1 of 50

Key concept: OTU or ASV

Do we really know what an OTU is or it should be?

2 of 50

What did the term OTU refer to before the era of DNA sequencing?

  • OTU -> operational taxonomic unit is an operational definition Introduced in 1963 by Robert R. Sokal and Peter H.A. Sneath to define simply the group of organisms currently being studied. An OTU is a pragmatic definition to group individuals by similarity, equivalent to but not necessarily in line with classical Linnaean taxonomy or modern evolutionary taxonomy.

  • The definition is intentionally really vague: the object of the observation could be an organism or a taxonomic group. It is up to scientists to define what an OTU is.

3 of 50

What is an OTU in DNA metabarcoding era?

4 of 50

DNA metabarcoding data analysis

Every time we analyse NGS data we need to face the issue related to sequencing errors.

For instance in genomics and transcriptomics we usually trim-out low quality region of reads.

In DNA-metabarcoding we need to account for:

  • Amplification errors:
    • Substitution
    • Chimera
  • Sequencing errors

5 of 50

DNA metabarcoding data analysis

OTU picking is the bioinformatic procedure applied for several years to treat DNA-metabarcoding data, with two general purposes:

  • reduce the effect of amplification and sequencing noise
  • reduce computational requirements.

Historically, OTUs represent groups of sequences with a similarity over a defined threshold, usually 97%.

We will briefly discuss how this threshold was defined and the reason why we need to overtake the idea of a fixed threshold.

6 of 50

What is a species?

A species is often defined as the largest group of organisms in which any two individuals of the appropriate gender or mating type can produce a fertile offspring, typically by sexual reproduction.

Ciccarelli FD et al (2006) Toward automatic reconstruction of a highly resolved tree of life. Science. 3;311(5765):1283-7.

7 of 50

Some definitions

  • For many decades, bacterial species definition was based on genome similarity measured by using DNA-DNA hybridization (DDH). Two genomes are conspecific (belongs to the same species) if their DDH ≥ 70% (Wayne et al, 1987).
  • Today, the genome similarity is measured as ANI (average nucleotide identity). Two genomes are conspecific if their ANI ≥ 95%.
  • Actually, microbiologist use a combined approach to define bacterial species based on genetic and phenotypic tracts (polyphasic taxonomy)

8 of 50

OTU

  • With the advent of sequencing techniques and the availability of full-length 16S rRNA sequences, the OTU was defined using a 97% identity threshold.
  • This value has been shown to correspond to a DDH value ≥ 70% (Stackebrandt and Goebel, 1994) which has been previously accepted as working species for bacteria.

9 of 50

OTU

Relationship between the rRNA 16% identity percentage and genomic DNA hybridization between different couples of microorganisms belonging to the Bacteria domain.

Orange area: 16S identity % and % of genomic similarity, higher than 97% and 70% respectively. Same species.

Blue area: 16S identity % higher than 97% but genomic similarity lower than 70%. Species sharing a similar 16S but not the same genetic content.

Green area: 16S identity % and % of genomic similarity, lower than 97% and 70% respectively.

TAKE HOME MESSAGE: The 16S gene may be very similar but does not always reflect the entire genome.

10 of 50

TAKE HOME MESSAGE: The 16S rRNA gene may be very similar but does not always reflect the entire genome.

Thus be careful using predictive software such as:

-1) PICRUSt2 for prediction of metagenome functions

-2) Tax4Fun2: prediction of habitat-specific functional profiles and functional redundancy based on 16S rRNA gene sequences

11 of 50

OTU (Operational Taxonomic Unit)

First application for OTU definition were based to agglomerative algorithms, a bottom-up method for obtaining hierarchical clustering.

Observed sequences

12 of 50

OTU (Operational Taxonomic Unit)

In the first step the agglomerative algorithm creates a cluster per sequence:

13 of 50

OTU (Operational Taxonomic Unit)

Then it starts to cluster sequence according to the imposed similarity threshold (i.e. 97%):

14 of 50

OTU (Operational Taxonomic Unit)

Then it tries to find the closest clusters and combining them. This operation is called linkage.

15 of 50

OTU (Operational Taxonomic Unit)

Then it tries to find the closest clusters and combining them.

16 of 50

OTU (Operational Taxonomic Unit)

The agglomerative algorithm requires a definition of linkage: how to calculate the distance between when a cluster contains more than one sequence:

  • complete: the distance between two cluster is equal to the highest distance between the elements of each cluster
  • average (UPGMA (Unweighted Pair Group Method with Arithmetic Mean)): the distance between two cluster is equal to the mean of the distances between the elements of each cluster
  • single: the distance between two cluster is equal to the smallest distance between the elements of each cluster

17 of 50

OTU

The advent of NGS technologies and its usage in metabarcoding investigation led to the development of new approaches:

  • reference-based OTU picking
  • denovo OTU picking

Due to read length limitation, we are just analyzing portion of the 16S.

18 of 50

OTU

Reference-based OTU picking. Sometimes referred as “consistent labelling” or “Phylotypes”

Reference Collection

Amplicon sequences

Global Alignment

19 of 50

OTU

Reference-based OTU picking

  • OTU are generated by grouping all the sequences with an identity threshold ≥ 97% with a reference sequence
  • Sequences that do not get a significant identity % are not clustered.

20 of 50

OTU

denovo OTU picking: this approach doesn’t require any reference collection. It builds OTUs just considering the similarity between sequences.

Amplicon sequences

Pairwise Alignments

OTUs

singletons

21 of 50

OTU

Regardless to the applied OTU picking procedure the main result is an OTU table:

Moreover, for each OTUs a representative sequence is defined

sample 1

sample 2

sample 3

sample 4

sample 5

OTU 1

0

44

5

7

0

OTU 2

20

0

100

19

21

OTU 3

3

33

0

3

0

OTU 4

0

2

0

25

0

OTU 5

0

0

0

0

0

22 of 50

OTU

OTU1

OTU2

OTU3

OTU4

OTU5

OTU6

23 of 50

OTU

PRO

CONS

Reference-based

  • reproducible
  • it is not related to the starting dataset
  • different dataset are comparable
  • lack of reference results in loss of information
  • assigns a read to a reference doesn’t necessarily mean you’re observing that sequences.
  • highly influenced by the used reference database.

denovo

  • suitable when a reference collection is not available
  • it is strictly dependent on the starting dataset (sequencing depth and relative abundances)
  • not reproducible
  • not comparable between different analysis

24 of 50

OTU - drawback

  • by using a fixed identity % threshold in OTU checking, OTU-based approaches do not allow to distinguish between low divergent sequences (few nucleotides)
  • One of the goal of microbiome investigation is the identification of specific traits that can be used as biomarker.

Examples: OTU-based surveys of human microbiome are unable to discern among dangerous and not dangerous members belonging to the Neisseria genus.

25 of 50

ASV

Amplicon Sequence Variants (ASV) is a recently developed approach allowing to infer the “real sequences” in the analysed samples, without imposing any similarity threshold.

Actually, it represents an evolution of denoising method developed for 454 data (e.g. AmpliconNoise PMID: 21276213) treatment in order to be applicable to Illumina sequences

26 of 50

ASV

The real advantage of ASV relies in the fact that it does not produce OTU!!!

ASV methods infers the real sequences in samples generating the reads we are analyzing. This real sequences are called features.

This mean that are not related to:

  • any reference database: they are not inferred by comparing it to reference sequences;
  • any denovo OTU picking approach

Features are comparable between different studies/surveys because they refer to real object.

27 of 50

ASV

DADA2 builds an error model to explain if the difference observed between sequences are due to real variability or are PCR/sequencing artifacts.

The error model is not pre-calculated but it is strictly defined by observing the real data. It is inferred by comparing the sequences.

While it is observing the sequences to infer the error model it identifies and removes chimeric sequences.

PAY ATTENTION: different sequencing run must be analysed in different DADA 2 runs. This is due to the fact that different runs generate different error profiles.

The result of the analysis is a “features table” a table (the same structure of OTU table) and the “feature sequences”

28 of 50

ASV

DENOISING

29 of 50

ASV

Error model

Sequences Abundance

ASV1

ASV2

ASV3

ASV5

ASV4

A

T

TT

A

A

T

T

G

T

AA

T

TT

T

TT

C

CA

30 of 50

ASV Table

sample 1

sample 2

sample 3

sample 4

sample 5

ASV 1

0

54

6

7

0

ASV 2

20

0

96

19

21

ASV 3

3

33

0

3

0

ASV 4

0

2

0

24

0

ASV 5

0

0

0

0

0

31 of 50

Is Denoising suitable for all sequencing technologies?

  • DADA2 is able to reliably process Illumina, IonTorrent (also 454) data.
  • You just need to keep in mind that sequencing errors in Illumina are principally substitution, while in IonTorren/454 there are issues related to homopolymer.
  • DADA2 is also able to process PacBio long reads. Currently there is a workflow able to process PacBio CCS (Circular Consensus Sequences).

32 of 50

Chimeras�

When you are dealing with metabarcoding data, probably the most important issue is the chimeras treatment.

Chimeras are amplicon generated from at least two different templates.

  • If your pipeline lacks of a chimera checking step the risk is to overestimate the biodiversity.
  • Chimera removal may be obtained by using specific tools (e.g. ChimeraSlayer) or it is intrinsic into the OTU picking.

33 of 50

OTU

Chimera removal

  • In reference-based Chimera removal, an amplicon reads that share portion of its sequences with different references is considered a chimera and removed. This means that you reference database MUST be chimera checked!!!

Ref 2

Ref 1

Chimeric sequence

34 of 50

OTU

Chimera removal

  • In denovo Chimera removal, the chimera identification is performed while the algorithm is comparing the sequences.

Sequence 2

Sequence 1

Chimeric sequence

35 of 50

OTU - ASV comparison on a mock community�

We compared the two approach by using a mock community. The V5V6 hypervariable 16S regions were amplified and sequenced.

Species

%

Bacteroides fragilis (ATCC 25285D-5)

0.08

Phocaeicola vulgatus (ATCC 8482D-5)

0.093

Bifidobacterium adolescentis (ATCC 15703D-5)

0.067

Clostridioides difficile (ATCC 9689D-5)

0.16

Enterococcus faecalis (ATCC 700802D-5)

0.053

Lactiplantibacillus plantarum (ATCC BAA-793D-5)

0.067

Enterobacter cloacae (ATCC 13047D-5)

0.106

Escherichia coli (ATCC 700926D-5)

0.093

Helicobacter pylori (ATCC 700392D-5)

0.027

Salmonella enterica subsp. enterica (ATCC 9150D-5)

0.093

Yersinia enterocolitica (ATCC 27729D-5)

0.093

Fusobacterium nucleatum subsp. nucleatum (ATCC 25586D-5)

0.067

36 of 50

OTU - ASV comparison on a mock community�

Data import

Data quality visualization

Denoising

ASV table

PE merging

OTU Picking

Chimera Checking

OTU table

97%

98%

99%

Primer trimming

37 of 50

OTU - ASV comparison on a mock community

Input

Primer trimming

merging

Not Chimera

ASV

378,621

368,511

(92.67%)

294,317

(79.92%)

229,600

(62.3%)

OTU 97%

324,218

(87.98%)

274,564

(72.52%)

OTU 98%

273,623

(72.27%)

OTU 99%

266,906

(70.49%)

38 of 50

OTU - ASV comparison on a mock community

39 of 50

OTU - ASV comparison on a mock community

40 of 50

How to denoise with QIIME2

qiime dada2 denoise-paired \

--i-demultiplexed-seqs demux-paired-end.qza \

--p-trim-left-f 13 \

--p-trim-left-r 13 \

--p-trunc-len-f 150 \

--p-trunc-len-r 150 \

--p-n-threads 2 \

--o-table table_16S.qza \

--o-representative-sequences rep-seqs_16S.qza \

--o-denoising-stats denoising-stats_16S.qza \

--p-n-reads-learn 50000 \

--verbose \

--p-max-ee-f 2 \

--p-max-ee-r 2

Qiime2 implements dada2 to perform data denoising. 4 version are available:

  • denoise-paired
  • denoise-single
  • denoise-pyro
  • denoise-css

41 of 50

OK… This is the theory, but what about real samples?

42 of 50

OTU or ASV?

The answer is:

«It depends on your project and the target organisms»

To ensure accurate results when using DNA metabarcoding, it's important to have a solid understanding of the biology of your target group beforehand.

43 of 50

Challenges in ITS-based Fungal Community Studies�

Kauserud (2023) ITS alchemy: On the use of ITS as a DNA marker in fungal ecology. Fungal Ecology ISSN 1754-5048.

44 of 50

Challenges in ITS-based Fungal Community Studies�

  • Numerous ITS alleles within fungal species.

  • Caution against using Amplicon Sequence Variants (ASVs) as units in ITS-based fungal community studies.

  • Markers with high intraspecific variation lead to overestimation of diversity.

  • Treating each ITS haplotype as a biological entity exaggerates diversity in statistical analyses (Fig. 2b).

Kauserud (2023) ITS alchemy: On the use of ITS as a DNA marker in fungal ecology. Fungal Ecology ISSN 1754-5048.

45 of 50

DADA2 and ASVs�

  • DADA2 corrects sequencing artifacts in DNA metabarcoding data.

  • ASV term associated with DADA2, used to separate original DNA templates from sequencing noise.

  • Miscommunication that ASVs are different from Operational Taxonomic Units (OTUs).

  • Clarification: ASVs and OTUs are conceptually similar, both involve clustering sequences.

Kauserud (2023) ITS alchemy: On the use of ITS as a DNA marker in fungal ecology. Fungal Ecology ISSN 1754-5048.

46 of 50

The Myth of exactness

  • DADA2 branded as an "exact" approach, but not entirely accurate.

  • High correspondence observed in mock communities, but additional ASVs likely due to errors or intragenomic variation.

  • DADA2's likely reduced accuracy in real samples with higher richness and complexities.

  • Highlighting the need for caution in interpreting results from real-world samples.

Kauserud (2023) ITS alchemy: On the use of ITS as a DNA marker in fungal ecology. Fungal Ecology ISSN 1754-5048.

47 of 50

ASVs and OTUs Integration

  • Reiteration: ASVs are not conceptually different from OTUs.

  • Recommendation: Report OTUs generated by DADA2, specifying clustering thresholds used.

Kauserud (2023) ITS alchemy: On the use of ITS as a DNA marker in fungal ecology. Fungal Ecology ISSN 1754-5048.

48 of 50

The dream is to use DADA2 and then cluster my ASVs with a specific cutoff (e.g. 97%).

Is this possible?

49 of 50

ASV clustering with qiime vsearch

qiime vsearch cluster-features-de-novo

Data import

Data quality visualization

Denoising

ASV table

OTU (97%) table

97% 98% 99%

Primer trimming

ASV clustering

50 of 50

How to cluster at 97% the ASV with QIIME2

qiime vsearch cluster-features-de-novo \

--i-table table_16S.qza \

--i-sequences rep-seqs_16S.qza \

--p-perc-identity 0.97 \

--o-clustered-table table-16S-dn-97.qza \

--o-clustered-sequences rep-seqs-16S-dn-97.qza