How to Read Phylogenetic Trees for mtDNA

A phylogenetic tree summarizes relationships inferred from sequence data. It is not a literal family tree and should be interpreted in light of the alignment, model, sampling, and method used to build it.

For mtDNA, all sites on the mitochondrial molecule are inherited largely as one linked locus. A tree can reveal sequence clusters, outliers, possible sample problems, or broad haplogroup structure. It cannot by itself establish a recent pedigree, migration event, phenotype, or direction of ancestry.

Start with the research question

Tree design follows the question. A QC tree asks whether replicates cluster and whether a sample is unexpectedly divergent. A population analysis asks about relationships under an explicit sampling design. A haplogroup display may be better served by a curated mtDNA phylogeny than by de novo inference from a small project.

Define the unit at each tip: one accepted consensus per biological sample, a replicate, a haplotype collapsed across samples, or a reference. If related samples, duplicate sequences, and controls are included, identify them. Sampling determines which relationships the tree can show; an unsampled lineage cannot appear.

Audit inputs before inference

A tree faithfully amplifies upstream mistakes. Build an input table containing sample code, consensus version/checksum, source type, covered range, sequence length, ambiguity count, missing-data proportion, and inclusion decision.

Resolve these issues before alignment:

• Unequal ranges. A complete mitogenome and a short control-region sequence share little comparable information. Restrict all samples to a common defensible interval or analyze strata separately.
• Reference orientation and origin. Put sequences in the same orientation and linearization. Rotate molecules crossing the circular origin consistently.
• Low-quality calls. Do not convert uncertain positions to the expected allele. Use the stated ambiguity/missing-data policy or exclude the sample/site with a reason.
• Replicates and identical sequences. Decide whether to keep all tips, collapse haplotypes, or show counts. Multiple identical tips do not add independent substitutions.
• Contamination or NUMTs. Investigate unexpected divergence with source evidence and assay-specific checks rather than pruning it because it looks inconvenient.

Inspect the alignment, not just the tree

MAFFT and other aligners propose positional homology. For closely related mtDNA sequences, most columns should be straightforward, but homopolymers, indels, poorly covered ends, and circular-origin handling can create alternative alignments.

After alignment:

1. Verify sequence labels and count.
2. Confirm all records span the intended interval.
3. Inspect gap-rich ends and every internal indel.
4. Quantify invariant, variable, parsimony-informative, ambiguous, and missing sites.
5. Decide whether any columns are masked and record the exact mask.
6. Save the alignment used for inference, not only the unaligned FASTA.

Never remove columns solely because they weaken a preferred grouping. Exclusion criteria should be defined by data quality or homology and applied consistently.

Match inference to the purpose

FastTree estimates approximately maximum-likelihood trees efficiently and is useful for rapid exploration and project-level QC. Its speed does not remove the need to specify the nucleotide model, support method, and exact command. For a publication-grade analysis, evaluate whether a fuller model-selection and inference workflow, alternative topology search, and predefined support procedure are warranted.

Record at least:

• software and version for alignment and inference;
• complete commands and random seeds, if the tool uses them;
• nucleotide/substitution model and rate treatment;
• gap, ambiguity, and missing-data handling;
• support method and number of replicates, where applicable;
• rooting method and any outgroup;
• input alignment and output tree checksums.

Read a tree in the right order

Tips are observed sequences or taxa. Internal nodes are inferred splits, not necessarily observed ancestors. Branches connect nodes; when scaled, their lengths usually represent expected substitutions per site, not calendar time. Clades contain a node and all descendants under a chosen root.

Read topology from shared nodes, not from visual proximity. Rotating branches around a node changes the drawing but not the relationships. The order of tips on the page is usually arbitrary.

Then inspect branch lengths. A long terminal branch can reflect true divergence, a different sequence interval, poor-quality calls, alignment error, or contamination. A zero or near-zero branch can indicate identical sequences. Confirm the tree viewer has not hidden lengths or transformed the scale.

Finally inspect support labels. First determine what the number means. FastTree commonly reports local support values; they are not automatically equivalent to standard nonparametric bootstrap percentages or posterior probabilities. Support measures evidence for a split under the data and method. It does not validate sample identity, alignment, model fit, or the biological story attached to that split.

Rooting changes the narrative

An unrooted tree represents splits without a direction from ancestor to descendant. A Newick file can be displayed as though it has a left edge while still lacking a biologically justified root.

Rooting options include a justified outgroup, a known reference point, or midpoint rooting under restrictive rate assumptions. An outgroup must be homologous and alignable; one that is too distant can introduce artifacts. State whether the stored tree is unrooted and apply rooting as a documented analytical step. Do not infer that the tip drawn nearest the root is an ancestor.

Failure modes and responses

Reproducibility checklist

• Research question, inclusion/exclusion criteria, sampling unit, controls, and replicates
• Input consensus identifiers, versions, checksums, ranges, and missingness summary
• Reference accession/version, orientation, circular-origin, ambiguity, and site-mask policy
• Aligner/version/command and archived final alignment
• Inference software/version/command, model, support procedure, and seed
• Rooting method, outgroup rationale, label map, Newick file, and visualization settings
• Sensitivity analyses and interpretation limits

Where GeneFlow fits

GeneFlow can build a project tree when at least three samples have sequences. It prefers each sample’s stored consensus and otherwise uses AB1 base calls, writes a FASTA with sanitized unique labels, runs mafft --auto, then runs nucleotide FastTree, and stores the Newick result, tip labels, sequence count, method label, status/error, project, and creator.

This is a useful reproducible project view, not a complete phylogenetic study. The current task includes every sequence-bearing sample in the project and does not expose common-range selection, column masking, outgroup/rooting, model selection, resampling design, or manual alignment review. It stores the Newick tree but not the generated alignment or executable versions in the tree record. Researchers should curate inputs first and use a validated external workflow when the question requires explicit models, support analyses, rooting, sensitivity tests, or publication-ready provenance. The resulting artifacts and rationale should accompany the GeneFlow record.

Further reading

• Katoh and Standley, MAFFT version 7
• Price, Dehal, and Arkin, FastTree 2
• PHYLIP documentation for the Newick tree format
• PhyloTree: the global human mitochondrial DNA phylogeny