ran the typing tool on such slices of HG002, a public reference genome whose HLA types two independent methods agree on^[20]. The class II genes came out right every time. HLA-B did not: HG002 carries B*35:08 and B*38:01, and in all three ways we cut out the reads the tool returned B*38:01 and, in place of B*35:08, B*35:525 with a quality score of 0. Its second HLA-C allele changed identity from one cut to the next.

Given the whole 40 GiB file, the same tool typed all ten alleles correctly. Class I typing needs reads that a slice leaves out by design, among them reads the aligner could not place at all. Until that changes Deep Read reports no class I gene. A separate decision stands behind it too: a result that one Finn in seven would receive, and that for most of them is never followed by the disease, is a finding we chose not to present as a health risk.

Where HLA-B27 fits in a diagnosis

In the clinic HLA-B27 is one input among several. The 2009 classification criteria for axial spondyloarthritis, from the Assessment of SpondyloArthritis international Society, cover people with at least three months of back pain that began before age 45. They give two routes: inflammation of the sacroiliac joints on X-ray or MRI plus one other feature of spondyloarthritis, or HLA-B27 plus at least two other features^[21]. In both, HLA-B27 counts only alongside symptoms and other findings.

Reveille and Weisman recount how Calin and colleagues summed up the limit in the 1970s: as a test for ankylosing spondylitis, HLA-B27 typing alone was judged about 95 percent sensitive but only 20 percent specific^[5]. Most patients carry it, and so do many people who never become patients. A tag in a DNA file adds nothing to that, except the tag's own errors.