Interpreting AB-dominated Smudgeplots in tetraploid and octoploid individuals

[hungweichen0327]

Hi Smudgeplot developers,

Thank you for developing and maintaining this excellent tool.

I am analyzing Illumina whole-genome sequencing data from three individuals representing diploid (2n=2x=20), tetraploid (2n=4x=40), and octoploid (2n=8x=80) individuals of the same plant species from different collection sites. The species has been reported to be self-compatible and capable of autonomous self-pollination.

Published estimates suggest a monoploid genome size of approximately 0.3 Gb.

Chromosome-scale assemblies generated using ONT and Hi-C data recovered:

• 10 chromosome-length scaffolds for the diploid individual (~0.3 Gb assembly)
• 20 chromosome-length scaffolds for the tetraploid individual (~0.55 Gb assembly)
• 40 chromosome-length scaffolds for the octoploid individual (~1.2 Gb assembly)

These assembly sizes are broadly consistent with the expected increase in chromosome number and DNA content across cytotypes. In addition, chromosome staining of root tip cells confirmed 80 chromosomes (2n = 80) in the octoploid individual.

Below are the GenomeScope profiles for the diploid, tetraploid, and octoploid individuals:

The commands used for the Smudgeplot analysis are shown below:

# run FastK to create a k-mer database
nohup FastK -v -t4 -k25 -M64 -T32 ./*.fastq.gz -N./FastK_Table 1> FastK.out 2> FastK.err &

# Find all k-mer pairs in the dataset using hetmer module
nohup smudgeplot hetmers -L 12 -t 32 -o ./kmerpairs --verbose ./FastK_Table > smudgeplot_hetmers.log 2>&1 &

# use the .smu file to infer ploidy and create smudgeplot
nohup smudgeplot all -o ./k25_smudgeplot ./kmerpairs.smu > smudgeplot_all.log 2>&1 &

This is the 2x smudgeplot result:

This is the 4x smudgeplot result:

This is the 8x smudgeplot result:

My questions are:

1. The confirmed tetraploid cytotype produces a predominantly AB-type Smudgeplot that is remarkably similar to the diploid. Is this unusual, and could it be consistent with diploidized or disomic inheritance?

2. Is Smudgeplot informative regarding potential allopolyploid versus autopolyploid origins, or would such interpretations be beyond the intended scope of the method?

3. Could relatively low heterozygosity contribute to both the AB-dominated Smudgeplot patterns and the unexpectedly small GenomeScope estimate (~150 Mb for the tetraploid, compared with an expected monoploid genome size of ~0.3 Gb)? (I used p =4 for tetraploid individual and p=4 for the octoploid individual since the maximum p value is 6)

Thank you very much for your time.

[KamilSJaron]

Hi! Sorry for the long wait.

I think you are right, the low heterozygosity is messing up with the k-mer profiling analyses. I think the key is to pay attention to 1n coverages and genome size estimates. They all need to make sense and you can always rerun the model giving it coverage priors (and sometimes you need to set up number of iterations to 1.).

More detailed thoughts:

So, for your diploid. I imagine you manged to assemble both haplotypes and you have 10 more chormosomes in alt, right? In that case, genomescope looks right, haploid assembly ~0.3Gbp corresponds well to the model. Interestingly, smudgeplot shows double coverage estimate. That probably means there are very few heterogyzous sites in the genome (which kind of makes sense given genomescope) and moreover, there are plenty duplications causing that the only smudge you see is in fact AABB. Observing this is quite rare, but we have seen them in some parthenogenetic species and a few selfing plants.

Following similar logic, I think both your genomescope model and smudgeplot for the tetraploid is wrong. I think in both cases converged to double coverage than it should. Assuming you assembled 2 out of 4 haplotypes (sort of equivalent of collepsed diploid assembly, which I am guessing based on 20 chromosomes). Then the estimate haploid genome size should be 1/2 of what you assembled. You estimated 150Mbp and assembled 550Mbp -> SO I thin the coverage should be half of what was estimated and that would make the genome size estimate approximately double ~300Mbp which would be much closer to the ballpark you expected. Interesting aspect of this is that there would be strong AABB struture. Two pairs of diploid very homozygous haplotypes forming this tetraploid.

And in fact, the same applies to the oktoploid. You fit only 4x model (because there is no way of fitting 8x), but then you expect approximately double the genome size, but you ended up with ~300Mbp. If your coverage was truly half, and the model oktoploid, you would end up with haploid coverage 300Mbp. If, similarily to the previous two cases, we assume you assembled 4 haplotypes of 8, you expect 4*300Mbp = 1.2Gbp, which is exactly what you got. Same applies to smudgeplot, so your smudges indeed go all the way to 8x.

This allo- vs. auto- is hard to formally test. You are hinted:

• divergence of haplotypic sets is very high (there is no strong 4n peak in tetraploid, or 8n peak in oktoploid). So possibly Allo?
• divergence within homologous chromosomes (heterozygosity) is very very very low, so whatever the origin of polyploidy was, it was likely coming in sets of 2 (similar thing was seen in strawberries, but they did feature some heterozygosity, so all the tools converged better). This also makes me think it can't be auto- because it would take forever to degenerate this much between haplotypic sets. You can probably quite easily test this by reconstructing a few trees of 1:1 orthologs between haplotypes. In case of allo- origin, some of the copies will be more closely related between species that between haplotypes of single species.

[hungweichen0327]

Thank you very much for the detailed explanation. This is very helpful.

Your interpretation that low heterozygosity may cause the k-mer models to converge to an incorrect coverage estimate makes sense, especially considering the self-compatible and predominantly selfing nature of this species.

I just have one point that I would like to clarify. For the diploid individual, GenomeScope estimates the main k-mer peak at ~38×, and Smudgeplot also reports 1n = 38. When you mentioned that Smudgeplot appears to have converged to double coverage, do you mean that the true 1n coverage should actually be closer to ~19×, or that the AB smudge is actually an AABB-like signal caused by duplicated regions?

I may be misunderstanding this point, so I would appreciate any clarification.

FYI:
Chromosome-scale assemblies generated using ONT and Hi-C data using hifiasm, and then yahs for Hi-C scaffolding.

• 10 chromosome-length scaffolds for the diploid individual (~0.3 Gb assembly)
• 20 chromosome-length scaffolds for the tetraploid individual (~0.55 Gb assembly)
• 40 chromosome-length scaffolds for the octoploid individual (~1.2 Gb assembly)

These were all primary assemblies (p_ctg in hifiasm output) rather than haploid assemblies since I can't get the haploid assembly with the reliable genome size (~0.3 Gb).

Thanks again for your suggestions.

[KamilSJaron]

I just have one point that I would like to clarify. For the diploid individual, GenomeScope estimates the main k-mer peak at ~38×, and Smudgeplot also reports 1n = 38. When you mentioned that Smudgeplot appears to have converged to double coverage, do you mean that the true 1n coverage should actually be closer to ~19×, or that the AB smudge is actually an AABB-like signal caused by duplicated regions?

Ya, exactly that. We used to have a flag in smudgeplot that would assume homozygous sample and double all the labels (I must admit I don't know if it's still there, hopefully it is). Alternatively, you can give smudgeplot the coverage prior too and hopefully it will respct it.

Chromosome-scale assemblies generated using ONT and Hi-C data using hifiasm, and then yahs for Hi-C scaffolding.

When you originally posted, you wrote that 2n=2x=20, which to me suggest that your 10-chromosome assembly, whatever mode it was in, ended up being a 1n representation of your genome and it seems to be the case for the tetraploid and octoploid too (I guess could be summarised as: For diploid: 1n=1x=10, for tetraploid 1n=2x=20, and for octoploid 1n=4x=40, right?)