Selection of Enzymes for Terminal Restriction Fragment Length Polymorphism Analysis of Fungal Internally Transcribed Spacer Sequences

Abstract

Terminal restriction fragment length polymorphism (TRFLP) profiling of the internally transcribed spacer (ITS) ribosomal DNA of unknown fungal communities is currently unsupported by a broad-range enzyme-choosing rationale. An in silico study of terminal fragment size distribution was therefore performed following virtual digestion (by use of a set of commercially available 135 type IIP restriction endonucleases) of all published fungal ITS sequences putatively annealing to primers ITS1 and ITS4. Different diversity measurements were used to rank primer-enzyme pairs according to the richness and evenness that they showed. Top-performing pairs were hierarchically clustered to test for data dependency. The enzyme set composed of MaeII, BfaI, and BstNI returned much better results than randomly chosen enzyme sets in computer simulations and is therefore recommended for in vitro TRFLP profiling of fungal ITSs.

Terminal restriction fragment length polymorphism (TRFLP) profiling was originally developed as a means of genotyping mixed DNA samples (30) and is currently being employed in fungal community ecology studies (3, 5, 6, 7, 10, 13, 19, 22, 26, 27, 29, 33, 38), despite a number of technical and conceptual difficulties (11). Briefly, TRFLP profiling involves amplifying the DNA in pools of mixed genetic material with fluorescently labeled primers, digesting the products with restriction endonucleases, and sizing the labeled terminal fragments in a sequencer. The difference in the positions at which the different restriction enzymes cleave DNA is thought to provide enough variability for such DNA mixtures to be characterized and the contributing organisms to be identified.

However, the technique is not without its problems. DNA extraction and PCR amplification biases burden most modern molecular techniques, including TRFLPs (18, 25). Additionally, concerns exist regarding the ability of the differences between primer-enzyme pairs (PEPs) to generate sufficiently different fragment sizes (2), the success of enzymatic cleavage (2), the dependency on the detection threshold of the sequencer (4), and the accuracy of DNA sizing (1). The choice of the primer pairs and restriction enzymes to be used has also been a matter of concern since the appearance of TRFLP profiling. Liu et al. (30) performed virtual digestion of all the bacterial RNA sequences in the Ribosomal Database Project database (release V) with 10 different enzymes and four primer pairs. This pioneering work showed the importance of avoiding enzymes with highly conserved target motifs, something that later became recognized as a major source of TRFLP bias (2, 14, 16, 32). Similar studies have been performed by Osborn et al. (36), Dunbar et al. (12), Engebretson and Moyer (15), and Cardinale et al. (8).

The first virtual TRFLP analysis involving a database of fungal DNA sequences was performed by Edwards and Turco (14). This consisted of virtual digestion, by use of six restriction endonucleases, of 316 internally transcribed spacer (ITS) sequences belonging to a number of ectomycorrhizal genera. Avis et al. (2) found only small differences in the diversity of the TRFLPs produced in silico by three PEPs when using their own fungal ITS database, although these differences increased with sample number in iterative analysis. Recent advances using automated resources, such as REPK software (9), have allowed optimal enzyme selection for TRFLP profiling of previously defined communities of organisms. This software selects up to four restriction endonucleases capable of discriminating a desired number of sequence groups. However, this system relies on a priori information, which in real biological communities may not available.

The aim of the present work was to improve selection of restriction enzymes for use in the TRFLP profiling of the ITS sequences of unknown fungal communities.

MATERIALS AND METHODS

Sequence acquisition and processing.

The International Nucleotide Sequence Database (INSD) was searched via the NCBI web service (http://www.ncbi.nlm.nih.gov/) in November 2007 for ITS sequences of dicaryal fungi thought to anneal with the ITS1-ITS4 primer pair. While these are not fungal DNA-specific primers, many more fungal ITS sequences contain their complementary sequences than those of the truly fungal specific primer pair ITS1F-ITS4. The diversity of the fragment sizes obtained with these primer pairs should be the same as that obtained with ITS1F-ITS4 since the length of the region between ITS1F and ITS1 is conserved (in fact, it contains the ITS5 universal primer locus). The search strings included the following terms and their etymological variations: (“ascomycete” [organism] AND “internal” [all fields]) and (“basidiomycete” [organism] AND “internal” [all fields]). These were additively employed. The FASTA format files produced were exported to a Microsoft Excel spreadsheet and filtered for the presence of the ITS1 (5′-TCCGTAGGTGAACCTGCGG-3′) and ITS4 (5′-TCCTCCGCTTATTGATATGC-3′) sequences.

Extra bases beyond the primer complementary sequence positions were deleted. ITS sequences only identified above the genus level were discarded, while sequences annotated as cf. or aff. were assumed to be properly identified. The entire database was virtually digested individually with 135 commercially available type IIP restriction endonucleases (see Table S1 in the supplemental material), using the string-finding functions in Excel. If no target was found in an amplicon, its length was assumed to be the size that would be recorded by the sequence analyzer in in vitro analysis. Additionally, sequences were discarded if they contained ambiguous nucleotides (12). Sequences identified at the genus level were verified not to contain redundant information, i.e., no fully identified sequences of the same genus presented the same TRFLP size (see the supplemental material). Finally, each set of data obtained from virtual digestion of the entire database with each PEP was independently filtered so that, for size diversity analysis purposes, a single record was obtained for each size type in each taxon.

The influence of the filtered PEP data set size on the diversity of fragment sizes was corrected by means of random selection of data from each PEP data set to provide final data sets of equal size (i.e., containing the same number of data as the smallest of all PEP data sets). This rarefaction was performed in quintuplet for each PEP data set. The relative abundance (p_i) of each size type in each rarefacted PEP data set was obtained using equation 1, where n represents the number of different taxa sharing that TRFL, S_n is the number of different TRFLs in each of these taxa, and m is the number of different TRFLs in the PEP-TRFL data set. A final filtering step was performed to avoid analyzing fragments outside the size range of the internal standard employed (50 to 1,200 bp) (24). Outsiders were pooled in a single “zero” category.

(1)

Diversity measurements.

Hill's effective numbers of classes (20), which are calculated from classic ecological indices (21, 23), were used to measure the diversity of the fragments' sizes. Hill's indices were calculated for diversity orders representing α values of 0, 1, 2, and ∞ (Table 1) by using the fragment size frequency data set obtained from each of the five rarefacted PEP data sets; the median value was calculated for each index. The PEPs were then ordered according to each of these four values, and the 20 top-scoring PEPs within each diversity rank were pooled together in a top-50 group for further analysis. The square root of Jensen-Shannon's divergence (28) between the averages for the five frequency data sets for each PEP pair in this pool was calculated using equation 2, where JSπ represents Jensen-Shannon's divergence, p₁ and p₂ are two probability distributions, H is Shannon's entropy [], and π is the weighting of each probability distribution (here, 0.5) (28).

(2)

The resulting divergence matrix was subjected to multidimensional scaling in two dimensions by using the ALSCAL and PROXSCAL routines in the SPSS 11.5 software package (with 50,000 random starting points). Tentative sorting was performed by calculating the product of the arctangent-transformed multidimensional scaling coordinates of each PEP. This transformation was intended to standardize both coordinates in order to invest them with equal weight.

TABLE 1.

Hill's effective number of classes and related classical indices

Diversity order	Hill's number of classesa	Ecological index
α
0		Richness
1		Shannon
2		Simpson
∞		Berger-Parker

^an is the number of classes of a given distribution, p_i is the frequency of the class, and i, v_α (n, p) is the efficient number of classes of order α diversity for the given distribution.

Enzyme set selection.

PEPs were tested for data independence by hierarchical clustering of the original (unfiltered) TRFLP data since some enzymes have been reported to provide redundant data (14). Free hierarchical clustering was performed using interval distance measurements, and employing the unweighted-pair group method using average linkages for clustering. One enzyme in each of the main clusters was selected, and the optimal number of these in a putative optimal set was analyzed by comparing the ribotype profiles resulting from fingerprinting differently sized databases. The ribotype richness (the number of different profiles), the percentage of unique ribotypes, and the ratio between congeneric and noncongeneric shared ribotypes were calculated using optimal enzyme sets composed of one to six enzymes. The ITS1- and ITS4-primed TRFLP data for each enzyme were employed in this analysis.

TRAMPR simulations.

The fragments produced by the selected set of enzymes and five randomly generated sets of enzymes were compared by means of TRAMPR software simulations (11, 17). This software allows for computer-assisted matching of TRFLP profiles, relating them to a given molecular database organized by taxonomic and ribotypic sequence similarity. Matching stringency and algorithmic distance measurements can be set by the user. “Knowns” files were generated from the databases mentioned above and used as input for the TRAMPR program. “Sample” files with 30 replicate random communities of 10, 20, and 50 entries each were also generated and loaded. Matching was performed with an error of 0.5 bp and by using the “maximum” distance computing method. Results were exported to Excel and original taxon frequencies calculated using equation 1, where n is the number of original taxa in each TRAMPR grouping, S_n the number of taxa in each TRAMPR grouping, and m the number of TRAMPR groupings in each virtual community. Finally, the resolving power of each set was measured as the square root of Jensen-Shannon's divergence between the original frequency distribution (homogeneous) and the TRAMPR-biased frequency distribution of the original taxa.

RESULTS

The INSD search of fungal ITS sequences retrieved a total of 61,752 entries. Filtering for the presence of sequences complementary to primers ITS1 and ITS4 in two subsets of ITS sequences (totaling 11,298 and 12,716 entries, respectively) was performed. Sufficiently identified sequences with both complementary sequences (4,618 distinct entries) (see Table S2 in the supplemental material) were virtually digested with 135 different endonucleases (see Table S1 in the supplemental material), redundancy filtered, and randomly rarefacted to a common size of 1,659 entries (presented by ITS4-ITS1 MseI) in quintuplet. The top performers from the four different median diversity measurements (Table 1) were pooled together, giving 48 distinct PEPs (Table 2).

TABLE 2.

Pooled diversity for top-scoring PEPs, sorted by effective number of classes of Renyi's order 1 diversity^a

Name	Target	H0	H1	H2	H∞
ITS1 MaeIII	/GTNAC	562	384.4	265.9	36.55
ITS1 MaeII	A/CGT	544	370.9	253.8	37.74
ITS1 ApyI	/CCWGG	552	364.2	237.9	33.61
ITS1 BstNI	CC/WGG	543	363.1	240.0	35.22
ITS4 BstNI	CC/WGG	533	361.3	213.8	24.28
ITS4 ScrFI	CC/NGG	554	360.6	186.7	20.75
ITS4 ApyI	/CCWGG	538	359.5	202.6	21.96
ITS4 BfaI	C/TAG	532	354.8	263.8	68.59
ITS4 MaeII	A/CGT	524	349.1	236.2	48.82
ITS4 StyI	C/CWWGG	517	348.3	263.0	74.09
ITS4 RsaI	GT/AC	505	347.4	266.4	81.04
ITS4 DraII	RG/GNCCY	484	340.1	264.3	63.40
ITS1 StyI	C/CWWGG	504	338.3	261.8	73.77
ITS4 BmyI	GDGCH/C	496	337.4	243.0	41.39
ITS4 DdeI	C/TNAG	553	336.5	181.9	27.14
ITS4 AflI	G/GWCC	495	333.6	217.8	31.25
ITS4 BanII	GRGCY/C	488	331.7	243.3	49.20
ITS1 AcyI	GR/CGYC	511	331.6	224.4	33.15
ITS4 BstUI	CG/CG	515	329.2	230.1	40.56
ITS1 DdeI	C/TNAG	578	328.9	111.5	12.84
ITS1 BsiEI	CGRY/CG	481	325.9	253.5	93.78
ITS1 NspBII	CMG/CKG	477	325.2	252.0	80.70
ITS1 NgoMI	G/CCGGC	483	323.8	255.3	93.85
ITS4 AcyI	GR/CGYC	476	323.2	248.7	70.00
ITS1 BanI	G/GYRCC	485	323.2	248.8	62.65
ITS1 CfrI	Y/GGCCR	493	321.7	238.7	59.20
ITS1 HaeII	RGCGC/Y	479	318.4	241.7	66.89
ITS1 RsaI	GT/AC	526	317.6	169.1	21.06
ITS1 BstUI	CG/CG	552	315.8	140.1	16.47
ITS1 AflI	G/GWCC	515	314.9	127.9	14.29
ITS4 BsrFI	R/CCGGY	461	313.2	239.6	75.93
ITS4 DsaI	C/CRYGG	477	313.1	234.5	75.69
ITS1 ScrFI	CC/NGG	546	308.7	140.9	19.66
ITS1 AhaIII	TTT/AAA	461	305.9	232.9	80.02
ITS1 SfcI	C/TRYAG	461	298.6	230.7	88.51
ITS1 EagI	C/GGCCG	438	291.7	232.8	88.62
ITS4 MaeIII	/GTNAC	527	286.8	83.6	10.91
ITS1 BfaI	C/TAG	566	282.5	92.6	12.89
ITS4 KasI	G/GCGCC	415	275.7	214.3	77.64
ITS4 AatI	AGG/CCT	403	272.9	212.8	74.55
ITS1 PaeR7I	C/TCGAG	407	272.7	217.2	85.55
ITS1 SnaBI	TAC/GTA	405	268.7	213.6	82.19
ITS1 KspI	CCGC/GG	403	260.0	202.5	75.60
ITS4 BsePI	G/CGCGC	393	254.6	200.1	76.12
ITS1 AatII	GACGT/C	387	249.7	196.3	76.36
ITS1 NlaIV	GGN/NCC	515	245.6	60.6	8.680
ITS1 SspI	AAT/ATT	385	243.2	190.2	75.55
ITS4 PvuII	CAG/CTG	371	242.5	195.6	81.03
ITS1 BamHI	G/GATCC	374	239.7	189.6	75.45
ITS4 Bsp1407I	T/GTACA	366	235.2	186.7	76.55

^aH₁, exponential Shannon index; H₀, richness; H₂, reciprocal Simpson index; H_∞, reciprocal Berger-Parker index; /, enzyme's cleavage point.

As an indirect estimate of diversity, two-dimensional scaling of Jensen-Shannon's divergences (28) between the averaged fragment size frequency distributions was performed, resulting in a consistent pattern (Fig. 1) in which dimension 1 relates to TRFLP richness (order 0 diversity) and dimension 2 represents TRFL evenness (in the sense of the Hill series). Tentative final scores were calculated from the rescaled coordinates (Table 3). A putatively optimal group of enzymes formed by MaeIII, MaeII, BfaI, BstNI, StyI, and DdeI was selected from among the independently clustered candidates (see Fig. S1 in the supplemental material), although MaeIII was replaced by RsaI on the basis of cost. The group formed by the ITS1- and ITS4-primed MaeII, BfaI, and BstNI data sets showed the greatest increases in both the number of ribotypes and the number of unique ribotypes (Fig. 2), indicating that the inclusion of any more enzymes would be ineffective. Similarly, the trend of the noncongeneric/congeneric shared ribotype ratio (Fig. 2) suggests that TRFLP profiles are increasingly shared by phylogenetically related sequences when up to three enzymes are used; no major changes are seen if more are included.

FIG. 1.

Common space plot of PROXSCAL (distances as interval; 50,000 random starting points; SPSS 11.5) multidimensional scaling of the Jensen-Shannon square root divergence matrix computed from the average frequency distributions for the top 50 PEPs. Each point represents a different PEP.

TABLE 3.

Single PEP tentative final sorting^a

PEP	Target	MDSX	MDSY	Score
ITS1-ITS4 MaeIII	/GTNAC	−0.7423	−0.2583	0.1614
ITS1-ITS4 MaeII	A/CGT	−0.6506	−0.2471	0.1397
ITS4-ITS1 MaeII	A/CGT	−0.4964	−0.2542	0.1147
ITS4-ITS1 BfaI	C/TAG	−0.3415	−0.3564	0.1126
ITS1-ITS4 BstNI	CC/WGG	−0.6566	−0.1642	0.0946
ITS4-ITS1 RsaI	GT/AC	−0.1920	−0.3770	0.0684
ITS4-ITS1 StyI	C/CWWGG	−0.2076	−0.3394	0.0669
ITS1-ITS4 ApyI	/CCWGG	−0.6226	−0.1208	0.0669
ITS4-ITS1 BmyI	GDGCH/C	−0.2478	−0.1201	0.0290
ITS4-ITS1 DraII	RG/GNCCY	−0.0714	−0.3508	0.0240
ITS4-ITS1 BstNI	CC/WGG	−0.6872	−0.0322	0.0194
ITS4-ITS1 BanII	GRGCY/C	−0.1163	−0.1462	0.0168
ITS1-ITS4 StyI	C/CWWGG	−0.0448	−0.3259	0.0141
ITS4-ITS1 BstUI	CG/CG	−0.1410	−0.0513	0.0071
ITS1-ITS4 AcyI	GR/CGYC	−0.2136	−0.0309	0.0065
ITS4-ITS1 ApyI	/CCWGG	−0.7172	−0.0081	0.0050

^aMDS-X and MDS-Y are the abscissa and ordinate coordinates, respectively, obtained by multidimensional scaling. /, enzyme's cleavage point.

FIG. 2.

Comparisons of differently sized enzyme sets. Databases of 4,618 entries (squares), 2,500 entries (triangles), and 1,000 entries (diamonds) were constructed.

The enzyme set composed of MaeII, BfaI, and BstNI (using both ITS1 and ITS4 TRFLP data) outperformed other randomly selected sets at identifying the members of model communities via their TRAMPR profiles, irrespective of the number of community members. This was true despite the fact that random enzyme sets were expected to be more data independent, since they were formed using either ITS1 or ITS4 TRFLP data (but not both) of six different enzymes. The enzyme set composed of MaeII, BfaI, and BstNI was, in turn, slightly less accurate than a single-primed six-enzyme optimal set (Table 4), in accordance with the simulations shown in Fig. 2.

TABLE 4.

Square roots of Jensen-Shannon's divergence between original sample communities and TRAMPR-biased communities

Set	PEPa						Mean square root (SD) for indicated virtual community size
							10 sequences	20 sequences	50 sequences
3E	MaeII-14	MaeII-41	BfaI-14	BfaI-41	BstNI-14	BstNI-41	0.3084 (0.0622)	0.2900 (0.0448)	0.2803 (0.0411)
6E	MaeII-14	BfaI-41	BstNI-14	RsaI-41	StyI-41	DdeI-14	0.2520 (0.0546)	0.2471 (0.0420)	0.2501 (0.0283)
Rand1	BssGI-41	CfoI-41	EcoNI-41	PacI-41	SwaI-41	TaqI14	0.3072 (0.0532)	0.3129 (0.0380)	0.3253 (0.0199)
Rand2	ApyI-41	BanII-41	BsiWI-14	MfeI-41	NdeI-41	NruI-14	0.4136 (0.0601)	0.4114 (0.0419)	0.4145 (0.0249)
Rand3	AlwNI-14	AspI-41	Fnu4HI-14	HpaI-41	NarI-14	SrfI-14	0.4013 (0.0480)	0.4022 (0.0322)	0.4287 (0.1024)
Rand4	AspI-41	AspHI-14	HaeII-14	Psp1406I-41	ScaI-14	SfcI-14	0.4199 (0.0604)	0.4292 (0.0379)	0.4268 (0.0212)
Rand5	BsaI-41	BspLU11I-41	CfrI-14	HpaI-41	NciI-41	NlaIII-14	0.3772 (0.0477)	0.3629 (0.0346)	0.3776 (0.0201)

^a3E, three-enzyme, double-sided PEP set; 6E, six-enzyme, single-sided PEP set; Rand1 to Rand5, randomly chosen PEP sets. “14” represents the ITS1-ITS4 primer pair, and “41” represents the ITS4-ITS1 primer pair.

DISCUSSION

An optimal, maximally cost-effective set of enzymes for TRFLP analysis of fungal ITS, formed by MaeII, BfaI, and BstNI, is here proposed. The results of in silico TRFLP diversity measurements, hierarchical clustering, and TRAMPR simulations all support the choice of this set of enzymes. Some of the individually top-performing enzymes were the same as those reported in other simulations involving fungi and even bacteria. Moyer et al. (34) reported HhaI, RsaI, and BstUI to be the best enzymes for RFLP profiling when performed virtually using a local bacterial sequence database. MspI, HhaI, RsaI, and BstUI were reported as top performers for TRFLP of bacterial samples by Liu et al. (30), and later, BstUI, DdeI, Sau96I, and MspI were identified as such by Engebretson and Moyer (15). Edwards and Turco (14), whose work involved profiling fungal ITS sequences, identified HaeIII as the top-performing enzyme. The same was reported by Avis et al. (2) and Dickie and FitzJohn (11), together with HpyCH4IV. However, the most-used enzymes in real fungal TRFLP profiling have been HinfI (3, 10, 13, 19, 26, 27, 29, 38), HaeIII (3, 5, 6, 7, 10, 13, 27), AluI (6, 7, 29, 38), TaqI (13, 19, 26, 29, 33), CfoI (29), HhaI and MspI (31), BsuRI (29), and Hsp92II (22), among others. While the outstanding performance of HpyCH4IV (an isoschizomer of MaeII) is in accordance with the present results, a comparison of Tables 2 and 5 shows that most of the other most-used enzymes cannot be considered optimal. Interestingly, the present results for RsaI contrast with those obtained by Edwards and Turco (14), who declared this enzyme unsuitable for TRFLP profiling of fungal ITS sequences. This discrepancy might be due to differences in fragment size detection range; detection thresholds can impose critical limitations on measurement of diversity (4, 15).

TABLE 5.

Diversity values for the most-frequently used TRFLP enzymes reported to be employed with fungi

Name	Hill's effective no. of classes fora:
	D0	D1	D2	D∞
ITS1-ITS4 HinfI	350	210.1	119.6	17.37
ITS4-ITS1 HinfI	324	201.3	132.1	23.86
ITS4-ITS1 HaeIII	491	251.9	114.6	17.18
ITS1-ITS4 HaeIII	492	206.2	47.6	7.68
ITS1-ITS4 AluI	487	312.7	184.2	23.16
ITS4-ITS1 AluI	429	261.4	162.3	30.29
ITS4-ITS1 TaqI	332	194.0	119.5	22.03
ITS1-ITS4 TaqI	301	164.8	69.2	10.29
ITS4-ITS1 CfoI	376	240.1	173.5	41.59
ITS1-ITS4 CfoI	412	24.3	102.0	13.75

^aD₀, order 0 diversity; D₁, order 1 diversity; D₂, order 2 diversity; D_∞, order ∞ diversity.

As Marsh (32) and, later, Engebretson and Moyer (15) indicated for bacterial in silico TRFLP simulations, databases afford biased views of true diversity; not all organisms' DNA have received the same sequencing interest (37). Moreover, Nilsson et al. (35) reported a worrying percentage of misidentified fungal sequences in public databases, some 10 to 21% of all those deposited. Database bias may affect the present results in a PEP evenness-dependent manner, since overestimation of diversity more probably occurs in more-diverse PEP environments.

A more realistic community would show size frequencies dependent on abundance of local taxa, number of ITS copies per taxon, and success of DNA extraction and amplification (4, 6, 18, 36). Assuming the frequencies derived from redundancy filtering, a conservative estimate of the resolving power of the TRFLPs was made.

Real biological communities are available as population sets (PopSets) at the NCBI website, although few of them meet the requirements for use in the current simulation. Only 10 POP sets (search performed in December 2008) corresponding to dicaryal fungi have a minimum of 60% of sequences (a total of 77 sequences) simultaneously showing the complementary sequences to the ITS1 and ITS4 primers, lack ambiguous nucleotides, and have been sufficiently well identified. If sufficient POP sets could be obtained, it would be interesting to select enzyme sets for use in identifying the organisms present in broad ecological systems, e.g., European temperate forests or decaying meat.

It would certainly be possible to select a different optimal enzyme set. MaeIII was rejected on the basis of cost, but it was in fact the best enzyme tested in the current simulation. It has been shown that a set of enzymes selected using six one-sided PEPs can outperform the proposed double-sided three-enzyme set, but this would make the method too expensive, and the gain in accuracy would only be very small. Other factors, such as the enzyme's optimal buffer and working conditions, could be interesting too. If an alternative enzyme set based on richness was constructed, the size distribution of the fragments produced might be uneven and data dependent, while a set based on diversity alone might suffer the same problem. In either case, lower diversity values would be returned and the results would show greater variability. Some of the results of the present work bear this out (see Tables S3 and S4 in the supplemental material).

The proposed enzyme set is nonoptimal in two ways. First, the entire diversity of the INSD ITS database is not fully reproduced by the TRFLPs, and second, it is insufficiently large to include all real diversity. As shown above, two or three enzymes can reflect most of the variation between sequences in this database but still cannot reflect it all, probably due to similarities between close-relative and improperly identified data in the INSD. Greater efficiency might be achieved by using more enzymes (15), but broader in silico simulations searching for more-diverse and data-independent PEPs may lead to requirement of fewer enzymes for achievement of the same resolving power. The number of enzymes to be used is therefore open to discussion, but certainly there must come a point at which too many could be used if databases are only small (14), or too few could be used in an attempt to profile the huge diversity of the world's fungi (37). Thus, the number of enzymes required in TRFLP profiling depends on the combined efficiency of those selected.