Results

Validation sequence accession numbers, gene names, and comparison results appear in Table 2.3. Incorrect inferences are underlined. The word-counting method was generally quite reliable when tested against sequences of known origin, being wrong in 3 cases of 50. This indicates a failure rate of 6%, all false negatives under the null hypothesis that a transcript originates from the plant host. Performance of the method was not influenced by whether the isolated source of a sequence was an mRNA or DNA molecule, as indicated by the column labelled ``mRNA?''.

Table 2.3: Dissimilarity (D) comparison results from fifty validation sequences.

ACCESSION	GENE NAME	mRNA	L	$D(A)$	$D(B_1)$	$D(B_2)$
		?	(nt)	PLANTS	STRAMEN.	BACTERIA
Glomus versiforme
AJ009628	chitin synthase Gvchs1	N	638	2535.2	2468.6	2718.4
AJ009629	chitin synthase Gvchs2	N	481	2203.2	2050.0	2286.0
AJ009630	chitin synthase Gvchs3	N	4116	7205.9	5235.8	5985.8
U38650	phosphate transporter	Y	1833	3937.9	5702.3	6514.3
Glycine max
J01297	actin SAc3	N	1620	3322.0	4554.6	5329.7
K00821	lectin Le1	N	2152	4124.6	6558.3	7928.3
M64267	iron superoxide dismutase	Y	1056	2773.6	3761.2	4269.2
Medicago truncatula
AF000354	phosphate transporter MtPT1	Y	1920	3800.3	5630.7	6654.2
AF000355	phosphate transporter MtPT2	Y	1867	3673.9	5390.1	6424.0
AF055921	Mt4 genomic sequence	N	954	2631.9	4004.4	4539.1
AF106929	cell wall protein AM1	Y	885	3433.6	4200.0	4774.3
AF106930	translation initiation protein AM3-1	Y	3154	4557.6	5982.7	7212.8
AF106931	translation initiation protein AM3-2	Y	1384	3371.1	4130.0	4644.4
AJ132891	ha1 gene, exons 1-22	N	3620	4383.2	8683.6	10730.7
AJ388847	MtNo213 superoxide dismutase	Y	530	2110.2	2219.8	2367.0
AJ388865	MtNo233 triosephosphate isomerase	Y	563	2171.6	2405.6	2618.6
U16727	peroxidase precursor rip1	N	2603	4246.1	8210.0	9901.9
U38651	sugar transporter	Y	1728	3619.6	5128.4	5976.5
X57732	leghemoglobin Mtlb1	N	1073	3021.3	5029.1	5845.9
X57733	leghemoglobin Mtlb2	N	592	2045.9	3156.0	3568.2
X60386	lectin lec1	N	1363	3228.8	4935.4	5605.6
X60387	lectin lec2	N	1192	3142.8	4472.6	4985.9
X82216	lec3	N	1155	2928.4	4283.3	4930.8
X68032	ENOD12	N	772	2780.4	3679.7	4096.5
X99466	ENOD16	N	1142	3124.6	4535.5	5156.2
X99467	ENOD20	N	1405	4003.6	5294.7	5966.7
Y10267	glutamine synthetase	Y	1413	3116.1	4506.5	5292.1
Y10373	chitinase	Y	1305	3369.5	4090.4	4703.4
Phytophthora infestans
AF004951	surface glycoprotein elicitor, inf2A	Y	648	3428.4	2421.9	2589.1
AF004952	surface glycoprotein elicitor, inf2B	Y	701	3611.7	2514.5	2698.6
L23938	ipiO2	N	1556	4125.2	4339.5	4855.5
L23939	ipiO1	N	1826	4360.5	4580.9	5259.0
L24206	ipiB1	N	1726	6086.7	4584.3	5159.3
M59715	actin actA	N	1736	5137.1	3637.2	4420.0
M59716	actin actB	N	1405	4425.3	3569.5	4141.6
M83535	calmodulin calA	N	1358	4063.0	3724.9	4138.1
X64537	tigA	N	2448	6221.0	4193.8	5181.9
P. capsici
U42304	chitin synthase chs	N	449	2238.8	1882.7	1997.5
P. parasitica
X97205	cellulose binding-elicitor lectin	Y	918	3819.1	2876.0	3208.4
Sinorhizobium meliloti
AF040724	nodD	N	1776	5317.9	4197.8	4179.4
AF110770	superoxide dismutase sodA	N	1196	4898.2	3343.2	2916.0
M61753	exoD	N	858	4071.8	2847.2	2372.9
M68858	nodulation protein nodP & nodQ	N	3476	9992.3	5288.0	3954.7
M96261	phosphate regulators, phoU & phoB	N	1178	5332.7	3359.3	2866.4
U90221	syrA	N	1102	4176.2	3375.8	3220.8
X01649	nodA, nodB, & nodC	N	3373	7684.1	4819.5	4646.1
X03065	regulatory nitrogen fixation fixD	N	2111	5723.0	4249.8	4228.3
X17523	glutamine synthetase II	N	990	4303.5	2959.9	2720.4
Y08500	putA	N	3804	13623.3	6376.5	4212.0
Agrobacterium tumefaciens
U91632	sugar transporter gguA & membrane-
	spanning permeases gguB & gguC	N	4185	11132.6	5959.4	4551.4

Table 2.4 summarizes the percent GC content and hexamer composition among the ten most abundant hexamers for training sequences. Plants and fungi (pooled Zygomycetes and Chytridiomycetes) have below 50% mean GC composition, their means differing by less than 3%, and have large standard deviations (5-8%). Sequences from Stramenopiles (pooled with P. infestans ESTs) and from rhizobacteria have greater than 50% mean GC content. Thus, fungi and plants are least distinguishable based on GC content, whereas Stramenopiles and rhizobacteria are more readily distinguished from plants. Differences in GC content between taxa are also reflected in the common hexamers. Abundance of the poly-A and poly-T hexamers among plants and fungi is clearly evident, but less apparent among Stramenopiles and rhizobacteria (Table 2.4).

Table 2.4: Percent GC composition and ten most common hexamers in training sequences. For each library, mean ($\bar{x}$) and standard deviation (s) of percent GC content, and ten most abundant hexamers (by rank) in each training set, with percentage of all hexamers represented by that hexamer, are shown.

	PLANTS		STRAMENOPILES		FUNGI		RHIZOBIA
	$\bar{x}$	s	$\bar{x}$	s	$\bar{x}$	s	$\bar{x}$	s
% GC	40.7	5.18	55.8	4.24	43.6	8.47	60.0	3.57
RANK
1	0.250	AAAAAA	0.126	AAGAAG	0.257	AAAAAA	0.159	GCCGGC
2	0.182	TTTTTT	0.113	CAAGGA	0.184	TTTTTT	0.159	CGCCGC
3	0.118	TTTGTT	0.111	CAAGAA	0.114	CAAGAA	0.157	GGCGGC
4	0.117	ATTTTT	0.108	AAAAAA	0.110	GGTGGT	0.157	GCGGCG
5	0.117	TATTTT	0.105	GCTGCT	0.109	AATAAA	0.156	CGGCGC
6	0.111	TTGTTT	0.098	CTGCTG	0.108	AAGAAG	0.150	CGGCGA
7	0.110	ATATAT	0.096	AAGGAG	0.106	AAAGAA	0.147	TCGGCG
8	0.109	TTATTT	0.096	GCCAAG	0.104	AAAAAT	0.147	TCGCCG
9	0.109	GAAGAA	0.094	TGCTGC	0.104	AAATAA	0.141	GCGCCG
10	0.108	AAAAAT	0.092	GAGGAG	0.100	TGGTGG	0.141	CCGGCG

Distributions of GC content are approximately normal in two of three cases studied, those of axenic P. sojae cultures (Figure 2.1). For sequences from infected plant cultures, a bimodal distribution is apparent. Roughly 25% from a total of 927 infected G. max sequences contain less than 50% GC; most of these are likely plant transcripts [93]. This is considerably greater than in axenic P. sojae cultures, in which fewer than 5% of mycelia and zoospore isolates contain less than 50% GC.

Figure 2.1: Distribution of GC base composition in two axenic Phytophthora sojae mycelia (blue) and zoospore (cyan) cDNA libraries and infected Glycine max (green). (A) Probability densities for histogram bin sizes of 0.02 (2%) in base composition. (B) Cumulative probability distributions.

Several properties of cumulative distribution functions warrant comment, to help explain similar plots from word dissimilarity comparisons (Figures 2.1B and 2.2A). The median of a distribution occurs where the function reaches a cumulative probability of 0.5. Medians from all three P. sojae libraries are similar, varying by less than 4% GC (Figure 2.1B). Other moments of the distributions are readily apparent; the variance is inversely related to the slope at the median value of the function. A useful property of cumulative distribution functions is that any point on the y axis gives the integrated area (cumulative probability) under the curve. We use this property to test for significant dissimilarity differences (Figure 2.2A). In this case, $\alpha=0.088$ and $\beta=0.032$.

Figure 2.2: Cumulative distribution functions (cdfs) of hexamer dissimilarity. (A) Calculation of statistical parameters from cdfs A and B. Overlap in the upper tail of A with B and the lower tail of B with A are likely regions for error. We find the false positive rate $\alpha$ where $1 - cdf_A$ intersects 0 [ $cdf_A(0) = 1 - \alpha$], and the false negative rate $\beta$ where $cdf_B$ crosses 0. Also shown are the medians $\mu$ for each distribution, where $cdf(\mu) = 0.5$. (B) Calibration curves for plant (Glycine and Medicago, solid black line) and Stramenopile plus P. infestans EST (dashed black line) training sequences. Superimposed distributions of test results show dissimilarity differences for infected G. max (green) and axenic P. sojae mycelia and zoospore sequences (blue and cyan, respectively).

Calibration curves from hexamer dissimilarity tests, shown in Figure 2.2B as solid black lines for plant and dashed black lines for Stramenopile training sequences, are approximately normal. The medians differ considerably, with only about 10% overlap in the two distributions' tails about the neutral t value of 0. Superimposed are comparison curves from P. sojae test sets (Figure 2.2B), which parallel the GC composition curves in Figure 2.1B but show slightly less variance. Axenic sequences are clearly more like Oomycetes than plants in hexamer composition, with all but a small percentage having positive t values. Plant-like sequences are as abundant in the mixed library as detected by GC composition, about 23%. As expected, the two methods agree, having positively correlated values for GC and t ($r^2=0.852$, $P<2 \times 10^{-16}$, $\nu=2641$).

Looking in more detail at the paired dissimilarity values (Figure 2.3), we can see which individual sequences are more or less like plant and pathogen. The magnitudes of dissimilarity are also apparent, with longer sequences having larger dissimilarity values. BLASTX similarity searches against the protein sequences in nr (from ftp.ncbi.nlm.nih.gov/blast/db) revealed that none of the 12 plant-like mycelial transcripts significantly resemble known proteins ($E>10^{-4}$). Among the top ten most plant-like transcripts from the infected G. max library, 3 had no significant matches, 4 matched putative A. thaliana proteins, and 3 matched known G. max proteins: cytochrome P450 (accession AF022460, $E=9 \times 10^{-35}$), methylglyoxalase (accession P46417, $E=8 \times 10^{-35}$), and a ripening related protein (accession AF127110, $E=4 \times 10^{-71}$). Therefore, the majority (70%) of plant-like transcripts in the infected soybean library strongly resemble characterized plant sequences.

Figure 2.3: Paired dissimilarity test results for infected G. max (green) and axenic P. sojae mycelia and zoospore sequences (blue and cyan, respectively), compared with plant and Stramenopile plus P. infestans EST training sequences. The identity function indicates equal dissimilarity to both training sets, D(A)-D(B)=t=0.

Figure 2.4 shows that calibration curves from comparing plant and microbial symbiont training sets have good separation and minimal overlap (about 10%) in two of three cases, but not for the training set comprised of Zygomycetes and Chytridiomycetes, which overlaps considerably with plants (Figure 2.4B). The associated error rates are $\alpha=0.126$ and $\beta=0.207$. When comparing between plants and bacteria, the error rates are $\alpha=0.052$ and $\beta=0.084$, much lower than when comparing plants (Medicago) with fungi (Zygomycetes and Chytridiomycetes). Error rates for comparing Stramenopiles and P. infestans ESTs with plants are as in Figure 2.2.

Figure 2.4: M. truncatula and microbial symbiont comparisons. Calibration curves compare plant training sets (solid black lines) with one of three microbial symbiont training sets (broken black lines): (A) Stramenopile and P. infestans EST sequences, (B) pooled Zygomycete and Chytridiomycete coding sequences, and (C) sequences from the genera Rhizobium, Sinorhizobium, and Bradyrhizobium. Cumulative distributions of test results from M. truncatula axenic and microbial symbiont mixed cultures appear in each panel (colored lines).

Also shown in Figure 2.4 are cumulative distributions from comparisons with M. truncatula and microbial symbionts. All resemble calibration curves from plant sequences, having similar medians and slightly less variance than the plant calibration curves. Comparison curves show that the great majority of test sequences are more plant-like than otherwise, with 20% or less resembling microbial symbionts more closely than plants. A greater proportion of microbial sequences is present in the M. truncatula-G. versiforme interaction library (20%, Figure 2.4B) than in the P. medicaginis-infected M. truncatula library (5%, Figure 2.4A). Contrary to expectations, the negative control test set from S. Long's root-hair enriched library (MtRHE) [34] had a greater proportion of putative microbial sequences present (7% and 25%) than any of the libraries isolated from symbiont-associated cultures! The axenic and nodulating root libraries had the smallest portion of putative microbial transcripts (< 2%, Figure 2.4C), with the axenic library closely resembling nodulating root libraries. The method of preparing a library can affect the proportion of plant and non-plant sequences, as discussed later.

Paired dissimilarity values in Figure 2.5 show in greater detail which sequences are more or less like plant and symbiont. Sequences from an interaction library and axenic negative controls appear together for comparison. Considerable variation in the degree of dissimilarity to both training sets is clear, largely due to variation in the length of sequences within test sets. Consistent with the cumulative distributions of D(A)-D(B) in Figure 2.4, most sequences lie above the diagonal, and resemble the plant host more closely than the microbial symbiont. Mycorrhizal test sequences are more difficult to differentiate than sequences from the rhizobacterial or pathogenic associations, as seen by the diminished variation about the diagonal in mycorrizal comparisons (Figure 2.5B), contrasted with comparisons from pathogen-infected and nodulating root libraries (2.5A and Figure 2.5C, respectively).

Figure 2.5: Paired M. truncatula and microbial symbiont comparisons. Each point indicates the dissimilarity of a test sequence with a plant training set and one of three microbial symbiont training sets: (A) Stramenopile and P. infestans EST sequences, (B) pooled Zygomycete and Chytridiomycete coding sequences, and (C) sequences from the genera Rhizobium, Sinorhizobium, and Bradyrhizobium. Sequences from M. truncatula axenic (green) and microbial symbiont mixed culture libraries are represented in each panel. The color of a point indicates the library from which the transcript was sequenced. The identity function (y=x) is also shown.