Newly formed ECMA0423 out of Escherichia coli and representative choice

Bacterial Taxonomy
1

Hello GTDB team,

Thanks for this new release, I can only appreciate the amount of work and curation that goes into this essential resource!!

I was puzzled to notice that one of my isolates previously assigned to the species cluster of Escherichia coli in r226 is now assigned to the newly formed ECMA0423 species cluster ( GTDB - Loading... ). Given the large sizes of this cluster (13k) and the remaining Escherichia coli (36k), I was surprised that this split was not documented in the release notes.

I was especially concerned as to why a MAG of relatively medium quality was selected as the representative of this species cluster (among 13k genomes, including some isolates).

plot_ECMA0423
plot_ECMA04231500×1500 223 KB

In particular, I spotted these two genomes from isolates with few contigs and perfect Checkm scores:

I can only imagine that clustering such a large quantity of genomes must come with compromises, but I fail to understand the choices for this one. I would be very grateful for pointers to understand the decisions. Please find below the code to get the data and to reproduce the figure.

All the best,

Charlie

PS: can’t add all the links as a new user sorry.

aria2c -c -s 16 -x 16 -k 1M -j 1  https://data.gtdb.aau.ecogenomic.org/releases/release232/232.0/bac120_metadata_r232.tsv.gz
csvtk filter2 -t -o subset_ECMA0423.tsv -f '$gtdb_genome_representative=="GB_GCA_047199055.1"' bac120_metadata_r232.tsv.gz

library(readr)
library(ggplot2)
library(cowplot)
ecma <- read_tsv("/data/subset_ECMA0423.tsv")
p <- ggplot(ecma, aes(x = checkm2_completeness, y = checkm2_contamination))+
  geom_point(aes(color =gtdb_representative, shape = gtdb_representative ), alpha = 0.7, size = 3)+
  scale_shape_manual(values = c("TRUE"=1, "FALSE"=5))+
  scale_color_manual(values = c("TRUE"="#FF69B4", "FALSE"="black"))+
  geom_rug(alpha=0.1)+
  geom_hline(yintercept = 5, linetype = "dashed")+
  geom_vline(xintercept = 90, linetype = "dashed")+
  labs(x="Completeness (Checkm2)", y = "Contamination (Checkm2)", color = "ECMA0423 representative",
       shape = "ECMA0423 representative")+
  theme_cowplot()+theme(legend.position = "bottom")

ggsave("/data/plot_ECMA0423.png",p, width = 5, height = 5, units = "in", bg = "white")
2

Hi Charlie,

Apologies for the slow reply. May is a busy month for the GTDB team.

GTDB uses a strict ANI-based definition for defining species. You can read about this in the following two manuscripts:

This works well in general, but can lead to situations that are less than ideal. This has clearly happen here for E. coli, where the genome GCA_047199055.1 has 94.9% ANI to the type strain of E. coli (GCF_003697165.2). GTDB considers genomes with <95% ANI to be from different species. As such, the GCA_047199055.1 genome was selected as a type genome for a new species cluster which was given the name s__ECMA0423 sp047199055. GTDB also requires (guarantees) that genomes be assigned to the closest type genomes. As such, a large number of genomes previously classified as E. coli were reassigned to this new species clusters.

I agree this is not ideal and we are looking to improve how species clusters are created and updated in GTDB. This will take some time as any such improvements need to be applicable across the GTDB in a largely (ideally entirely) automated fashion to allow for yearly updates.

Cheers,

Donovan

3

I understand why the species was split, but why move these Escherichia to a new genus? It seems odd when they are just below the cutoff for to be the same species. A similar thing happened with some strains of E. coli K-12, i.e. GCF_000750555.1 is now classified as genus G047199095 (species sp047199095). It also seems that various K-12 strains are now assigned to different species, which seems strange (but I have not actually checked how similar those genomes are).

4

Hi Morgan,

Thanks again for flagging this and you are right to find it odd (moving the species to a new genus).

This is an unfortunate consequences of the aggressive WitChi filtering applied in R232. Later check of the alternative trees generated with different filtering thresholds place these strains within true Escherichia. So the split you are seeing (including the E. coli K-12 strain GCF_000750555.1 placed in genus G047199095) is an issue introduced by this specific filtering step.

We did not catch this during curation for R232, and we appreciate you bringing it to our attention as it affects a non-trivial number of genomes, and we plan to correct it in the next release. We will also look into adding a note about this on the website in the meantime, given how many genomes are affected.

Thanks again for the careful check and please let us know if you spot any other cases!

Best wishes,

Masha

5

Hi Donovan,

Thank you for taking the time to reply, I appreciate! I need to apologize for my delay now.

I understand your clarifications about the clustering and the split thank you.

However, could you please expand on the choice of the reference genome for the newly formed cluster s__ECMA0423 sp047199055. Was this because the GCA_047199055.1 was the first detected to be different from the Ecoli cluster? I completely acknowledge the amount of curation you guys provide to the community, but I was especially concerned as to why a MAG of relatively medium quality was selected as the representative of this species cluster among many suitably more trustworthy ones. Is your decision process (GTDB - Methods) to update representative not applied during new cluster formation?

all the best,

Charlie

6

Hi Charlie,

This is hard to explain in words, but I will try. The GCA_047199055.1 genomes being selected as the representative genomes is a result of the species clustering process. Essentially, this genome was found to by sufficiently divergent from all existing species representative genomes that it became a new species. After all species representatives have been established, genomes are then assigned to the representative genome they are most similar to. There is no additional step where the “best” genome in these clusters as now selected as a new representative since this would become circular. Unfortunately, this means an arbitrarily poor genome can become a species representative as a consequence of respecting previously selected genome representatives and not performing a full de novo clustering.

Please send me an email and we can arrange a time to talk if you want more details. This needs some visuals to appreciate. Short story is that the GTDB is operating as designed, but there are challenges with maintaining an annually updated taxonomy that can result in less than ideal outcomes when handling genomes that form a genetic continuum.

We have discussed having an additional concept in GTDB that would be the “reference” or “exemplar” genome for a species cluster. This would explicitly be the genome assembly deemed to be of highest quality.

Cheers,
Donovan

7

Hi again,

Little more color. The GTDB representative updating methodology you indicated does not get applied here. However, new genome added to the ECMA0423 sp047199055 species cluster in the next release will be subject to this update criteria. Perhaps there should be a step where the formation of a new species cluster also results in this updating criteria being applied to all genomes in the cluster. This is hard though, since this (will!) result in circular situations where applying this updating rule results in new representatives, which results in genomes being assigned to different clusters, which results in new representatives, …

Challenges aside, I fully agree with you that the selected representative for this species cluster would ideally be a different genome.

Cheers,
Donovan