ACMG secondary findings: the 84 genes and why only those

What makes a finding secondary

A clinical exome or genome is ordered for a reason: a child with seizures, a family with heart failure at forty. The lab reads the genes that bear on that question and reports on them. The data, though, cover thousands of other genes, and a variant in one of them can matter more to the patient than the answer to the question that was asked.

In 2013 the US Presidential Commission for the Study of Bioethical Issues sorted such results into kinds. An incidental finding turns up without being sought. A secondary finding is, in the Commission's words, "a finding that is actively sought by a practitioner that is not the primary target"^[3]. ACMG's first statement, published earlier that year, spoke of incidental findings; the college later moved to the Commission's term, because the genes on its list are searched deliberately^[4].

In practice the lab goes through the listed genes, applies each gene's reporting rule, and reports what it finds to the clinician who ordered the test. The 2013 working group expected about 1 percent of sequencing reports to carry such a finding^[1]. It also chose not to ask patients whether they wanted one. That was changed: people may now opt out of secondary findings, and ACMG's 2021 update keeps the opt-out as the standard^[4].

From 56 genes to 84

The first list came out of a year-long consensus process, and its authors were frank about the evidence behind it: "there are insufficient data on penetrance and clinical utility to fully support these recommendations"^[1]. Its table named 57 genes; the count later fell to 56^[7] when NTRK1 was taken off^[8]. In 2014 ACMG set up a standing working group to take nominations and review the list, and since the 2021 update the gene list has been published apart from the policy, so that it can change every year^[4].

The largest step was v3.0 in 2021, which added 14 genes and removed none^[10]. Among them were PALB2 for breast cancer, TTN for dilated cardiomyopathy and HFE for hereditary hemochromatosis, the list's one entry defined by a single variant.

What is on the list

Table 1 of the v3.3 statement sorts the 84 genes into four groups^[2]. Cancer predisposition has 28, among them BRCA1 and BRCA2, the four Lynch syndrome genes (MLH1, MSH2, MSH6 and PMS2), APC and TP53. Cardiovascular disease has 41, covering cardiomyopathies, inherited arrhythmias such as long QT syndrome, disease of the aorta, and familial hypercholesterolemia through LDLR, APOB and PCSK9. Seven genes cause inborn errors of metabolism, Fabry and Pompe disease among them. The other eight include RYR1 and CACNA1S for malignant hyperthermia, a dangerous reaction to some anaesthetics, and ATP7B for Wilson disease.

Most entries are whole genes: any pathogenic or likely pathogenic variant counts. A few carry rules of their own. In TTN only truncating variants count, and in LMNA only variants tied to heart disease. The nine recessive genes need two variants, and for HFE that means two copies of one variant, C282Y. Of the three X-linked genes, GLA and OTC are reported in women with a single variant; ABCD1 is not^[2].

Heart genes keep being added, and the v3.3 statement says why: heart failure and sudden cardiac death are serious, both can be treated or prevented with established interventions, many genes cause them, and the evidence linking those genes to disease is strong^[2]. PLN, the heart gene added in 2025, is an example. One Dutch founder variant in it may be present in up to 10 to 15 percent of patients with dilated or arrhythmogenic cardiomyopathy in the Netherlands^[2].

The other two additions of 2025 show how the case is argued. CYP27A1 causes cerebrotendinous xanthomatosis, a rare recessive disease of chronic diarrhoea from childhood, early cataracts and progressive neurological decline, which an oral treatment, chenodeoxycholic acid, halts in many cases^[2]. ABCD1 causes X-linked adrenoleukodystrophy, whose childhood brain form is treated best when caught early. Newborn screening normally keeps a condition off the list, but screening for this one began only in late 2013 and some US states still do not do it. ABCD1 is reportable in a male with one variant or anyone with two, not in a female with one, because a single copy is less likely to cause the treatable forms^[2].

The bar a gene has to clear

ACMG's 2021 policy update sets out what a gene needs^[4]. The variants that matter have to be detectable in routine sequencing, without custom laboratory methods. The condition should do its harm with little warning, and a medical intervention, whether surveillance, medication or surgery, should be available to change the outcome. Penetrance should be high. Two exclusions are written into the criteria. Lifestyle changes do not count as an intervention; the update's example is avoiding smoking. And how common a disease is decides nothing: a rare condition qualifies if it is actionable.

The college calls the result a minimum list and means it. Laboratories may report more, and the 2021 update describes the list as "not an inclusive list for any genetic result that could be actionable"^[4]. For v3.3 the working group reviewed seven nominated genes, added three, turned down three and put off a decision on the seventh^[2].

GCK shows where the line sits. The list has generally not taken genes whose only action is to avoid a treatment or an exposure. RYR1 is the standing exception: the action there is avoiding certain anaesthetics, but the consequence of exposure can be sudden death^[2]. ADA2 shows the other limit: everyone with two variants would be kept on lifelong treatment while healthy, and without penetrance data from people found by sequencing rather than by illness, the group would not vote on it^[2].

How often a secondary finding turns up

The 2013 group guessed about 1 percent of reports^[1]. Measured rates have come in higher: ACMG's 2021 update counts studies reporting between 1 and 6 percent, depending on who was sequenced and how variants were judged^[4].

In the UK Biobank survey, 992 people carried 548 such variants. Cancer genes led, BRCA2 most of all with 166 carriers, and LDLR, the main gene of familial hypercholesterolemia, came next with 68^[6]. In eMERGE, a network of ten clinical sites, cancer findings came to 1.38 percent of participants, cardiovascular ones 0.87 and lipid disorders 0.50. Counting genes beyond the 59 as well, chiefly HFE and PALB2, which joined the list later, raised the total to 3.02 percent^[15].

The UK Biobank figure also shows how much rides on method. Of its 548 variants, 315 were already classified pathogenic in ClinVar with at least two review stars and no conflicts. The other 233 were loss-of-function changes the authors judged likely pathogenic under ACMG's classification rules, in genes where such changes are known to cause disease^[6]. A screen that only restates ClinVar's two-star entries starts from the first group.

What those 20 would go on to develop is the hard part. For most inherited conditions, penetrance has been measured in families found because someone was ill, and ACMG writes that in people with no family history it "is likely lower than published penetrance estimates"^[2]. The UK Biobank survey gives one measure of the gap: it estimated a breast cancer risk of around 20 percent in carriers of actionable BRCA1 or BRCA2 variants, at a median age of 58, against earlier estimates of 30 to 40 percent by 60^[6]. Our hemochromatosis guide follows the same problem through one gene.

What a clear result leaves out

Several gaps follow from the rules themselves. Only pathogenic and likely pathogenic variants are reported, so a variant of uncertain significance in BRCA2 never appears^[2]. Recessive genes need two variants, so carrying one in ATP7B or MUTYH, which is carrier status, goes unreported^[2]. And large structural changes were deliberately left aside: the 2013 group excluded disorders caused mainly by large deletions, rearrangements or repeat expansions, and did not ask labs to search the listed genes for them by other methods^[1].

Finland has a clear case of that last gap. In 1995, two founder variants in MLH1 explained 19 of 30 Finnish families meeting the criteria for hereditary nonpolyposis colorectal cancer, now called Lynch syndrome^[16]. Mutation 2, c.454-1G>A, changes one letter at a splice site^[17], and ClinVar's expert panel classifies it as pathogenic^[18]. Mutation 1 is a 3.5-kilobase deletion that removes exon 16^[16]. A list of small variants records mutation 2 like any other change. It shows mutation 1 only if whoever made the file also ran a caller for large deletions, a gap our file guide covers.

Europe has been more cautious

The list is American, and European genetics bodies have taken a different line. The European Society of Human Genetics recommended in 2013 that genome analysis stay with the original health problem, at least for the time being. Its 2021 recommendations call the deliberate search for secondary findings opportunistic screening, find it "too early to recommend" as a professional standard, and keep "a generally cautious approach", with informed consent at the centre; looking for later-onset conditions in children it calls premature^[7]. The French Society of Predictive and Personalized Medicine drew up its own list of 36 cancer genes for adults, with no heart genes, and asks for explicit consent where ACMG relies on an opt-out^[7].

Finnish adults have been asked about such results. In four focus groups of 23 people, shown example letters reporting familial hypercholesterolemia, long QT syndrome, Lynch syndrome and Li-Fraumeni syndrome, participants were willing to receive the findings and worried about being left alone with them; they wanted timely preventive care and an expert to talk to^[19].

What a raw DNA file can show

A genotyping chip, the kind behind 23andMe, AncestryDNA and MyHeritage exports, reads a fixed set of positions chosen in advance, mostly common ones. Pathogenic variants in genes such as BRCA1 and BRCA2 are individually very rare, and chips read rare variants badly. In UK Biobank, only 16 percent of chip calls for variants rarer than 1 in 100,000 were confirmed by sequencing, and for BRCA1 and BRCA2 a positive chip result was right 4.2 percent of the time^[20]. An authorised consumer test covers a sliver: the FDA's 2018 decision on 23andMe's BRCA report notes that it detects 3 of more than 1,000 known BRCA mutations, so a negative result does not rule out the others^[21].

ACMG does not expect secondary findings from tests that are not genome-wide, such as gene panels, partly because a panel covering only the cancer genes or only the heart genes could leave people unsure whether they had been tested for the full list^[4]. A chip is a sparser case of the same problem.

On a VCF or gVCF our report covers the whole list: 83 genes in its clinical findings section and HFE on a card of its own. It restates ClinVar. A variant is reported when ClinVar classifies it pathogenic or likely pathogenic with two or more review stars, which means agreeing submitters, an expert panel or a practice guideline^[22]. One-star entries, from a single submitter, are listed apart as lower-confidence leads, and entries with no stars are dropped. In TTN only truncating variants count, as the list requires. In the recessive genes a single variant is not a finding, and two different two-star variants in one gene are flagged together, because a variant file cannot tell whether they sit on the same copy of the gene. Four of those genes, ATP7B, BTD, GAA and CYP27A1, are also on the carrier panel, which shows one copy as carrier status; for MUTYH, CASQ2, TRDN and RPE65 a single copy is not shown.

What it leaves out matters as much. It does not classify variants ClinVar has not, so a truncating variant ClinVar has yet to classify, the kind behind most of the 233 in the UK Biobank survey, goes unreported^[6]. It has no caller for large deletions in these genes. With a plain VCF, a position without a record is read as the reference letter and labelled as an inference, a problem we have written about. Beyond the list, a genome report also reads ClinVar's two-star variants at exact positions in 52 more dominant genes whose conditions ClinGen's actionability panel scores at least 7 out of 12, under a heading of their own.

By the Commission's definitions, a result from a file sent in to be read is not secondary at all: these genes are the target. ACMG says its list "was not validated for general population screening" and that its recommendation applies only in the clinical setting^[4]. A consumer re-analysis sits closer to screening than to a test ordered for a reason. We use the list anyway, as the most closely argued short list of conditions where knowing early changes what medicine can offer. The sample report shows the section as a customer sees it.