You may have come across the claim that a few percent of our protein-coding genes are involved in carbohydrate metabolism. It is a striking idea: of everything the genome does, a meaningful slice is devoted to handling sugars and starches. But how true is it, and what does it really mean? This post tries to pin the number down, and to explain why it is both interesting and easy to over-interpret.

First, how many genes are there?

A percentage needs a denominator, and the denominator is itself uncertain. Round figures of about 20,000 protein-coding genes are common, but the real catalogue is still being edited. One study used protein-detection data to flag 2,001 annotated "protein-coding" genes that probably do not code for proteins, and concluded the true number may be as low as about 19,000.[1Genetic associationMultiple evidence strands suggest that there may be as few as 19,000 human protein-coding genesEzkurdia et al. 2014 · Human Molecular Genetics 23(22):5866–5878Using protein-detection data, the authors flagged 2,001 annotated "protein-coding" genes as probably non-coding, suggesting the true catalogue may be nearer 19,000 genes than the higher figures often quoted.Click for the full reference] So any "X% of genes" claim moves by a few tenths of a percent depending on which total you use.

What the best count says

The most systematic count I could find built a database of carbohydrate and glucose metabolic genes in five vertebrates. For humans it lists 698 genes.[2ReviewA comparative genomics study of carbohydrate/glucose metabolic genes: from fish to mammalsZhang et al. 2018 · BMC Genomics 19:246Built a database of 698 human carbohydrate/glucose metabolic genes (791 zebrafish, 666 mouse, 593 frog, 523 chicken); most are predicted to encode enzymes, over 57% are related to type 2 diabetes, and more than 70% are conserved across the five species.Click for the full reference] Divided by 19,000–20,000 genes, that is about 3.5–3.7%.

So "a few percent" is fair. I could not trace a single canonical source for an exact figure such as 5%, and the number depends heavily on how the list is built. The count above is a bit below 5%, and a looser definition that also includes regulators and signalling genes would give a larger number.

Some other details from that study are worth knowing:

• Most of the 698 genes are predicted to encode specific enzymes that perform defined reactions.
• More than 57% of them are related to type 2 diabetes.
• More than 70% of the carbohydrate genes are conserved across all five species (human, mouse, chicken, frog and zebrafish), which suggests the core machinery is ancient.
• The counts for the other species were 791 in zebrafish, 666 in mouse, 593 in frog and 523 in chicken. Humans do not have the most.

What those genes actually do

It helps to picture them in layers, which also shows why the list is long.

1. Digestion. Enzymes that break starch and sugars into absorbable units. Salivary amylase (AMY1) varies in copy number between people and populations.[3Genetic associationDiet and the evolution of human amylase gene copy number variationPerry et al. 2007 · Nature Genetics 39:1256–1260Copy number of the salivary amylase gene (AMY1) correlates with salivary amylase level, and high-starch populations carry more copies on average.Click for the full reference] Lactase (LCT) can stay switched on into adulthood through a regulatory variant,[4Genetic associationIdentification of a variant associated with adult-type hypolactasiaEnattah et al. 2002 · Nature Genetics 30:233–237A DNA variant upstream of the lactase gene (C/T-13910) tracks with whether adults keep digesting lactose.Click for the full reference] and sucrase-isomaltase finishes digesting sucrose and starch fragments; reduced-function variants are over-represented in people with irritable bowel syndrome.[5Genetic associationFunctional variants in the sucrase-isomaltase gene associate with increased risk of irritable bowel syndromeHenström et al. 2018 · Gut 67:263–270Variants that reduce sucrase-isomaltase function were over-represented in people with irritable bowel syndrome (1,887 cases and controls).Click for the full reference] (See our post on the genetics of macronutrient digestion.)
2. Transport. Sugars must cross membranes. Humans express 14 GLUT proteins, encoded by the SLC2 genes, each with its own tissue distribution and preferences.[6ReviewThe SLC2 (GLUT) family of membrane transportersMueckler et al. 2013 · Molecular Aspects of Medicine 34(2–3):121–138Fourteen GLUT proteins are expressed in humans, in three classes; GLUTs 1–5 are the best studied glucose and/or fructose transporters.Click for the full reference]
3. Core pathways. Glycolysis, glycogen storage and release, gluconeogenesis and the pentose phosphate pathway, each a chain of enzyme-encoding genes.
4. Regulation. Hormones and signalling proteins such as insulin and its receptor, which tune all of the above. The five-species study even highlights one human-specific gene, CALML6, as a possible regulator of glucose use.[2ReviewA comparative genomics study of carbohydrate/glucose metabolic genes: from fish to mammalsZhang et al. 2018 · BMC Genomics 19:246Built a database of 698 human carbohydrate/glucose metabolic genes (791 zebrafish, 666 mouse, 593 frog, 523 chicken); most are predicted to encode enzymes, over 57% are related to type 2 diabetes, and more than 70% are conserved across the five species.Click for the full reference]

A few percent is not the whole story

The list is a catalogue, not a measure of variation. Being on the list says a gene takes part in the pathway; it does not say that variation in it matters for anyone's diet. Which variants matter for a given person is a separate question.

Glucose control is a team sport. A large genome-wide study of type 2 diabetes, with 180,834 cases and more than a million controls, found 237 genome-wide significant loci, resolving into 338 distinct association signals.[7Genetic associationMulti-ancestry genetic study of type 2 diabetes highlights the power of diverse populations for discovery and translationMahajan et al. 2022 · Nature Genetics 54(5):560–572In 180,834 people with type 2 diabetes and 1,159,055 controls (48.9% non-European), 237 loci reached genome-wide significance, resolving into 338 distinct association signals.Click for the full reference] The genetic influence on glucose control is spread across hundreds of signals rather than concentrated in a few genes.

Having many genes does not mean carbohydrate advice is gene-specific. In the DIETFITS trial of 609 adults, genotype pattern did not predict who lost more weight on a healthy low-carbohydrate versus a healthy low-fat diet.[8Randomized trialEffect of low-fat vs low-carbohydrate diet on 12-month weight loss in overweight adults and the association with genotype pattern or insulin secretion (the DIETFITS randomized clinical trial)Gardner et al. 2018 · JAMA 319:667–679In 609 adults there was no difference in 12-month weight loss between a healthy low-fat and a healthy low-carbohydrate diet, and genotype pattern did not predict who did better on which.Click for the full reference] We discuss what that means in personalized nutrition based on genetics.

How to read an "X% of genes" claim

• What is the list? Enzymes only, or everything touching the pathway?
• What is the denominator? 19,000, 20,000 or something else, and from which annotation?
• Is it conserved or human-specific? Most of these genes are shared with fish, mice and chickens, which tells you the percentage reflects ancient biology rather than something unique to human diets.
• Does it translate into advice? A gene list tells you what is possible, not which variants matter for you.

Take-home

• Roughly 700 human genes, about 3.5% of protein-coding genes by one systematic count, are involved in carbohydrate and glucose metabolism.
• The exact figure depends on definitions, and "about 5%" is only a ballpark.
• Most of these genes are ancient, shared with other vertebrates, and encode enzymes.
• The size of the gene list does not mean that dietary advice should be gene-specific.

This post is general information, not medical advice.