Appendix B
Thresholding and Quasispecies Diversity

Should classification aim to emphasize relationships? If so, then one tends to be a lumper. Or should classification reflect the power of evolutionary processes to produce differences? Then one will tend to be a splitter.

Most taxonomists try to strike a balance, but in cases when it is not clear where the balance is, sometimes it is necessary to make a philosophical decision, and not everyone will agree on which philosophy to use.
--P. Regal

Sequence similarity searches enable comparative analysis of sequences in a pairwise manner, resulting in optimal global [81], optimum local [111], or near-optimal local [4,5] alignments. This section compares plant and fungal sequences using the BLAST family of similarity search algorithms [4,5]. The purpose is to evaluate alternative parameter sets and justify the parameters used to compute quasispecies diversity in Chapters 3 and 4.

Table B.1 summarizes the sequences on which we will focus. These sequences were analyzed in detail in Chapter 4, and represent several gene family members from a plant, Medicago truncatula, and from two species of arbuscular mycorrhizal fungi from the genus Glomus.

Following procedures described in Chapter 3, I performed two searches with blastn and one search with tblastx [5]. Two different expect value cutoffs were used, such that either $E<10^{-10}$ or $E<10^{-2}$. The tblastx search, which compares nucleotide query and subject sequences as amino acid translations in every possible reading frame, was run with a threshold $E<10^{-2}$. All other parameters were set at their default values, as provided by the pre-compiled binary executable obtained from ftp.ncbi.nlm.nih.gov/blast/executables.

In these searches, both the query and subject set consisted only of those sequences listed in Table B.1. For each match having an expect value smaller (closer to zero) than the threshold, I computed percent identity scores as the ratio of the raw score for matching a subject sequence to the query sequence's self-score, obtained from matching the query sequence with itself. This yields percent identities from zero through 100%. Each sequence always matches itself perfectly, resulting in 100% identity. Partial matches for overlapping fragments have intermediate identity values.

Identity matrices summarize identity values for any pair of query and subject sequences. Figures B.1, B.3, and B.5 illustrate identity matrices that resulted from the three searches described above.

In each case, we are interested in knowing how the number of distinct transcripts varies with the degree of stringency required to consider two transcripts the same quasispecies. Percent identity varies continuously, so a percent identity threshold is used as a criterion for lumping two individual transcripts into a common quasispecies.

Dendrograms constructed from each identity matrix appear in Figures B.2, B.4, and B.6. These were made using the hclust method in R [62]. How does quasispecies diversity vary with the identity threshold, using varied search algorithms and parameters?

In the first case, using BLASTN with $E<10^{-10}$ (Figures B.1 and B.2), quasispecies observed diversity at 90%, 70%, 50%, and 30% threshold identity values is 22, 21, 21, and 20, respectively.

This is the case for which results are summarized in Chapters 3 and 4. Let us now consider two simple alternatives.

In the second case, using BLASTN with $E<10^{-2}$ (Figures B.3 and B.4), quasispecies diversity is the same as in the previous case. However, one sequence (GvCHS1) has joined a cluster of other chitinases from Glomus, having a very weak similarity that was excluded when using a higher expect value threshold. A weak match between this sequence and a plant chitinase (raw score = 30, E=0.008) is also observed in the identity matrix, though it does not appear in the dendrogram. The two sequences match perfectly over a span of 14 nt.

Here, the two BLASTN searches resulted in the same observed diversity. It is easy to imagine that other weak matches might occur in large-scale comparisons. Whether or not these would affect the observed diversity is unclear. I chose to use the more conservative of the two parameter choices, to minimize spurious clustering based on weak matches.

In the third case, using TBLASTX with $E<10^{-2}$ (Figures B.5 and B.6), we observe many more matches. This indicates that identities as amino acids are more readily identified than as nucleic acids. In the dendrogram, clusters have more constituents; fewer singleton clusters are apparent. All but one of the chitin synthases from G. intraradices and G. versiforme cluster together. The fungal phosphate transporter clusters with the two plant transporters, though the degree of identity is too low to be counted as a single quasispecies. Quasispecies diversity ranges from 24 at 90% identity to 16 at 30% identity.

Comparing nucleotides as amino acid translations results in greater sensitivity than comparing them as nucleotides. However, because of the tendency to join transcripts that originated from the genomes of different, reproductively isolated species, considering them as constituents of the same transcript quasispecies is a dubious procedure.

Because of observations such as those described here, I chose to use the first set of parameters in BLAST comparisons to compute observed diversity of transcript quasispecies. Different proteins evolve at different rates [48,69,70], so I thought it appropriate to report results at varied percent identity thresholds.

Table B.1: Fungal and plant sequences subjected to various thresholding criteria. The ID column gives the symbol by which sequences are identified in identity matrix figures; the LOCUS column gives the symbol by which sequences are identified in dendrograms.

ID	$L$	LOCUS	ACCESSION	GENE NAME
Glomus intraradices
a	1532	GiHB1	AF110198	homeobox protein HB1
b	858	GiMYC2	AF110197	MYC2
c	1453	GiMYC1	AF110196	MYC1
d	610	GiCHS1	L77908	chitin synthase
e	617	GiBCHS1	AF260996	chitin synthase, isolate GiBCHS1
f	614	GiCHS3	AF260993	chitin synthase, isolate GiCHS3
g	617	GiCHS2	AF260986	chitin synthase, isolate GiCHS2
h	617	GiBCHS2	AF260985	chitin synthase, isolate GiBCHS2
i	617	GiVCHS2	AF260983	chitin synthase, isolate GiVCHS2
j	617	GiWCHS2	AF260982	chitin synthase, isolate GiWCHS2
Glomus versiforme
k	4116	GvCHS3	AJ009630	chitin synthase, clone Gvchs3
l	481	GvCHS2	AJ009629	chitin synthase, clone Gvchs2
m	638	GvCHS1	AJ009628	chitin synthase, clone Gvchs1
n	1833	GvPT	U38650	phosphate transporter
Medicago truncatula
o	1867	MtPT2	AF000355	phosphate transporter MtPT2
p	1920	MtPT1	AF000354	phosphate transporter MtPT1
q	954	Mt4	AF055921	Mt4
r	1305	MtCHI1	Y10373	chitinase
s	181	MtCHI08g	AF167329	chitinase, clone T130008g
t	265	MtCHI07g	AF167328	chitinase, clone T130007g
u	188	MtCHI06g	AF167327	chitinase, clone T130006g
v	188	MtCHI05g	AF167326	chitinase, clone T130005g
w	191	MtCHI04g	AF167325	chitinase, clone T130004g
x	197	MtCHI03g	AF167324	chitinase, clone T130003g
y	260	MtCHI02g	AF167323	chitinase, clone T130002g
x	245	MtCHI01g	AF167322	chitinase, clone T130001g

Figure B.1: Identity matrix using BLASTN and $E<10^{-10}$. Area of a box is proportional to the percent identity.

Figure B.2: Dendrogram using BLASTN and $E<10^{-10}$. Quasispecies diversity at 90%, 70%, 50% and 30% identity thresholds are 22, 21, 21, and 20, respectively.

Figure B.3: Identity matrix using BLASTN and $E<10^{-2}$. Area of a box is proportional to the percent identity.

Figure B.4: Dendrogram using BLASTN and $E<10^{-2}$. Quasispecies diversity at 90%, 70%, 50% and 30% identity thresholds are 22, 21, 21, and 20, respectively.

Figure B.5: Identity matrix using TBLASTX and $E<10^{-2}$. Area of a box is proportional to the percent identity.

Figure B.6: Dendrogram using TBLASTX and $E<10^{-2}$. Quasispecies diversity at 90%, 70%, 50% and 30% identity thresholds are 24, 22, 22, and 16, respectively.