Tuberculosis genomic surveillance gets a broader lineage framework
Whole-genome sequencing has become central to tuberculosis surveillance because it can connect cases, detect drug resistance, and separate local spread from repeated importation. Those uses depend on a lineage framework that is broad enough to classify both common and uncommon isolates without forcing them into the nearest available label. In a disease caused by a genetically structured pathogen such as Mycobacterium tuberculosis, a narrow typing scheme can distort epidemiology before a public health analyst even begins to interpret the data.
Pathotypr was built to address that gap. The open-source system accepts assembled genomes or raw short reads and returns lineage assignments, World Health Organization–cataloged resistance-associated variants, and a closely related reference genome for downstream analysis. The design matters because surveillance laboratories do not all work from the same input format, and the path from sequence to usable result still varies widely across settings. A tool that can handle both assemblies and reads without a conventional alignment workflow reduces friction at the point where genomic data meet routine surveillance.
A single misassigned lineage can weaken downstream inference about transmission, importation, and resistance clustering.
A phylogeny-driven marker set built from 26,813 genomes
The lineage framework behind Pathotypr came from a maximum-likelihood phylogeny of 26,813 MTBC genomes and 609,003 polymorphic sites. From that tree, the investigators selected single-nucleotide polymorphisms on the stem branch leading to each monophyletic lineage or clade, then removed 3,039 missense variants and 454 homoplastic nonmissense variants. The final marker set contained 2,145 lineage-defining SNPs drawn from 5,638 candidates.
That filtering strategy is clinically and epidemiologically sensible. Lineage markers need to be stable enough to distinguish evolutionary branches rather than transient mutations that appear in multiple parts of the tree. When homoplasy is allowed to creep into a genotyping framework, the same change can appear to support unrelated branches and blur the structure that surveillance depends on. Pathotypr’s marker selection therefore leans on phylogenetic traceability rather than sheer marker count.
The resulting framework covered the recognized human-adapted and animal-adapted diversity of MTBC, including L1–L10 and A1–A4. Coverage across rare groups remains an important issue because some lineages were represented by only small numbers of genomes, leaving open the possibility that additional diversity has not yet been sampled. Even so, the marker set extended lineage recognition beyond the narrower scope of several existing tools, including uncommon groups that other schemes either missed or left unresolved.
Assembly, read, and reference workflows all pointed in the same direction
Pathotypr is organized into four modules. The Predict module uses a k-mer-based Random Forest model for genome assemblies, while the Classify module uses SNP markers for assemblies and Split-FASTQ applies the same lineage framework directly to raw sequencing reads. A fourth module, Match Ref, selects a lineage-aware reference genome for read mapping.
Validation was unusually clean. Against 498 complete RefSeq genomes, Predict matched the phylogenetic lineage in all 498 and Classify matched 497; the single unassigned genome lacked the complete marker path required for a call, which is consistent with an assembly gap or a problem at a diagnostic marker position. On 254 chromosome-level assemblies not used to train the Random Forest, Predict and Classify returned the same major lineage for every sample. Split-FASTQ correctly assigned all 162 clinical read datasets to the expected lineage.
The reference-selection step produced a concrete downstream benefit. Across those 162 clinical read sets, mapping to the reference chosen by Pathotypr reduced nonfixed SNPs by a median of 89% relative to the MTBC ancestral reference and increased median coverage breadth by 0.84 percentage points. In practical terms, a more appropriate reference can clean up spurious variation and improve the interpretability of variant calls, which matters when the same data will be used for both surveillance and resistance interpretation.
Rare lineages emerged when the framework was scaled to 88,071 samples
The surveillance-scale test involved 88,071 UShER-TB samples processed with Split-FASTQ in about 24 hours using four threads. After excluding 1,290 genomes with multiple lineage calls from the direct comparison, Pathotypr and TB-Profiler assigned the same major lineage to every directly comparable sample. The catch was coverage: only 698 of the TB-Profiler lineage markers overlapped with Pathotypr, and the new framework identified additional uncommon isolates that TB-Profiler either misclassified or left unassigned.
Three points stand out. Pathotypr identified 2 L10, 6 A1, and 23 A2 isolates that were not recovered by the TB-Profiler marker set. TB-Profiler misclassified one L10 isolate as La1.5/_M. bovis_ and left another L10 isolate plus the A1 and A2 isolates unassigned. For a surveillance program, these are not trivial bookkeeping differences. Rare lineage detection affects how laboratories recognize introductions from underrepresented evolutionary groups and how they decide whether a cluster is really local or the product of repeated importation.
Resistance calls can travel with lineage calls without losing clinical relevance
Pathotypr also reports resistance-associated variants alongside lineage assignments. In the 7,148-isolate CRyPTIC dataset, categorical resistance profiles based on World Health Organization confidence groups 1 and 2 agreed with phenotypic drug-susceptibility results for 86.4% of isolates. Severe disagreement, defined as a difference of three levels between genotypic and phenotypic profiles, occurred in 1.4% of isolates. For multidrug-resistant or pre–extensively drug-resistant status, the system reached 93.5% sensitivity and 92.3% positive predictive value.
Agreement was weaker for bedaquiline, clofazimine, delamanid, and linezolid. Most phenotypically resistant isolates in those categories carried variants of uncertain significance or no cataloged variant at all. This is the expected pressure point for resistance prediction in a pathogen where newer drugs have fewer validated genetic correlates than older agents. The biologic problem is straightforward: if the resistance mechanism is not yet well represented in curated catalogs, the genotype may lag behind the phenotype until the underlying mutation spectrum is better mapped.
Lineage and resistance outputs were also combined for a surveillance analysis of probable cross-border introductions into Germany, Italy, or Ukraine. Among 590 descendant isolates from 135 inferred transitions, 195, or 33.1%, had multidrug-resistant or pre–extensively drug-resistant genotypes, compared with 5.5% of the remaining Europe isolates in the dataset. Because CRyPTIC is enriched for drug-resistant isolates and is not population-based, those proportions cannot be treated as national estimates. They do, however, show how lineage-resistance pairing can flag clusters that deserve epidemiologic follow-up.
Why broader lineage coverage changes the public health workflow
Pathotypr’s main contribution is not a single statistic. It is the combination of breadth, speed, and traceability. A framework that covers L1–L10 and A1–A4, accepts assemblies and reads, reports multiple supported lineages in mixed samples, and chooses a closely related reference genome gives laboratories a more consistent route from sequence data to interpretation. Low memory requirements and both desktop and command-line access further widen its use in settings where bioinformatics infrastructure remains limited.
For tuberculosis control programs, the value lies in reducing unresolved calls and preventing uncommon isolates from disappearing into generic categories. The same framework can also help place resistance results in an evolutionary context, which is essential when clusters cross borders or when phenotypic resistance emerges without a clear cataloged marker. The principal limitation is built into the biology and the sampling: the rarest lineages remain sparsely represented, so future sampling could still reveal marker diversity not captured here.