Skip to content

How to Check Your Own MTHFR Genotype From Raw Data

Oak
A lab instrument stretching a glowing DNA strand, with two distant base positions pinched by calipers and lit yellow.

By the end of this you will have two genotypes written down with their genome build, their strand, their read depth, and their genotype quality: rs1801133 (the variant usually called C677T) and rs1801131 (A1298C). You will also know whether your data can tell you which chromosome each variant sits on, which is the part most free upload tools skip. You need one of two inputs. The first is a consumer array raw data file from 23andMe, AncestryDNA, MyHeritage, or FTDNA, which is a tab-separated text file of about 600,000 to 700,000 lines. The second is a VCF from whole-genome or exome sequencing, ideally with the aligned CRAM or BAM alongside it. Tooling is bcftools, samtools, and standard Unix text tools. Everything below runs on a laptop in under a minute, except the coverage checks on a CRAM, which need the reference FASTA.

MTHFR encodes methylenetetrahydrofolate reductase. That enzyme reduces 5,10-methylenetetrahydrofolate to 5-methyltetrahydrofolate, the methyl donor used to remethylate homocysteine into methionine. The reaction is old and conserved. The bacterial homolog MetF carries the same step in Mycobacterium tuberculosis, where disrupting it blocks methionine biosynthesis and impairs survival in the host.1 That conservation is a reason the gene gets attention. It is not a reason to over-read a single genotype.

1. Know exactly which variants you are looking for

The two variants that free MTHFR reports are built on have confusing names, and the confusion causes real errors. Get the identifiers straight before you touch a file.

rs1801133 is HGVS c.665C>T, protein p.Ala222Val, in NM_005957.5. It is still widely called C677T after an older transcript numbering. MTHFR sits on the minus strand of chromosome 1, so on the forward genome strand this variant is a G to A change. Its coordinate is chr1:11856378 on GRCh37 and chr1:11796321 on GRCh38.

rs1801131 is c.1286A>C, protein p.Glu429Ala, commonly called A1298C. On the forward strand it is a T to G change. Its coordinate is chr1:11854476 on GRCh37 and chr1:11794419 on GRCh38.

The two sites are 1,902 base pairs apart. Remember that number. It decides whether short-read data can phase them.

The strand detail matters because consumer arrays report forward-strand alleles. A 23andMe file showing AA at rs1801133 means homozygous for the 677T allele, even though the letter is an A. A file showing GG at rs1801131 means homozygous for the 1298C allele. If you read the letters as if they were the cDNA names you will invert both results.

2. Pull the genotypes out of a consumer array file

Array raw data is the fastest path, and for these two positions it is reliable, because both are common enough that every major array carries a direct probe. Start by confirming the build. Recent 23andMe and AncestryDNA exports are GRCh37, and the header says so.

head -20 genome_raw.txt

Then extract the two lines:

awk -F'\t' '$1=="rs1801133" || $1=="rs1801131" {print}' genome_raw.txt

A typical result looks like this:

rs1801131	1	11854476	TT
rs1801133	1	11856378	AG

Read that as heterozygous for 677T and homozygous reference at 1298. Check the position column against the coordinates in step 1. If the numbers match GRCh37 but your file header claims something else, trust the coordinates and re-derive the build.

If either line is missing or shows --, the probe failed for you. That is a no-call, not a negative result, and no amount of reformatting will recover it. Arrays no-call a low single-digit percentage of sites per sample, and the rate is higher on degraded saliva.

AncestryDNA files use the same four columns but split the genotype into two columns, so adjust the awk accordingly. MyHeritage files are comma-separated with a quoted header. The rsIDs are the same everywhere.

3. Pull them out of a VCF, which is the version we would trust

If you have sequencing data, use it instead of the array. A VCF gives you depth, genotype quality, and allele balance. An array file does not. First make sure the file is indexed and that you know whether contigs are named 1 or chr1.

bcftools index -t sample.vcf.gz
bcftools view -h sample.vcf.gz | grep -m1 contig

Then query both sites in one pass. This example assumes GRCh38 and chr prefixes:

bcftools view -r chr1:11796321,chr1:11794419 -Ou sample.vcf.gz \
| bcftools query -f '%CHROM\t%POS\t%ID\t%REF\t%ALT\t[%GT\t%DP\t%GQ\t%AD]\n'

Expected output for a 677 heterozygote at reasonable depth:

chr1	11794419	rs1801131	T	G	0/0	38	99	38,0
chr1	11796321	rs1801133	G	A	0/1	41	99	20,21

Two habits are worth building. Normalize before you query, so that multi-allelic records and left-alignment differences do not hide a call:

bcftools norm -f GRCh38.fa -m -both -Oz -o sample.norm.vcf.gz sample.vcf.gz

And check that a missing line means reference rather than missing data. A single-sample VCF from most pipelines only emits variant sites. Absence of rs1801133 is ambiguous unless the pipeline also wrote a gVCF with reference blocks. If you have the gVCF, confirm the site is covered:

bcftools view -r chr1:11796321 sample.g.vcf.gz | bcftools query -f '%POS\t%END\t[%GT\t%DP]\n'

If you only have the VCF and the BAM or CRAM, go to the alignment directly:

samtools mpileup -f GRCh38.fa -r chr1:11796321-11796321 sample.cram
samtools depth -a -r chr1:11794419-11794419 --reference GRCh38.fa sample.cram

We want at least 20 reads at each site with a balanced allele fraction near 0.5 for a heterozygous call. Anything under 10 reads should be treated as provisional.

4. Decide whether you can phase the two variants

This is the step that separates a useful analysis from a generic report. When someone carries one copy of 677T and one copy of 1298C, the two can sit on the same chromosome (in cis) or on opposite chromosomes (in trans). The functional consequence differs, because in trans both copies of the enzyme carry a substitution. Free upload tools almost never address this, and many of them label the combination with confident-sounding language they cannot support from unphased genotypes.

The two sites are 1,902 bp apart. A 150 bp paired-end library will not produce a read pair spanning both, so read-backed phasing fails on standard short-read data. You can confirm that quickly rather than assume it:

whatshap phase --reference GRCh38.fa -o phased.vcf.gz \
  --chromosome chr1 sample.vcf.gz sample.cram
bcftools query -r chr1:11794419-11796321 -f '%POS\t[%GT\t%PS]\n' phased.vcf.gz

If both sites come back as 0/1 with no phase set (PS) in common, they are unphased. Three things can fix that. Long reads (Oxford Nanopore or PacBio HiFi) span the interval easily and phase it directly. Linked-read or Hi-C data will also work. Statistical phasing against a reference panel using SHAPEIT5 or Beagle gives a probabilistic answer over this short distance, which is usually accurate but is an inference rather than an observation.

Population data also helps. The 677T and 1298C alleles are in strong linkage disequilibrium, and the haplotype carrying both is rare in most populations. A person heterozygous at both sites is therefore very likely to carry them in trans. That is a prior, not a measurement, and we would write it down as such.

5. Put the allele frequency in the right population context

A genotype means little without a denominator. Pull the frequency for your ancestry group rather than the global average, because the 677T allele varies widely across populations. It is common in European and East Asian samples and much less common in West African ones. A cross-sectional study of 564 school-aged children in Beijing reported allele distributions for MTHFR C677T, MTHFR A1298C, and MTRR A66G together. That is a useful reminder that these variants are ordinary polymorphisms with population-specific frequencies rather than rare findings.2

The quickest route is the gnomAD GraphQL API:

curl -s https://gnomad.broadinstitute.org/api \
  -H 'Content-Type: application/json' \
  -d '{"query":"{variant(variantId:\"1-11796321-G-A\",dataset:gnomad_r4){genome{populations{id ac an}}}}"}'

Divide ac by an per population to get the frequency. If you prefer to work offline, annotate your VCF with a downloaded gnomAD sites file:

bcftools annotate -a gnomad.genomes.sites.chr1.vcf.bgz \
  -c INFO/AF,INFO/AF_nfe,INFO/AF_eas -Oz -o annotated.vcf.gz sample.norm.vcf.gz

When you see that a genotype is present in a double-digit percentage of the population you were born into, the framing of the result changes. This is variation. It is not a rare finding.

6. Connect the genotype to something you can measure

Genotype is a fixed input. Enzyme function and folate status are the outputs, and those you can measure directly, which is far more informative than reading a report generated from two SNP calls.

In vitro work on the 677 variant has shown reduced enzyme activity and thermolability, with homozygous carriers retaining roughly a third of reference activity in assay conditions and heterozygotes somewhere in between. Assay conditions are not a person. The measurable downstream consequence in people is plasma total homocysteine, which most laboratories report in µmol/L with an adult reference interval in the region of 5 to 15 µmol/L. Serum folate, red blood cell folate, and serum B12 complete the picture. RBC folate reflects longer-term status than serum folate does.

Order these as a fasting panel, and repeat once before drawing any conclusion from a single value. If your results fall outside a laboratory reference interval, that is a matter for a physician. Do not start, stop, or change any supplement or medication based on a genotype or on a homocysteine value read off a PDF. Interpretation of these labs, and any decision about B vitamins, belongs with a clinician who can see your full history.

7. Widen the query to the rest of the pathway

Once you have a working extraction command, running it over the neighbouring genes costs nothing and gives a much better picture than MTHFR alone. Build a small BED or position list and query it in one pass:

cat > folate_sites.txt <<'EOF'
chr1	11796321	rs1801133	MTHFR_c.665C>T
chr1	11794419	rs1801131	MTHFR_c.1286A>C
chr1	236885200	rs1805087	MTR_c.2756A>G
chr5	7870860	rs1801394	MTRR_c.66A>G
EOF

bcftools view -R <(cut -f1,2 folate_sites.txt) -Ou sample.norm.vcf.gz \
| bcftools query -f '%CHROM\t%POS\t%ID\t%REF\t%ALT\t[%GT\t%DP\t%GQ]\n'

Verify each coordinate against dbSNP for your build before trusting the output, because positions in hand-written lists rot. The MTRR A66G variant in that list is the one reported alongside the MTHFR variants in the Beijing cohort.2 There is also a 19 bp insertion-deletion in the DHFR intron 1 region that gets discussed in folate genetics. Short-read callers handle it inconsistently, and arrays do not type it at all, so treat any report claiming a DHFR indel result from array data with suspicion.

If you have RNA sequencing from whole blood, two more measurements become available. Gene-level expression tells you how much MTHFR transcript is present, which you can read straight out of a salmon quant or featureCounts table as TPM. Allele-specific expression at rs1801133 tells you whether the two haplotypes are transcribed at comparable rates:

gatk ASEReadCounter \
  -R GRCh38.fa -I rnaseq.bam -V het_sites.vcf.gz \
  -O ase_counts.tsv --min-depth-of-non-filtered-base 10

MTHFR expression in whole blood is modest, so depth at the site is often thin. Check the totalCount column before reading anything into the ratio.

Common problems

Strand inversion is the most frequent error, and it silently produces the opposite answer. Anchor every result to the forward-strand reference and alternate alleles from the VCF (G>A for rs1801133, T>G for rs1801131) rather than to the cDNA letters in the variant’s common name.

Build mismatch is the second. A GRCh37 coordinate queried against a GRCh38 file returns nothing, or worse, returns a different variant. Always print the %ID column so you can confirm the rsID matches what you asked for.

A missing line in a VCF is ambiguous. It can mean homozygous reference or it can mean the region was never covered. Resolve it with the gVCF or with samtools depth before recording a reference genotype.

Imputed genotypes get reported as hard calls. Many free upload services impute millions of extra sites from a 700,000-marker array and present them without a quality score. Both MTHFR sites are directly typed on the major arrays, so this is usually not a problem here. Imputation is a real problem for the rest of the pathway. If a service reports a variant your raw file does not contain, it was imputed.

Unphased compound heterozygosity gets stated as fact. If you are heterozygous at both sites and your data is short-read, you do not know the phase from the data alone. Record the inference and its basis separately from the observed genotypes.

Consumer array data and research-grade sequencing are not clinical results. Neither one is a diagnostic test, and confirmation in a clinical laboratory is required before any result is used in medical decision-making. Take the numbers to a clinician and let them decide what confirmation is warranted.

Finally, calibrate the interpretation. These are common polymorphisms with frequencies that vary by ancestry.2 The genotype is one fixed input into a pathway whose output you can measure directly in blood. Measure the output.

Footnotes

  1. Ikue Tosa, Tomoki Kitahara, Katsuki Takebe, et al. MetF‐Dependent Methionine Biosynthesis Is Required for Mycobacterium tuberculosis Survival and In Vivo Persistence. MicrobiologyOpen, 2026. https://doi.org/10.1002/mbo3.70370 ↩

  2. Xinrui Peng, Wangxin Ding, Xin Kang, et al. Distribution of ACTN3 R577X, MTHFR C677T, MTHFR A1298C, and MTRR A66G Polymorphisms in 564 School-Aged Children in Beijing: A Cross-Sectional Study. Genes, 2026. https://doi.org/10.3390/genes17080886 ↩ ↩2 ↩3

Pothos builds longitudinal molecular profiles of individuals: whole-genome sequencing, single-cell RNA sequencing, and plasma proteomics, delivered as a detailed report, every data file, and stored cells. Build your profile.