Within-host competition between Borrelia afzelii ospC strains in wild hosts as revealed by massively parallel amplicon sequencing

Abstract

Infections frequently consist of more than one strain of a given pathogen. Experiments have shown that co-infecting strains often compete, so that the infection intensity of each strain in mixed infections is lower than in single strain infections. Such within-host competition can have important epidemiological and evolutionary consequences. However, the extent of competition has rarely been investigated in wild, naturally infected hosts, where there is noise in the form of varying inoculation doses, asynchronous infections and host heterogeneity, which can potentially alleviate or eliminate competition. Here, we investigated the extent of competition between Borrelia afzelii strains (as determined by ospC genotype) in three host species sampled in the wild. For this purpose, we developed a protocol for 454 amplicon sequencing of ospC, which allows both detection and quantification of each individual strain in an infection. Each host individual was infected with one to six ospC strains. The infection intensity of each strain was lower in mixed infections than in single ones, showing that there was competition. Rank-abundance plots revealed that there was typically one dominant strain, but that the evenness of the relative infection intensity of the different strains in an infection increased with the multiplicity of infection. We conclude that within-host competition can play an important role under natural conditions despite many potential sources of noise, and that quantification by next-generation amplicon sequencing offers new possibilities to dissect within-host interactions in naturally infected hosts.

1. Introduction

Pathogenic microbes and other parasites often consist of a number of different strains, and hosts are commonly infected with more than one strain simultaneously [1]. Experimental infections under controlled laboratory conditions have shown that such ‘mixed’ or ‘multiple’ infections can result in competition between co-infecting strains, so that the infection intensity of at least one of the strains is reduced in mixed as compared with single infections [2]. For example, a series of experiments with the rodent malaria parasite Plasmodium chabaudi in laboratory mice have shown severe competition in infections with two strains, and that the strain with the highest infection intensity in single infections suffers least from competition in double infections [3,4]. Such competition can have a number of important epidemiological and evolutionary consequences [2,5]. In particular, given that infection intensity is correlated with transmission success and Darwinian fitness of the pathogen, within-host competition can select for more virulent pathogens [2,5].

Infection experiments have, however, also shown that the extent of competition is highly dependent on factors such as relative inoculation dose of co-infecting strains, synchronicity of inoculations and host phenotype [6–9]. In rodent malaria, competition can be alleviated or even eliminated when infections are asynchronous [8]. In fact, double infections with P. chabaudi can sometimes result in facilitation instead of competition, so that at least one of the strains in double infections performs better than in single infections [6,7]. In an experiment with two mouse strains, double infections in a relatively resistant mouse strain resulted in competition, but in a relatively susceptible mouse strain double infections lead to facilitation during the chronic phase of the infection [7]. A similar pattern has been found in an insect–virus system [9]. In natural infections, there will often be considerable variation in inoculation dose, timing and host phenotype, but there are as yet few studies that have investigated the nature of within-host interactions under the more noisy conditions in the wild. Hence, the actual importance of within-host competition is not yet clear. The lack of studies of within-host competition in natural infections is at least partly a methodological issue; it is often difficult to resolve the strain composition of mixed infections when there is no a priori knowledge about which strains might be involved, and even more difficult to get unbiased estimates of the infection intensities of each strain.

Here, we used the tick-transmitted bacterium Borrelia afzelii to test for within-host competition in naturally infected wild hosts. Borrelia afzelii belongs to a species complex including 12 named genospecies [10]. At least three of these cause Lyme disease in humans (besides B. afzelii also Borrelia burgdorferi and Borrelia garinii [11]). Borrelia afzelii occurs throughout Eurasia [11] and is a multihost pathogen, infecting a wide range of small mammals such as voles, mice and shrews [12,13]. Borrelia afzelii is highly clonal, that is, it consists of a number of strains with limited recombination [13]. Different strains are often distinguished based on the immunodominant outer surface protein C, encoded by the plasmid-borne single-copy gene ospC [10]. The pairwise difference in DNA sequence between ospC alleles is highly bimodal, with differences either less than 2.6% or more than 5.7% ([11]; electronic supplementary material, figure S1). A similar pattern occurs in B. burgdorferi [14]. We refer to genotypes with ospC alleles that differ by more than 5% as different ‘strains’. Previously sampled populations (e.g. in a small forest) of B. afzelii hosts harboured five to eight different ospC strains each [13,15].

In an infection experiment with B. burgdorferi in white-footed mice (Peromyscus leucopus), transmission success of each ospC strain from vertebrate to tick was lower in infections with two strains as compared with single strain infections [16], showing that there is potential for competition in Borrelia infections. However, a study of within-host interactions in natural B. afzelii infections in one of its main hosts, the bank vole, found that strains were highly aggregated (i.e. there were more multiple infections than expected from the prevalence of each strain), suggesting facilitation for infectivity and/or persistence rather than competition [17]. Moreover, overall infection intensities increased with number of ospC strains in an infection in a largely additive way, indicating that competition affecting infection intensities did not occur (although the statistical power of this test might have been low; [17]). Taken together, the results of the bank vole study suggest that, if anything, mixed infections with B. afzelii in wild naturally infected hosts result in facilitation rather than competition.

To be able to analyse interactions between co-infecting B. afzelii strains in more detail—in particular whether there is competition affecting infection intensities or not—we developed a protocol for 454 amplicon sequencing of ospC which allows both detection and quantification of each individual strain in an infection. The accuracy of detection and quantification was validated by sequencing two different loci (besides ospC also the 16S–23S rRNA intergenic spacer (rrs-rrlA IGS), which is in strong linkage disequilibrium (LD) with ospC). We then used this method to analyse within-host interactions in B. afzelii infections in three different host species sampled in the wild: bank voles Myodes glareolus, yellow-necked mice Apodemus flavicollis and common shrews Sorex araneus. Specifically, we tested if infection intensities of individual strains are affected by co-infection.

2. Material and methods

(a). Small mammal sampling and DNA extraction

The samples used in this study were collected from naturally B. afzelii infected bank voles, M. glareolus (N = 50), yellow-necked mice, A. flavicollis (N = 50), and common shrews, S. araneus (N = 24), trapped at Kalvs mosse, Revingehed, Southern Sweden (55°42.470′ N, 13°29.216′ E) during 2006–2010, as part of a long-term project on the evolutionary ecology of B. afzelii. These species are the most common small mammals in the study area, and together make up approximately 90% of the small mammal community [13]. Animals were caught with live traps (Ugglan Special, GrahnAB, Sweden), baited with oat and apple. From each animal, a skin biopsy (2 mm diameter) was collected from the ear. Biopsies were stored in 70% ethanol until DNA extraction. DNA was extracted following the protocol of Laird et al. [18] and kept at −20°C until used.

(b). Optimization of amplification

To quantify relative amounts of different strains in B. afzelii infections, a protocol for 454 amplicon sequencing of the outer surface protein C gene (ospC), and the nuclear rrs-rrlA IGS region of B. afzelii was developed. The optimal number of PCR cycles for downstream quantification of separate sequences (strains) was established by quantitative real-time PCR (qPCR) using a subset of the DNA samples [bank voles (N = 8), yellow-necked mice (N = 6) and common shrews (N = 1)]. Each qPCR reaction (25 µl) contained 37.5 ng of DNA-template, 0.2 µM of each primer, 0.05 µM of ROX reference dye and 1× Platinum SYBR Green qPCR SuperMix-UDG (Invitrogen, Carlsbad, CA, USA). Partial ospC fragments (618 bp) were amplified with the inner primers (ospC Fn and ospC Rn) in the nested protocol by Bunikis et al. [19], applying the following temperature programme: 50° for 2 min, 95° for 2 min followed by 40 cycles of 95° for 30 s, 52° for 30 s and 72° for 45 s. Primers for amplification of a 340bp-long rrs-rrlA IGS fragment were designed with the software Geneious (v. 5.0, Biomatters, Auckland, New Zealand) using previously described rrs-rrlA IGS sequences [13]. The forward primer IGS_F1 (P202) 5′-TCGTACTGGAAGTGTGGCT-3′ was combined with the reverse primer IGS_R1 (P203) 5′-TTTGTCAATTTCGATGTTAGRGT-3′and the cycling parameters were: 50° for 2 min, 95° for 2 min followed by 40 cycles of 95° for 30 s, 61° for 30 s and 72° for 45 s. These trials and an additional test using 454 fusion primers in standard PCR showed that 38 amplification cycles were optimal for accurate downstream detection and quantification of strains for both ospC and rrs-rrlA IGS targets. The Qiagen Multiplex PCR mastermix (Qiagen, Hilden, Germany) was used (a non-high-fidelity DNA polymerase) to allow good amplification from samples with low infection loads.

(c) Amplification and 454 sequencing

Partial ospC gene fragments and the rrs-rrlA IGS region were amplified separately from all the 124 B. afzelii infected small mammal samples. In addition, 20 of these samples [bank voles (N = 8), yellow-necked mice (N = 6) and common shrews (N = 6)] were technically duplicated. Amplifications were performed with individually labelled HPLC purified 454 fusion primers (Eurofins MWG, Ebersberg, Germany) with gene-specific sequences specified above. For ospC, 2 × 72 individually 10 bp labelled (Roche's extended MID set, MID-1—MID-74, excluding MID-9 and MID-12, TCB no. 005–2009) Lib-L libraries (one library per sample) were prepared for unidirectional forward amplicon sequencing. Each PCR reaction (15 µl) contained 25 ng of DNA-template, 0.2 µM of one MID-tagged forward 454 fusionprimer, 0.2 µM of the reverse 454 fusionprimer and 1× Qiagen Multiplex PCR mastermix. The cycling conditions were: 95° for 15 min followed by 38 cycles of 95° for 30 s, 52° for 90 s and 72° for 60 s. A final extension step at 72° for 10 min was applied. For rrs-rrlA IGS, Lib-A libraries with 6 bp individual tags in both the 5′ and 3′ ends (12 tags making up in total 144 combinations) were prepared for bidirectional 454 amplicon sequencing. Each PCR reaction (15 µl) contained 25 ng DNA-template, 0.2 µM of each forward and reverse 454 fusionprimer and 1× Qiagen Multiplex PCR mastermix. The cycling conditions were: 95° for 15 min followed by 38 cycles of 95° for 30 s, 61° for 90 s and 72° for 60 s. A final extension step at 72° for 10 min was applied.

PCR products (amplicon libraries) were verified on a 2% agarose gel. The PCR products (5 µl from strong and 10 µl from weaker products) were purified in pools of eight samples (to increase the proportion of recovered PCR product) on MinElute PCR Purification Kit spin columns (Qiagen) according to the manufacturer's protocol. The DNA concentration of each purified pool was quantified on a NanoDrop 2000/2000c (Thermo Fisher Scientific, Wilmington, DE, USA) and equimolar amounts of DNA were combined to make up two final pools of 72 ospC samples each (the numbers of samples per species were evenly distributed between pools and technical replicates were included in separate pools) and one pool of 144 rrs-rrlA IGS samples. Each pool was run in an individual region of a 4-region 454 sequencing run.

Sequencing was conducted at Lund University Sequencing Facility (Faculty of Science). Pooled amplicons were reduced to short fragments by using Agencourt AMPure XP (Beckman Coulter, Brea, CA, USA) and then inspected using a DNA 1000 kit on a 2100 Bioanalyzer (Agilent, Santa Clara, CA, USA). Amplicons were quantified using the Quant-iT dsDNA assay kit (Invitrogen) and a Quantiflour fluorometer (Promega, Fitchburg, WI, USA), and pools were diluted to 1 × 10⁷ copies per microlitre. Titration and library production (aiming at 10–15% enrichment) were performed using emulsion PCR and the Lib-L kit (for ospC sequencing) or the Lib-A kit (for rrs-rrlA IGS sequencing) (Roche, Penzberg, Germany). DNA positive beads were enriched, counted on an Innovatis CASY particle counter (Roche), processed using an XLR70 sequencing kit (Roche) and loaded onto a picotiter plate for pyrosequencing on a 454 Life Sciences Genome Sequencer FLX+ machine (Roche).

Sequences were obtained for ospC and rrs-rrlA IGS from 123 out of 124 amplicons each. The ospC and rrs-rrlA IGS assays failed on different samples so there were 122 samples with data on both ospC and rrs-rrlA IGS.

The overall B. afzelii infection intensity in each of the 124 samples was measured by qPCR targeting flaB as described in [20]. All samples were run in duplicate on separate plates.

(d). 454 Sequence annotation and quantification of Borrelia afzelii strains

We obtained 171 016 and 40 591 raw sequences from the two regions with ospC, and 231 665 raw sequences from the region with rrs-rrlA IGS. Sequences were quality-trimmed with default settings in the 454 Sequencing System Software (Roche). Mean read lengths after quality trimming were 217 and 264 bp for the two regions with ospC, and 160 bp for the region with rrs-rrlA IGS. The quality of the sequences was inspected with FastQC [21] and the first 245 bp of ospC that were critical for annotation (see below) had a satisfactory average quality score per base of 30 (range 20–38). For rrs-rrlA IGS, the average quality score per base for the full length sequences was above 30 (range 25–40). Demultiplexing of individual samples for the ospC dataset was done with the software Geneious (v. 5.0, Biomatters, Auckland, New Zealand) that handles Roche's MID set. For the rrs-rrlA IGS dataset, demultiplexing was done with the software jMHC [22] that handles extraction of forward and reverse individually tagged samples. To establish the composition of B. afzelii strains in an amplicon, BLASTN (v. 2.2.27) searches (using default settings) were performed with the obtained reads against two local databases containing all unique European B. afzelii ospC and rrs-rrlA IGS sequences available in GenBank. The BLAST approach allowed us to (i) assign sequences to strains even in cases where the naturally occurring high homopolymer content in ospC and rrs-rrlA IGS sequences lead to homopolymer sequencing errors, and (ii) filter out any chimeric sequences and sequences with PCR errors from the final dataset.

The ospC database contained 42 unique sequences (electronic supplementary material, figure S1a). As in previous studies of B. burgdorferi [14], the frequency distribution of pairwise distances between reference sequences was clearly bimodal (electronic supplementary material, figure S1b). Based on the frequency distribution, the reference sequences were divided into major groups, with pairwise distances less than 2.6% within groups and more than 5.7% between groups. The resulting 19 major groups, referred to as ‘ospC strains’, each with 1–10 variant alleles (indicated in electronic supplementary material, figure S1a) were named according to Hellgren et al. [13] and Bunikis et al. [19]. Stringent criteria for reliable annotation of ospC were set in the BLAST process. Sequences were kept when BLAST hit alignment length exceeded 245 bp and BLAST hit identity was at least 98%. BLAST hit alignment length was calculated as the difference between the last hit position in the subject and the first hit position in the subject + 1. BLAST hit identity was calculated as nucleotide identities between query and subject divided by BLAST hit alignment length. These criteria allowed us to distinguish all ospC alleles that had previously been found in the study area using Sanger sequencing [13,17].

For the less variable rrs-rrlA IGS region the criteria in the BLAST process for annotating and keeping sequences were necessarily stricter than for ospC (read full length = 360 bp and BLAST hit identity = 100%). The rrs-rrlA IGS database consisted of 14 unique sequences (all from [13,19]) with pairwise differences ranging from 0.35 to 3.1%.

Relative measures of infection intensity were calculated per strain and host individual from the number of annotated sequences per strain and the total number of annotated sequences from each individual. Absolute measures of infection intensity per strain and host individual were obtained by multiplying the relative measure with host individual overall B. afzelii infection intensity as measured by qPCR targeting flaB [20]. Infection intensities are expressed as number of ospC copies per nanogram of host DNA.

(e). Statistical analyses

Infection intensities were analysed by means of general linear mixed models, with infection intensity of each ospC strain in an infection as dependent variable, using proc mixed in SAS v. 9.3 (SAS Inc., NC, USA), and Satterthwaite approximation of denominator d.f.; p-values for fixed effects (number of ospC strains in an infection; host species; ospC strain, year) were determined by F-test (method = type3). Diagnostic plots were checked to ensure that residuals were normally distributed.

3. Results

(a). Evaluation of 454 sequencing of Borrelia afzelii strains

We obtained between 2 and 2204 reads (mean = 427) per ospC amplicon and between 2 and 210 reads (mean = 60.3) per rrs-rrlA IGS amplicon that fulfilled the annotation criteria specified in §2d. The difference in coverage between loci can be attributed to the sequencing design (two sequencing regions for ospC versus one region for rrs-rrlA IGS), but also to the stricter BLAST criteria needed for annotating/keeping the rrs-rrlA IGS strains (see §2d).

The accuracy of quantification by amplicon sequencing was evaluated. The results from the two loci—ospC and rrs-rrlA IGS—were compared and found to be very similar. In this analysis, we took advantage of the fact that B. afzelii (like B. burgdorferi; [23,24]) has a highly clonal genetic structure so that ospC and rrs-rrlA IGS are in strong LD, especially within populations [13,19]. A previous study showed that among the ospC and rrs-rrlA IGS alleles detected in this study, there are five ospC alleles that can be expected to be in perfect LD (LD = 1) with a single rrs-rrlA IGS allele at our main study site at Kalvs mosse, Revingehed (ospC 1-rrs-rrlA IGS 1, 2–2, 4–4, 7–11 and 9–9; cf. fig. 1d,e in [13]). An ANCOVA with infection intensity of a given ospC as dependent variable (calculated as the proportion of reads of that ospC allele times the overall B. afzelii infection intensity as measured by qPCR), infection intensity of the corresponding rrs-rrlA IGS as covariate, ospC—rrs-rrlA IGS linkage group as factor, and the interaction between linkage group and the covariate, showed the infection intensity of a strain as measured by ospC and rrs-rrlA IGS to be strongly correlated (F_1,224 = 391, p < 0.0001, Pearson r = 0.86), and that there was no difference in intercepts or slopes between the different linkage groups (ospC—rrs-rrlA IGS linkage group: F_4,224 = 0.89, p = 0.47; ospC—rrs-rrlA IGS linkage group × rrs-rrlA IGS infection intensity: F_4,224 = 1.43, p = 0.22; figure 1). Hence, quantification based on ospC or rrs-rrlA IGS gave very similar results. We base our analyses on ospC as coverage was higher for this locus.

Figure 1.

Correlation between strain-specific infection intensities measured from two linked B. afzelii loci (ospC and rrs-rrlA IGS) in naturally infected small mammals (bank voles (N = 50), yellow-necked mice (N = 50) and common shrews (N = 24) sampled at Kalvs mosse, Sweden). Five linkage groups of the two loci [13] are indicated with different colours. The infection intensity of each strain was estimated by 454 amplicon sequencing of the two loci in combination with qPCR of B. afzelii flaB.

(b). 454 Sequence data of ospC

Data from samples with at least 100 reads per ospC amplicon were included to allow reasonable resolution of strain composition (although analyses based on the whole ospC dataset, including samples with lower coverage, yielded conclusions identical to the analyses presented in §3d–f). This restriction left us with 45 vole samples, 27 mouse samples and 22 shrew samples. The average number of reads per ospC amplicon in this dataset was 575 and the technical reproducibility of detection of ospC strains (number of strains detected in both replicates/total number of strains in a pair of replicates) was 0.58 (N = 20). The relatively low reproducibility was mainly owing to low detection reproducibility of strains constituting less than 1% of the total number of reads in one of the replicates. These low abundance strains could be false positives. However, given that total infection intensities span more than three orders of magnitude (electronic supplementary material, figure S2), it does not seem unlikely that there could be two orders of magnitude difference in abundance between strains within an infection, and that the low abundance strains are true positives. In any case, analyses of within-host interactions based on datasets with and without strains constituting less than 1% of the reads yielded identical conclusions. Below we present analyses based on a dataset where strains constituting less than 1% of the reads in an amplicon were excluded. In this dataset, the technical reproducibility of ospC strain detection was 0.9 (N = 20). The technical repeatability of the infection intensity of a given ospC strain in an infection (calculated as the proportion of reads of that ospC allele times the overall B. afzelii infection intensity as measured by qPCR) was 0.99 (N = 20).

(c). Distribution of strains

We detected seven different ospC alleles (ospC 1, 2, 3, 4, 7, 9 and 10). These are the same alleles as previously found in the study area (by Sanger sequencing [13,17]). Each strain occurred in 30.9–77.7% of the samples and was found in all three host species. Full details of the distribution of ospC alleles across host species are shown in electronic supplementary material, table S1. We detected six different rrs-rrlA IGS alleles (rrs-rrlA IGS 1, 2, 3, 4, 9, 11). Again, these are the same alleles as previously found in the study area (by Sanger sequencing [13,17]). The distribution of rrs-rrlA IGS alleles across host species is shown in electronic supplementary material, table S2.

The average number of ospC strains per individual was 2.17. Shrews and mice had one to four strains, while voles had up to six. The distribution of number of ospC strains per individual host for each host species is shown in figure 2. The number of ospC strains per individual did not differ between host species (F_2,91 = 1.48, p = 0.23).

Figure 2.

Frequency distribution of infections with different number of B. afzelii ospC strains in three species of naturally infected small mammals (bank voles (N = 45), yellow-necked mice (N = 27) and common shrews (N = 22)) sampled at Kalvs mosse, Sweden. The number of ospC strains per individual was determined from 454 amplicon sequencing data.

(d). Effects of host species and multiplicity of infection on total infection intensity

The total infection intensity differed between host species (shrew: 2.98 ± 0.17 (mean ± s.e.); vole: 2.47 ± 0.13; mouse: 1.81 ± 0.17, F_2,86 = 23.8, p < 0.0001), indicating that host species differed in resistance. There was no effect of the number of ospC strains on total infection intensity (F_5,86 = 1.27, p = 0.29; electronic supplementary material, figure S2).

(e). Effects of multiplicity of infection on infection intensity of each strain

Median infection intensities of each strain in infections with different numbers of strains and in different host species are shown in figure 3 (infection intensities of N = 204 ospC strains from 94 individuals, between one and six strains per individual). To test for effects of within-host interactions on the infection intensity of individual ospC strains, we performed a general linear mixed model with infection intensity against number of ospC strains in an infection, ospC strain, host species, year and their interactions as fixed factors, and individual host as a random effect. There was a significant effect of the number of ospC strains on infection intensity of each strain (F_5,61.5 = 6.71, p < 0.0001; figure 3). Post hoc tests showed that in infections with 2, 3 and 4 ospC strains, each strain had lower intensity than in infections with one strain (Dunnett: p ≤ 0.0066), indicating that strains in multiple infections suffered from competition. There was no main effect of ospC strain, but the interaction between host species and ospC strain approached significance (F_12,153 = 1.63, p = 0.089). Full details of the statistical analysis are shown in electronic supplementary material, table S3.

Figure 3.

Box plot of infection intensity per B. afzelii ospC strain in infections with different number of strains in three small mammal species (bank voles (N = 45), yellow-necked mice (N = 27) and common shrews (N = 22)) sampled at Kalvs mosse, Sweden. The box plots indicate the median, first and third quartiles, and range of the data. Relative infection intensity of each ospC strain in an infection was determined from 454 amplicon sequencing of ospC. Total infection intensity was estimated with qPCR of B. afzelii flaB.

(f). Effects of multiplicity of infection on proportion of each strain

To investigate how the different strains in an infection shared the overall infection intensity, we produced rank-abundance plots for infections with different number of ospC strains (figure 4). This showed that there was typically one dominant strain. For example, in infections with two strains, the most abundant strain had on average 86% of the overall infection intensity. However, the evenness of relative abundance of the different strains in an infection increased with increasing multiplicity of infection, and in the only infection with six strains the most abundant strain had only 25% of the total infection intensity.

Figure 4.

Rank-abundance plot of natural B. afzelii infections with different number of strains in small mammals (N = 94). The box plots indicate the (log₁₀) median, quartiles and range of the proportions of the most abundant strain in each infection (abundance rank = 1), the second most abundant (abundance rank = 2), and so on, in infections with two to six ospC strains. The slope of the relationship between relative abundance and rank indicates the evenness of relative abundance in infections with a certain number of strains; a more shallow slope means higher evenness. Relative abundance of the strains was determined by 454 amplicon sequencing of B. afzelii ospC.

4. Discussion

Here, we developed a protocol for detection and quantification of individual strains of B. afzelii based on 454 amplicon sequencing, and applied this to samples collected from naturally infected wild hosts. This technique has recently been used for estimating the diversity of infections by a given pathogen [25,26], but as far as we are aware not previously for quantification of absolute infection intensities of different genotypes. There is an ongoing debate about the utility of 454 amplicon sequencing (and other next-generation sequencing techniques) for quantification of microbes in other contexts, in particular in metagenomic analyses of environmental bacterial communities [27–30]. One potential problem stems from PCR amplification; even if primer-binding sites are identical, variable amplicon length or GC content can affect the efficiency of amplification, and slight differences in efficiency can generate large differences in the amount of end product. To test hypotheses of within-host interactions and virulence evolution, it is important to get unbiased estimates of infection intensity of each strain. Here, we took advantage of the high clonality of B. afzelii to evaluate the accuracy of 454 amplicon sequencing-based quantification. Our analyses showed that independent quantifications based on two loci in perfect LD were highly correlated. Most importantly, the different linkage groups differed in neither intercept nor slopes (although at low infection intensities, there was considerable uncertainty regarding the reliability of strain detection). This shows that there was no difference in the ability to detect different strains, and no difference in the quantification of different strains. We thus conclude that next-generation amplicon sequencing can give unbiased estimates of infection intensities of each strain, something that opens up new possibilities to test hypotheses of host–parasite interaction in the wild.

Our analysis of B. afzelii infections revealed that the infection intensity of individual ospC strains was lower in mixed infections than in single infections, showing that strains in mixed infections suffered from competition. Within-host competition has previously been demonstrated in a variety of host–pathogen systems in controlled laboratory infection experiments, but this study shows that competition between pathogen strains can be intense too in natural infections in the wild, despite noise in the form of variable inoculation doses, asynchronous infections and host variation in resistance. The transmission success of B. afzelii from vertebrate to tick increases steeply with the total infection intensity (i.e. summed over all strains in an infection) [20]. Assuming a similar relationship for each individual strain in an infection, the competition observed in this study may have severe effects on the Darwinian fitness of B. afzelii, although analyses of transmission success of individual strains are required to ascertain this.

In each infection, there was typically one dominant strain with more than 50% of the total infection intensity (figure 4). This effect was most pronounced in infections with few strains. For example, in infections with two strains, the dominant strain obtained on average 86% of the total infection intensity. However, the evenness of the relative infection intensity of the different strains in an infection increased with the multiplicity of infection. In other ecological communities, both positive and negative relationships between species richness and evenness have been reported, and the factors determining the sign of this relationship are controversial [31]. One potential explanation for the higher evenness in B. afzelii infections with more strains is that the competitiveness of strains follows a non-transitive hierarchy so that strain A dominates strain B, B dominates C, but C dominates A. Such ‘competitive intransitivity’ [32] should lead to a clear winner in infections with few strains, but not in infections with many strains. Another potential explanation for the relationship between evenness and multiplicity of infection is that the relative abundance of strains is more dynamic over time in infections with few strains. The infection intensities measured here represent a snapshot in time; infection intensities of individual strains could vary through time, so that a strain that is dominant at one point in time is minor at another, as occurs in human malaria parasites [33]. One could imagine that the relative abundance of different strains is more stable over time in infections with many strains, for example if it is difficult for the immune system to mount specific immune responses to each strain in very diverse infections.

What causes the competition between B. afzelii strains? Within-host competition can in principle be mediated by three different mechanisms [2]. First, there can be interference competition, if strains produce allelopathic substances. Second, co-infecting strains may compete over some limiting resource. Third, there can be immune-mediated competition, if the immune response is strain-transcending and multiple infections induce a stronger immune response than single infections (for example, because multiple infections have higher infection intensities during the early phase of the infection, before infection intensities are regulated by the immune response) [34,35]. Which of these mechanisms is most important in the case of B. afzelii? Interference competition occurs in some bacteria, including Escherichia coli and Haemophilus influenzae [1], but has, as far as we are aware, not been described in Borrelia. Antibody responses to OspC are largely strain-specific, but responses to other outer surface proteins (e.g. OspA) are cross-reactive, so there is scope for antibody-mediated competition between strains [36]. Borrelia spirochaetes respond adaptively to nutrient deprivation, suggesting that availability of nutrients may sometimes be a limiting factor for their replication [37]. It is difficult to elucidate the relative importance of different types of competition from the current dataset. Instead, infection experiments where different aspects of the host phenotype are modified are required [34,35].

This study also showed that the infection intensity differed between host species, with shrews having more than ten times higher bacterial loads than mice, while voles were intermediate. A similar pattern was found in a previous study of bank voles and yellow-necked mice [20]. As there was no difference in average number of strains or composition of strains between host species, the difference in infection intensity indicates that host species differ in resistance. This interpretation fits with the observation that yellow-necked mice mount a more vigorous antibody response to Borrelia than bank voles [38]. The difference in resistance between host species could affect the expression of fitness costs of Borrelia infection. It has been difficult to demonstrate fitness effects in naturally infected wild hosts, but existing studies have focused on mice [39]. Perhaps fitness effects are more pronounced in less resistant host species, such as voles and shrews.

A previous study of B. afzelii in bank voles investigated the nature of within-host interactions by analysing the distribution of strains [17]. Competition would result in fewer mixed infections than expected by chance (based on the prevalence of each strain), whereas facilitation would lead to an aggregated distribution with more mixed infections than under a random distribution [40]. In the bank vole study, strains were highly aggregated, which is consistent with facilitation of infectivity (i.e. establishment of a strain is enhanced by the presence of other strains). Aggregation could also have other causes, for example variation in prevalence among age classes [40], but there was no evidence that such effects could explain the aggregation of ospC strains in bank voles [17]. Based on the results of Andersson et al. [17] and this study, we propose that there are opposing effects of multiplicity of infection on transmission of B. afzelii at different transmission stages: the aggregated distribution of strains [17] suggests the transmission success of each ospC strain from tick to vertebrate is enhanced if the inoculum contains more than one strain, perhaps because mixed infections are more difficult for the immune system to control during the early stage of infection. By contrast, the lower infection intensity of each strain in mixed infections (as found in the present study) indicates that once infections are established, strains in mixed infections suffer from competition which should reduce transmission from vertebrate to tick [16].

In conclusion, this study shows that within-host competition can have important effects on the transmission potential of a pathogen strain under natural conditions. It also highlights the utility of next-generation sequencing to dissect within-host dynamics.

Ethics

The study was approved by the ethical committee for animal experiments in Malmö/Lund, Sweden (permit nos. M101-06 and M141-10).

Data accessibility

Data has been deposited at Dryad (http://dx.doi.org/10.5061/dryad.vf83f) and SRA (accession no. SRP055811).

Competing interests

We declare we have no competing interests.

Funding

This work was funded by the Swedish Research Council and the Crafoord Foundation.