GLS2, SLC39A8 and SLC5A11: three genes behind Nightingale's new blood markers

Twenty new markers, and the genes the first scans found

Nightingale's panel measures about 250 substances in one blood sample by nuclear magnetic resonance (NMR) spectroscopy. The expansion adds six polar metabolites (glutamic acid, lysine, serine, succinate, myo-inositol and 1,2-propanediol), seven fatty acids (alpha-linolenic, eicosapentaenoic, docosapentaenoic and arachidonic acid, and omega-5, omega-7 and omega-9 fatty acids) and a score that predicts lipoprotein(a). The company says the new measures reach all its customers from early 2027^[1].

The announcement names the genes beside the strongest signals. For myo-inositol they sit at SLC5A11, a transporter, and ISYNA1, the enzyme that starts making it. For glutamic acid they sit at SLC1A3, GLS2 and PRODH. The fatty acids mostly found genes already known for lipids, FADS1 and FADS2, LPA and APOE, plus an omega-9 signal near SLCO1B1. The post also reports that rs13107325, a protein-changing variant in SLC39A8, was strongly associated with lower glutamic acid^[1].

Several of the named genes already have their own entries in the Aimosti report: lipoprotein(a), APOE, the FADS1 omega-3 and omega-6 conversion card, the HSD17B13 liver card and SLCO1B1. GLS2, SLC39A8 and SLC5A11 are the three new ones, because published studies give each of them a variant, a direction and a size that a raw data file can read. We found no published blood-level variant for SLC1A3, ISYNA1 or PRODH that a raw file reads cleanly, and the report has no card for them.

GLS2 and blood glutamine

GLS2 makes the liver form of glutaminase, an enzyme inside mitochondria that splits glutamine into glutamate and ammonia^[2]. rs2657879 changes one letter of the gene and one amino acid of the enzyme, leucine to proline at position 581^[5]. Glutamine is among the amino acids Nightingale already measured, and the 2026 analysis below used those measurements^[6]; glutamate, which the enzyme makes, is one of the new ones^[1].

The clearest early numbers come from Finland. Kettunen and colleagues measured blood metabolites by NMR in up to 24,925 people from 14 datasets, eight of them Finnish with 13,906 people between them. Each copy of the G version went with glutamine 0.22 SD lower (P = 3 × 10⁻⁷⁰), and the paper listed GLS2 among loci found before^[7]. In 8,299 Canadians measured by mass spectrometry, a different method, the shift was 0.29 SD per copy^[8]. In the 2026 analysis of the Estonian Biobank and UK Biobank, 599,249 people of European ancestry, it was 0.25 SD, with a P value below 10⁻³⁰⁰^{[6, 9]}.

Miller and colleagues tested the variant in liver cells from donors and in cells made to carry it, and found the proline version more active, a gain of function. They describe L581P as the polymorphism most strongly associated with higher fasting plasma glucose and lower plasma glutamine, a glucose link they draw from earlier genetic studies^[5]. An enzyme that breaks down more glutamine in the liver would leave less of it in the blood, which is the direction every study above found.

SLC39A8 A391T: one change, many traits

SLC39A8 makes ZIP8, a transporter that carries zinc and manganese into cells^{[3, 11]}. rs13107325 swaps alanine for threonine at position 391, so the change is written A391T, and the threonine version is the T allele in a raw data file. In mice, the liver uses ZIP8 to reclaim manganese from bile: without it, manganese fell in the blood and organs, and enzymes that need manganese worked less well, including the ones that build sugar chains on proteins^[3].

The human studies point the same way. In 949 older adults in Uppsala, Sweden, each copy of the minor allele went with whole-blood manganese 0.77 SD lower (P = 5 × 10⁻¹¹)^[12]. Mealer and colleagues genotyped 117 people in a US biobank, 25 of them with two T copies, and found serum manganese lower with the T version while no other trace element moved; the sugar chains on their plasma proteins were simpler and less branched^[13]. The rare people with two broken copies of the gene have serum manganese below detection and a severe disorder of protein glycosylation^[14]. The common variant sits at the mild end of the same mechanism, in Mealer's words a spectrum of hypofunction^[13].

The schizophrenia signal is the one that draws attention. The Psychiatric Genomics Consortium found the locus among 108 in 2014, with rs13107325 as the protein-changing variant there^[20], and its 2022 analysis called the variant strongly supported as causal^[21]. A cell study found that the threonine version took up less zinc and weakened glutamate receptor signalling in neurons^[11], which is the link Nightingale draws to its blood glutamic acid result^[1].

In Finland the threonine version is rare: 1.3 percent of chromosomes in gnomAD's Finnish group, against 7.4 percent in other Europeans and 13 percent in Ashkenazi Jewish people^[22]. About one Finn in 40 carries a copy, and about one in 5,000 carries two.

SLC5A11: myo-inositol seen from blood and from urine

Myo-inositol is a sugar alcohol, abundant in the kidney, where it is reabsorbed and broken down^[1]. SLC5A11 makes SMIT2, a transporter that moves it into cells together with sodium^[23]. In the kidney the gene is switched on in the S3 segment of the proximal tubule, the part of the tubule that takes filtered substances back^[4].

In blood, Shin and colleagues measured metabolites in 7,354 people from two European studies, TwinsUK and KORA. Each T at rs4788439 went with myo-inositol about 6 percent lower (P = 7 × 10⁻¹³), and each study found it on its own^[23]. Nine years later, 8,299 Canadians showed the same direction at rs72770483, a variant beside the gene (0.19 SD per copy, P = 2.5 × 10⁻¹⁰)^[8]. The two markers travel together: r² is 1.00 in Finns and 0.89 across Europeans^[24].

Urine shows the mirror image. In 3,860 people from the general population of Pomerania, Germany, a urine signal matched to myo-inositol was about 52 percent higher per copy of the minor allele at the same locus (P = 9 × 10⁻⁶⁶)^[25]. In 1,627 people with chronic kidney disease, rs72770483 again went with more myo-inositol in the urine^[4]. A 2023 study of 5,023 people with kidney disease measured blood and urine from the same people and found both directions again, more in urine and less in plasma, the plasma result short of that study's own significance threshold^[26].

The opposite directions are what makes this locus easy to read: a transporter that reclaims a little less sends more out in the urine and leaves less in the blood. The blood results agree in direction and are modest in size; the 2023 Canadian one fell just short of that paper's own strict threshold for a whole metabolome^[8].

How common each version is

Two columns need a word. The SLC39A8 variant is about six times rarer in Finns than in other Europeans, so most Finnish readers will see the common pair on that card. The SLC5A11 marker is the common version in African ancestry, at 65 percent, and across all 1000 Genomes populations it barely tracks rs72770483 (r² of 0.09)^{[27, 24]}. The blood studies behind that card were of European ancestry, so its reading holds for European ancestry and is unknown elsewhere.

How much of a blood level one marker explains

Per copy, 0.25 SD is a quarter of the typical spread between people. The figure draws the three GLS2 genotype groups as curves of equal width, their averages 0.25 SD apart per G copy. Most of each curve overlaps the others: plenty of people with two G copies have more glutamine than plenty with none.

A standard formula puts a number on it. For a variant with frequency q and an effect of b SD per copy, the share of the spread between people it accounts for is 2 x q x (1 - q) x b². With the Finnish frequency of 0.18 and the published effects of 0.22 and 0.25 SD, GLS2 accounts for 1.4 to 1.9 percent of the person-to-person variation in blood glutamine^{[7, 9, 10]}. The rest comes from other genes, diet, fasting, exercise, illness and the moment the sample was taken.

What a raw data file can read

All three markers are single-letter changes, and none is a strand-ambiguous A/T or C/G pair, so a chip's letters can be oriented. We checked four chip exports we test against, one each from 23andMe (v5), AncestryDNA, FamilyTreeDNA and Living DNA, on 20 October 2026. rs2657879 is on all four. rs13107325 is on the AncestryDNA and Living DNA exports and missing from the other two, and none of its close neighbours is on them either. rs72770483, the 2023 lead beside SLC5A11, is on none of the four; rs4788439 is on all four.

That is why the SLC5A11 card reads rs4788439. It is a published lead for blood myo-inositol in its own right, significant in two cohorts separately^[23], and in Finns it travels perfectly with rs72770483^[24]. A sequenced genome reads all three positions directly.

# rsid      chromosome  position   genotype
rs2657879   12          56865338   AG
rs13107325  4           103188709  CC
rs4788439   16          24931798   CT

Other files leave different gaps. A plain VCF lists only positions that differ from the reference genome, so a missing row reads as the common version, with a caveat on the card that the position may not have been covered. A gVCF says which positions were read. A chip export with no row for a marker leaves that card unresolved: the report says the marker is not on the array and shows no genotype.

What the Aimosti report shows

The cards sit in the Frontier, the exploratory tier below the clinical findings, where every entry carries a label for how strong its evidence is. Each card states what its studies found, in how many people and by how much per copy, and each genotype's line says what that group averaged in the studies. None of them turns a genotype into a predicted level, and none reads a disease into a result.