Amplified products from 397 trap gene clones were subjected to RFLP analysis using three enzymes (EcoRI, HpaII, and Taq^αI) and were analyzed by either agarose or acrylamide gel electrophoresis. A total of 88 distinct RFLP patterns (haplotypes) were detected, of which 28 distinct haplotypes were found among clones originating from multiple birds. Because these 28 distinct haplotypes were found independently in two or more birds, they are referred to as being "confirmed" trap haplotypes and represent the minimum number of haplotypes that are likely present in these infected birds (Table 2). One predominant RFLP haplotype (77) was found in all birds, and comprised 72% of all clones. Subsequent SSCP analysis of several clones with this predominant RFLP pattern revealed multiple additional banding patterns indicating further diversity among these RFLP-defined haplotype 77 clones (data not shown). Clone data from birds 15 and 18 were removed from further numerical summation due to minimal data collection with only 2 clones and 1 clone, respectively, obtained from these birds. In the 16 remaining birds, the minimum number of distinct trap haplotypes per bird ranged from 3 to 10, with an average of 5.9 haplotypes per bird. The average frequency of haplotype 77 was 0.63, but ranged from 0.17 in a wild 'Elepaio from Kauai to 0.80 in an 'Amakihi from Hawaii. We estimated diversity using the Simpson's index of diversity (D) 22 and the Shannon-Wiener index (H) [reviewed in 23], and results are shown in Table 2. Diversity estimates by both indices produced parallel values and ranged from D = 1.49–6 and H = 0.69–1.79. These (generally) low parasite diversity values estimated in most individual hosts reflect the unequal abundances of haplotypes detected. One exception is 'Elepaio 17 from Kauai with D = 6, in which each haplotype was detected equally. Since the total number of distinct haplotypes detected is an indicator of within-host parasite variability and is likely dependent on the number of clones evaluated, we include the number of haplotypes/total clones ratio (hap/clones) as an additional reference indicator of variability. The hap/clone ratio varies among hosts ranging from 0.15 (in 'Amakihi 6 and 'Amakihi 11) to a value of 1.00 (in 'Elepaio 17). A rarefaction curve was generated in Primer v.5 24 to provide comparable estimates of expected number of haplotypes (or expected species richness, ES_n) based on number of clones evaluated (Figure 1). Mean values of ES_n were plotted in Systat (vers 11.00.01), with 95% confidence intervals as indicated.

Average clonal diversity of isolates K1, KV115 and wild isolates, as well as those of clones originating from survivors and non-survivors are presented in Table 3. The diversity estimates of isolates K1 and KV115 are only marginally distinct from one another and this difference is not statistically significant by any index (described in Table 3, legend). Among the five wild isolates the overall average hap/clones ratio was 0.37, with those originating from Kauai having 10 haplotypes among 20 clones (hap/clones = 0.50), and those from Hawaii having eight haplotypes among 18 clones (hap/clones = 0.44). While diversity of clones originating from surviving birds are marginally higher than those of non-surviving birds based on all indices using similarily-generated clones (80 cycles, PCR+1 and Promega Taq), this difference was not statistically significant (Table 3, legend). Unique haplotypes were detected among isolates. Four of the 28 total haplotypes in Table 2 were unique to Kauai isolate K1 (3, 4, 27, and 58), and another three were detected in only K1 or wild Kauai isolates (10, 36 and 73). Only two of the 28 total haplotypes were unique to Hawaii KV115 (47 and 49) and not detected in any other isolate; only one of the haplotypes (44) was detected among wild isolates but not in either KV115 or K1. Five haplotypes (1, 41, 51, 56, and 62) were unique to non-survivors but not survivors or wild isolates, and an additional six haplotypes were present in non-survivors that were also present in wild isolates (8, 25, 30, 34, 36, and 59). Only two haplotypes (10 and 73) were found in survivors but not non-survivors and these were also detected among wild isolates.

To examine the possible effects of PCR enzyme fidelity or specificity we performed side by side amplifications with 35 total cycles in two 'Amakihi (6 and 11) experimentally infected with isolates K1 and KV115, respectively, using standard Taq polymerase (Promega) and the FastStart High Fidelity PCR System (Roche) (Table 4). In both 'Amakihi 6 (isolate K1) and 'Amakihi 11 (isolate KV115), we observed diversity estimates greater (by all indices) among clones amplified with Promega Taq than in clones produced using the Roche system when other variables (35 cycles, PCR +1 methods) are held constant (Table 4, Trial 3 vs. Trial 4). In fact, the haplotype/clone ratio was reduced by 50% in both birds using the Roche enzyme system. We investigated the effect of cycle number on diversity by comparing 80 cycles vs. 35 cycles while holding other variables (Taq Promega polymerase, and PCR+1 methodology) constant (Trial 2 vs Trial 3). In both birds, we observed diversity values lower among clones generated with 35 cycles than among those generated by 80 cycles (by all indices). We also compared PCR + 1 amplification methods 2125 with standard amplification and cloning methods in these two birds (Trial 1 vs. Trial 2). We observed greater diversity estimates of clones generated using PCR+1 methods as compared with those generated using standard methodology based on the Simpson's and Shannon-Wiener indices, but not the haplotype/clone ratio. Statistical analysis was not completed for any of these estimates in Table 4 due to low sample sizes (n = 2 birds) and are reported simply as empirical observations.

Sequence information was obtained across the entire trap fragment (1624 bp, not including primer sequences) from 33 clones originating from 12 individuals (Table 1) and assigned Genbank nos. (EU760855-EU760887). Eighteen clones of the RFLP "dominant" haplotype 77 and fifteen clones representing 12 different haplotypes were sequenced. Two clones each were sequenced representing RFLP haplotype 12, and three clones of haplotype 59.

All sequences appeared highly similar to previously obtained Plasmodium relictum trap sequences 1819. Average nucleotide composition is 26% T, 16% C, 41% A, 17% G with an overall GC content of 33%. Overall π = 0.006 (0.001) (Tamura 3-parameter) (n = 33), with the ratio of nonsynonymous to synonymous nucleotide substitutions (d_N/d_S) = 1. Diversity of sequences representing clones of the dominant RFLP pattern variant 77 (n = 18) is π = 0.003 (0.001). Within-individual diversity is estimated at π = 0.003 (0.001) in both 'Amakihi 11 (n = 10, isolate KV115, Roche enzyme 35 cycles) and 6 (n = 10, isolate K1, Roche enzyme 35 cycles). In comparing the two experimental isolates, diversity of sequences originating from the K1 isolate (n = 13 among 4 birds) is π = 0.005(0.001), and of those from the KV115 isolate (n = 16 among 4 birds) is π = 0.006 (0.001); sequences from the four wild isolates (n = 4 among 4 birds) is π = 0.013 (0.002).

A schematic of the trap gene of P. relictum is presented in Figure 2. These trap sequences are composed of functional domains similar to those previously described in other Plasmodium species 2627 and in P. relictum 1819. The amino acid sequence corresponds to amino acids 56–533 of the P. falciparum clone T9/96 sequence 9. An alignment file (Additional File 1) of the amino acid sequences (543 amino acids) is provided, and the nucleotide alignment file is available upon request. The core of Region II (which is also found in many other proteins) is WTPCSVTCG, which differs by one amino acid from that of other species of Plasmodium including P. falciparum 9 and P. gallinaceum 27 (WSPCSVTCG). The repeat region consists of 9 highly conserved repeats of 12 or 13 amino acids followed by a region of semi-conserved repeats between the repeat region and transmembrane region in Figures 2, and 3. The amino acid sequences (frequency) of repeats are as follows with the most highly conserved sites underlined and the cell-recognition signal RGD 9 conserved in three of the nine repeats in bold: repeat 1

PE

VPD

PE

VPD

PE

VPD

PE

VPD

PE

VPD

PE

VPD

PE

VPY

PE

VPD

PE

VPD

Diversity across the entire sequence includes a total of 143 polymorphic nucleotide sites resulting in 87 amino acid changes. Deletions are detected at two sites in the 5' region (amino acids 124 and 130) and at one site in the repeat region (amino acid 170). The 13 polymorphic amino acid sites occurring in two or more sequences are of particular interest. Of these, five sites are dimorphic, involving only two amino acid substitutions. Site 539 bp (amino acid 179) is either K or E, site 570 bp (amino acid 189) is either H or R, sites 1178 bp (amino acid 392) and 1349 bp (amino acid 449) are either I or V. Of particular interest is site 1289 bp (amino acid 429), a G to C mutation, which results in a change from proline to alanine at the amino acid level. All sequences from a given individual, regardless of polymorphism at other sites, appear to be either proline or alanine at this site (specified in Table 2). All 10 sequences from 'Amakihi 11 and all 4 sequences from 'Amakihi 9 have alanine at this site, and all 10 sequences from 'Amakihi 6 have proline at this site.

The 33 sequences are presented in Figure 3 as a phylogenetic network [TCS,28]. The TCS software collapses sequences into haplotypes and calculates their frequencies. These frequencies are used to estimate haplotype outgroup probabilities, which correlate with haplotype age 28. Ancestral haplotypes are identified by the internal positioning (shown boxed in Figure 3), and determined by degree of commonness and number of lineages derived from them 29. Tip clades connected by only one connecting branch to the rest of the network are considered as derived. Based on these 33 input sequences, the ancestral haplotype is identified as originating from K1 (Kauai) isolate from 'Amakihi 6 and is represented by sequences 6.165 and 6.182. Three single step haplotypes or clades are derived from this ancestral haplotype, one of which (represented by sequences 11.175 and 11.172) forms a clade with most all other sequences from Hawaii (isolates KV115, and 4 wild Hawaii isolates). All other haplotypes involve multiple mutational differences (indicated by circular connections) with a maximum of 17. Within host diversity appears high with a total of nine haplotypes represented by 10 sequences from 'Amakihi 6 infected with isolate K1. Among birds infected with isolate KV115, seven haplotypes were represented by 10 sequences from 'Amakihi 11 and four haplotypes were represented by four sequences from 'Amakihi 9. The G/C polymorphism at nucleotide 1289 appears very influential in the network estimation as all haplotypes on the left side of Figure 3 have a C at this position and all those on the right side have a G in this position. Those on the left side consist of haplotypes from four 'Amakihi infected with K1 (Kauai), and one haplotype each from wild isolates from an Elepaio from Kauai and an I'iwi from Hawaii. Haplotypes on the right side originate from four 'Amakihi infected with isolate KV115 (Hawaii), and one haplotype each from wild isolates from an Elepaio from Kauai and an 'Amakihi from Hawaii.

The two laboratory isolates of P. relictum that were used in this study originated from wild hosts on Kauai or Hawaii Islands. The KV115 isolate was obtained in 1992 from a wild 'Apapane (Himatione sanguinea) captured at Kilauea Crater in Hawaii Volcanoes National Park 33. The K1 isolate was obtained from pooled blood samples from three Elepaio (Chasiempis sandwichensis) captured in the Alakai Wilderness Preserve in August 1994 20. Blood from only one of the three Elepaio was later found to have a detectable parasitemia by microscopy. Fresh, heparinized blood from wild hosts was inoculated intravenously into individual canaries (passage 1) to amplify the number of parasites prior to aliquoting and cryopreservation of the isolates.

The KV115 isolate was passed once in a canary prior to freezing. Several weeks prior to infection of 'Amakihi in this study, an aliquot was thawed, deglycerolized, and inoculated intravenously into an uninfected canary (passage 2). When parasitemia was detectable by blood smear, approximately 0.1 cc of heparinized whole blood was passed intravenously into a one week-day-old Pekin Duckling (passage 3), passed two more times into Pekin Ducklings to amplify parasitemia and then inoculated into three Pekin Ducklings (passage 6) that were used to infect mosquitoes (Culex quinquefasciatus) as described by Atkinson et al. 33.

After initial recovery in a canary (passage 1), The K1 isolate was passaged into a second canary (passage 2), and 4 Pekin Ducklings (passages 3–6) prior to aliquoting and cryopreservation. Several weeks prior to infection of 'Amakihi, an aliquot was thawed, deglycerolized and inoculated intravenously into a Pekin Duckling (passage 7) and subsequently passaged two more times in Pekin Ducklings (passages 8 and 9) to increase parasitemia prior to infection of mosquitoes.

Genomic DNA was extracted from tissue samples using standard techniques. Briefly, samples were digested in 1 ml of 50 mM Tris-HCl, 100 mM sodium EDTA, 2% SDS with 0.5 mg/ml proteinase K overnight at 55°C. DNA extractions were completed with one phenol, two phenol/dimethyl chloride 1:1, and one dimethyl chloride/isoamyl alcohol (24:1) extraction steps, and dialyzed against 10 mM Tris-HCl, pH 8.0, 1 mM EDTA. DNA concentration was quantified by spectrophotometry or agarose gel electrophoresis.

Gel-purified PCR products or PCR + 1 products were ligated into the pCR 2.1-TOPO vector and transformed into TOP10F' E. coli using the manufacturer's suggested protocol (TOPO TA Cloning Kit, Invitrogen). Colonies were screened by PCR amplification using primers P5 and P6 for regular PCR products or plasmid primers M13F20 and M13R (Invitrogen) for PCR + 1 products. PCR + 1 colonies were further screened by digestion of the M13 PCR product with the restriction enzyme MluI. Plasmid DNA was obtained using the QIAprep Mini Kit (Qiagen) according to the manufacturer's suggested protocol.

For non-PCR+1 products, the cloned trap gene fragment was amplified directly from the colonies. Cells picked directly from a single colony were added to a PCR reaction containing 1.6 μM each primers P5 and P6, 1× reaction buffer, 3 mM MgCl₂, 0.8 mM total dNTPs, and 1.67 units Taq DNA polymerase (Promega). The reactions were heated at 94°C for 1 min, then subjected to 40 cycles of 94°C for 30 seconds, 50°C for 1 minute, and 72°C for 2 minutes. For PCR+1 products, the cloned trap gene fragment was amplified from the M13 PCR product used for colony screening or from plasmid DNA. 1 μl of PCR product diluted 1:100 or 1 μl of plasmid DNA diluted 1:200 was used as template in a 50 μl reaction containing 0.4 μM each primers P5 and P6, 1× reaction buffer, 2 mM MgCl₂, 0.6 mM total dNTPs, and 1.25 units Taq DNA polymerase (Promega). Cycling parameters consisted of 5 cycles of 94°C for 30 seconds, 65°C for 1 minute, and 72°C for 1 minute, followed by 5 cycles with an annealing temperature of 63°C and 20 cycles with an annealing temperature of 60°C. Denaturing and extension times and temperatures did not change.

P5 and P6 PCR products (5 μl) amplified from clones were digested in a total volume of 10 μl with EcoRI (10 units), HpaII (10 units), or Taq^αI (20 units) in corresponding 1× reaction buffer (New England BioLabs). The Taq^αI digestion also contained BSA at a concentration of 100 μg/ml. Digestions were incubated at 37°C (EcoRI, and HpaII) or 65°C (Taq^αI) for 90 minutes, then stored at 4°C prior to RFLP analysis. All digested trap fragments were run on 2% agarose gels (3 NuSieve:1 SeaKem, FMC Bioproducts) in 1× TBE buffer at 100 V for 1 hour and 15 minutes with 100 bp DNA ladder (New England Biolabs) every 13 lanes. Further resolution of some clones was achieved by electrophoresis on 10% non-denaturing acrylamide gels in 1× TBE (PAGEr Gold, Cambrex) at 150 V for 1 hour and 20 minutes. TIFF images of ethidium bromide-stained gels were scanned and imported into BioNumerics v. 3.5 (Applied Maths). The gels were normalized using the ladder and common bands. Automatic band calling and band matching were hand corrected.

In most cases, clones were chosen for sequencing based on variable RFLP banding patterns. In the case of the Roche PCR products, all clones showing RFLP variation were sequenced, and an additional subset of clones was randomly selected so that a total of 10 clones from each individual were sequenced. Sequencing reactions contained 50–100 ng plasmid DNA. Sequencing primers were: M13F20 and M13RV, modified plasmid primers M13F20X and M13RVX 45, trap primers P3, P5, P6 18, TPSEQ4.5 (5' GACTGTTCTCCTTAT 3'), TPSEQ5 (5' GCCTTAAGTAGAGTATT 3'), CEQTP3.1 (5' CCTTATAATCATTTCTCCTTGGTCGT 3'), CEQTP5.1 (5' AGGAAGAGAAGACGCGGTTCAA 3') and CEQTP5.2 (5' GATGTTCTGATCACATGAATGCTCTT 3'). Sequencing was completed on an ABI 373-cycle sequencer (BMBITF/PBRC, University of Hawaii at Manoa, Honolulu, HI) or a CEQ 8000 capillary electrophoresis system (Hilo Genetics Core Facility, University of Hawaii at Hilo, Hilo, HI). Sequences were proofread, assembled and aligned using Sequencher 3.0 (GeneCodes Corp). Alignments were imported into MEGA version 3.1 46 for further analysis.

For RFLP data, diversity was estimated using the Simpson's Index 22 with the formula D = 1/∑p_i² and the Shannon-Weiner Index H = -∑p_i log_e p_i reviewed in 23. A rarefaction curve was generated in Primer v.5 24 with mean values of expected species richness ES_n plotted in Systat (vers 11.00.01). Statistical comparisons of diversity estimates were also completed in Systat (vers 11.00.01). Sequence data was used to produce a phylogenetic network using statistical parsimony to connect the haplotypes based on 95% confidence interval (TCS 1.21) 28. Distance analyses (Tamura 3-parameter with pairwise deletion) of sequence data were conducted using MEGA version 3.1 46.

Jarvi et al. clearly show there is significant genetic diversity in Hawaiian P. relictum and that most infections contain multiple parasite clones. However, greater understanding of the TRAP gene, or of the population genetics of Hawaiian P. relictum will require more data. However the reliable data produced are rather limited, and some of the data analysis is handled inappropriately, in my view. I detail these problems below.

First, the authors own controls indicated that their methods (PCR amplification with a non proof-reading Taq for 80 PCR cycles) produced some PCR-induced polymorphism artefacts. Although PCR-induced polymorphism is not expected to produce patterns such as geographic or phenotypic associations (rather just to add noise), it will increase measures of genetic diversity. The majority of data was produced with the non proof-reading Taq using 80 PCR cycles, however there is no consideration of this in results/discussion sections.

Authors' response: We completed this small trial to evaluate these variables in our system. We don't believe that PCR error is significantly impacting estimates of genetic diversity in this study as we have included only clones found independently in two or more separate birds. Please also see previous response to similar concerns of Reviewer 1.

Also, RFLP haplotype data from two birds were effectively removed from the study because they contained unique haplotypes. This will also have the effect of increasing estimates of diversity.

Finally there was a non-random selection of which clones to sequence. Proportionally more clones were chosen from the rarer RFLP haplotypes. This choice will also have the effect of increasing measures of genetic diversity from the sequence data. The result is that while estimates of genetic diversity are high from sequence (genetic diversity =~ 0.06) it is difficult to know how much of this value is artefact, and hence how to compare these values directly to other studies.

Authors' response: We removed the birds from the estimates of genetic diversity because of the minimal data collected (only 2 clones and 1 clone were obtained from each of these birds). We did not try to produce more clones from these birds, but kept them in the overall study as we had obtained sequence data from one of them. To include them in the estimates would bias them in the other direction. We sequenced 18 clones from the most common RFLP haplotype (77) and 15 clones representing 12 other haplotypes. I don't believe we disproportionately selected the more rare haplotypes. Since this is not a population-level study, these results may not be directly comparable to other studies.

Another limitation of this study is that apart from the RFLP analysis, the sequence data produced in this study is relatively small. Only 14 TRAP sequences were produced from one of the cultured isolates (10 of which from a single bird infection), 16 from another (again, 10 of which from a single bird infection) and 5 from wild isolates (none of which are derived from high fidelity Taq). The result is that few of the more interesting conjectures in the discussion have significant support. For example there was no significant difference between diversity estimates of the two mosquito-derived isolates, or the wild isolates, nor between those birds that survived experimental infections vs. those that did not. Collection of more sequences from more isolates would allow analysis of the frequency distributions of polymorphisms, and be much more informative. To be fair the authors state that population level genotyping is underway. I would caution readers from drawing too many conclusions until this analysis is available.

For example, I would regard any clustering of TRAP isolates with respect to mortality with strong suspicion. While the TRAP gene may contribute to mortality, an association between mortality was not shown (recall that most sequence data are from two isolates from experimentally-infected birds from cultured parasites), and even if it were shown we would need evidence that this association was not due to population subdivision or to linkage with another causal variant.

In summary I think that the data has been extended beyond what it is capable of describing, and may also be limited by PCR-artefacts. Certainly there is no subterfuge about this in the manuscript (the authors for the most part describe the limitations), but the results could well have been limited to a simpler story. Further work on this system, particularly with the improvements that have been shown to be necessary with the data in this work, should provide a much more informative picture.

Authors' response: I think it is important to realize that this manuscript is only intended as an initial attempt to begin to genetically characterize diversity in a previously uncharacterized species of Plasmodium. It not intended as an analysis of survivors and non-survivors of avian malaria. However, as noted above, many of the individuals under study were involved in experimental infection studies and we feel compelled to include this data here. We have reworked the discussion to reflect this concern. We also included a more detailed description of the isolates. Thank you for your suggestions.

Review 3: (Susan Perkins, American Museum of Natural History, Assistant Curator, Division of Invertebrate Zoology, Central Park West at 79^th Street, New York, NY 10024: nominated by Eugene Koonin, Senior Investigator NCBI, NLM, NIH Bethesda, MD 20894)

I found this manuscript, by Jarvi, Farias, and Atkinson to be very interesting and thorough. It also represents an important addition to our knowledge of the nonmammalian and/or non-model system malaria parasites with respect to the genetic diversity that is naturally present in these systems. When I first began reading this manuscript, I immediately worried that some of the high amount of genetic variability might have been caused by heteroduplex formation and bacterial repair during the cloning process, so was very relieved when the authors said that they also incorporated PCR+1 methodology into this study. That said, though, I found it slightly unusual that instead of bumping up the reaction to 80 cycles, they did not also choose to try a scenario of using a very small number of cycles and then pooling these products – a method that is commonly used among microbiologists who are trying to assess diversity in an environment in an attempt to recover rarer organisms (sequences) that might be swamped out by more common ones through the PCR amplification process. My second suggestion is that I think that their sequence data would be far better represented by a haplotype network rather than a traditional, bifurcating tree. As the authors admit, the bootstrap support values on most nodes are very low and this is no doubt a result of a more network-like scenario of evolution that resulted from one or a few haplotypes founding these populations and then accumulating point mutations as the populations expanded and dispersed. A bifurcating, distance-based tree such as the one in Figure 2, will be complicated and distorted by this pattern of evolutionary change. Representing the data as such, along with perhaps shading to depict those sequences found in survivors vs. non-survivors might reveal more interesting and clear-cut patterns than this NJ tree could ever do.

Authors' response: Yes, I wish I had thought of that at the time (i.e., pooling reactions using smaller numbers of cycles). Thank you for your suggestion of a haplotype network. We completed this analysis and it is much more informative than a NJ tree. We have incorporated these data into the manuscript.

At times, this paper also seemed like a mixture of what would be two or perhaps even three different publications. There was the central issue of how much genetic diversity of the TRAP gene – and ergo, number of parasite clones, exists within infected birds and whether or not this diversity is related to host or geography. There was also the attempt to look at the correlation of the presence of certain TRAP sequences to survival in infected birds and this was discussed, in part, with particular changes within the TRAP molecule. Finally, there was the embedded study of enzyme, cycle number, and methodology (PCR+1 or not) and their relationships to the number of clones recovered. Perhaps the authors should consider if these three things might not be better split as such to show that first, the optimum method for detecting clonal diversity was found and used and then that the resulting data on the diversity of genotypes was then analyzed in a context to allow for the most powerful ability to correlate virulence with genotype (ideally by sequencing all genotypes present in these infections).

Authors' response: Yes, we are packing a lot in a single paper, but individually, the pieces do not seem substantial enough to warrant a publication on their own.

You may want to explain the origin of the K1 and KV115 isolates – these are not "cloned isolates" in the sense that human and rodent malaria researchers are used to thinking about them.

I would advocate for including the primer sequences here, even if they are published elsewhere. This will facilitate the ability of others to use them, if desired, without having to track down multiple other publications, which they may or may not have easy access to.

Authors' response: We have included a section describing the origin of the isolates, as well as the sequences of all primers with their references. Thank you for your suggestions.