A mitochondrial variant has no complete meaning without a reference. The familiar statement A263G encodes a coordinate, the base expected in a declared reference, and the base observed in the sample. Change the reference and the sample sequence remains the same, but the list of differences can change.
Human mtDNA research commonly encounters two reference systems: the revised Cambridge Reference Sequence (rCRS) and the Reconstructed Sapiens Reference Sequence (RSRS). They answer different questions. The rCRS provides a historically stable, widely used modern sequence. The RSRS represents a reconstructed state at the root of the phylogeny of living human mtDNA lineages. Neither choice changes the specimen; each changes the comparison.
A reference performs several jobs
“Reference” can mean at least three things in one workflow:
- Coordinate framework: where position 263 or 16,519 lies.
- Alignment target: the sequence against which reads or a consensus are aligned.
- Notation baseline: the sequence from which differences are reported.
Those jobs are often coupled, but they need not be. A study might align reads to rCRS coordinates, infer a consensus, and then express phylogenetic changes relative to RSRS. Software may accept an RSRS-relative profile while retaining conventional rCRS position numbers. A report that says only “the mitochondrial reference was used” leaves all of this unresolved.
For reproducibility, identify the actual sequence or reconstruction, the alignment target, the notation baseline, and the coordinate convention separately.
What the rCRS is
The original Cambridge Reference Sequence came from the first complete human mitochondrial genome published in 1981. Andrews and colleagues reanalyzed and corrected that sequence in 1999, producing the revised Cambridge Reference Sequence. Its RefSeq accession is NC_012920.1 , it is 16,569 bases long, and it belongs to haplogroup H2a2a1.
The rCRS is not a consensus of humanity, the most common sequence, or an ancestral human sequence. It is one modern mitochondrial lineage adopted as a standard. Consequently, some positions at which another sample differs from rCRS reflect a derived state on the sample’s lineage, while others reflect a derived state on the lineage leading to rCRS. “Different from rCRS” and “mutated in this sample” are therefore not equivalent statements.
The rCRS remains operationally important. Long-established population, forensic, and variant resources use its coordinates and notation; VCFs are commonly represented relative to it; and many tools align or classify against rCRS-based phylogenies. Compatibility is a scientifically legitimate reason to use it, provided its role is described accurately.
What the RSRS is
The Reconstructed Sapiens Reference Sequence was published by Behar and colleagues in 2012. It was inferred for the root between the deepest branches of extant human mtDNA variation, using the human mitochondrial phylogeny and outgroup information. It is a reconstruction, not DNA obtained from a single ancient person and not a claim that every base is known without uncertainty.
RSRS notation makes branch direction more intuitive: differences from the reconstructed root are intended to represent changes accumulated along paths to living lineages. This helps prevent a common conceptual error in rCRS-relative lists, where a reference-derived allele can make another lineage look “mutated” simply because it retains the ancestral state.
RSRS retains the established position numbering so researchers can discuss the same loci, but reference alleles and therefore variant strings differ at positions where rCRS and RSRS differ. Indels and repetitive regions require particular care because equivalent alignments can generate different textual representations.
The same sequence, two difference lists
Let a sample sequence be $S$, rCRS be $R$, and RSRS be $A$. The reported difference sets are:
$$D_R = \{i : S_i \neq R_i\}$$$$D_A = \{i : S_i \neq A_i\}$$Where $R_i \neq A_i$, membership in the two sets can reverse even though $S_i$ is unchanged. If a sample matches RSRS at such a position, it appears in the rCRS-relative list but not the RSRS-relative list. If it matches rCRS, the opposite occurs. A simple count of reference differences is thus not a reference-independent measure of evolutionary distance.
Conversion is not accomplished by changing the heading on a column. It requires the sample sequence or a complete, correctly normalized variant profile, the two reference sequences, and explicit handling of positions absent from the profile. A VCF containing only alternate sites cannot always distinguish “matches reference” from “not callable” unless coverage or genotype information is retained.
Alignment and notation hazards
Human mtDNA has recurrent substitutions, homopolymer tracts, length-variable regions, and a circular coordinate boundary. These create several sources of non-equivalent-looking but biologically compatible calls:
- Insertions and deletions can be left- or right-aligned differently in repetitive sequence.
- Control-region poly-C tracts can be difficult to sequence and normalize consistently.
- Reads crossing positions 16,569/1 need circular-aware mapping or a shifted reference.
- A bare string such as
263Gmay assume the reference base, whereasA263Gstates it. - Strand orientation and mitochondrial naming conventions can be confused with nuclear HGVS conventions.
Do not compare two exported mutation lists until reference, coverage, normalization, and no-call conventions have been reconciled.
Choosing a reference for the research question
| Objective | Usually practical baseline | Reason | Required caveat |
|---|---|---|---|
| Compare with rCRS-based archives or VCFs | rCRS | Maximum interoperability | Reference differences are not all derived sample mutations |
| Describe changes from the reconstructed human mtDNA root | RSRS | Direction is phylogenetically intuitive | RSRS is inferred and tool support varies |
| Haplogroup classification | Tool’s supported reference/tree pair | Caller scoring expects a specific representation | Report both tree build and input reference |
| Reanalyze legacy control-region profiles | Original reference first | Avoids lossy conversion assumptions | State range and normalization before harmonization |
| Publish reusable consensus sequences | Full FASTA plus declared coordinates | Others can derive either notation | Also publish callable range and ambiguity policy |
There is no requirement that every project choose the same baseline. There is a requirement that all samples within a comparison be harmonized before distances, frequencies, or phylogenies are interpreted.
Common interpretation mistakes
Calling rCRS “normal” or “wild type.” It is one haplotype from haplogroup H2a2a1, not a biological normal against which all other lineages are abnormal.
Calling RSRS a sequenced ancestor. It is a phylogenetic reconstruction of the root state.
Comparing mutation counts across reference systems. Counts are conditional on the baseline and on which positions were callable.
Assuming a haplogroup changed after conversion. A correct conversion changes notation, not the underlying sequence or its phylogenetic placement. A changed call signals differences in tree, coverage, normalization, or software behavior that need review.
Omitting the phylogeny version. A reference sequence and a haplogroup tree are different versioned objects. “RSRS analysis” does not identify which tree release or scoring method was used.
Reporting checklist
- Name the alignment reference and accession or exact sequence.
- Name the variant-notation reference separately if it differs.
- State that positions follow the conventional 1-to-16,569 coordinate system and explain handling of the circular boundary.
- Report sequenced and callable ranges, not only nominal assay targets.
- Define substitutions, insertions, deletions, ambiguous bases, and no-calls.
- Record normalization software and version for VCF or indel comparisons.
- Record the haplogroup caller, version, tree name/build, parameters, and quality output.
- Preserve a consensus FASTA or other representation from which the profile can be independently reconstructed.
Where GeneFlow fits
GeneFlow pipeline definitions can record either rCRS or RSRS as the intended reference, keeping that choice with a project’s analysis setup. This is useful provenance, but the field is not itself a liftover or variant-conversion engine. Changing it does not rewrite an uploaded VCF or prove that all downstream steps used the selected sequence.
The current automated haplogroup path runs HaploGrep3. Sequence inputs are aligned and classified with the configured HaploGrep3 tree, which defaults in the shipped setup to an rCRS-based PhyloTree build; uploaded VCF profiles are likewise expected to be rCRS-relative. Researchers using RSRS-relative data must verify compatibility and perform any required transformation before treating outputs as comparable. GeneFlow preserves source files, sample sequences, classification output, pipeline runs, and reports so that reference assumptions can be audited, but it does not remove the need for reference-aware methods.