doi
stringlengths 28
28
| title
stringlengths 19
311
| abstract
stringlengths 217
5.08k
| plain language summary
stringlengths 115
4.83k
| article
stringlengths 3.87k
161k
|
|---|---|---|---|---|
10.1371/journal.pntd.0001190 | Polyandry Is a Common Event in Wild Populations of the Tsetse Fly Glossina fuscipes fuscipes and May Impact Population Reduction Measures | Glossina fuscipes fuscipes is the main vector of human and animal trypanosomiasis in Africa, particularly in Uganda. Attempts to control/eradicate this species using biological methods require knowledge of its reproductive biology. An important aspect is the number of times a female mates in the wild as this influences the effective population size and may constitute a critical factor in determining the success of control methods. To date, polyandry in G.f. fuscipes has not been investigated in the laboratory or in the wild. Interest in assessing the presence of remating in Ugandan populations is driven by the fact that eradication of this species is at the planning stage in this country.
Two well established populations, Kabukanga in the West and Buvuma Island in Lake Victoria, were sampled to assess the presence and frequency of female remating. Six informative microsatellite loci were used to estimate the number of matings per female by genotyping sperm preserved in the female spermathecae. The direct count of the minimum number of males that transferred sperm to the spermathecae was compared to Maximum Likelihood and Bayesian probability estimates. The three estimates provided evidence that remating is common in the populations but the frequency is substantially different: 57% in Kabukanga and 33% in Buvuma.
The presence of remating, with females maintaining sperm from different mates, may constitute a critical factor in cases of re-infestation of cleared areas and/or of residual populations. Remating may enhance the reproductive potential of re-invading propagules in terms of their effective population size. We suggest that population age structure may influence remating frequency. Considering the seasonal demographic changes that this fly undergoes during the dry and wet seasons, control programmes based on SIT should release large numbers of sterile males, even in residual surviving target populations, in the dry season.
| Glossina fuscipes fuscipes is the most common tsetse species in Uganda where it is responsible for transmitting Trypanosoma brucei rhodensiense and Trypanosoma brucei gambiense parasites causing sleeping sickness in humans in addition to related trypanosomes that cause Nagana in cattle. An understanding of the reproductive biology of this vector is essential for the application of sustainable control/eradication methods such as Sterile Insect Technique (SIT). We have analysed the number of times a female mates in the wild as this aspect of the reproductive behaviour may affect the stability and size of populations. We provide evidence that remating is a common event in the wild and females store sperm from multiple males, which may potentially be used for insemination. In vector eradication programmes, re-infestation of cleared areas and/or in cases of residual populations, the occurrence of remating may unfortunately enhance the reproductive potential of the re-invading propagules. We suggest that population age structure may influence remating frequency. Considering the seasonal demographic changes that this fly undergoes during the dry and wet seasons, control programmes based on SIT should release large numbers of sterile males, even in residual surviving target populations, in the dry season.
| Tsetse flies (Diptera: Glossinidae) are the sole vectors of pathogenic trypanosomes in tropical Africa, where they cause Human African Trypanosomiasis (HAT), or sleeping sickness, one of the most seriously neglected tropical diseases. HAT is a zoonosis caused by the flagellate protozoa Trypanosoma brucei rhodesiense in East and Southern Africa and by T. b. gambiense in West and Central Africa [1]. The only country with known infection foci of both parasites is Uganda [2]. The World Health Organization (WHO) has estimated that there are around 10,000 cases of HAT as the recent epidemics are beginning to decline, but 60 million people continue to live at risk in 37 countries covering about 40% of Africa [3]. In addition to HAT, trypanosomes transmitted by tsetse cause a fatal disease in livestock, called Nagana, which represents a major impediment to agricultural development in Africa. No vaccines exist to prevent the disease and drugs currently available to treat HAT are expensive, can cause severe side-effects, and are difficult to administer in remote villages. As a consequence, an effective alternative for controlling the disease is to target the tsetse vector [1], [4]. In 2001, the African Union launched the Pan African Tsetse and Eradication Campaign (PATTEC) to increase efforts to manage this plague, which is considered one of the root causes of hunger and poverty in most sub-Saharian African countries [5].
Glossina fuscipes fuscipes, a member of the palpalis complex, is one of the most important vectors of human and animal trypanosomiasis in Africa. It is a riverine species confined to forested patches along rivers and lacustrine environments [6]. Its range extends across the central part of the African continent from Sudan, Democratic Republic of Congo to Uganda. As a trypanosome vector, G. f. fuscipes is exposed to a large reservoir of parasites, as it feeds on both domestic and wild animals in addition to humans.
Attempts to control/eradicate tsetse require in-depth information about their population characteristics such as dispersal rates, distribution, densities and reproductive biology. The riverine nature of G. f. fuscipes has resulted in a patchy distribution of its populations and as a consequence of drift, populations arising from historical colonization events show a considerable population structure [7]. Nevertheless, Beadell et al. [8] inferred a high dispersal capacity for G. f. fuscipes, demonstrating ongoing gene flow among apparently isolated populations, with an equilibrium between drift and gene flow in western and south-eastern Uganda. Since populations undergo seasonal contractions during the year due to changes in water availability, Krafsur [9] suggests that high levels of genetic drift during the dry season could be masking effects due to gene flow.
The capacity of G. f. fuscipes to disperse and colonize may also depend on the number of times a female mates in the wild and whether the matings are with the same or different males. This specific mating behaviour influences the effective population size, and may constitute a critical factor in determining the success of control methods [10], [11]. Some aspects of mating behaviour, such as the effect of age on mating competitiveness, have been studied in laboratory colonies [12], but to date, the polyandrous behaviour of G. f. fuscipes has not investigated in the laboratory or in the wild.
Data on the proportion of tsetse females that mate more than once can be obtained in two ways: through the number of fathers (male genotype) represented in her offspring [13], [14] or through genotyping stored sperm in the spermatheca of the female. In the first case, the genotyping of offspring can reveal the minimum number of males that sire a brood, but not necessarily the number of males with which a female had mated, as females may bias paternity towards one or a few of their mates, resulting in an underestimation of the actual level of polyandry [15]. In the second case a more accurate estimate of the number of mates can be obtained, through the genotyping of the female's stored sperm supply [16], [17].
Using microsatellite markers to genotype sperm, we ascertained the minimum number of males that were able to transfer sperm to a female's spermatheca in two Uganda populations. The interest in Uganda is based on the fact that eradication efforts by PATTEC are at the planning stages in this country. The results obtained in two sites, which are eco-geographically differentiated, are of particular interest, as in both populations a large proportion of females were found to have mated more than once. The remating frequencies, validated with probability values obtained with two inference statistical models, are relevant for interpreting the reproductive biology of the species but may also have an immediate impact on the strategy to be employed for eradication success.
Natural populations of G. f. fuscipes were sampled from two localities in Uganda: Kabunkanga (KB, Western territory, 0°58′37.88″N, 30°32′47.40″E) and Buvuma Island in Lake Victoria (BV, Southern zone, 0°15′23.15″N, 33°12′22.86″E) (Figure 1). Both sites are favourable for this riverine species and harbour well established populations. Males and females were collected using biconical traps located 500 m apart at both sites. The traps were checked daily and the average daily fly catch per trap was recorded. The collections from Kabunkanga were made in November 2008, at the end of the dry season from four traps with an average of 15 flies/day/trap. The collections from Buvuma Island were made at the beginning of April 2008, during the wet season, from five traps with an average of 58 flies/day/trap. Individuals of each sex were removed from the traps and placed in tubes containing 95% ethanol. The Kabunkanga (KB) sample was composed of 20 males and 29 females, while for Buvuma Island (BV) 20 males and 40 females were analyzed. The number of males and females in each sample mirrored the sex-ratio observed in the collections. The age and the reproductive history of the sampled flies were unknown, but all the 29 Kabukanga females and the 40 females collected in Buvuma had mated as their spermathecae contained sperm. More precise information about the age structure of the flies collected in each sample could have been obtained from ovarian inspection and/or wing fray analysis [18]; however the extent of damage observed in the wings due to trapping and EtOH preservation, did not permit wing fray analysis. Ovarian age was not assessed. For each site, all of the collected flies were considered to compute allele frequencies and variability estimates. For the remating analysis, the 29 females from KB and 30 females, randomly chosen from the BV collection, were examined.
For sperm isolation, the ethanol preserved female body was rehydrated in physiological solution (0.9% NaCl) for 24–48 h before dissection. The spermathecae were easily isolated from the abdomen, stored in 70% ethanol to permit the sperm to coagulate in a “sperm bundle” [19] and then dissected in a drop of 1× PBS (Phosphate buffered saline). The sperm bundle was isolated and DNA extraction was performed using QIAamp DNA Micro Kit (Qiagen, Valencia, CA). DNA extraction from the legs was performed using the protocol described in Baruffi et al. [20]. The DNA extracted from legs and sperm was used as PCR template for the amplification of microsatellite markers (SSRs).
Nineteen SSR loci were previously isolated from a G. f. fuscipes SSR enriched library [7]. For eight of these loci (A06, A09, A112, B05, C7, C107, D06, and D109) the described primer sequences were adopted [7], [8], [21]. For the remaining 11 loci (A03, B03, B06, B11, B109, C104, D3, D05, D12, D101 and D103) primer sequences and amplification conditions were determined using DNA extracted from Kabunkanga flies as PCR template. Amplification reactions were performed in 15 µl volumes containing 1 µl of genomic DNA, 1× reaction buffer, 1.5 mM MgCl2, 25 µM dNTP, 1 U Taq polymerase (Invitrogen, Carlsbad, CA) and 10 pmol of each primer. Reactions were performed with an Eppendorf MasterCycler Gradient thermocycler. After an initial denaturing step of 10 min at 96°C, the PCR consisted of 40 cycles of 1 min at 96°C, 1 min at optimal annealing temperature, and 1 min at 72°C, followed by a final extension step of 15 min at 72°C. Microsatellite loci were analyzed using an ABI PRISM 310 Genetic Analyzer and the GeneScan program (Applied Biosystems). An individual was declared null (non-amplifying allele) after at least two amplification failures.
Mitotic chromosome spreads were obtained from freshly deposited larvae obtained from the Slovakia laboratory strain. Briefly, brain tissues were incubated in 1% sodium citrate for 10 min at room temperature and transferred to methanol-acetic acid 3∶1 solution for 4 min. The material was disrupted in 100 µl 60% acetic acid and dropped onto clean slides and dried. Pre-hybridization was performed according to Willhoeft [22]. In situ hybridization was performed using the following protocol: the probe DNA was labelled using the Biotin High Prime kit (Roche, Basel, Switzerland) and detection of hybridization signals was performed using the Vectastain ABC elite kit (Vector Laboratories, Burlingame, CA, USA) and Alexa Fluor 594 Tyramide (Invitrogen). Chromosomes were DAPI stained and the slides were mounted using the VECTASHIELD mounting medium (Vector Laboratories, Burlingame, CA, USA). Chromosomes were screened under an epiflorescence Zeiss Axyoplan microscope; images were captured using an Olympus DP70 digital camera. For the chromosomal location of SSRs on mitotic chromosomes the karyotype description in Willhoeft [22] has been adopted.
The polymorphic information content (PIC) of each of the 19 SSR loci was determined using the program Cervus 3.0 [23]. For each locus and population, the number of alleles (Na), frequency range, observed heterozygosity (HO) and expected heterozygosity (HE) were estimated using the program Genepop version 4 [24]. The same software was also used to test for linkage disequilibrium between pairs of loci in each population (100 batches, 1000 interactions per batch) and for deviations from Hardy-Weinberg (HW) equilibrium, at each locus/population combination, using Fisher's exact test. The Bonferroni correction was used for all tests involving multiple comparisons [25]. The average exclusion probability (Excl.), i.e. the probability of excluding a single unrelated candidate parent from the parentage of a given offspring, knowing the genotype of the second parent, was estimated using the program Cervus 3.0. For each locus and population, the frequency of null alleles was calculated using the Brookfield estimation [26] in Micro-Checker 2.2.3 [27]. For the X-linked loci the number of alleles and the frequency range were evaluated using the data from both males and females, whereas heterozygosity, exclusion tests and frequency of null alleles, were calculated using the data obtained from only the females. Microsatellite Analyser (MSA) software, version 4.05 [28] was applied to determine the degree of genetic differentiation between Kabunkanga and Buvuma in terms of Fst [29].
There are three potential sources of errors associated with the genotyping of the sperm stored in the spermathecae [30], [31];
Two different approaches were used to determine the minimum number of mates per female. The first is a simple descriptive method, based on direct count, which does not involve any probabilistic model. The second approach, which incorporates information derived from the allele frequency in each population using the Hardy-Weinberg principle, provides expected values of multiple matings. This information would be lost if one followed only the first approach. It is worth noting that the expected values of multiple matings also take into account cases in which both males and females, in the population, share the same alleles for each locus. These cases are not recognizable as rematings in the direct count. For the second approach two different viewpoints were adopted: (a) the maximum likelihood technique and (b) the Bayesian analysis. For elementary explanations of these methods see [35]–[37].
The characteristics of the 19 identified SSR loci, in terms of primer sequence, amplification conditions and PIC values, are summarized in Table 1.The characterization was performed on DNA from single flies (29 females and 20 males) collected in KB. Eleven of these loci are X-linked while the remaining eight are spread along the L1 and L2 autosomes, as assessed by chromosomal in situ hybridization analyses (Figure 2). Out of these 19 loci, 4 autosomal (A03, B11, C7, D101) and 2 X-linked (C107 and D3) loci appear to be good candidates for sperm genotyping in remating studies, as they display high PIC values and are easy to score.
The variability estimates describing the suitability of the six loci: A03, B11, C7, D101, C107 and D3, for remating analysis in KB and BV, are shown in Table 2. The number of alleles per locus ranged from 6 to 12 with a mean of 8.83 in the KB population, and from 3 to 11 with a mean of 7.00 in the BV population. After Bonferroni correction [25] for multiple comparisons, Fisher's exact test revealed that the six loci are in Hardy-Weinberg equilibrium in both populations. No significant genotypic linkage was detected between the six loci (Fisher's exact test, Genepop) and therefore they can be considered as independent loci. Analyses performed with Micro-Checker [27] indicated that the average frequency of null alleles is low, 0.02 in KB and 0.01 in BV. The accuracy of these six loci for assessing remating is measured by their combined probability of excluding (Excl) an unrelated candidate parent from parentage when the genotype of the mother is known. The combined exclusion value is 0.99 in KB and 0.93 in BV. The different levels of variability between KB and BV populations is accompanied by a significant level of differentiation [38], as the estimate of FST is equal to 0.174 between the two populations.
The six microsatellite loci were successfully amplified from sperm DNA isolated from the spermathecae of 29 KB and 30 BV females.
From the 19 analyzed microsatellites, we chose the six most informative loci to estimate the number of matings per female, through the analysis of sperm preserved in the spermathecae in samples from two Uganda populations. The chosen loci are polymorphic for a large number of alleles, which differ in repeat number, making them easy to score for sperm genotyping. The loci are spread over the autosomes L1, L2 and the X chromosome of G. f. fuscipes, therefore they assort independently and no linkage disequilibrium has been assessed, providing further statistical power for the mating/remating analyses. The use of X-linked loci in association with autosomal loci provided a more sensitive estimate of number of matings, increasing, for instance, the power of identifying cases of triple matings as shown in Table 3. The direct count of remating estimates has been compared to probability estimates obtained with two inference methods, incorporating population allele frequencies. The mean number of matings per female, obtained from sperm genotyping in Kabunkanga and in Buvuma (Ncount = 1.61 and 1.33 respectively), were very similar to both probability estimates (Table 4), confirming that the use of the six microsatellites did not result in an under-estimation of remating events.
Our deductions are based on molecular data, which provide information on the number of males that were able to transfer sperm in a PCR-detectable quantity to a female's spermathecae. Consequently, a conservative (minimum) estimate of the number of males with which a female had mated, was determined in the Kabunkanga and Buvuma wild populations. Although our conditions were able to detect the presence of a second male sperm at a ratio as low as 1∶10, an undetected sperm contribution cannot be excluded. Furthermore cases of failure of sperm transfer, apparently after normal copulations, have been reported [39], [40].
Our results provide the first direct evidence that remating is a common event in the wild and what is more, females of G. f. fuscipes may store sperm from different males. These are biologically relevant data for interpreting the reproductive biology of this tsetse species, as it appears that many females preserve sperm from different mates, that could potentially be used for insemination. It is also known that this fly is able to maintain the sperm alive for long time [41]. The simultaneous presence of sperm derived from each mating suggests that one of the potential mechanisms of cryptic female choice, such as sperm displacement, [42]–[45] is not operating in this species. On the other hand, the storage of sperm from more than one male generates the opportunity for sperm competition for fertilization. Whether post-copulatory specific events/mechanisms are operating in the female storage organs to control or drive sperm use, is an important open question, which may clarify how the copulations are translated into fertilization in this fly. It is noteworthy that in G. austeni twice mated females utilize sperm from both matings for fertilization of oocytes [10]. If this is the case also for G. f. fuscipes, considering the high frequency of remating, this sperm use by polyandrous females may have a strong impact on the effective population size of the population.
Both direct count estimates of remating and probability estimates, obtained with the two inference methods, are significantly lower in Buvuma than in Kabunkanga: more than fifty per cent (57%) of females mated more than once in Kabunkanga while a smaller proportion (33%) remated in Buvuma. Various factors, which may be interrelated, could be responsible for the observed difference. First, the lower genetic variability in Buvuma, with respect to Kabunkanga, diminishes the discriminatory power of the six SSR loci in this island population, as revealed by the lower combined exclusion probability estimate (Excl 0.93 versus 0.99). Probably this observation is not related to the choice of loci, as Beadell et al. [8] demonstrated that in Uganda there is a significant decline of microsatellite allelic richness from West to East: Kabunkanga and Buvuma are located at a great geographic distance in the West and East, respectively, of the predicted range of the species (Figure 1). Thus, considering that the Excl estimate is related to the level of genetic variability, with an Excl value of 1.00, we would have increased our remating estimates, obtaining an expected value of 0.58 for Kabukanga and 0.36 for Buvuma. Since there is still a difference in the remating frequencies between the two populations, other interrelated eco-geographic and demographic factors must account for the difference. The average age structure may have played a role. In Buvuma Island, flies were caught in April, at the beginning of the rainy season when the population was expanding as also confirmed by the high fly density in the traps, which is about four times greater than the density in Kabunkanga. The Kabunkanga flies were collected in November, at the end of the cooler dry season, when the population undergoes seasonal demographic contractions with a high level of mortality particularly among the young teneral flies while the remaining flies concentrate in moist refugia. In the absence of objective observations regarding the age, such as ovarian measurements and wing-fray analysis [18], we can speculate that in an expanding population, such as Buvuma, the proportion of young flies may be greater than that in a residual population after a seasonal bottleneck, such is the case of the Kabunkanga sample [9], [46]–[51]. It is a reasonable assumption that the surviving flies collected in Kabunkanga at the end of the dry season, had more time and opportunity to remate, than those from Buvuma. In addition, according to Abila et al. [12], male mating competitiveness increases with age, i.e. older males copulate significantly more frequently than younger flies and the peak of female receptivity is between the 8th–13th day after emergence [52]. It has been also reported that Glossina females tend to mate more than once with no apparent difference in receptivity and the number of matings appears to be directly related to the amount of semen in the spermathecae: young females contain less semen than older ones [53]. On the basis of these observations, it can be speculated that a demographic parameter such as age could be the cause of the observed difference in remating frequency between Kabunkanga and Buvuma. However, this hypothesis must be confirmed by appropriate analyses. Finally, as the two study sites, Kabunkanga and Buvuma island harbour well established populations which show a significant level of genetic differentiation (Fst = 0.174), we cannot exclude that the distinct genetic backgrounds of the two populations had an effect on the extent of the observed remating estimates.
Several considerations concern the applied aspects of the present findings. As the Sterile Insect Technique (SIT) is being entertained for tsetse population control, the presence of remating and the fact that females maintain sperm from different mates, potentially available for insemination, may constitute a critical factor for the success of eradication programmes. Although specific experiments would be necessary to assess the sperm use and the possible presence of paternity skew in populations, multiple mating may potentially help maintain genetic variability and increase the effective population size. Thus polyandry may affect the long-term stability and effective size of G. f. fuscipes populations. In cases of eradication programmes, re-infestation of cleared areas and/or in cases of residual populations, the occurrence of remating may, unfortunately, enhance the reproductive potential of the re-invading propagules in terms of their effective population size. The comparison of two populations highlights another important factor, which, if confirmed, influences the remating frequency, i.e. the population age structure. Consequently, any vector control programme for G. f. fuscipes, according to the present results, must address the greater dimension of the young expanding population in the early wet season, and the increased rate of remating of the fewer, remaining adults after the bottleneck in the dry season. For instance in the case of SIT, a large number of sterile males should be released, also in a population with a reduced number of individuals because of the high rate of remating. These considerations agree with the recommendation to release aged, more competitive, sterile males in all cases [12].
Finally, analyses have identified the presence of parasitic Wolbachia infections in some individuals of natural populations of G. f. fuscipes, including those from Uganda described here. As it has been suggested that Wolbachia-associated incompatibilities may promote polyandry [54], future studies can now investigate the potential influence of Wolbachia in the remating phenomenon described here. As Wolbachia infections are entertained as a tool to drive genetically desirable phenotypes into natural populations [55], female mate choice and remating may also have an impact on strategies of population replacement.
|
10.1371/journal.pgen.1004498 | Microsatellite Interruptions Stabilize Primate Genomes and Exist as Population-Specific Single Nucleotide Polymorphisms within Individual Human Genomes | Interruptions of microsatellite sequences impact genome evolution and can alter disease manifestation. However, human polymorphism levels at interrupted microsatellites (iMSs) are not known at a genome-wide scale, and the pathways for gaining interruptions are poorly understood. Using the 1000 Genomes Phase-1 variant call set, we interrogated mono-, di-, tri-, and tetranucleotide repeats up to 10 units in length. We detected ∼26,000–40,000 iMSs within each of four human population groups (African, European, East Asian, and American). We identified population-specific iMSs within exonic regions, and discovered that known disease-associated iMSs contain alleles present at differing frequencies among the populations. By analyzing longer microsatellites in primate genomes, we demonstrate that single interruptions result in a genome-wide average two- to six-fold reduction in microsatellite mutability, as compared with perfect microsatellites. Centrally located interruptions lowered mutability dramatically, by two to three orders of magnitude. Using a biochemical approach, we tested directly whether the mutability of a specific iMS is lower because of decreased DNA polymerase strand slippage errors. Modeling the adenomatous polyposis coli tumor suppressor gene sequence, we observed that a single base substitution interruption reduced strand slippage error rates five- to 50-fold, relative to a perfect repeat, during synthesis by DNA polymerases α, β, or η. Computationally, we demonstrate that iMSs arise primarily by base substitution mutations within individual human genomes. Our biochemical survey of human DNA polymerase α, β, δ, κ, and η error rates within certain microsatellites suggests that interruptions are created most frequently by low fidelity polymerases. Our combined computational and biochemical results demonstrate that iMSs are abundant in human genomes and are sources of population-specific genetic variation that may affect genome stability. The genome-wide identification of iMSs in human populations presented here has important implications for current models describing the impact of microsatellite polymorphisms on gene expression.
| Microsatellites are short tandem repeat DNA sequences located throughout the human genome that display a high degree of inter-individual variation. This characteristic makes microsatellites an attractive tool for population genetics and forensics research. Some microsatellites affect gene expression, and mutations within such microsatellites can cause disease. Interruption mutations disrupt the perfect repeated array and are frequently associated with altered disease risk, but they have not been thoroughly studied in human genomes. We identified interrupted mono-, di-, tri- and tetranucleotide MSs (iMS) within individual genomes from African, European, Asian and American population groups. We show that many iMSs, including some within disease-associated genes, are unique to a single population group. By measuring the conservation of microsatellites between human and chimpanzee genomes, we demonstrate that interruptions decrease the probability of microsatellite mutations throughout the genome. We demonstrate that iMSs arise in the human genome by single base changes within the DNA, and provide biochemical data suggesting that these stabilizing changes may be created by error-prone DNA polymerases. Our genome-wide study supports the model in which iMSs act to stabilize individual genomes, and suggests that population-specific differences in microsatellite architecture may be an avenue by which genetic ancestry impacts individual disease risk.
| Over 3% of the human genome consists of microsatellites, defined as short tandem repeats of 1–6 bases per motif unit, interspersed throughout the genome [1]. Strand slippage during DNA synthesis is facilitated by the presence of tandem repeats, and has been proposed to be the dominant mutational mechanism for microsatellites [2], [3]. Perfect microsatellites contain repeats of a single motif sequence, whereas interrupted microsatellites (iMSs) include tandem repeats of a single motif interrupted by other bases. Many microsatellites are located within coding and regulatory sequences [4], and can be important modifiers of gene expression, affecting transcription rate, RNA stability, splicing efficiency, and RNA-protein interactions [5]–[7]. Because microsatellite alleles are highly polymorphic, they may provide a large pool of heritable, phenotypic variants for subsequent selection [8]–[10]. Length variation at certain microsatellites contributes to natural variation in brain development and behavioral traits [11], and may modulate neurodegenerative disease risk [12].
Microsatellite interruptions also are known to have important consequences for human health and disease. For instance, germline interruptions of disease-causing microsatellite alleles act as a disease modifier for spinocerebellar ataxia type 10 [13], and alter the age of onset of spinocerebellar ataxia type 1 [14]. Importantly, the presence of interrupted alleles at the FMR gene (Fragile X syndrome) microsatellite diminishes the likelihood of repeat-expansion to disease length alleles in the next generation [15], [16]. Similarly, the presence of multiple interruptions at the DM-1 gene microsatellite decreases the probability of both germline and somatic expansions [17], [18]. Furthermore, a population-specific, single nucleotide polymorphism within the APC gene coding region converts an iMS (AAATAAAA) to a perfect microsatellite (A)8, leading to an increased risk of somatic APC mutation and colorectal cancer in Ashkenazi Jews [19]. Biomedical interest in microsatellite interruptions has been renewed recently by the demonstration that iMSs within the ATXN2 (SCA2) gene are associated with a different disease presentation than perfect expanded alleles [20]. These studies demonstrate that a complex relationship exists between microsatellites and disease, that involves not only length but also sequence polymorphisms. Importantly, iMSs might represent a reservoir of mutable alleles that can expand in subsequent generations, as was shown for SCA2 [21] and myotonic dystrophy type 2 [22].
Microsatellite interruptions are major contributors to the microsatellite life cycle. According to the life cycle hypothesis, a microsatellite locus undergoes stages of birth, adulthood and death during its evolution [23]. Microsatellites are “born” from short tandem repeats (proto-microsatellites) when they reach a threshold length that alters their mutational behavior [24], [25]. Microsatellites display a characteristically high frequency of motif-based insertion/deletion (indel) mutations that drive high germline microsatellite mutation rates; this is in contrast to proto-microsatellites that have lower indel mutation frequencies than microsatellites [25], [26]. Microsatellites “die” when the length of the tandem repeat falls below the threshold, and interruptions are the major cause of microsatellite death [27], [28]. Some interruptions can persist for millions of years (MYs), e.g., for 19–35 MYs at one locus studied in artiodactyls [29]. These features can serve as an advantage when using iMSs as markers in population genetics, since interrupted repeats exhibit lower homoplasy than uninterrupted MSs. Indeed, for iMSs, the probability of acquiring an interruption by two independent events (i.e. the probability of a homoplasy) is much lower than the probability of inheriting this interruption from a common ancestor. Because of this, iMSs might be more appropriate markers than perfect microsatellites for studying population differentiation [30]. Interrupted microsatellites are more stable genetically (less mutable, but still polymorphic) than perfect repeats in natural chicken populations [31], and interruptions can reduce the mutability of specific microsatellite sequences [32]–[34]. However, the quantitative effects of interruptions on decreasing human microsatellite mutability have never been evaluated previously in a genome-wide study.
The significant role of iMSs in modifying the clinical manifestations of disease and their important contributions to genome evolution warrant a detailed understanding of iMSs. Specifically, the architecture of human genomes with regard to iMSs has not been previously investigated, and the mechanism by which interruptions arise has not been extensively studied. We used a multi-disciplinary approach combining computational and biochemical methods to address three biologically important questions regarding microsatellite interruptions. First, what is the quantitative effect of microsatellite interruptions on microsatellite mutability genome-wide? Second, how common are microsatellite interruptions within the human genome, where do they occur, and how often are human populations polymorphic for the presence/absence of interruptions? Third, what are the possible biochemical pathways giving rise to microsatellite interruptions? Our results reveal the highly dynamic nature of microsatellite mutagenesis in the human genome, one that includes a robust level of interruption variation, and demonstrate that iMSs provide a source of population-specific genetic modifiers potentially affecting the stability of individual human genomes.
To understand the impact of microsatellite interruptions on human genome stability, we first set out to determine the genome-wide magnitude of microsatellite mutability reduction due to the presence of interruptions. For this analysis, we studied high-quality primate genome alignments using a comparative genomics approach. Mono-, di-, tri- and tetranucleotide microsatellites above the threshold repeat number were identified in human, chimpanzee, orangutan, macaque, and marmoset reference genomes (Table S1; penta- and hexanucleotide microsatellites were omitted due to their lower abundance and algorithmic difficulties in specifying all possible interruptions). iMSs were identified as microsatellites in which at least one perfect repeat stretch extended beyond the threshold repeat number. An interruption was required to be shorter than or equal to the microsatellite's motif size. For each of the five primate genomes examined, iMSs were more abundant than perfect microsatellites (Table S1). When only orthologous iMSs with one or two interruptions were considered (see Materials and Methods for details), iMSs numbered from 6,000–38,000, while perfect microsatellites numbered from 8,000–48,000, depending on the primate genome analyzed.
The mutability, or the average squared difference in repeat number (allele length) between two species [35], was contrasted for all perfect versus interrupted microsatellites present in human-chimpanzee genomic alignments. Namely, we performed a genome-wide comparison of the mutability of microsatellites with the same repeated motif that were perfect in both human and chimpanzee to that of microsatellites that were interrupted (with the same interruption(s)) in both of these species. For microsatellites of all motif sizes examined, short microsatellites with one interruption were less mutable than perfect microsatellites with the same overall repeat number (Figure 1A). The average, genome-wide mutability difference for mononucleotides was ∼two-fold at 12 repeat units, and up to ∼six-fold for di-, tri-, and tetranucleotide microsatellites with 6, 5, and 4 units, respectively. Microsatellites with two interruptions were, on average, one to two orders of magnitude more stable than uninterrupted microsatellites with the same repeat number (Figure 1A). The mutability difference between perfect and iMS loci was highest at shorter repeat numbers for all motifs. Thus, the quantitative effect of a single interruption on an individual microsatellite locus can be substantial. For example, more centrally located interruptions have a strong effect on mutability, dramatically lowering microsatellite mutability up to two to three orders of magnitude, whereas interruptions located on the microsatellite fringes have only a marginal effect (Figure 1B). The identity of the interrupting base has a non-significant effect on mononucleotide microsatellite mutability (Figure S1).
Armed with the knowledge that interruptions significantly stabilize microsatellites genome-wide, we next examined individual human genome microsatellites for the presence of interruption polymorphisms. We found such polymorphisms to be highly abundant and informative for predicting population-specific microsatellite stabilization. In this analysis, we identified 1,814,151 perfect mono-, di-, tri-, and tetra-nucleotide microsatellites above the threshold length within the reference human genome (UCSC build hg19) [25]. Here, we imposed an upper limit on the microsatellite lengths analyzed (10, 9, 8, and 7 units for mono-, di-, tri- and tetranucleotide repeats, respectively), because we found next generation sequencing data at longer repeats to be biased due to sequencing errors and/or read-length limitations [25]. For microsatellites that are perfect in the reference genome, we analyzed the frequency of iMSs within four human population groups (African, European, East Asian, and American), using the 1000 Genomes Phase-1 variant call set [36]. Interruptions were defined as single nucleotide polymorphisms (SNPs) or indels leading to a sequence within the microsatellite that differs from the full motif unit. All indel and SNP variants (with allele frequency ≥0.05) were identified, and considered to be interruptions if they were located within a microsatellite but not at the starting/ending repeat unit. In this manner, we identified ∼26,000–40,000 polymorphic iMSs, depending on the population group (Table 1, Figure 2A; Datasets S1, S2, S3, S4, S5). A substantial number of interrupted alleles were present in all four population groups with different allele frequencies, corresponding to a fixation index (FST) of 0.061 (range: 0.000–0.590; sd: 0.062; median: 0.041), which falls well within the range of SNP FST values (0.052–0.083) derived from pair-wise population comparisons of the 1000 Genomes Phase-1 project [36] (Dataset S6). Despite such low observed average level of population differentiation, numerous interruptions were shared by two or three population groups, or unique to a single population group (referred to as ‘population-specific’ interruptions henceforth)(Figure 2A). The greater number of interruptions within Africans compared to other population groups is likely due to a higher number of the 1000 Genomes variants in Africans, reflecting their high diversity [36], [37]. We also identified genes that encode polymorphic exonic iMSs. Among the four population groups studied, ∼3,000–4,000 genes contained polymorphic interruptions within exonic microsatellites (Table 2). Several genes encoding exonic iMS alleles are specific to only one population, or are shared by two or three populations (Figure 2B; Dataset S1, S2, S3, S4, S5). These data demonstrate that iMSs can provide an abundant source of population-specific alleles potentially stabilizing individual genomes by lowering microsatellite mutation rates.
We performed more in-depth analyses of the polymorphic exonic iMSs identified above in the four human population groups to determine the potential functional impact of iMS presence on genome function. Only a few of the iMSs identified are predicted to cause frameshifts or nonsynonymous mutations (Figure S2, Table S2); the vast majority of population-specific interruptions are not expected to alter protein sequence. Thus, the primary effect of iMS may be to modulate the mutation rate of the underlying microsatellite. To gain further insight into the potential biological relevance of the iMSs, we performed Gene Ontology (GO) analyses for each set of genes encoding population-specific iMS alleles. The significantly (p<0.01) enriched GO terms are distinct for each population. For example, the GO terms enriched in the African-specific iMS genes included several neurological and organ development terms (Table S3), while those for the European-specific iMS genes were predominantly immunological terms (Table S4). Since the GO vocabularies are structured such that they can be queried at different levels, we examined the smallest sized GO terms, identified the associated genes containing the iMS, and queried these genes for clinical associations using Online Mendelian Inheritance in Man (www.omim.org). Several genes that we identified in this manner are associated with familial disease or disease susceptibility (Table 3). For example, we discovered three, African-specific interrupted mononucleotide microsatellites within the HTT (Huntington's) gene, which correspond to perfect microsatellites in European, Asian and American populations (Table 3). It is important to bear in mind that although the genes identified by this analysis are implicated in disease, the associated microsatellites have not been shown to play a causal role. Therefore, these iMSs will have to be studied further for their potential role in modulating disease risk.
We also examined polymorphisms in 15 genes containing exonic (coding and UTR) iMS alleles that are well known to be associated with microsatellite expansion diseases [38]. Eight loci (ARX, CBFA1, FMR1, FMR2, HOXA13, OPMD, SCA3, and ZIC2) contained no differences in microsatellite sequence from the reference genome in any of the four population groups studied. Four genes (AIB1, SCA2/ATXN2, SCA17, and HOXD13) contained iMS alleles that differed from the reference genome sequence, and the variants were present in all four population groups at differing allele frequencies (Table 4). For some loci/populations, the reference genome sequence is not the major allele (e.g., SCA17). The genetic consequences of the iMS variants include both sequences that are expected to increase mutability, and sequences expected to decrease mutability. For example, the HOXD13 variant iMS allele is expected to have lower mutability than the reference genome iMS due to the presence of a third interruption that decreases the perfect tandem (GCG)5 repeat to a length below the mutability threshold (four units for trinucleotide repeats [25]). The frequency of this triply-interrupted allele varies from 0.76 in the African population to 0.26 in European and American populations. The AIB1 locus contains four alternative iMS alleles present at varying frequencies among the populations, one of which is a doubly interrupted allele, leading to greater stabilization of the repeat due to disruption of the (CAG)6 array. For three loci (DRPLA, SCA1, and FOXL2), we observed instances of population-specific iMS alleles. DRPLA contained variant alleles in only two of the four population groups studied (African and American), both of which decrease the number of interrupting bases, relative to the reference genome, potentially increasing mutability of the repeat. Finally, we noted an increased number of interruptions within polyglutamine repeats compared to polyalanine repeats, consistent with previous observations about the high propensity of polyglutamine repeats to acquire length and nucleotide polymorphisms [39].
Low indel mutation rates of iMSs (Figure 1) also are expected to be reflected in their low indel polymorphism levels. To test this, we investigated the levels of heterozygosity and the presence of linkage disequilibrium (LD) between interrupted microsatellite alleles caused by indels and neighboring, population-matched SNPs from the 1000 Genomes Phase-1 data. Approximately 30–40% of iMSs display low levels of heterozygosity (below 0.2; Figures S3A–D). In fact, we observed a skew towards lower heterozygosity for iMSs as compared to that for perfect microsatellites (p = 0.028 for Asians; p = 0.066, p = 0.057, and p = 0.072 for Africans, Americans, and Europeans, respectively; Kolmogorov-Smirnov test).
In each of the four populations studied, 4,400 to 5,000 interruption-causing alleles (36–49% of the alleles investigated) were found to be in moderate LD (R2>0.80) with SNPs (Figure S4, Table S5), and 686 to 990 alleles (6–10%) were in perfect LD (R2 = 1) with SNPs. Interestingly, certain interruption alleles displayed perfect LD in some, but not all, populations (Table S5). Generally, iMS alleles in the African population displayed lower levels of LD compared to the other three populations (Figure S3), likely due to the abundance of low-frequency variants in Africans compared to non-African populations [36].
The exonic iMSs in perfect LD with neighboring SNPs were examined in more detail. Within each population, 6 to 11 of such alleles were identified (Table S6). For each allele, we examined the phenotype and disease relationships of the linked SNPs using SNPnexus web browser [40]–[42], and found associations with cancer, neurological, immune, cardiovascular, and metabolic disorders (Table S6). These associations reiterate a potential for iMSs to modulate disease risk in a population-specific manner.
We sought to directly verify the quantitative effect on mutability of a single base substitution interruption within an exonic microsatellite encoded within a human disease gene. We chose the well-established biological model of a population-specific iMS encoded within the APC tumor suppressor gene. In 6% of the Ashkenazi Jewish population, a centrally located iMS (AAATAAAA) within an exon of the APC gene is present in the germline as a perfect A8 microsatellite (AAAAAAAA); this nonsynonymous SNP leads to an I1307K variant, but has no effect on APC protein function [19]. Nevertheless, this population has a greater chance of producing an inactive APC gene in somatic tissues, which increases the risk of colorectal cancer [43]. The proposed mechanism accounting for this observation is the enhanced somatic mutability of the perfect A8 sequence, relative to the interrupted sequence [19], [44]. We modeled the germline sequences of the perfect and interrupted APC microsatellites, and measured DNA polymerase strand slippage error rates using our established in vitro assay. Briefly, in this analysis, defined tandem repeat sequences are inserted in-frame within a reporter gene. Vectors containing these reporter cassettes are used as templates for in vitro DNA synthesis reactions, and DNA polymerase errors that result in gene inactivation (frameshift, nonsense or missense mutations) are scored by genetic selection in E. coli [45], [46]. To determine the specificity of polymerase errors, independent mutants are isolated, and the DNA sequence changes within the reporter region are determined by dideoxy DNA sequence analysis of purified vector DNA [47].
For these experiments, we examined three DNA polymerases, representing distinct polymerase families and postulated to be required for distinct genome maintenance functions: Pol α, DNA replication; Pol β, DNA repair; and Pol η, translesion synthesis. The accuracy of each polymerase was measured on four DNA templates, representing the complementary strands of the perfect (A8 and T8) and iMS (A3TA4 and T3AT4) alleles in APC (Figure 3A). For the perfect allele templates, the polymerases created +1 A/T insertions, −1 A/T deletions, and A:T to T:A tranversions that lead to TAA nonsense codons (data not shown), which also are the types of inactivating APC somatic mutations observed within tumors from I1307K carriers [44]. For the iMS allele templates, the polymerase indel error frequency was five- to 50-fold lower than that for the perfect allele, depending on the polymerase, demonstrating strand slippage stabilization by this single interruption (Figure 3A; Table S7). We observed that the interrupting base is rarely removed by these polymerases; the predominant errors (>95%) are indels within the remaining perfect tandem repeat tracts (Figure 3B). The frequency of deleting the interrupting base to create a perfect allele was very low (9.2×10−6 and 2.2×10−5 for Pol α and Pol β, respectively), relative to other types of polymerase errors (Table S7). Moreover, the polymerase error frequencies at the residual repeats within the iMS alleles were similar to the error frequencies at similar short tandem repeats located elsewhere within the HSV-tk gene coding sequence (data not shown). These analyses strongly suggest that the single nucleotide interruption within the APC gene leads to the mutational death of the microsatellite.
DNA sequence analyses of Pol η errors produced on the interrupted templates emphasized the novel mutational signature of this enzyme within this specific microsatellite motif (Figure 3C). Intriguingly, Pol η has the unique ability to litter this iMS with additional errors, often creating a DNA synthesis product that is more random in sequence than the starting iMS template sequence. Despite this ability, the original interrupting base is maintained in the majority (79%) of Pol η synthesis products.
Despite the clear biological significance of iMSs on human genome stability and disease risk, very little is known regarding the biochemical pathways by which interruptions arise in microsatellites. Mutational events to create interrupted alleles could be produced during several cellular mutagenesis pathways, including cytosine deamination events, the creation of abasic sites, endogenous DNA damage-induced mutations and DNA polymerase errors, among others. We used two complementary approaches to gain insight into the potential pathways underlying the production of iMS in the human genome. First, the abundance of polymorphic interruptions and the short evolutionary time since divergence of the four 1000 Genomes population groups allowed us to examine the types of mutations leading to population-specific microsatellite interruptions in detail. (We observed a high degree of interruption gain/loss event saturation along primate phylogenetic branches, precluding us from deciphering interruption pathways in this data set. For instance, the resulting numbers of interruptions along the human or chimpanzee lineages since their ∼6 MY split were similar to that along the orangutan lineage since its ∼12 MY split from the human lineage (Figure S5)). Second, the fact that DNA polymerases can create interruption errors during in vitro synthesis of microsatellite-containing templates [45], [46], [48] afforded us the opportunity to examine one biochemical pathway- namely, polymerase errors during DNA synthesis.
In this study, we answered three biologically significant questions regarding mono-, di- tri- and tetranucleotide microsatellite interruptions in the human genome. First, using primate genome alignments, we quantified the genome-wide effect of interruptions on decreasing microsatellite mutability, and found it can be significant and strong – from several fold to several orders of magnitude, compared with perfect repeats. Second, utilizing the 1000 Genomes Phase-1 dataset, we found iMS polymorphisms to be highly abundant and informative for predicting population-specific microsatellite stabilization, especially for exonic loci. The vast majority of the population-specific, exonic iMSs we identified are not expected to alter protein sequence; thus, the primary effect of interruptions may be to modulate the mutation rate of the underlying microsatellite. Third, we discovered that base substitutions are the primary type of interruption among MSs in all population groups, and for the four microsatellite classes examined. We surveyed five mammalian DNA polymerases involved in DNA replication, repair, and specialized functions, and found that, for the mono- and dinucleotide microsatellite sequences analyzed, iMSs are created most frequently by error-prone polymerases. Pol η is notable among the enzymes examined in that the microsatellite DNA synthesis products are characterized by a high degree of sequence diversity.
Early studies of microsatellite interruptions demonstrated reduced mutation rates at a few iMS loci, as compared with perfect alleles of the same repeat number [28]–[30], [44]. A higher mutability of microsatellites was observed for interruptions closest to the repeat tract ends, as compared with centrally located interruptions [31], [33], [53], [54]. Such studies suggested that interruptions might effectively divide microsatellites into shorter repeat runs. Within the interrupted repeat itself, the mutation rates of the individual arms depend on the lengths of perfect tracts remaining within the iMS allele [55].
Here, we provide a detailed, genome-wide analysis of the mutability of perfect and interrupted MSs in completely sequenced primate genomes. For the four motif sizes examined, interruptions significantly reduced mutability when present (a) within shorter microsatellites, (b) in multiple numbers (i.e., two interruptions per microsatellite), or (c) near the center of the microsatellite (Figure 1) – all of which give rise to a shorter perfect repeat tract. Importantly, the magnitude of the effect of interruptions on microsatellite allele length variation ranged from a few-fold to several orders of magnitude for loci across the genome.
We also report here that the perfect microsatellites in the human reference genome analyzed here (≤10 units in length) are frequently found as iMS polymorphisms within the genomes of individuals from four population groups. Although the majority of iMS alleles were shared among all groups, many of the iMS alleles we detected were specific to only one population group, or shared between subsets of population groups (Figure 2). Our quantitative results for the stabilizing effects of interruptions in short microsatellites are biologically relevant here, as the vast majority of iMSs we identified in human genomes are within short microsatellites, just above the length threshold. Therefore, interruptions are expected to have a strong effect on stabilizing such microsatellites. Thus, iMSs are a likely source of population-specific genetic variants that can affect the stability of individual genomes by reducing the mutability of microsatellites. To the best of our knowledge, this is the first report of iMSs as an abundant source of population-specific genetic modifiers in the human genome. The full abundance of iMSs within the human genome must await future studies, when improvements in sequencing technology read length and accuracy will allow the interrogation of all microsatellite motif sizes, lengths, and sequences that are present within individual genomes.
The APC tumor suppressor gene illustrates a provocative example in which a single, population-specific, germline SNP can affect disease risk by altering the mutagenic potential of a microsatellite sequence. Our data directly support the previous model that the perfect [A8/T8] allele creates a hypermutable region within the APC gene, leading to cancer predisposition [19]. We measured DNA polymerase strand slippage error rates that are up to 50-fold lower during replication of the iMS sequences [A3TA4/T3AT4], compared to the perfect sequences [A8/T8] (Figure 3). Previous biochemical studies of trinucleotide microsatellites have shown that interruptions decrease slipped strand formation [56] and decrease the thermostability of secondary structures formed by repetitive sequences [57]. Our results advance these studies by demonstrating that the mechanism of reduced mutability by an interruption within a mononucleotide A/T allele is lowered polymerase strand slippage errors during DNA synthesis.
Expanding on the APC gene observation that SNPs can create perfect microsatellites and hypermutable sequences in disease states, we identified ∼3,000–4,000 genes (depending on the population group) that are perfect in the reference genome, but contain iMS within exonic regions (Figure 2). The exonic iMS alleles that are specific to only one or two populations likely represent a pool of genes that are at a risk for increased mutation in the other population groups. Madsen and colleagues reported that short tandem repeats/microsatellites in exons are overrepresented among human genes associated with cancer and immune system diseases [58]. We observed that while European-specific iMSs are enriched in genes associated with immunological function, African-specific iMSs are enriched in genes associated with neurological function. Thus, population-specific differences in microsatellite architecture (perfect versus interrupted) may be a widespread mechanism by which genetic ancestry impacts individual disease risk. While our focus has been on comparing population groups, our FST analysis indicated that many iMS alleles are not fixed within population groups, thus potentially providing a rich source of individual genetic variability.
Perfect microsatellites are at a higher risk for microsatellite expansion mutations that are causative for numerous neurological/neurodegenerative diseases [3], and the presence of interrupted alleles has been well documented to decrease disease risk. We investigated several genes previously described as harboring disease-associated, coding iMS alleles [38]. The genetic consequences of the iMS variants we identified include both sequences that are expected to increase mutability, and sequences that are expected to decrease mutability. Various AIB1 iMS alleles have been noted previously in a survey of European DNA samples [59], consistent with the allelic distribution we observed for the 1000 Genomes European population group. One of the iMS variants we identified within AIB1 occurs at a much higher allele frequency in the African population, and is expected to display higher mutability than the reference sequence, due to an increased perfect tandem repeat tract length. The two HOXD13 iMS alleles we identified were observed previously in a pedigree analysis of 16 synpolydactyly families [60]. Importantly, repeat expansions in these families segregated with the disease phenotype; however, the iMSs were retained in all of the expanded alleles. Recently, amyotophic lateral sclerosis patients have been described as having moderately expanded SCA2 iMS alleles that retain at least one of the interruptions [20], [61]. Both microsatellite length and purity (interruption) SCA1 and SCA2 polymorphisms have been described among unaffected individuals [62], [63], consistent with the iMS variant alleles we detected in this study.
The pathways by which iMSs arise in genomes have not been extensively studied. Several cellular mechanisms could account for the production of iMS alleles in genomes, including (but not limited to) endogenous DNA damage-induced mutations and DNA synthesis errors during DNA replication, repair and/or recombination. The types of iMS ultimately observed in human genomes will be further shaped by DNA repair pathways and selection, which will serve to reduce the number of and narrow the types of mutational events within microsatellites. We demonstrate here that base substitutions are the primary type of iMS present in individual human genomes. We also used our established biochemical assay to determine the potential contribution of errors created by three distinct DNA polymerase families to the formation of iMS alleles. For the microsatellite templates and types of detectable errors examined, we observed that genome stabilizing microsatellite interruptions are created most frequently in vitro by error-prone, specialized Pols η and κ, while replicative Pols α and δ rarely created interruptions (Figure 5). The generality of our observations for all microsatellite sequences and human polymerases is not known, and must await future experimental analyses. Nevertheless, we observed that DNA Pol η is very efficient at making interruptions within perfect microsatellites and creates multiple errors within a single DNA synthetic event. Pol η also creates base substitution errors within the tandem repeat tracts of iMS templates, with the net result being a more random sequence. DNA Pol η serves several important functions in human genome stability. Germline mutations leading to loss of Pol η activity causes the cancer predisposition syndrome, xeroderma pigmentosum-variant [64], and enhanced cellular UV sensitivity [65]. Pol η has been well- characterized biochemically, and is capable of accurate translesion synthesis across UV photoproducts and other DNA lesions [64], [66]. Human Pol η also is required for the maintenance of common fragile sites and prevention of chromosomal rearrangments [67], [68]. On the other hand, Pol η performs a key role in targeted mutagenesis during somatic hypermutation of immunoglobulin genes, primarily targeting mutations to A:T basepairs [69]–[71]. Here, we show in vitro that Pol η litters mononucleotide A/T microsatellites with many base substitution errors (Figure 3C and Table 5), an error characteristic that is highly reminiscent of somatic hypermutation.
Previous studies of primate MSs reported that point mutations occur more frequently than expected within microsatellites, based on the overall genome divergence [72], and that there is a two-fold higher rate of base substitutions within coding microsatellites relative to other coding sequences [73]. In a study of microsatellite births and deaths, we observed that substitutions were the leading cause of death, and that the density of births/deaths is non-random throughout the genome [27]. Although interruptions can be removed from microsatellites, restoring long perfect repeat stretches and high mutability of microsatellites [27], our in vitro results suggest that this may be a rare event during DNA synthesis based on the small number of microsatellites examined.
Our discovery that interruptions are created more frequently by low fidelity repair and specialized polymerases than by high fidelity replicative polymerases suggests one potential mechanistic explanation for these observations. Based on our data to date, we would predict that the frequency of interruptions among microsatellites in the genome (of the same motif and number) will depend upon the relative activities of replication, repair and recombination DNA synthesis pathways, such that more iMSs are expected in genomic regions where either repair or specialized polymerases, such as Pols η, κ and β, are more frequently engaged. DNA synthesis by these polymerases would have the consequence of speeding up microsatellite death and impeding microsatellite resurrection [74]. For example, specialized polymerases may be engaged at the replication fork more often during synthesis of highly repetitive microsatellite sequences than of coding sequences, because replicative polymerases are inhibited [46], [68], [75]. Indeed, Pol κ was recently implicated in the synthesis of DNA at stalled replication forks in unstressed human cells [76]. Alternatively, an increased level of DNA damage within microsatellites, relative to coding sequences, would necessarily engage repair and specialized polymerases during the downstream pathways of gap-filling or translesion synthesis, respectively. A noncanonical pathway of mismatch repair that is activated by DNA lesions was shown to recruit Pol η to chromatin in a replication-independent manner [77]. Finally, Pol η activity may be targeted to specific genomic sequences, such as the highly mutable hotspots identified for somatic hypermutation of immunoglobulin genes.
Microsatellites present within regulatory regions of the genome can affect gene expression, and allele length polymorphisms are increasingly recognized as contributing to phenotypic variation and disease risk [5], [10], [12]. Indeed, it has been previously proposed that polymorphic microsatellite alleles present within candidate genes associated with a disease or trait should be considered as contributing to the trait [11]. Genomic microsatellites display genetic variation that includes both allele length and sequence polymorphisms. The genetic architecture of microsatellites can include stabilizing, interrupted alleles. Our study advances our understanding of the impact of microsatellite sequence variation by illuminating the sheer abundance of iMS alleles within individual human genomes and the magnitude of the genome stabilization effects. We have identified genes encoding exonic microsatellites that are present as protective, interrupted alleles in only one of four human population groups. These population-specific, iMS-containing genes are enriched in distinct functional pathways, suggesting that microsatellite sequence variation may contribute to the effects of genetic ancestry on disease risk. Importantly, our analyses demonstrate that many iMS alleles are not fixed within population groups, suggesting that microsatellite interruptions could be a source of genetic variability impacting individual phenotypic variation.
We identified perfect as well as interrupted microsatellites in human (hg18), chimpanzee (panTro2), orangutan (ponAbe2), macaque (rheMac2) and marmoset (calJac1) genomes using Sputnik [78] and a computational pipeline that we developed for proper extraction of iMSs (see below). In this approach, Sputnik is utilized to perform a genome-wide search for microsatellite ‘seeds’ (see Table S1 for search parameters) i.e., stretches of perfect mono-, di-, tri- and tetra-nucleotide repeats at or above the threshold repeat lengths of 9, 5, 4 and 3 units, respectively (following [24], [79]). Each seed's (e.g. [AC]6) flanking sequences are examined for the presence of (a) any additional seeds of any motif, or (b) additional instances of the repeat motif (e.g. [AC]2) with the intervening non-repeat nucleotides extending to not more than the length of the repeat motif itself (here, 2 bp). If additional complete repeats of the repeating motif or seeds composed of the same repeat motif are identified in the neighborhood of the seed, then the focal seed and the discovered extensions are merged into a single microsatellite. To complete the above example, if the focal seed [AC]6 exists such that (a) on its 3′ end, following a dinucleotide GT, there was discovered another seed [AC]7, and (b) on its 5′ end an immediately adjacent instance of [CA]2 is found, then the resultant focal seed is extended to include these additional repeats such that the final repeat becomes [AC]7GT[AC]6[CA]2. This extension process is continued iteratively into the flanking regions until no more additional instances of the focal motif are identified, or if the terminal additions to the microsatellites are composed of repeat instances that are smaller than two repeats long. After the extension process is terminated, each repeat is classified as an iMS if the above microsatellite extension process was possible, and as a perfect microsatellite if the extension was not possible. Compound microsatellites, created when adjacent seeds were composed of different motifs, are discarded.
We then identified orthologous microsatellites using the publicly available multiZ alignments of primate genomes [80]. From the identified set of orthologous microsatellites, we removed those that (1) were located within 25 bp of each other; (2) possessed at least one nucleotide of low sequence quality (namely, with PHRED score below 20); (3) had low-complexity flanking (20 bp upstream and 20 bp downstream) sequences; (4) had flanking sequence identity below 85% between any species pair; (5) differed in nucleotide sequence of the repeating motif, (6) had more than two interruptions in any species; (7) were interrupted microsatellites but differed in the sequence of the interrupting nucleotide(s) between species; (8) were interrupted microsatellite loci that differed in the context of the interruption (i.e., the repeat nucleotides immediately flanking the interruption) between species (Table S1). Our final set of microsatellite loci consisted of 30,715 perfect orthologous microsatellite loci and 46,356 orthologous microsatellites with one or two interruptions in the studied species.
The size of each iMS was measured in terms of repeat numbers and was calculated by dividing the total length of microsatellite-native sequence (i.e., all sequence other than the interrupting nucleotides) by the size of the repeating motif. Mutability values and their respective 95% confidence intervals (CI) were measured at multiple repeat numbers for microsatellites with 0, 1 and 2 interruptions separately, using methods previously implemented in [50].
We obtained variant calls (SNPs and indels) from the 1000 Genomes Phase-1 Project [36] for four population groups – Africans, Europeans, Asians and Americans. These calls were intersected with perfect microsatellites (mono-, di-, tri-, and tetra-nucleotide repeats of length ranges 8–10, 10–18, 12–24, and 16–28 bp respectively) identified from the human reference genome (UCSC build hg19) – the lower bounds of the chosen length ranges represent microsatellite thresholds and the upper bounds represent the length up to which indel calls generated from short-reads are reliable (see [25] for details). All indel and SNP variants present at an allele frequency ≥0.05 were identified separately for each population group. These variants were considered to be interruptions if they were located within a microsatellite but not at the starting/ending repeat unit. Additionally, for indels, only those indels that did not include a whole-motif insertion/deletion were considered to be interruptions. We next compared the list of iMS loci across populations to identify microsatellites interrupted in all populations and in subsets of populations. Population-specific interruptions were defined as those that are interrupted in one population, but remain perfect in the other three. We obtained coordinates of disease-associated loci [38] from the UCSC Genome Browser [81], [82], and intersected the 1000 Genomes Phase-1 Project variant calls to identify interruptions at these loci across the four population groups. Again, we used the allele frequency cut-off of 0.05 and the aforementioned filters to identify interruptions.
For interruptions present in all four population groups, the frequencies of the interruption variant alleles (p) were extracted for each of the four population groups. For each interruption, heterozygozity (H = 2pq) values were computed separately for each population group, where q = 1-p denotes the frequency of the reference allele. The average of these population heterozygozities was computed as HS. Next, the average allele frequencies for the total population (P, Q) were computed by averaging the allele frequencies (p and q) over the four populations. Next, total heterozygosity was estimated as HT = 2PQ. FST was then estimated as FST = (HT−HS)/HT [83].
Population allele frequencies for the variant iMSs as well as perfect microsatellites (those without interrupting variants) were obtained from the VCF files, and heterozygosity was estimated as 2pq, where, p = allele frequency of the variant and q = 1-p. Frequencies of iMSs and perfect microsatellites were estimated at different heterozygosity bins (ranging from 0 to 0.5, with bin-size equal to 0.02), and the distributions of these frequencies were compared against each other using two-sample bootstrap Kolmogorov-Smirnov test with 10,000 iterations from the R “Matching” package [84].
Pairwise correlation coefficient, R2 (proxy for LD), was calculated between interruption-causing indels and neighboring (located within a 1-Mb window around the indel), population-matched SNPs from the 1000 Genomes Phase-1 dataset using PLINK v1.07 (http://pngu.mgh.harvard.edu/purcell/plink/) [85]. For each indel, SNPs with the maximum R2 values were chosen for subsequent analysis. Indel-SNP pairs that showed a perfect LD (R2 = 1) were selected and intersected with a list of exon coordinates to identify exonic indel-SNP pairs in perfect LD using Galaxy. The SNPs from such perfect LD pairs were submitted to SNPnexus to obtain phenotype and disease associations.
iMS loci were intersected with exon coordinates obtained from the UCSC Genome Browser [81], [82] using Galaxy [86], [87], [88] and HUGO gene names [89] were obtained for exonic iMS. Using functions from the R package “GOstats” [90], we compared the exonic iMS-containing genes with all other genes in the genome to determine an over/underrepresentation of GO molecular functions, biochemical processes and cellular components in the selected gene set.
Purified calf thymus pol α-primase complex (pol α) was kindly supplied by Dr. Fred Perrino or the human complex was purchased from Chimerx (Madison, WI). Recombinant DNA pol β was purified as described [91]. The 4-subunit recombinant human Pol δ4 was purified as described [92] and was a generous gift of Dr. Marietta Lee. Purified full-length human pol κ and pol η were purchased from Enzymax (Lexington, KY). [GT]n and [TC]n microsatellite-containing herpes simplex virus type 1 thymidine kinase (HSV-tk) vectors have been previously described [26], [45]. Dinucleotide microsatellites were inserted in-frame between positions 111 and 112 of the HSV-tk sense strand. Additional vectors were constructed with in-frame inserts in the same position as above and the final sequences of [T]8, [A]8, [T]3 A [T]4 and [A]3 T [A]4. These sequences model the perfect and interrupted (iMS) alleles found within the APC gene (positions 3917–3924) of the Ashkenazi Jewish and non-Ashkenazi populations, respectively [19].
Linear DNA fragments and ssDNA were used to construct MluI (position 83) to StuI (position 180) gapped duplex (GD) molecules, as described [47], [93]. In vitro polymerase reactions for pol α [94], pol β [45], and pols δ, κ, and η [46] at dinucleotide microsatellite templates were previously described. For the APC gene model templates, polymerase reactions contained 1 pmol of oligonucleotide-primed ssDNA at 20 nM concentration. Reaction conditions were the same as in the references above except 20 units of Chimerx human pol α, 15 pmol of pol β, and 1–2 pmol of pol η were used. To sample reaction products for mutations, small fragments were prepared by MluI and StuI digestion and hybridized to the corresponding GD molecule as described [45]. Successful hybridization was verified by agarose gel analysis as described [52]. An aliquot of DNA from the final hybridization was used to transform E.coli strain FT334 for mutant frequency determination on VBA selective media [47]. The presence of 50 µg/mL chloramphenicol (Cm) selects for progeny of the polymerase-synthesized strand and the presence of 40 µM 5-fluoro-2′-deoxyuridine (FUdR) selects for bacteria carrying HSV-tk mutant plasmids. The observed HSV-tk mutant frequency (MF) is the number of FUdRRCmR colonies divided by the number of CmR colonies. To control for pre-existing mutations, we also determined the HSV-tk MF for each ssDNA used to construct the GD molecules. Independent mutants for DNA sequence analyses were isolated as described [47] from two polymerase reactions per template. The DNA sequence of the HSV-tk gene in the MluI-StuI region of each mutant was determined by dideoxy DNA sequence analysis of plasmid DNA as described [45].
Pol η and Pol κ produce multiple mutational events per target sequence. In order to properly compare polymerase error frequencies (Pol EFs) among polymerases, we identified those mutational events that were detectable as single mutational events, and adjusted the observed HSV-tk MF to reflect multiple errors per target. First, Pol EFs were determined by the following equation: Pol EF = (Observed MF) − (ssDNA background MF) − (Outside target MF), where outside target MF is the frequency of errors occurring outside the gap target. Next, each mutational event was scored as detectable or undetectable. All frameshifts and those base substitutions that caused an amino acid change or a stop codon within coding sequences were considered detectable. Base substitutions within microsatellite sequences were only considered detectable when a stop codon was produced. Only detectable events were used for determining Pol EFest. Each mutational event was also scored as tandem or nontandem. Tandem events were those adjacent to one another, whereas nontandem were errors >1 nt apart. Pol EFs were then corrected for the existence of multiple nontandem mutations as described [46]. The Pol EFest obtained is the overall Pol EFest and includes mutational events within the microsatellite sequence and within the adjacent HSV-tk coding sequence (see Table S5 and accompanying footnotes). The Pol EFest of a specific type of mutational event was calculated from the proportion of the specific mutational event (among the total analyzed) multiplied by Pol EFest. For analyses presented herein, we further subdivided the microsatellite Pol EFest into unit-based indel Pol EFest or interruption Pol EFest. A unit-based indel is an error that occurs when an entire microsatellite unit or units are inserted or deleted (i.e., [GT]10→[GT]9). An interruption is an indel or base substitution that disrupts the repetitive nature of the microsatellite sequence (i.e., [GT]10→[GT]5T[GT]5).
|
10.1371/journal.pcbi.1000662 | Network-Based Elucidation of Human Disease Similarities Reveals Common Functional Modules Enriched for Pluripotent Drug Targets | Current work in elucidating relationships between diseases has largely been based on pre-existing knowledge of disease genes. Consequently, these studies are limited in their discovery of new and unknown disease relationships. We present the first quantitative framework to compare and contrast diseases by an integrated analysis of disease-related mRNA expression data and the human protein interaction network. We identified 4,620 functional modules in the human protein network and provided a quantitative metric to record their responses in 54 diseases leading to 138 significant similarities between diseases. Fourteen of the significant disease correlations also shared common drugs, supporting the hypothesis that similar diseases can be treated by the same drugs, allowing us to make predictions for new uses of existing drugs. Finally, we also identified 59 modules that were dysregulated in at least half of the diseases, representing a common disease-state “signature”. These modules were significantly enriched for genes that are known to be drug targets. Interestingly, drugs known to target these genes/proteins are already known to treat significantly more diseases than drugs targeting other genes/proteins, highlighting the importance of these core modules as prime therapeutic opportunities.
| Many human diseases are related to each other through shared causes or even shared pathology. Knowledge of these relationships has long been exploited to treat similar diseases with the same therapies. However, most of the traditional approaches to discover these relationships have depended on subjective measures, such as similarity in symptoms, or incomplete knowledge, such as genes with mutations. Here we present the first approach integrating high-throughput datasets such as mRNA expression and large-scale protein-protein interaction networks to discover human disease relationships in a systematic and quantitative way. We discover 138 significant pathological similarities between 54 human diseases ranging from lung cancer, schizophrenia, and malaria. We also discovered a set of common pathways and processes within the cell that are dysregulated in at least half of the diseases. We infer that these processes correspond to a common response of the human system to a disease state. Interestingly, we find that many of the proteins in these pathways are already known to be targets of existing drugs. In fact, the drugs corresponding to these proteins are known to treat significantly more diseases than expected by chance highlighting the importance of these common molecular pathological pathways as prime therapeutic opportunities.
| Our understanding of the human disease state is incomplete without the knowledge of how various diseases relate to each other. Relationships between diseases have been used to gain insights into the etiology and pathogenesis of similar diseases [1]. Study of disease similarities has also led to the discovery of new causal genes for diseases [2],[3]. Moreover, similarities between biological concepts such as genes have been used successfully in gene function prediction [4]. However, most of the early work on finding disease-similarity has been limited to studying the clinical phenotypes of the diseases. For instance, similarities in disease symptoms and pathological results have been used to ascertain similarities between Alzheimer's disease and vascular dementia [1]. These methods are not quantitative and cannot be used to compare the relative similarities between diseases. More recently, scientists have been able to explore the genetic similarity between diseases because of the availability of large-scale knowledge-bases such as the Online Mendelian Inheritance in Man (OMIM) [5]. In 2007, Goh and colleagues created the first “Diseasome”, a network of human diseases [6]. This network consisted of human diseases/disorders as nodes and two diseases were joined by a link if they shared known disease genes (data obtained from OMIM). Van Driel et al. [7] inferred disease-disease associations by an automated text mining of OMIM descriptions. Liu et al. [8] mined for disease etiologies from the Medical Subject Headings (MESH) [9] vocabulary and used it to reveal similarities between diseases. Although the above studies provided comprehensive views of disease interrelationships, they were mainly studying monogenic disorders and generally ignored the effect of the environment on these and other, more complex, diseases. They also relied heavily on information that is already known, such as known disease genes or known pathways. As a result, they were limited in their ability to uncover hitherto unknown relations between diseases.
Advances in high-throughput molecular assay technologies, accompanied by declining per-sample costs, have given rise to numerous public repositories of biomolecular data such as mRNA expression profiles and protein interaction networks. In particular, the availability of these datasets for many different diseases presents a ripe opportunity to use data-driven approaches to advance our current knowledge of disease relationships in a systematic way. As a matter of fact, very recently, Hu and Agarwal [10] presented an approach to determine disease relationships using only gene expression data. In order to obtain the disease correlations, the authors excluded genes which don't change meaningfully using an arbitrary threshold. They also did not take advantage of the plethora of protein interaction data available for the human system. Protein networks represent the physical processes taking place inside a cell and are essential to acquire a complete understanding of any biological condition such as disease.
Therefore, just as sequencing of genomes has enabled the reorganization of many species and provided quantitative metrics to appreciate their relationships, we believe an integrated approach combining both mRNA expression and protein interaction data will provide us a quantitative way to assess the correlation between diseases. Here, we present the first such systematic and integrated approach to explore the architecture of human diseases. In particular, we identified 4,620 functional modules analogous to important complexes and pathways in the human protein network and recorded how they varied in each of the 54 diseases using the mRNA expression data. This process provided a quantitative measure to describe the overall response of the human system to a given disease. Subsequently, we used these measures to identify 138 significant associations between diseases. We also discovered functional modules that are common to at least half of the diseases representing a common “disease-state” signature. These common disease-state modules were not only significantly enriched for genes that were known drugs targets, but their corresponding drugs were known to treat significantly more diseases than expected by chance highlighting their importance as therapeutic opportunities.
Protein complexes and pathways are accountable for most processes in the cell. Accordingly, we can gauge the response of a cell system to a certain perturbation (such as disease) by the measuring changes in the expression levels of various functional modules of the system. To this end, we first generated a catalog of 4,620 functional modules by querying the large-scale human protein interaction network (see Methods). We then collected the mRNA expression arrays associated with each disease from the Gene Expression Omnibus (GEO) [11]. After several rounds of filtering the gene expression data for accuracy, reliability, and experimental context, we had microarrays representing 54 human diseases (see Methods and Table S1). Next, we combined the gene expression data and the 4,620 functional modules to generate a Module Response Score (MRS) for each module in each disease-state representing its activity level (see Methods). Specifically, positive MRS values correspond to modules that are up-regulated and negative MRS values identify modules that are down-regulated in the disease-state as compared to the control (healthy). Figure 1 gives an overview of the process to compute the MRS values for a given disease. In the end, we generated a matrix containing the MRS values for each module in each of the 54 diseases considered in this study. The relationships between different diseases were then ascertained by the Partial Spearman correlation coefficient of their MRS values (see Methods and Figure S1). Specifically, we calculated the Spearman correlation between two diseases conditioned on the responses of the functional modules in their respective control samples. The use of the Partial Spearman correlation coefficient instead of the generic Spearman correlation coefficient not only provided a quantitative metric to assess disease similarity but also explicitly factored out the possible dependencies between different gene-expression experiments due to their underlying tissue or cell types.
Figure 2(A) is the hierarchical clustering of diseases based on the correlations generated above. To assign significance to these associations, we randomized the gene to module assignments as well as the control and disease labels 100 times to generate a background distribution of disease correlations (see Methods). We then selected only those disease correlations that passed the p-value threshold of 0.01 (FDR = 10.37%) resulting in 138 significant disease-disease similarity relationships. Immediately, we see that many expected disease associations such as the brain disorders like Alzheimer's disease, Bipolar disorder and Schizophrenia are pooled together in one sub-branch. We also see many novel and hitherto unknown significant correlations such as the similarity between uterine leiomyoma and lung cancer. We also created a network representation to display all the 138 significant disease correlations (Figure 2(B)). In this network, the nodes are diseases, while the thickness of the edges between two diseases represents their strength of correlation. This abstraction allows us to pick additional significant disease associations that were missing in the hierarchical clustering. For example, Crohn's disease and Malaria share a significant disease correlation. A listing of all the significant disease correlations is provided in Table S2.
Although the 54 diseases considered in this study cover many categories of diseases ranging from cancers to cardiomyopathies, some categories of diseases such as cancer are over-represented as opposed to others such as infectious diseases. Ideally, we would like to explicitly correct for this bias by down-weighing over-represented classes. However, the principle behind organizing diseases into categories such as cancers, infectious diseases and others is not the same. For instance, diseases are classified as cancers if their underlying pathology consists of a group of cells that show uncontrolled growth, invasion of nearby cells and metastasis. On the other hand, infectious diseases relates to diseases which are caused by pathogens and have the potential to spread from person to person. Lack of a common organization scheme prevents us from explicitly correcting for the observed over-representation. Moreover, there is considerable heterogeneity even among diseases of the same category. For instance, the category of cancers covers a wide variety of diseases affecting many different cell types and having many different biological causes ranging from mutations caused by chemical carcinogens to bacterial and viral infection. This heterogeneity is seen even at the transcriptional level [12]. We also have observed this heterogeneity in the results of our study as all the 17 cancers considered in our analysis did not cluster together (Figure 2(A)). By combining both mRNA expression and protein interaction data, we are providing one of the first ways to compare and classify diseases systematically. The common organizing principle here is the molecular pathology of a given disease.
At the outset, we explored the genetic basis of the diseases in our study to explain and validate the observed disease correlations. Specifically, we aimed to test the hypothesis that diseases which are significantly associated through the MRS-based correlation coefficient also significantly shared disease genes. For this purpose, we collected a list of genes known to be associated with diseases, hereinafter as the Disease Gene List (see Methods). We found known gene variants associated with only 31 of the 54 diseases in our study resulting in an overall total of 465 possible pair-wise disease comparisons. A pair of diseases was considered to significantly share disease genes only if the Hypergeometric p-value of the overlap was less than 0.01. Eighty-two of the overall 465 comparisons significantly shared disease genes. On the other hand, only 73 of the 465 disease pairs were significantly associated using the MRS-based correlation coefficient. This gives rise to a contingency table as shown in Table 1 with a one-sided Fisher's Exact Test p-value of 0.033. It suggests that the genetic similarity between diseases significantly contributes to the molecular pathological disease similarity observed in this study. Lack of a strong p-value might be explained by the fact that the number of known disease genes are much higher for well-studied diseases like Schizophrenia (345 genes) as opposed to less well-studied diseases like Mixed hyperlipidemia (4 genes). Mapping of genes to diseases was also hindered due to fact that we used a very strict vocabulary to define diseases (see Methods). Finally, this result might also allude to the role of environment in disease causation and similarity. A few of the significant disease correlations which also significantly shared disease genes is provided in Table 2 and the complete list is provided in Table S3.
In order to further understand the biology behind the observed disease correlations, we examined some of their underlying functional modules. First, we analyzed the sub-branch of brain disorders, Alzheimer's disease (ALZ), Bipolar disorder (BIP), Schizophrenia (SCHZ), and Glioblastoma (GLIO), in the hierarchical representation of the disease correlations (Figure 2(A)) in more detail. Figure 3(A.i) corresponds to the synaptic vesicle and was one of most down-regulated modules in all four diseases (second lowest average MRS value). This module is a secretory organelle that stores neurotransmitters and releases them into the synapse. Loss of synaptic functions and more specifically, decreased expression of synaptic vesicle proteins such as SNAP-25 is one of the main effects of ALZ [13],[14]. Decreased synaptic function has also been observed for both BIP and SCHZ [15],[16]. In particular, the levels of protein SNAP-25 was shown to be reduced in both BIP and SCHZ [17]. The function of this module in GLIO is still to be explored. Uterine leiomyomas (UTL) are benign tumors affecting the uterus. As shown in Figure 2(A), UTL shares a strong correlation with lung cancers. Figure 3(A.ii) corresponds to the DNA repair pathway which had the highest average MRS value for the three diseases. Polymorphisms in the genes involved in the DNA repair pathway such as PCNA, POLB have been associated with increased risk of lung cancer [18]. Moreover, the Arg399Glu allele of the XRCC1 gene has been shown to be a risk factor for lung adenocarcinoma [19] and lung squamous cell carcinoma [20]. Surprisingly, the same Arg399Glu polymorphism in the XRCC1 gene has also been associated with an increased risk of UTLs [21] giving causal genetic evidence for the correlation we observed between the diseases using microarray-based molecular pathological measurements.
Knowledge of a comprehensive disease-similarity tree (network) based on molecular data could possibly be used in finding new uses for existing drugs. Similar diseases share similar molecular phenotypes and could potentially be treated by similar drugs. To explore this avenue, we collected a list of drugs, their corresponding target genes and the diseases they are known to treat (US FDA approved indications) or off-label uses. This information was obtained from the RxNorm from National Library of Medicine [22], DrugBank [23], National Drug File Reference Terminology (ND-FRT) [24] and MicroMedex [25]. Overall, 17 of the 138 significant disease correlations shared at least one drug in common and 14 of them had a significant Hypergeometric p-value less than 0.01 (Table 3, Table S4). For instance, we found that the FDA approved drug Flouroucil, used to treat Actinic keratosis, has been shown to have positive indications for treating Malignant tumor of the colon [25]. Similarly, the drug Doxorubicin is FDA approved to treat both Urothelial carcinoma and Acute myeloid leukemia [25]. This number is a conservative estimate as the list of drugs used here is incomplete. Moreover, we used a very specific vocabulary to define diseases (see Methods) and accordingly mapped drugs to them. For instance, we found many drugs treating lung cancer; however in many cases, our combined knowledge base doesn't specify whether the cancer was an adenocarcinoma or a squamous cell carcinoma. In those cases, we excluded the drug from our consideration. A caveat to this approach is that drugs can be shared between diseases mainly because the corresponding diseases belong to the same category. For instance, drugs can be shared between two cancers etc. As a result, it is difficult to differentiate whether two diseases shared drugs due to the similarity in their molecular pathology or due to their underlying disease type. Moreover, the chemical similarity between drugs can also affect the reported p-values.
Another consequence of elucidating and quantifying the response of the cell system to a disease is that we can use this methodology to find modules that are generally dysregulated (activated or repressed) in the disease-state. In other words, we used the MRS values to characterize a common “signature” across disease-states. In order to generate the set of modules that are commonly dysregulated in the 54 diseases considered in this study, we used a two-fold approach. Firstly, a module was selected if the median of its absolute MRS values across all diseases was significantly higher than expected at random. We generated a random background distribution of median scores by shuffling the gene to module assignments (see Methods). Overall, at a p-value of 0.01 and associated FDR of 16.15%, we selected 286 modules. We then filtered the above set of 286 modules to only include those modules which were significantly differentially expressed in many diseases. A module was determined to be significantly differentially expressed in a given disease if the absolute value of its MRS was above 1.5 (p-value = 0.028). Finally, we selected 59 modules that were significantly differentially expressed in 20 or more diseases as the common disease state signature. These modules were not only dysregulated in at least half of the diseases each but were also significantly differentially expressed in more than 20 diseases. Moreover, these 59 modules taken together were dysregulated in 45 of the 54 diseases in our study. Figure S2 shows the combined illustration of all the 59 modules. They were mainly enriched for the functions of immune system response (p-value = 6E-70) and DNA repair (p-value = 4.1E-30). A representative sample of 7 modules is shown in Figure 3(B.iii–vii).
We investigated the 59 modules further by searching for known drug target genes/proteins. We obtained the list of drugs and their corresponding targets from the DrugBank database [23]. Overall, 70 genes/proteins within the 59 signature pathways were identified as targets of known drugs giving a Hypergeometric p-value of 1.8E-11. Thus, the set of the signature modules was significantly enriched for drug target genes compared to that expected by chance. We then predicted that other genes/proteins in these modules would also serve as prime candidates for designing new drugs. Most existing drug target genes usually fall into a comparatively small set of gene families such as G protein coupled receptors, serine proteases etc [26]. Hence, new drug targets can be found by exploring other members of the protein families of the existing drug targets. We explored the 59 signature modules for genes which belonged to the same protein families as known drug target genes. For that purpose, we obtained a list of genes and their corresponding families and sub-families from the PANTHER database [27]. Overall, we found 241 genes among a total of 450 genes in the signature modules sharing the same protein families as the known drug target genes compared to a total of only 3,520 such genes in the whole human PPI giving a Hypergeometric p-value of 1.47E-12. Therefore, the 59 signature modules were also significantly enriched for druggable genes. Further, we also counted the number of distinct diseases that are known to be treated by the drugs corresponding to each of the 70 known drug targets. We observed that drugs targeting these 70 genes are known to treat an average of 65 diseases each compared to an average of ∼42 diseases for all known drug targets (p-value = 0.02). These results provide evidence that the genes in the signature modules are more likely to be good drug targets and drugs that target these proteins are more likely to treat many diseases. Yildirim et al. [28] showed that most drugs seemed to be palliative and only cured the symptoms of the diseases rather than the diseases themselves. Therefore, the enrichment for drug target genes which treat many diseases might be due to the shared symptoms of the diseases.
In summary, this study demonstrates the value of an integrated approach in revealing disease relationships and the resultant opportunities for therapeutic applications. Looking forward, we aim to incorporate more gene expression data from GEO and other similar repositories, and expand the set of diseases in our disease-similarity network.
The gene expression data used in this analysis was obtained from the NCBI Gene Expression Omnibus (GEO) [11]. In this study, we restricted to using only those microarrays that were curated and reported in the GEO Datasets (or GDS). We selected for microarrays that were assigned to human disease conditions. These assignments were made by the method explained in Butte et al. [29]. Briefly, the experimental context of a collection of microarrays from GEO (or GEO Series, GSE) can be obtained from the Medical Subject Headings (MeSH) [9] terms associated with the records of corresponding publications in PUBMED. Subsequently, the MeSH terms were connected to disease concepts using the Unified Medical Language System (UMLS) [30]. The GDS curation provided more details such as the tissue or biological substance from which the samples were derived. We only included those GSEs in which both disease as well as their corresponding control condition was measured in the same tissue (cell type) in the same experiment, using a previously described method [31]. We also manually selected for GSEs in which the control was a healthy sample. Further, we removed all GSEs that included time-series data to avoid complications arising due to temporal changes in gene expression. For consistency, we also restricted the GSEs to only those arrays which used Affymetrix Gene Chip Human Genome U133 Array Set HG-U133A or U95 Version 2 platforms, which are among the most commonly used platforms, mapping to current gene identifiers as previously described [32]. As both of these platforms have shared probe-sets, the bias of the platform used on the overall analysis would be reduced considerably. We subsequently selected GSEs that had at least two disease samples and two control samples. GEO contains some experiments (GSEs) that have gene expression measurements for more than one disease but share the same control measurements. Such measurements might induce correlations between their component diseases, which are not necessarily biological. Thus, to avoid bias, in all such cases, we included only one representative disease for each set of control samples in contrast to Hu et al. [10]. This entire process yielded 54 diseases for our final analysis.
The protein-protein interaction (PPI) data for human was obtained from the Human Protein Reference Database (HPRD) [33]. This database contains PPI obtained from the two high-throughput yeast two-hybrid experiments [34],[35] as well as through literature curation. Further, HPRD contains the maximum number of PPI of any of publicly available literature-derived databases for human PPI [36]. We filtered the PPI for human proteins that had a corresponding Entrez Gene ID, yielding 34,998 unique protein interactions spanning 9303 proteins in human. Previously, Sharan et al. [37] presented the PathBLAST family of network alignment tools. Briefly, these methods help identify conserved modules between protein networks of two (or more) species. Suthram et al. [38] also used it effectively to identify dense subnetworks corresponding to functional modules within a protein network of a single species. Applying the same approach here, we identified 4,620 functional modules in the human PPI network.
First, we normalized the gene expression data in each microarray sample (disease state or control) using the Z-score transformation. This transformation allows for the direct comparison of gene expression values across various microarray samples and diseases. Next, we computed the activity level of a gene i in disease k as the t-test statistic (Sik) of its Z-transformed score between the disease and the control samples for each disease. In cases where there was more than one experiment (or GEO Series) for a given disease, we employed a meta-analysis technique using linear regression to obtain a combined t-test statistic. This approach takes into account the variations between different experiments in the calculation of the gene activity score (Sik) (see section below). Finally, the module response score (Mik) for each module i in a disease k is assigned to be the mean of the gene activity scores (Sik) of its component genes. In the end, we obtained a vector of module response scores (Mik) for each disease.
The t-test statistic between two conditions can be represented using linear regression. For instance, let Yi and Xi be gene expression values and disease state (disease has a value of 1 and control a value of 0), respectively. Then, we have a linear regression model as follows:where and are the parameters of the model. The t-test statistic when estimating the value of is the same as the standard t-test statistic between disease and control states. The advantage of the linear regression model is that we can add more terms to the model to account for other sources of variation such as the experiment number. In the present work, we expanded on the above model as follows:where n is the number of different experiments for a given disease and, is an indicator variable which is 1 if the i'th gene expression measurement is from the experiment number k. Again, the and are the parameters of the model which need to be estimated. The addition of the new terms allows for explicitly accounting for the effect of the experiment on the gene expression value. This approach is similar to a mixed effects model for adjusting for the within-experiment dependencies, but is more aggressive in removing such effects. Consequently, the t-test statistic in the estimation of will now be a combined metric for the different studies.
The partial correlation coefficient gives the correlation between two variables, say x and y, keeping a third variable, z, constant. This method tries to measure the similarity between x and y, over and above that caused by their common dependency on z. The partial correlation can be calculated as follows:
The above formula can be expanded to condition on two variables as follows:In this study, we calculated the Partial Spearman correlation between two diseases conditioned on the responses of the functional modules in their respective control samples. The response of a functional module in the control samples for a given disease was calculated as the mean of the z-transformed scores of its component genes. As a result, the Partial Spearman correlation coefficient provided a quantitative metric to assess disease similarity and also explicitly factored out the possible dependencies between different gene-expression experiments due to their underlying tissue or cell types [39]. Our approach is consistent with the findings by Dudley et al. [40] that the disease signal in the GEO datasets is stronger than the tissue signal and hence, implying that the observed disease correlations reflect true biology. We used the R script provided by Kim et al. [41] for calculating the Partial Spearman correlation between two diseases.
To assign significance to the observed disease correlations, we created a background distribution of disease correlations expected at random. First, we randomized the gene to module assignments. We envisioned the gene to module assignments as a bi-partite graph (Figure S3) where there exists a link between a gene and a module if that gene is a member of that module. We then randomized the graph by randomly swapping links. This process preserved the number of modules, the number of genes assigned to a module as well as the number of modules a given gene belongs to. Next, we also shuffled the disease and the control sample labels. We then calculated the MRS values for the modules using the randomized data and computed the corresponding disease correlations. Finally, we repeated the whole process 100 times to create a background distribution of disease correlations.
We built a comprehensive disease-associated gene database, referred to as the Disease Gene List, by collecting genes known to be associated with various diseases from literature curation and large databases. In particular, we first curated 37,953 disease Single Nucleotide Polymorphism (SNP) associations from 2,679 papers, mapping 10,167 specific SNPs from the SNP Database (dbSNP) to 748 diseases and phenotypes. We then annotated each SNP with its corresponding gene(s) using dbSNP (Chen and Butte, unpublished data). Next, we extracted genes that are significantly associated with diseases in Genetic Association Database (GAD) [42]. These consisted of associations that were reported to be positive at least once. We also collected genes that are associated with disorders in the Online Mendelian Inheritance in Man (OMIM) [5]. Lastly, we retrieved genes that are associated with diseases in the professional version of Human Gene Mutation Database (HGMD) [43]. Finally, we combined disease genes obtained from the above four different sources by relating them to Entrez gene IDs and removing outdated Gene IDs using AILUN [32].
The module figures in the paper were drawn using the Cytoscape software [44].
|
10.1371/journal.pgen.1008248 | A secretion-enhancing cis regulatory targeting element (SECReTE) involved in mRNA localization and protein synthesis | The localization of mRNAs encoding secreted/membrane proteins (mSMPs) to the endoplasmic reticulum (ER) likely facilitates the co-translational translocation of secreted proteins. However, studies have shown that mSMP recruitment to the ER in eukaryotes can occur in a manner that is independent of the ribosome, translational control, and the signal recognition particle, although the mechanism remains largely unknown. Here, we identify a cis-acting RNA sequence motif that enhances mSMP localization to the ER and appears to increase mRNA stability, and both the synthesis and secretion of secretome proteins. Termed SECReTE, for secretion-enhancing cis regulatory targeting element, this motif is enriched in mRNAs encoding secretome proteins translated on the ER in eukaryotes and on the inner membrane of prokaryotes. SECReTE consists of ≥10 nucleotide triplet repeats enriched with pyrimidine (C/U) every third base (i.e. NNY, where N = any nucleotide, Y = pyrimidine) and can be present in the untranslated as well as the coding regions of the mRNA. Synonymous mutations that elevate the SECReTE count in a given mRNA (e.g. SUC2, HSP150, and CCW12) lead to an increase in protein secretion in yeast, while a reduction in count led to less secretion and physiological defects. Moreover, the addition of SECReTE to the 3’UTR of an mRNA for an exogenously expressed protein (e.g. GFP) led to its increased secretion from yeast cells. Thus, SECReTE constitutes a novel RNA motif that facilitates ER-localized mRNA translation and protein secretion.
| Proteins destined for secretion from the cell, including soluble secreted and membrane proteins (SMPs), are translocated into the endoplasmic reticulum (ER) either directly upon translation on the ER surface or post-translationally, as in the case of type II membrane proteins. Interestingly, several studies have demonstrated that mRNAs encoding SMPs (mSMPs) are also enriched on the ER, yet how they target this organelle is less clear. The signal recognition particle (SRP), which recognizes N-terminal hydrophobic signals of nascent polypeptides and targets them to an ER-localized receptor, was proposed to mediate the co-translational ER targeting of mRNA. However, more recent studies show that SRP inactivation, as well as the inhibition of translation, do not prevent targeting. Thus, how mSMPs reach the ER and whether the process is translation-independent remain open. Here we identify a cis-acting sequence element in mSMPs that appears to facilitate mRNA stability and localization to the ER and, more importantly, enhances protein secretion. This motif, entitled “SECReTE” (secretion-enhancing cis regulatory targeting element) is enriched in nearly all mSMPs in eukaryotes and its addition or removal from mRNAs results in either enhanced or reduced protein secretion, respectively. Thus, SECReTE is a RNA sequence motif that regulates protein translation and secretion.
| mRNA targeting and localized translation is an important mechanism that provides spatial and temporal control of protein synthesis. The delivery of mRNA to specific subcellular compartments has a major role in the establishment of polarity in various organisms and cell types, and was shown to be crucial for the proper function of the cell [1,2]. Interestingly, the localization of mRNA is often governed by cis-acting elements (“zipcodes”) embedded within the mRNA sequence [1,2]. RNA-binding proteins (RBPs) recognize such sequences and act together with molecular motors to direct the mRNA to its final destination.
The endoplasmic reticulum (ER) is the site of synthesis of secreted and membrane (SMP; secretome) proteins. Moreover, mRNAs encoding for SMPs (mSMPs) are thought to localize to the ER membrane by a distinct translation-dependent mechanism, termed the signal recognition particle (SRP) pathway [3–5]. According to this model, protein translation begins in the cytoplasm and when SMPs undergo translation, a signal peptide present at their amino terminus emerges from the exit channel of translating ribosome and is recognized by the SRP. The SRP then is recruited to its receptor on the ER membrane and translocation of ribosome-mRNA-nascent polypeptide chain complex from the cytoplasm to the ER occurs. There, translating ribosomes interact with the translocon to enable co-translational protein translocation and mRNA anchoring [6,7]. Thus, the SRP model describes mSMPs as components with no active role in the ER translocation process.
However, multiple lines of evidence suggest that there are additional pathways for the delivery of mRNAs to the ER [8,9]. First, attenuation of the SRP pathway did not result in lethality of yeast [10] and mammalian cells [11], and did not have a significant effect upon membrane protein synthesis and global mRNA distribution between the cytoplasm and the ER [11]. Second, genome-wide analyses of the distribution of mSMPs between cytosolic polysomes and ER-bound polysomes demonstrated a significant overlap in the composition of the mRNA in the two fractions and also showed that cytosolic protein-encoding mRNAs are broadly represented on the ER [12–16]. This means that mRNAs lacking an encoded signal sequence or transmembrane domain can also localize to the ER. In agreement with these findings, removal of the signal sequence and the inhibition of translation did not disrupt mSMP localization to the ER [14,17,18]. Third, subsets of secretome proteins are known to localize to the ER in an SRP-independent pathway [19,20]. These proteins are thought to translocate into the ER after translation in the cytosol [21]. In a study that utilized a technique for a specific pull-down of ER-bound ribosomes [22], it was found that there is no significant difference in the enrichment of mRNAs encoding SRP-dependent proteins in comparison to mRNAs encoding SRP-independent proteins on ER membranes. In addition, a subset of ribosomes managed to reach the ER before the emergence of the signal sequence. A possible explanation for these observations could be that mRNAs reach the ER before the ribosomes in an SRP-independent mechanism. If mRNA targeting to the ER does not begin until signal peptide emergence, membrane-bound ribosomes should not be translating portions of the transcript upstream of the signal peptide. However, this was not the case, as translating membrane-bound ribosomes were found to be evenly distributed across entire transcripts in another study [23]. While it is possible that pre-signal sequence-encoding transcripts could arise from translating polysomes, altogether the various findings strongly suggest a scenario whereby mRNAs can localize to the ER prior to translation initiation. Lastly, a recent study demonstrated that conditional SRP depletion from yeast does not necessarily block the co-translational ER targeting of mRNA, especially of those transcripts encoding predicted SRP-independent proteins [24]. Thus, mRNA targeting to the ER likely involves different and, perhaps, multiple paths.
Although it has been difficult to identify cis-elements within mRNA that direct it to the ER [25–28], specific sequence characteristics of mSMPs have been identified. For example, sequence analysis of the region encoding the signal sequence revealed a low usage of adenine to create no-A stretches within this sequence [29]. Additionally, mRNAs encoding membrane proteins have a high degree of uracil enrichment, as well as pyrimidine usage, in comparison to mRNAs encoding cytosolic proteins [8,27,30,31]. These findings raise the possibility that the motif resides in a general, more diffuse, fashion within the nucleotide composition of the mRNA molecule.
By examining the sequences of transmembrane domain (TMD)-containing regions in mRNAs, we have now identified high content stretches of pyrimidine (C and U) repeats every third base (NNY, N–any nucleotide, Y–pyrimidine) that can be ≥10 nucleotide triplets in length. Analysis of the transcriptomes of several eukaryotic organisms (e.g. S. cerevisiae, S. pombe, and H. sapiens), revealed that this sequence pattern is significantly over-represented in mRNAs encoding for secretome proteins, that typically localize to the ER. The location of the motif is not restricted to the coding region but can be present in the untranslated regions (UTRs). Although we originally found the motif by analyzing the sequences of TMDs in secreted membrane proteins, in fact it is enriched at a higher level in mRNAs encoding secreted proteins that lack TMDs. We utilized both computational and experimental tools to establish the existence and significance of this motif. Computational analysis verified that mSMPs are the group most enriched with the motif, while synonymous mutations that either elevated or decreased motif strength (i.e. number of consecutive pyrimidine repeats) in mRNAs encoding yeast invertase, SUC2, as well as cell wall proteins, CCW12 and HSP150, enhanced or reduced protein synthesis and secretion, respectively. This motif, which appears to facilitate mRNA stability, localization and translation at the ER, we have named the secretion-enhancing cis regulatory targeting element (SECReTE). Importantly, we show that SECReTE is enriched in secretome-encoding transcripts in all organisms examined, from prokaryotes to both lower and higher eukaryotes. This suggests that SECReTE may have a conserved role in the translational control of mRNAs either as a targeting motif or in other processes such as translation efficiency, mRNA processing (i.e., polyadenylation, capping, splicing), mRNA decay, and secondary structure, etc. We propose that SECReTE is important, not only to understand how mRNAs may reach the ER in eukaryotes, but may have practical applications in the field of biotechnology.
RNA codons for the major hydrophobic residues (e.g. valine, isoleucine, leucine, methionine, and phenylalanine) are enriched in uracil in their second position [31], although others like proline, alanine, serine and threonine, which can also be found in TMDs, contain cytosine at this position. We speculated, therefore, that triplet repeats of the form, NYN (where N = any nucleotide and Y = pyrimidine—U or C), might be common to proteins destined for translation at the ER (i.e. secreted and membrane proteins). We further speculated that uninterrupted repeats of this nature might be indicative of an RNA localization motif that could, potentially, exist in any frame. Thus, we examined mRNAs encoding yeast secretome proteins for the presence of consecutive pyrimidine repeats every third nucleotide (i.e. YNN, NYN, or NNY) in both the coding and UTR regions, using computational analysis.
We first determined how many such repetitive nucleotide triplets might best differentiate mSMPs from non-mSMPs (i.e. other nuclear-encoded mRNAs). For that, the number of consecutive triplet repeats along an mRNA transcript was scored according to a defined threshold (e.g. 5, 7, 10, 12, and 15 repeats). For a random motif we expected to see a linear correlation between the probability of its appearance(s) in a gene with gene length, as shown in Fig 1A. We examined SECReTE lengths between 5 and 15 (e.g. nucleotide triplet lengths of 5,7,10,12, and 15) and indeed observed a direct correlation between SECReTE count and gene length for SECReTE5 and SECReTE7 (Fig 1A). However, the dependency on gene length becomes significantly weakened for SECReTE10 and above, where motif occurrence at ≥10 triplets correlates poorly with gene length (Fig 1A). This implies the presence of ≥10 consecutive repeats is not a random phenomenon and may be important.
If SECReTE repeats ≥10 (e.g. termed here as “SECReTE10”) play a role in protein secretion, we expect them to be more abundant in mRNAs encoding secretome proteins, as defined according to Ast et al. [19]. To test this possibility, we divided the complete yeast genome into two gene sets: secretome and non-secretome, and calculated the fraction of transcripts that contain at least one instance of SECReTE per gene in each gene set. We found transcripts coding for secretome proteins are enriched with SECReTE motifs of length >7 (Fig 1B), in comparison to transcripts encoding non-secretome proteins. To test the motif length that gives the most significant separation between secretome and non-secretome transcripts, we evaluated the different thresholds for their ability to classify mSMPs using receiver operator characteristics (ROC) analysis. Bona fide secretome protein-encoding transcripts were used as a true positive set and non-secretome protein-encoding transcripts were defined as true negatives. As seen (Fig 1C), motifs with ≥10 repeats (i.e. SECReTE10 and above) maximally differentiated (in terms of ROC area under the curve) secretome transcripts from non-secretome transcripts. As the occurrence SECReTE10 did not show a dependency upon gene length and gave the most significant separation between secretome and non-secretome transcripts, we used it as the threshold by which to define motif presence in subsequent analyses. Previous studies have used high throughput analyses to quantify the level of enrichment of transcripts on yeast ER-bound ribosomes and ER membranes [22,23]. By comparing the cumulative distribution of the ER enrichment value of SECReTE10-containing transcripts to transcripts lacking SECReTE10, we could verify that a higher fraction of SECReTE10-containing transcripts is indeed enriched on ER-bound ribosomes (S1A and S1B Fig) and ER membranes (S1C Fig). In contrast, SECReTE10-containing transcripts are not enriched on mitochondrial ribosomes, in comparison to transcripts lacking SECReTE10 (S1D Fig).
To ascertain whether SECReTE enrichment in mSMPs is not merely due to its presence in encoded TMDs, we determined at which position of the nucleotide triplet in SECReTE10 elements is the pyrimidine (Y) is located: i.e. first (YNN); second (NYN); or third (NNY). We calculated SECReTE10 abundance separately for each position using only the coding sequences (i.e. from start codon to the stop codon; CDS) and without the UTRs. While the signal is present in the second position (Fig 2A; NYN), as expected, it is also abundant in the third position of the codon (Fig 2A; NNY). In contrast, the SECReTE10 element is poorly represented in the first position, YNN (Fig 2A). Importantly, the abundance of NNY-based motifs suggests that the TMD is probably not the only element that confers SECReTE enrichment in the coding sequences of yeast mSMPs. A list of all SECReTE motifs (i.e. ≥10 repeats, either NYN- or NNY-based) present in the coding sequences of the yeast genome and their position within the mRNA is listed in S3 Table.
Next, we checked for the presence of SECReTE10 in mRNAs coding for TMD-containing proteins and soluble secreted proteins separately. As expected, more transcripts encoding TMD-containing secretome proteins contain SECReTE10 (i.e. ≥1 SECReTE10) in the second position (NYN) than transcripts that encode soluble secreted proteins (Fig 2B). However, the fraction of transcripts coding for soluble secreted proteins that contain at least one SECReTE10 in the third position (NNY) is even higher. This provides compelling evidence for SECReTE10 enrichment in transcripts that is independent of encoded TMD regions. Correspondingly, when we removed the TMD sequences from mRNAs encoding membrane proteins, we found that these transcripts were no longer enriched with the NYN-based form of SECReTE10 (Fig 2C). In contrast, SECReTE10 remained abundant at the third position, NNY, after TMD removal (Fig 2C). Thus, TMD sequences contribute to NYN-based SECReTE abundance. Finally, we note that the removal of TMD sequences from genes encoding secretome and non-secretome proteins did not alter the overall enrichment of SECReTE in those secretome messages versus non-secretome messages, nor did it change the threshold for differentiating between these groups (S1E Fig).
As contiguous stretches of codons for hydrophobic amino acids (i.e. TMDs) foster SECReTE abundance, we also examined whether the removal of signal sequence coding regions (SSCRs), which encode signal peptides, from secretome genes had an effect upon the computational analysis of the SECReTE score (Fig 2D). However, the results indicate that no significant change in the overall fraction of secretome genes bearing SECReTE is incurred upon SSCR removal and this is not altered by TMD presence or motif position (compare Fig 2D to 2B). Thus, SSCRs do not contribute extensively to SECReTE abundance.
Since SRP depletion does not block the co-translational ER targeting of mRNAs encoding predicted SRP-independent proteins [24], we examined whether SRP-independent transcripts on yeast ER-bound ribosomes are more enriched with SECReTE10 than SRP-dependent ones (S2A–S2D Fig). Using this dataset [24], we found that both SRP-dependent and -independent transcripts contain SECReTE10 (S2B Fig). However, SRP-dependent (i.e. TMD-containing) transcripts essentially bear only the NYN-based motif, whereas both NYN- and NNY-based motifs appear in SRP-independent transcripts (S2B and S2C Fig). Moreover, the subset of mRNAs that remain ER-bound after SRP depletion all appear to be enriched in NNY-based SECReTE (S2D Fig).
Repetitive nucleotide triplet sequence elements [i.e. NNX; where X = is any of five other dinucleotide combinations: K (T/G), M (C/A), R (A/G), S (G/C), or W (A/T)] other than NNY might exist in secretome transcripts and distinguish them from non-secretome transcripts. Thus, we examined for the presence of ≥10 uninterrupted NNX repeats in the coding sequences of secretome and non-secretome proteins. Examination of the yeast genome did not reveal any significant repetitive NNX-based triplet repeats in mSMPs, except for that seen for pyrimidine (i.e. NNY). Moreover, this did not change upon removal of the TMD sequences from the analysis, indicating that concatenation of the regions flanking the TMD does not create SECReTE motifs de novo (S3A and S3B Fig). In contrast, we did find that R (purine) and W repeats are enriched in the third position of a large fraction of non-secretome transcripts, especially after TMD removal (S3A and S3B Fig). This indicates that SECReTE is the principal, if not sole, nucleotide triplet motif of ≥10 repeats in the secretome protein-encoding genes of yeast, although a repetitive NNR motif could be identified in the third position of a subset of the non-secretome genes. No additional motifs in either the first or second positions (i.e. XNN, NXN) were identified in non-secretome genes.
Next, we examined the distribution of SECReTE in the different regions (i.e. 5’UTR, CDS, 3’UTR) of yeast genes. We found that the large majority (>90%) of SECReTE motifs (8211 out of 9003) are present in the CDS regions (S4A Fig, left), however, the overall distribution is biased to the 5’ and 3’UTRs, when normalized for the mean length of these smaller regions (S4A Fig, right). Secretome transcripts (1144) contained ~35% of the total SECReTE motifs, an amount proportionally larger than that of non-secretome transcripts (4760), although the motif is more or less evenly represented in both the CDS and separate UTR regions after normalization for length. Mapping of motif distribution along the entire gene length (after normalization) revealed a uniform distribution in TMD-containing transcripts, but also showed that a number of transcripts encoding soluble proteins (~60) have a preference for SECReTE at the 5’ end, as this distribution could be eliminated upon SSCR removal (S4B Fig). Thus, despite the fact the SSCRs do not highly contribute to overall SECReTE abundance (Fig 2D), this does not exclude the possibility that the motif cannot be present therein. Examination of both NYN- and NNY-based SECReTE motifs in the coding regions showed that both contribute to motif presence at the 5’ end of the same subset of transcripts encoding soluble secreted proteins (S4B Fig).
When comparing SECReTE motifs residing in the CDS to those in the UTRs, we found that CDS motifs tend to consist of RRY repeats rather than NNY-based repeats (S4C Fig). On the other hand, UTR-residing SECReTE motifs are statistically more pyrimidine-rich and, thus, are biased towards the NNY pattern of repeats (S4C Fig). Next, we checked if the UTRs of the secretome transcripts are enriched with pyrimidines in general. Indeed, we found that secretome transcripts have a slightly higher Y content in their UTR (S4D Fig), however, this enrichment disappears after removal of the SECReTE-containing UTRs from the analysis (e.g. 43 and 99 transcripts for the 5’ and 3’UTRs, respectively) (S4E Fig). This implies that SECReTE motifs contribute to the pyrimidine-enrichment of UTRs in secretome transcripts.
There is a possibility that SECReTE enrichment results from codon usage of the transcript. To check this possibility, we performed permutation test analysis. In this case, each gene sequence was randomly shuffled (1000 times), while codon usage remained constant. We then calculated the Z-score (i.e. number of standard deviations from the mean) of SECReTE10 for each gene to evaluate the probability of the signal to appear randomly. By looking at Z-score distribution in secretome and non-secretome genes, it can be concluded that SECReTE enrichment in mSMPs is not a random consequence of codon usage (S5A Fig). This conclusion is valid for mSMPs encoding both membrane and soluble proteins (S5B Fig). We also conducted the analysis for each nucleotide position of the codon separately (i.e. for the YNN, NYN, and NNY versions of the motif). For that, we calculated the fraction of genes with a significant Z-score (≥1.96) for each position separately. The fraction of genes with a significant Z-score was larger in secretome genes than in the non-secretome genes at both the second and third positions of the codon (S5C Fig), strengthening the notion that SECReTE is significantly more enriched in those positions. This finding is not dependent on the presence of TMDs, since the fraction of genes with a significant Z-score was larger for both soluble and TMD-containing secretome transcripts, rather than for soluble and TMD-containing non-secretome transcripts (S5D Fig).
To determine those gene categories that are overrepresented with SECReTE-containing genes, gene ontology (GO) enrichment analysis was conducted. When genes that contain at least one occurrence of SECReTE10 in any of its YNN-, NYN- or NNY-based forms were searched for GO enrichment (using all yeast genes as a background), unsurprisingly, membrane proteins were found to have a high enrichment score (fold enrichment = 1.67) (Fig 3A). The most SECReTE-enriched gene category was that comprising cell wall proteins (fold enrichment = 1.8) (Fig 3A). When 15 NNY repeats served as a threshold, the fold-change enrichment of the cell wall protein category increased to 4.8-fold (Fig 3B). To further characterize the mRNAs enriched with SECReTE, we divided the secretome and non-secretome into subgroups and calculated the fraction of transcripts containing SECReTE10 in each category. In agreement with the GO analysis, more than 90% of mRNAs coding for cell wall proteins possess SECReTE10 (and above) motifs and the cell wall proteins were the most SECReTE-rich overall (Fig 3C). Interestingly, this group also comprised the principal set of transcripts that remain associated with ER-bound ribosomes after SRP depletion (S2E and S2F Fig). In addition, we found that 86% of mRNAs of proteins encoding both TMD and signal-sequence (SS) regions, as well as 84% of TMD-encoding secretome mRNAs, contain SECReTE10 (Fig 3C). Of these, mRNAs encoding tail-anchored (TA) proteins contain the lowest number of transcripts with SECReTE10 in the secretome (Fig 3C). TA proteins are known to translocate to the ER through an alternative pathway (GET) after being translated in the cytosol [32–34], and their transcripts are not enriched on ER membranes [22,23] either before or after SRP depletion [24]. This implies that SECReTE is more abundant in mRNAs undergoing translation on the ER. In contrast, transcripts for non-secretome proteins (i.e. mitochondrial and cytonuclear) have the lowest abundance of SECReTE elements (Fig 3C).
Since SECReTE is highly enriched in mRNAs coding for cell wall proteins, we wanted to check if it could be discovered using an unbiased motif search tool. For that, we analyzed the mRNA sequences of cell wall proteins using MEME to identify enriched mRNA motifs. The most significant result obtained highly resembled the SECReTE10 repeat with either U or C (Fig 3D). Importantly, we did not detect a protein motif within this mRNA motif, eliminating the possibility that the SECReTE element is dependent on a specific protein sequence.
Conservation or convergence in evolution are often strong indications of functional significance. To check whether SECReTE enrichment in mSMPs is found in additional organisms (e.g. humans and B. subtilis) we analyzed other genomes. In humans, as in S. cerevisiae, SECReTE10 gave the most significant separation between RNAs encoding secretome and non-secretome proteins, based on ROC analysis (Fig 4A). After verifying that SECReTE10 does not correlate with gene length, SECReTE10 served as a threshold to define presence of the SECReTE motif. As in yeast, SECReTE is enriched in the second and third codon positions of secretome transcripts, in comparison to non-secretome transcripts (Fig 4B). Also, a larger fraction of secretome transcripts that lack TMDs contain the NNY-based SECReTE, as compared to non-secretome transcripts bearing TMDs (Fig 4C). Thus, the SECReTE motif is present in higher organisms. A list of all SECReTE10 and higher motifs found in human genes is given in S4 Table. Interestingly, unlike yeast, we found that a disproportionally large majority of motifs (29753 out of 52,047) are present in the UTRs instead of being in the CDS, especially after normalization for length, and this phenomenon is observed for both secretome and non-secretome transcripts (S6A Fig). Therefore, in contrast to yeast, the UTRs and especially the 3’UTR are preferential sites for SECReTE location in human transcripts. Like yeast, however, RRY enrichment is observed for SECReTE motifs in the CDS regions, while high pyrimidine content is observed in the UTRs (S6B Fig). Therefore, although both yeast and human share the same SECReTE motifs, their distribution over gene region appears different. Interestingly, human transcripts encoding glycophosphatidylinositol (GPI)-anchored proteins, which are equivalent to cell wall proteins, were found to be highly enriched with SECReTE. In fact, a SECReTE-like motif was previously shown to confer the translation-independent localization of a transcript encoding human GPI-anchored protein, placental alkaline phosphatase, to the ER [35] In contrast, tail-anchored genes, as well as mitochondrial and cytonuclear genes, have a low SECReTE abundance as seen in yeast (Fig 4D). Finally, we also detected a high abundance of SECReTE10 in genes encoding secretome proteins from B. subtilis, in comparison to those encoding non-secretome proteins (Fig 4E). Thus, SECReTE motifs are also present in prokaryotic genomes.
We next asked whether SECReTE is conserved evolutionarily via inheritance. To differentiate between conservation and possible convergence we analyzed the genome of the fission yeast, S. pombe, for the presence and position of SECReTE in secretome and non-secretome transcripts. As found for S. cerevisiae, SECReTE is enriched (in both NYN- and NNY-based forms) in a larger fraction of S. pombe mSMPs that lack TMDs, as compared those containing TMDs or to non-secretome transcripts that either bear or lack TMDs (Fig 4F). Next, we aligned orthologous genes encoding secretome proteins from S. cerevisiae to those of S. pombe (457 genes total), and examined whether SECReTE is found in the same (i.e. aligned) position within the gene. We found that the coordinates of SECReTE motifs in the large majority (e.g. 393 out of 457) of ortholog pairs were non-aligned. This might imply that the majority of SECReTE motifs arose through convergent evolution, although we cannot rule out drift of the motif after species divergence. Nonetheless, it is clear that SECReTE is present in all species examined by us, from prokarya to eukarya, the latter including yeasts and mammals.
To further understand the significance of SECReTE and validate its importance to yeast cell physiology, we examined its relevance by elevating or decreasing the signal in selected genes. Three representative genes were chosen, based on their relatively short gene length, a detectable phenotype upon their deletion, and their function in different physiological pathways. These genes included: SUC2, which encodes a soluble secreted periplasmic enzyme; HSP150, which encodes a soluble media protein; and CCW12, which encodes a GPI-anchored cell wall protein. The overall SECReTE signal of the genes was increased by substituting any A or G found in the third codon position with a T or C, respectively, thereby enriching SECReTE presence along the entire gene [(+)SECReTE]. The reverse substitution, converting T to A or C to G, decreased the overall SECReTE signal [(-)SECReTE]. We note that we added or removed only NNY-based triplet motifs, in order not to change the amino acid sequence of the encoded protein. The number of motifs present in each gene before and after SECReTE addition/reduction is shown in S5 Table. Thus, in the case of HSP150 several NYN-based SECReTE motifs remain in the (-)SECReTE mutant. Changes in the stability of the mRNA secondary structure (free energy) and the Codon Adaptation Index (CAI) [32] were kept to within a similar range (S5 Table). SECReTE mutations in SUC2, HSP150, and CCW12 are shown along the length of the gene, using a minimum threshold of either 1 NNY repeats or 10 NNY repeats, as shown in S7 Fig (S7A–S7C Fig; upper and lower parts, respectively).
SUC2 codes for different forms of invertase translated from two distinct mRNAs, short and long, which differ only at their 5’ ends. While the longer mRNA codes for a secreted protein that contains a signal sequence, the signal sequence is omitted from the short isoform, which codes for a cytoplasmic protein. Secreted Suc2 expression is subjected to glucose repression; however, under inducing conditions (i.e., glucose depletion), Suc2 is trafficked through the secretory pathway to the periplasmic space of the cell. There, it catalyzes the hydrolysis of sucrose to glucose and fructose, this enzymatic activity being responsible for the ability of yeast to utilize sucrose as a carbon source and can be measured by a biochemical assay (i.e. invertase activity), both inside and outside of the cell. The effect of SECReTE mutations on Suc2 function was tested by examining the ability of mutants to grow on sucrose-containing media by drop-test. Interestingly, the growth rate of SUC2(-)SECReTE on sucrose plates was decreased, while the SUC2(+)SECReTE mutant exhibited better growth in comparison to WT cells (Fig 5A), even though no growth change was detected on YPD plates. These findings suggest that SECReTE strength affects the secretion of Suc2. These changes in Suc2 secretion could result from changes in SUC2 transcription, Suc2 production, and/or altered rates of secretion. To distinguish between possibilities, WT cells, suc2Δ, and SUC2 SECReTE mutants were subjected to invertase assays. The invertase assay enables the quantification of secreted Suc2, as well as internal Suc2, by calculating the amount of glucose produced from sucrose. As expected, under glucose repressing conditions (e.g. 2% glucose) the levels of both secreted and internal Suc2 were very low. When cells were grown on media containing low glucose (e.g. 0.05% glucose) to promote the expression of the secreted enzyme, secreted Suc2 levels were altered due to changes in SECReTE. Corresponding to the drop-test results (Fig 5A), a significant decrease in secreted invertase was detected with SUC2(-)SECReTE cells, while a significant increase was detected with SUC2(+)SECReTE cells, in comparison to WT cells (Fig 5B). Importantly, SUC2(+)SECReTE cells were found to secrete nearly 2-fold (92.2±9.2%, p <0.016) more invertase than SUC2(-)SECReTE cells, while no Suc2 secretion was detected from suc2Δ cells (Fig 5B, secreted). If SECReTE mutations affect the efficiency of Suc2 secretion, but not its synthesis, then Suc2 should accumulate in SUC2(-)SECReTE cells corresponding to the difference secreted from SUC2(+)SECReTE cells. However, this was not the case as the internal amount of Suc2 decreased in SUC2(-)SECReTE cells and slightly increased in SUC2(+)SECReTE cells (Fig 5B, internal). These findings suggest that SECReTE alterations in SUC2 might affect the level of protein production. We next examined the rate of invertase secretion for WT, SUC2(+)SECReTE, and SUC2(-)SECReTE cells shifted to low glucose medium for varying amounts of time (Fig 5C). The results show that the average maximal rate of secretion from SUC2(+)SECReTE cells is slightly higher than for WT cells (i.e. 0.479±0.016 vs. 0.432±0.013 units per min per O.D.600 unit of cells; ±standard deviation, n = 3 experiments), and was significantly (62.4±7.8%; p <0.0001) higher than of SUC2(-)SECReTE cells (0.295±0.022 units/min per O.D.600 unit of cells; ±standard deviation, n = 3 experiments). In contrast, the time required to achieve half-maximal secretion between SUC2(+)SECReTE and SUC2(-)SECReTE cells was relatively unchanged under the experimental conditions (i.e. ~74 min; R2 values = >94). Thus, the presence of SECReTE affects not only invertase production and overall secretion, but also its rate of secretion from yeast.
Next, we wanted to study the importance of SECReTE in HSP150. Hsp150 is a component of the outer cell wall and while the exact function of Hsp150 is unknown, it is required for cell wall stability and resistance to cell wall-perturbing agents, such as Calcofluor White (CFW) and Congo Red (CR). While hsp150Δ cells are more sensitive to cell wall stress, the overproduction of Hsp150 increases cell wall integrity [36]. Hsp150 is secreted efficiently into the growth media and its expression is increased upon heat shock [37,38]. The effect of modifying SECReTE in HSP150 was examined via drop-test by testing the sensitivity of HSP150(-)SECReTE and HSP150(+)SECReTE cells to added CFW, in comparison to WT and hsp150Δ cells. As can be seen from Fig 5D while the HSP150(-)SECReTE strain was more sensitive to CFW as compared to WT cells, the HSP150(+)SECReTE strain was more resistant to CFW. As expected, hsp150Δ cells are the most susceptible to CFW (Fig 5D). HSP150 strains were also subjected to Western blot analysis to measure levels of the mutant proteins. Since HSP150 secretion is elevated upon heat-shock [37,38], cells were shifted to 37°C before protein extraction. Protein was extracted from both the growth medium and cells to detect both external and internal protein levels, respectively. The amount of Hsp150 secreted to the medium was decreased in HSP150(-)SECReTE cells and elevated in HSP150(+)SECReTE cells, in comparison to WT cells (Fig 5E). Similar to Suc2, the internal amount of Hsp150 was decreased in HSP150(-)SECReTE cells, relative to WT cells, and showed a greater reduction than that seen in the external form (Fig 5E), despite the fact that several NYN-based motifs remain in the gene. As the internal level of Hsp150 in HSP150(+)SECReTE cells was more or less unchanged relative to WT cells, we concluded that SECReTE alteration in HSP150 may also affect protein production.
CCW12 encodes a GPI-anchored cell wall protein that localizes to regions of the newly synthesized cell wall and maintains wall stability during bud emergence and shmoo formation. Deletion of CCW12 was shown to cause hypersensitivity to cell wall destabilizing agents, like hygromycin B (HB) [39,40]. Since the SECReTE score is very high in CCW12, it was not possible to further increase the signal. Therefore, we generated only CCW12(-)SECReTE cells and tested their ability to grow on HB-containing plates. As seen with HSP150(-)SECReTE (Fig 5D), we found that the CCW12(-)SECReTE mutation rendered cells sensitive to cell wall perturbation, in comparison to WT cells (Fig 5F).
The ability of SECReTE addition to improve the secretion of an exogenous protein would not only be substantial evidence for its importance, but also could be a useful, low-cost, industrial tool to improve the secretion of recombinant proteins without changing protein sequence. To test that, we employed a GFP transcript construct bearing the encoded signal sequence (SS) of Gas1 (SSGAS; SSGas1) at the 5’ end. SSGas1 addition enables the secretion of GFP protein to the medium, although its secretion was not as efficient in comparison to other SS-fused GFP proteins, such as SSKar2 (Fig 5G). To potentially improve the secretion of SSGAS-GFP, we added an altered 3’UTR sequence of GAS1 that contained SECReTE [i.e. in which all A’s and G’s were replaced with T’s and C’s, respectively; SSGAS-GFP-GAS13’UTR(+)SECReTE]. We then tested the effect of SECReTE addition upon the secretion of GFP into the media. We found that the addition of SECReTE to the 3’UTR of SSGAS-GFP improved the secretion of GFP secretion into the media, in comparison to SSGAS-GFP, and was similar to that of SSKar2-GFP construct (Fig 5G). GFP expression without the signal sequence was unable to be secreted (Fig 5G).
As protein levels may be altered by (-)SECReTE and (+)SECReTE mutations (Fig 5B, 5E and 5G), we examined whether changes in gene transcription or mRNA stability are involved. Quantitative real-time (qRT) PCR was employed to check whether mRNA levels of SUC2, HSP150, and CCW12 are affected by SECReTE strength. We found that SUC2(–)SECReTE mRNA levels were almost 30% lower than in SUC2 WT cells, while SUC2(+)SECReTE levels were ~50% higher than WT (S8A Fig). This change in mRNA levels might contribute to the ability of SUC2(+)SECReTE mutant to increase protein production and, therefore, grow better on sucrose-containing medium (Fig 5A–5C).
The effect of SECReTE mutation on HSP150 mRNA levels was also studied. We found that the mRNA level of HSP150(-)SECReTE was similar to WT, while that of HSP150(+)SECReTE was slightly decreased (S8B Fig). Thus, the change in Hsp150 protein levels and sensitivity to CFW due to SECReTE alteration (Fig 5D and 5E) is not entirely explained by changes in mRNA levels. SECReTE mutations in CCW12(-)SECReTE did not cause a significant change in its mRNA level (S8C Fig), therefore, the increased sensitivity of CCW12(-)SECReTE to HB (Fig 5F) is probably not due to a decrease in CCW12 mRNA.
To test whether SECReTE has a role in dictating mRNA localization, we visualized the SUC2 and HSP150 mRNAs by single-molecule FISH (smFISH) using specific fluorescent probes and tested the influence of SECReTE alteration on the level of mRNA co-localization with the ER. We used Sec63-GFP as an ER marker and calculated the percentage of mRNA granules (spots) per cell that co-localized with cortical and perinuclear ER (cER and nER, respectively) or were not localized to the ER. We note that probes to the native gene sequences were used to measure the level of mRNA localization in WT cells as well as in (+)SECReTE or (-)SECReTE cells. The number of FISH spots per cell was variable for both mRNAs, with (-)SECReTE cells having less spots than WT cells, while (+)SECReTE cells had more spots per cell than WT cells (S9A Fig). This largely reflects the results obtained by qRT-PCR for SUC2 (S8A Fig), however, we cannot discount the possibility that SECReTE alterations lessen the level of mRNA hybridization with the probe set and, thus, underestimate RNA localization to some degree.
We found that the level of co-localization between SUC2(-)SECReTE mRNA granules and Sec63-GFP was significantly less in comparison to SUC2(+)SECReTE mRNA granules (e.g. 56.7±1.5% vs. 74.1±1.7% co-localization, respectively; p = 2.0E-13) (Figs 6A, 6B and S9B), while being similar to native SUC2 (e.g. 56.3±2.4% ER co-localization). This finding suggests that the number of SECReTE motifs influences mRNA localization to the ER, in addition to enhancing secretion. We also found that there were fewer granules present in SUC2(-)SECReTE cells than observed in either SUC2(+)SECReTE or WT cells (S9A Fig), which corresponds with the qRT-PCR results (S8A Fig). Finally, we note that no specific ER subdomain (i.e. cER or nER) was preferentially labeled upon the increase in SECReTE motifs (Fig 6B).
We next examined the level of HSP150 mRNA localization to the ER in HSP150(+)SECReTE or HSP150(-)SECReTE cells. We found that as with SUC2, the addition of SECReTE motifs increased the level of ER localization (Fig 6B and S9C Fig) from 63.7±2.0% to 77.9±1.6% over native HSP150 localization (p = 7.0E-8). In contrast, no change in the level of HSP150(-)SECReTE mRNA co-localization was observed (e.g. 64.0±1.6%), which perhaps reflects presence of the NYN-based motifs that could not be mutated without altering the amino acid sequence. Overall, however, both sets of results show that SECReTE addition to an mRNA increases the pattern of ER localization. To substantiate the smFISH results for SUC2, we also performed the subcellular fractionation of cells expressing native SUC2, SUC2(-)SECReTE, or SUC2(+)SECReTE to obtain crude membrane (containing ER) and cytosolic fractions and quantified the distribution of mRNA using qRT-PCR (Fig 6C). After normalization using actin mRNA as a control, the results indicated that SUC2(-)SECReTE mRNA is less abundant overall (as observed above in S8A Fig and S9A Fig) and appeared to be less membrane-associated than either native or SUC2(+)SECReTE mRNA by ~40%. Taken altogether, our results imply that SECReTE presence/addition stabilizes secretome mRNAs, increases mRNA localization to the ER, and enhances both protein production and secretion.
To further elucidate the role of SECReTE it is essential to identify its binding partners, presumably RBPs. Large-scale approaches were previously used to identify mRNAs that are bound >40 known RBPs in yeast [41–43]. To obtain a list of potential SECReTE-binding proteins (SBPs) we searched the datasets for RBPs that bind mRNAs highly enriched with SECReTE. For each RBP, we calculated what fraction of its bound transcripts contain SECReTE10. RBPs found to bind large fractions of SECReTE10-containing mRNAs included Bfr1, Whi3, Puf1, Puf2, Scp160, and Khd1 (Fig 7A), and were all previously shown to bind mSMPs [41–43]. To test which of these candidates bind SECReTE, each of the genes these RBPs was deleted in either WT or HSP150(+)SECReTE cells. We hypothesized that the deletion of a genuine SBP might confer hypersensitivity to CFW and eliminate the growth rate differences between WT and HSP150(+)SECReTE cells observed on CFW-containing plates (Fig 5D). When PUF1, PUF2, or SHE2 were deleted we found that HSP150(+)SECReTE strain was still more resistant to CFW than WT cells (S10 Fig). One possible explanation for this lack of effect is that these RBPs either do not bind HSP150 or that they are redundant with other SBPs. However, we did find that the deletion of either WHI3 or KHD1 eliminated the differences between WT and HSP150(+)SECReTE strains on CFW-containing plates (Fig 7B). This suggests Whi3 and Khd1 bind HSP150 mRNA and possibly other secretome mRNAs, and even WT cells alone were rendered more sensitive to CFW in their absence (Fig 7B).
The correct sorting of proteins within the cell is crucial for cellular organization and normal function. While the information for protein localization can reside within the protein sequence (e.g. protein targeting sequences), the spatial localization of an mRNA may also be important for protein proper targeting cell [1,2]. For example, mSMPs localize to the surface of the ER independently of translation and that localization requires elements within the transcript that are presumably recognized by an ER-localized RBP (see reviews [8,9,44]). It was shown previously that ER-targeted TMD-containing proteins are highly enriched with amino acids containing uracil-rich codons [31] and, thus, their ORFs are enriched with pyrimidines [27]. Nevertheless, mRNAs coding for secretome proteins that do not contain TMDs were also found to be enriched on ER membranes [2,14,45]. Therefore, an additional mechanism or element appears necessary to confer mSMP localization. Here, we identify features that characterize all mSMPs, either encoding a TMD or not, and discovered a repetitive motif consisting of ≥10 consecutive NNY repeats. This motif, termed SECReTE, is not restricted to transcripts coding for TMD-containing proteins, but can be found in higher abundance in all secretome transcripts, from prokaryotes (e.g. B. subtilis) to yeast (S. cerevisiae and S. pombe) to humans (Figs 1 and 4). By analyzing the S. pombe genome it was discovered that SECReTE tends to be positioned differently than in orthologous S. cerevisiae genes encoding secretome proteins. This implies that SECReTE enrichment in mSMPs may have evolved in a number of different ways (e.g. conservation, drift, or convergence). Correspondingly, we found that SECReTE is preferentially located in the 3’UTR of human transcripts, while being present mainly in the CDS of budding yeast (S4A and S6A Figs). The idea that SECReTE motifs are present throughout evolution likely emphasizes its significance and functionality.
To better characterize SECReTE, we first determined the number of NNY repeats that can serve as a threshold to verify its presence and found that ten (i.e. SECReTE10) constitute a genuine motif, rather than a random occurrence, and enabled significant separation between secretome and non-secretome mRNAs (Fig 1). Importantly, no other repetitive motif was identified in secretome transcripts (S3A and S3B Fig). SECReTE abundance was calculated separately for each position of the codon and while being barely present in the first position (Fig 2A, YNN), it was highly represented in the second and third positions in mSMPs (NYN and NNY, respectively), in comparison to non-mSMPs. Interestingly, the SECReTE10-containing fraction of transcripts coding for soluble secreted proteins is larger than that of mRNAs encoding secreted membrane proteins, suggesting that SECReTE enrichment is not merely due to the high fraction of TMD-containing genes in the secretome (Fig 2B). Importantly, when encoded TMD sequences were removed from the analysis, SECReTE10 was found to be more abundant in the third position of the codon (NNY) in secretome transcripts (Fig 2C). In contrast, no significant change in SECReTE abundance was observed upon removal of the SSCR regions from the computational analysis of secretome genes that encode signal peptides (Fig 2D). Thus, it is the TMD regions that contribute to NYN-based SECReTE motif enrichment.
By analyzing the ribosome profiling datasets of both Jan et al. [22] and Chartron et al [23], we verified that a higher fraction of SECReTE10-containing transcripts is enriched on ER-bound ribosomes (S1A Fig) and in polysomes extracted from the membrane fraction (S1B Fig), as well as in the membrane fraction itself (S1C Fig). In contrast, transcripts with SECReTE10 were not enriched on mitochondria-bound ribosomes (S1D Fig). Moreover, analysis of a recent dataset by Costa et al [24] revealed that conditional SRP depletion strongly affects the association of predicted SRP-dependent (i.e. TMD-containing) transcripts with ER-bound ribosomes, but had less effect upon SRP-independent transcripts that are more enriched with NNY-based SECReTE motifs (S2 Fig). Permutation analysis confirmed that SECReTE enrichment in mSMPs is not arbitrary and demonstrated that it is independent of codon composition (S5 Fig). Altogether, SECReTE motifs can be found in both TMD-containing and -lacking transcripts, whereby NYN-based motifs are contributed principally by the TMD regions of SRP-dependent secretome proteins. In contrast, NNY-based SECReTE motifs are enriched in soluble/SRP-independent secretome proteins. Finally, we note that SECReTE motifs are equally distributed to UTRs in the case of yeast and preferentially in the case humans, after normalization for gene length (S4A Fig and S6A Fig).
Although SECReTE10 enables the classification of mSMPs (Fig 1C and Fig 4C), the separation between secretome and non-secretome is not absolute and mRNAs coding for non-secretome proteins may also contain SECReTE sequences. While this might suggest that the motif is not completely defined, it might also imply that SECReTE plays a role in non-secretome mRNAs, perhaps in ER localization. There is an ongoing debate regarding whether mRNAs encoding cytosolic proteins localize to the ER and undergo translation by ER-associated ribosomes [46,47]. The idea that ER can support the translation of both secretory and cytosolic proteins was initially proposed by Nicchitta and colleagues [14–16,44,48]. Furthermore, they suggested that since translation initiation can start before the emergence of the signal sequence, ER-bound ribosomes would not distinguish between mRNAs and, therefore, mRNAs encoding cytosolic proteins can tether to ER membranes [16,23,44]. The fact that a large fraction of mRNAs encoding cytosolic proteins also contain SECReTE raises the possibility that their targeting to the ER is intentional and that this motif plays a role in it, even if the protein is not destined for secretion. Thus, SECReTE presence could be an organizing principle for transcript localization to the ER, while the presence or absence of an ER translocation signal (e.g. signal peptide, TMD, GPI anchor) in the polypeptide is the determinant for either secretion or cytosolic localization, respectively.
Gene ontology analysis revealed that genes encoding cell wall proteins are the most enriched with SECReTE (Fig 5A–5C). In contrast, TA-protein encoding transcripts show less (Fig 5C), perhaps since they are not enriched on ER membranes [22,23] and their translation products translocate to the ER only after full translation in the cytosol [32–34]. This finding implies that SECReTE is more abundant in mRNAs that are meant to be translated on (or near) the ER. Importantly, SECReTE was also identified with an unbiased method for motif discovery using the MEME server to find common sequence elements in cell wall genes. This parallel methodology supports our original identification of SECReTE and its importance is further enhanced by the discovery that it is present from bacteria to humans (Fig 4). As in yeast, human mSMPs are more enriched with SECReTE than non-secretome transcripts and this is independent of TMD presence (Fig 4B and 4C). Unlike yeast, however, human transcripts contain much larger UTR sequences and SECReTE elements appear to be more abundant therein, both in number and distribution (S6A Fig).
The physiological relevance of SECReTE was explored by altering its enrichment in three mSMPs: SUC2, HSP150, and CCW12 (Fig 5 & Fig 6, and S7 Fig, S8 Fig, & S9 Fig). Although the amino acid sequences were not altered by motif mutation, the functionality of these genes was. SUC2 SECReTE mutants exhibited altered growth rates on sucrose-containing medium in comparison to WT cells, i.e. reduced growth when motif score (number) was decreased and better growth when motif score was elevated (Fig 5A). Moreover, either the decrease or increase of motif score corresponded directly with a decrease or increase in invertase synthesis, invertase secretion, and the rate of secretion, respectively (Fig 5A–5C). HSP150 SECReTE mutants also behaved differently, i.e. HSP150(-)SECReTE cells exhibited higher sensitivity to CFW in comparison to WT cells, while HSP150(+)SECReTE cells were more resistant (Fig 5D). Similarly, CCW12(-)SECReTE cells exhibited hypersensitivity to HB (Fig 5F). These findings strengthen the notion that SECReTE may play an important role in regulating the amount of protein secreted from cells. This idea was verified using an exogenous substrate, SSGAS-GFP, whose secretion was significantly enhanced upon addition of the GAS1 3’UTR containing the SECReTE motif (Fig 5G). The number of SECReTE motifs not only increased protein production and secretion, it also enhanced the localization of SUC2 and HSP150 transcripts to the ER (Fig 6 and S9B and S9C Fig). Thus, it would seem clear that SECReTE motifs enhance mRNA localization to the ER and subsequent secretome protein production and secretion, although the mechanism is not entirely known. It may be that SECReTE abundance helps stabilize secretome mRNAs as observed (S8 Fig and S9A Fig), perhaps by increasing their localization to the ER (Fig 6A–6C and S9B and S9C Fig), and through this mechanism yields higher amounts of protein translation and secretion. Higher levels of translation may promote the elevated rate of secretion afforded by (+)SECReTE mutations, although further work is necessary to fully resolve the function(s) of SECReTE.
Although SECReTE is present throughout evolution, it is not a strict sequence-based motif since a wide variety of pyrimidine-rich sequences fit its demands. This variability might allow for the preferential binding of specific mSMPs (or non-secretome-encoding mRNAs that contain SECReTE elements) to different SBPs under different conditions, depending upon secretory needs of the cells. While it is generally assumed that mRNA localization is required for local translation and proper positioning of the translated protein, SBP binding post-export may provide spatial and temporal regulation of mRNA stability and protein synthesis [49,50]. Moreover, additional features within secretome mRNAs may also influence both protein synthesis and secretion. For example, Palazzo et al (2007) previously showed that the low usage of adenine residues created no-A stretches within the signal sequence of SSCR-encoding proteins and the addition of adenines could affect nuclear export of the mRNA [29]. Thus, multiple cis RNA elements appear to impinge upon the translational control of secreted proteins.
Correct mRNA localization is not redundant to protein localization, but is yet another level of regulation that affects protein production. Supportive of this model is Puf3, an RBP that targets its associated mRNAs to the surface of the mitochondria [51]. In addition to its mitochondrial targeting role, Puf3 binding regulates the translational fate of mRNAs. Specifically, Puf3 binding leads to mRNA decay and repressed translation on high glucose, but becomes phosphorylated and promotes translation under low glucose conditions [52,53]. Interestingly, alterations in SUC2 and HSP150 SECReTE motifs also support this model, as mutations altered the amount of secreted protein, but not necessarily the ratio between secreted and non-secreted protein (Fig 5B and 5D). As Suc2 and Hsp150 both contain a signal peptide, SECReTE alteration does not necessarily affect protein targeting, but only mRNA targeting. Yet, if localizing mRNAs to the ER is important for conferring efficient translation, either through mRNA stabilization or the regulation of protein production, then SECReTE presence and strength (in terms of length or number) is expected to fill such a regulatory role. If SECReTE affects mRNA stabilization, this might well explain why we observed a decrease in SUC2(-)SECReTE mRNA levels and an increase in SUC2(+)SECReTE mRNA levels, in comparison to WT cells (Fig 6C and S8A Fig & S9A Fig). Moreover, we found that SECReTE abundance in SUC2 and HSP150 led to enhanced ER localization and membrane association (Figs 6 and S9B), suggesting that ER localization and mRNA stability are likely to be interconnected. Taken together, our results suggest that SECReTE abundance affects the localization, stability, and translation of secretome mRNAs (Fig 5, Fig 6, S8 Fig & S9 Fig).
If SECReTE is a cis regulatory element, the question is who are its trans-acting partners? Large-scale approaches have been used to identify mRNAs that interact with known RBPs in yeast [41–43]. These analyses enabled the identification of Bfr1, Whi3, Puf1, Puf2, Scp160, and Khd1 as potential SBPs, based upon their ability to interact with known SECReTE-containing transcripts (Fig 7A). As a means of verification, we first deleted individual RBPs and determined whether this alleviated the growth differences between WT and HSP150(+)SECReTE cells on CFW-containing medium, as might be expected upon the removal of a bona fide SBP. While the deletion of PUF1, PUF2, or SHE2 did not alter the increased resistance of HSP150(+)SECReTE cells to CFW, those of KHD1 and WHI3 did (Fig 7D and S10 Fig). This suggests that they may be SBPs and several indications support the idea that Whi3 and Khd1 serve in this regard. For example, Whi3 possesses an RNA recognition motif and was already identified as preferentially binding mSMPs, including HSP150 [41,54]. Whi3 also binds CLN3 mRNA and is important for the efficient retention of Cln3 at the ER [55], as well as to destabilize CLN3 and other mRNA targets [54]. In addition, the whi3 deletion mutant is sensitive to cell wall perturbing agents, such as CFW and congo-Red [41], and is synthetic lethal with the deletion of CCW12 in a synthetic genetic analysis screen [40]. Thus, Whi3 is an attractive candidate SBP. The same can be said for Khd1, which interacts with hundreds of transcripts including many mSMPs [42], and contains 3 K homology (KH) RNA-binding domains suggested to cooperatively recognize triplets of C/U-rich sequence elements [56]. These transcripts include CCW12 [42] and, correspondingly, Khd1 plays a role in the cell wall integrity signaling pathway [57]. However, Khd1 is not essential and is best known for its association with ASH1 mRNA and is required for both its translational repression and efficient localization to the bud tip [58]. ASH1 mRNA, as well as mRNAs encoding polarity and secretion factors (e.g. SRO7), are physically bound to cortical ER and both are delivered to the bud tip via the same mechanism involving She2, She3, and Myo4/She1 [59,60]. Importantly, both ASH1 and SRO7 have SECReTE10 motifs (S3 Table). Thus, Khd1 interactions with SECReTE-containing mRNAs might potentiate their targeting to the ER, although this remains to be proven. Further work is required to identify SBPs and determine their role in protein secretion.
Although the mechanism is not entirely clear, SECReTE binding to ER-associated SBPs is likely to enhance transcript interactions with the ER and, thereby, increase mRNA stabilization, with the result being either increased translation efficiency and/or number of mRNAs translated on ER-bound ribosomes (see model, Fig 8). Our model supports the idea that mRNA plays an active role in its own targeting and this does not necessarily contradict the importance of co-translational localization, but rather provides another level of regulation. Thus, we believe that SECReTE plays an important physiological role in the fine-tuning of cellular secretion.
Being both a unicellular and eukaryotic organism, S. cerevisiae is advantageous for the production of recombinant proteins as it grows quickly, is easy to culture, and secretes post-translationally modified proteins into the extracellular medium, which can facilitate their purification. Moreover, S. cerevisiae is a generally recognized as a safe (GRAS) organism, which makes it favorable for use in the production of biopharmaceuticals [61,62]. Unfortunately, the natural capacity of S. cerevisiae secretory pathway is relatively limited and, thus, mechanisms that improve secreted protein production would be of significant benefit. Since SECReTE abundance increases protein production and secretion its use as an added RNA motif may prove to be a simple low-cost tool to improve recombinant protein production.
The yeast strains and plasmids used in this study are listed in S1 Table and S2 Table, respectively.
Yeasts were grown at the indicated temperature either in a standard growth medium (1% Yeast Extract, 2% Peptone, 2% Dextrose) or synthetic medium containing 2% glucose [e.g., synthetic complete (SC) and selective SC dropout medium lacking an amino acid or nucleotide base] [63]. Deletion strains using the NAT antibiotic resistance gene in WT (BY4741) cells were created using standard LiOAc transformation procedures and with nourseothricin (100μg/ml) for selection on synthetic solid medium. For the creation of SECReTE mutant strains, SECReTE gene fragments were designed with the appropriate modifications, from the first to the last mutated base, and synthesized either as a gBlock (Integrated DNA Technologies, Inc., Coralville, IA, USA) or cloned into a pUC57-AMP vector (Bio Basic Inc.). Both (-)SECReTE and (+)SECReTE strains were generated. SUC2(-)SECReTE, SUC2 (+)SECReTE and CCW12(-)SECReTE strains were constructed in the BY4741 background genome using the delitto perfetto method for genomic oligonucleotide recombination [64], in which the CORE cassette from pGKSU [64] was integrated first into the genomic region corresponding to site of the SECReTE gene fragment. The CORE cassette contains the URA3 selection marker with an I-SceI homing endonuclease site and a separate inducible I-SceI gene. The SECReTE gene fragment for CCW12(-)SECReTE was amplified from the synthetic gBlock using primer sequences containing 20 bases of homology to both the region outside of the desired genomic locus and the CORE cassette. The amplified SECReTE gene fragment subsequently replaced the CORE cassette in the desired genomic site through an additional step of integration. CRISPR/Cas9 was utilized instead to generate the HSP150 mutant strains. HSP150(-)SECReTE and HSP150(+)SECReTE were created in the BY4741 genome. The CRISPR/Cas9 procedure involved deletion of the native genomic region corresponding to the SECReTE gene fragment, using the NAT cassette from pFA6-NatMX6. A CRISPR/Cas9 plasmid vector was designed to express the Cas9 gene, a guide RNA that targets the NAT cassette, and the LEU2 selection marker. The CRISPR/Cas9 plasmid was co-transformed with the amplified SECReTE gene fragment to replace the NAT cassette. Standard LiOAc-based protocols were employed for transformations of plasmids and PCR products into yeast. Transformed cells were then grown for 2–4 days on selective media. Correct integrations were verified at each step using PCR and, at the final step, accurate integration of the (-)SECReTE or (+)SECReTE sequences was confirmed by DNA sequencing.
RNA was extracted and purified from overnight cultures using a MasterPure Yeast RNA Purification kit (Epicentre Biotechnologies). For each sample, 2μg of purified RNA was treated with DNase (Promega, Madison, WI, USA) for 2hrs at 37°C and subjected to reverse transcription (RT) using Moloney murine leukemia virus RT RNase H(-) (Promega) under the recommended manufacturer conditions. Primer pairs were designed, using NCBI Primer-Blast [65], to produce only one amplicon (60-70bp). Standard curves were generated for each pair of primers and primer efficiency was measured. All sets of reactions were conducted in triplicate and each included a negative control (H2O). qRT-PCR was performed using a LightCycler480 device and SYBR Green PCR Master Mix (Applied Biosystems, Waltham, Massachusetts, USA). Two-step qRT-PCR thermocycling parameters were used as specified by the manufacturer. Analysis of the melting curve assessed the specificity of individual real-time PCR products and revealed a single peak for each real-time PCR product. The ACT1 or UBC6 RNAs were used for normalization and fold-change was calculated relative to WT cells.
Drop test assays were performed by growing yeast strains in YPD medium to mid-log phase and then performing serial dilution five times (10-fold each) in fresh medium. Cells were spotted onto plates with different conditions and incubated for 48hrs, prior to photo-documentation. Calcofluor White (CFW) or Hygromycin B (HB) sensitivity was tested by spotting cells onto YPD plates containing either 25μg/ml HB or 50μg/ml CFW (dissolved in DMSO, prepared as described) [66], following the protocol as mentioned above.
For the induction of Hsp150 secretion, strains were grown in YPD overnight at 26°C, diluted in YPD medium to 0.2 O.D.600 units, and then incubated at 37°C and grown until log-phase. For GFP secretion, yeast were grown O/N to 0.2 O.D.600 at 30°C in synthetic selective medium containing raffinose as a carbon source, then diluted to 0.2 O.D.600 units in YP-Gal and grown to mid-log phase (0.6–0.8 O.D.600) at 30°C. Next, 1.8ml of the culture was taken from each strain and centrifuged for 3mins at 1900 x g at room temperature. Trichloroacetic Acid (100% w/v) protein precipitation was performed on the supernatant and protein extraction, using NaOH 0.1M, was performed on the pellet [67]. Samples were separated on SDS-PAGE gels, blotted electrophoretically onto nitrocellulose membranes, and detected by incubation with rabbit anti-Hsp150 [1:10,000 dilution; gift from Jussi Jäntti (VTT Research, Helsinki)] or monoclonal mouse anti-GFP (Roche Applied Science, Penzberg, Germany) antibodies followed by visualisation using the Enhanced Chemiluminescence (ECL) detection system with anti-rabbit peroxidase-conjugated antibodies (1:10,000, Amersham Biosciences). Protein markers (ExcelBand 3-color Broad Range Protein Marker PM2700, SMOBiO Technology, Inc., Hsinchu, Taiwan) were used to assess protein molecular mass.
Invertase secretion was measured as described previously [68]. Cell preparation for the invertase assay was performed as described in [69]. The protocol was optimized based on previous work [70]. Internal and external activities were expressed in units based on absorption at 540 nm (1 U = 1 μmol glucose released/min per OD unit).
Yeast cells expressing Sec63-GFP were grown to mid-log phase and shifted to low glucose–containing medium [0.1% glucose] for 1.5 h to induce SUC2 expression. Cells were fixed in the same medium upon the addition of paraformaldehyde (4% final concentration) and incubated at room temperature (RT) for 45min with rotation. Cells were gently washed three times with ice cold Buffer B (0.1M potassium phosphate buffer, pH 7.5 containing 1.2M sorbitol), after which cells were spheroplasted in 1ml of freshly prepared spheroplast buffer [Buffer B supplemented with 20mM ribonucleoside vanadyl complexes (Sigma-Aldrich, St. Louis, MO), 20mM β-mercaptoethanol, and lyticase (Sigma-Aldrich, St. Louis, MO) (25 U per O.D.600 unit of cells)] for 10min at 30°C. The spheroplasts were centrifuged for 5min at 1300 × g at 4°C and washed twice in ice cold Buffer B. Spheroplasts were then resuspended in Buffer B and approximately 2.5 O.D.600 units of cells were placed on poly-L-lysine coated coverslips in 12-well plates and incubated on ice for 30 min. Cells were carefully washed once with Buffer B, then incubated with 70% ethanol for several hours to overnight at -20°C. Afterwards, cells were washed once with SSCx2 (0.3M sodium chloride, 30mM sodium citrate), followed by incubating with Wash buffer (SSCx2 with 10% formamide), for 15 min at room temperature (RT; ~23°C). Next, 45μl of hybridization buffer (SSCx2, 10% dextran sulfate, 10% formamide, 2mM ribonucleoside vanadyl complexes, 1mg/ml E. coli tRNA, and 0.2mg/ml BSA) containing 250nM of the TAMRA-labeled Stellaris probe mix for native SUC2 or HSP150 (Biosearch Technologies, Novato, CA)) was placed on parafilm in a hybridization chamber. Coverslips with the immobilized cells were placed face down on top of the hybridization buffer and were incubated overnight at 37°C in the dark. After probe hybridization, cells were incubated twice in Wash buffer for 15min at 37°C. Cells were then washed once with SSCx2 containing 0.1% Triton X-100 and incubated with SSCx2 supplemented with 0.5μg/ml DAPI for 1min at RT and finally washed with SSCx2 for 5min at RT. Cells were mounted with Prolong Glass (Thermo Scientific) mounting media on clean microscope slides. Samples were imaged using a Zeiss AxioObserver Z1 DuoLink dual camera imaging system equipped with an Illuminator HXP 120 V light source, PlanApo 100× 1.4 NA oil immersion objective, and Hamamatsu Flash 4 sCMOS cameras. Incremental (0.2μm) z-stack images were taken using a motorized XYZ scanning stage 130x100 PIEZO and ZEN2 software at 0.0645μm/pixel. Images were processed by deconvolution. At least 50 cells each showing both cER and nER labeling, as well as mRNA spots, were scored for ER localization per cell type [e.g. native, (+)SECReTE, and (-)SECReTE] examined. Scoring of mRNA granules (spots) was performed using the FISH-quant program [71] (https://bitbucket.org/muellerflorian/fish_quant) to analyze deconvolved images of single cells. mRNA co-localization to cER or nER was scored manually and was defined by overlap between the deconvolved signals.
Yeast cells were grown to mid-log phase (O.D.600 = 0.6–0.8) and shifted to low glucose–containing medium (0.1% glucose) for 1.5hrs to induce SUC2 expression. Cultures (400ml) were centrifuged for 3min at 500 × g, resuspended in a buffer solution containing 50mM Tris (pH 7.6), 150mM NaCl, 200U of RNasin Ribonuclease Inhibitor (Promega, Madison, WI), Complete Protease Inhibitor Cocktail (Roche Diagnostics, Basel, Switzerland), and cycloheximide 100μg/ml, and disrupted using glass beads and vortexing for 10min at 4°C. Crude lysates were centrifuged for 10min at 1000 × g to remove cell debris, and then 1ml of each lysate was subjected to ultracentrifugation for 1 h at 48,000 × g. The resulting pellet was then resuspended in 500μl of buffer containing 50mM Tris (pH 7.6), 150mM NaCl, 80U/ml RNasin Ribonuclease Inhibitor, Complete Protease Inhibitor Cocktail, and 100μg/ml cycloheximide. Next total RNA from both the membrane fraction (resuspended pellets) and the cytosolic fraction (supernatants) was isolated using the MasterPure Yeast RNA Purification Kit (including DNase I treatment) according to the manufacturer's recommendations.
To calculate the SECReTE count for each gene, the length of an uninterrupted run of NNY triplet repeats was counted separately for each of the three different codon positions (i.e. YNN, NYN, NNY, where N is any nucleotide and Y is a pyrimidine) for every gene. A SECReTE motif defined by the number of NNY triplet repeats where the threshold is ≥10 (see Results). Note that a gene may have more than one SECReTE motif along its length, each potentially in a different frame (relative to the coding sequence). The SECReTE count is defined as the number of SECReTE motifs present in a transcript.
For permutation analysis, the codon order of each gene sequence was randomly shuffled 1,000 times to evaluate the statistical significance of the SECReTE count in random sequences with unchanged codon usage. A Z score was calculated for each motif according to the formula: Z = (Observed–mean)/STD. Observed is the length of the motif in the real gene sequence. Mean is the average SECReTE length for all shuffled sequences of the gene. STD is the standard deviation of the SECReTE length from all shuffled sequences of the gene.
In order to differentiate between SECReTE motifs that follow an NNY pattern and those that are simply pyrimidine-rich, an RRY [R (purine); G or A)] score was calculated for each defined motif. Since poly-Y conforms to NNY, RRY was used instead. The score is calculated for every motif by representing each nucleotide of the motif that follows the RRY pattern as ‘1’ and each nucleotide that defies this pattern as ‘0’, and taking the average score over the entire length of the motif. The score is calculated for all three possible frames and the maximal score between them is taken for each motif. The UTR sequences for the yeast genome were obtained from: https://downloads.yeastgenome.org/sequence/S288C_reference/.
A motif search was performed by MEME suites [73] (http://meme-suite.org/tools/meme) to identify RNA motifs in genes encoding cell wall proteins in yeast.
|
10.1371/journal.ppat.1006991 | Cross-sectional analysis of CD8 T cell immunity to human herpesvirus 6B | Human herpesvirus 6 (HHV-6) is prevalent in healthy persons, causes disease in immunosuppressed carriers, and may be involved in autoimmune disease. Cytotoxic CD8 T cells are probably important for effective control of infection. However, the HHV-6-specific CD8 T cell repertoire is largely uncharacterized. Therefore, we undertook a virus-wide analysis of CD8 T cell responses to HHV-6. We used a simple anchor motif-based algorithm (SAMBA) to identify 299 epitope candidates potentially presented by the HLA class I molecule B*08:01. Candidates were found in 77 of 98 unique HHV-6B proteins. From peptide-expanded T cell lines, we obtained CD8 T cell clones against 20 candidates. We tested whether T cell clones recognized HHV-6-infected cells. This was the case for 16 epitopes derived from 12 proteins from all phases of the viral replication cycle. Epitopes were enriched in certain amino acids flanking the peptide. Ex vivo analysis of eight healthy donors with HLA-peptide multimers showed that the strongest responses were directed against an epitope from IE-2, with a median frequency of 0.09% of CD8 T cells. Reconstitution of T cells specific for this and other HHV-6 epitopes was also observed after allogeneic hematopoietic stem cell transplantation. We conclude that HHV-6 induces CD8 T cell responses against multiple antigens of diverse functional classes. Most antigens against which CD8 T cells can be raised are presented by infected cells. Ex vivo multimer staining can directly identify HHV-6-specific T cells. These results will advance development of immune monitoring, adoptive T cell therapy, and vaccines.
| This paper deals with the immune response to a very common virus, called human herpesvirus 6 (HHV-6). Most people catch HHV-6 in early childhood, which often leads to a disease known as three-day fever. Later in life, the virus stays in the body, and an active immune response is needed to prevent the virus from multiplying and causing damage. It is suspected that HHV-6 contributes to autoimmune diseases and chronic fatigue. Moreover, patients with severely weakened immune responses, for example after some forms of transplantation, clearly have difficulties controlling HHV-6, which puts them at risk of severe disease and shortens their survival. This can potentially be prevented by giving them HHV-6-specific "killer" CD8 T cells, which are cells of the immune system that destroy body cells harboring the virus. However, little is known so far about such T cells. Here, we describe 16 new structures that CD8 T cells can use to recognize and kill HHV-6-infected cells. We show that very different viral proteins can furnish such structures. We also observe that such T cells are regularly present in healthy people and in transplant patients who control the virus. Our results will help develop therapies of disease due to HHV-6.
| Human herpesvirus 6 (HHV-6) may be among the most prevalent persistent viruses in the human population. Antibodies to HHV-6 are present in 95–100% of healthy adults [1,2]. Like other herpesviruses, HHV-6 establishes a lifelong infection. HHV-6 is a group of two virus species known as HHV-6A and HHV-6B. Primary infection with HHV-6B, the more widespread species of the two, usually occurs before two years of age, and often causes a common childhood disease known as three-day fever or exanthema subitum [3,4]. The first infection with HHV-6A is thought to occur later and appears mostly asymptomatic [5].
Later in life, HHV-6 may be involved in a variety of diseases. HHV-6A is suspected of contributing to the pathogenesis of thyreoiditis Hashimoto [6] and to neuroinflammatory diseases such as multiple sclerosis [7]. HHV-6B is related to severe complications in immunocompromised patients. After allogeneic hematopoietic stem cell transplantation (allo-HSCT), HHV-6B reactivation is associated with increased all-cause mortality, delayed engraftment, graft-versus-host disease, and damaging infection of the central nervous system [8,9]. Since no HHV-6-specific antiviral agents are available, treatment of infection after allo-HSCT usually involves drugs approved for use against cytomegalovirus (CMV), but these come along with significant side effects such as kidney failure or bone marrow depression [5]. A potentially more efficacious and tolerable form of therapy aims at restoring antiviral T cell immunity, which is defective in patients who reactivate HHV-6 [10]. For other viral infections after allo-HSCT, many clinical investigations have shown that adoptive transfer of donor-derived virus-specific T cells is safe and effective [11]. Most of these studies focused on the herpesviruses CMV and Epstein-Barr virus (EBV), but some have recently included HHV-6-specific T cells [12].
Further development of such immunotherapies and of HHV-6 vaccines will require a detailed understanding of the virus-specific T cell response in health and disease. Information on HHV-6-specific T cell responses is still limited, in particular regarding CD8 T cells [13]. It was shown early that healthy virus carriers have CD4 T cells that respond to HHV-6 lysate or infected cells [14,15]. Target antigens and epitopes of the specific CD4 T cell response were identified first in a study on six selected structural proteins [16], and more recently by a proteomic approach that has identified ten viral antigens targeted by CD4 T cells [17]. Information on the targets of CD8 T cells has remained much more limited. Responses to five HHV-6B proteins have been investigated so far, and a number of epitopes from these proteins that are presented by infected cells were identified [18–21]. These proteins were chosen because of their (mostly distant) homology to CMV proteins that elicit CD8 T cell responses. However, HHV-6B encodes approximately 98 unique proteins [22], and the hypothesis remains unproven that T cell responses to HHV-6 and CMV are similarly structured or directed to corresponding antigens. The biological differences between these viruses are significant despite their evolutionary relationship as β-herpesviruses, and widespread cross-reactivity of T cells to HHV-6 and CMV seems unlikely considering that most of their proteins have quite divergent sequences [21]. Individual HHV-6 epitope-specific CD8 T cell responses were described to be of low frequency in peripheral blood [18–21], and it has remained unknown whether stronger responses exist.
These open questions prompted us to devise a method to analyse the CD8 T cell response to HHV-6 in a more comprehensive, cross-sectional fashion. Screens with libraries of peptides have been particularly efficient in obtaining copious information on the CD8 T cell repertoire against complex viruses [23–25]. However, due to the large number of possible targets, each such study has necessarily neglected some aspects of analysis, either regarding antigen coverage, HLA allotype coverage, precision of epitope identification, or verification of T cell function in the context of infection. Since detection of ex vivo responses to artificial peptides is not sufficient to prove the presence of T cells that recognize functional viral epitopes [26], it is of particular importance to verify recognition of infected cells by individual peptide-specific T cells.
To obtain a cross-sectional overview of the truly functional repertoire of HHV-6B-specific CD8 T cells and their target antigens and epitopes, we chose to base our approach on the entirety of HHV-6B proteins, but to focus on only one HLA class I allotype. We considered HLA-B*08:01 to be particularly suitable for such a study, because of the clarity of its peptide anchor motif [27] and its tendency to present dominant CD8 T cell epitopes in human viral infections [23,28–30]. To verify T cell specificity and function, we established specific T cell clones wherever possible, and used these to verify HLA restriction and recognition of infected cells. Our results show that the HHV-6-specific CD8 T cell repertoire targets multiple epitopes from all phases of the viral life cycle. We identify potent epitopes and track them in patients. We discuss implications for improved immune monitoring, studies of viral pathogenesis, and immunotherapy designs.
We wished to obtain a cross-sectional overview of HHV-6B antigens targeted by CD8 T cells. The reference sequence for HHV-6B strain Z29 contains 98 unique protein-coding genes or annotated ORFs with a total of 43,836 amino acids. For reasons of feasibility, we decided to screen the viral proteome for specific T cells with only one representative HLA class I restriction. We chose HLA-B*08:01, the second most frequent HLA-B allotype in populations of European origin [31,32]. We had two more reasons for this choice. First, T cell responses to HLA-B*08:01-restricted viral epitopes are often among the strongest that are observed in a particular virus. Examples of such epitopes are shown in Table 1. Second, B*08:01-presented peptides recognized by such T cells mostly conform to a clear-cut consensus motif [27,33,34]. This motif demands basic anchor residues (arginine or lysine) in positions 3 and 5 and an aliphatic residue (leucine, isoleucine, valine, or methionine) in the C-terminal position of an octameric or nonameric peptide. The HHV-6B reference sequence (strain Z29, GenBank NC_000898) contains 146 octameric and 153 nonameric peptides with this B*08:01 epitope motif, and 77 of 98 nonredundant ORFs contained at least one candidate (see Supporting S1 Table for a full list). These peptides were synthesized and used in the following experiments.
First, we attempted to determine the frequency of T cells specific for these HHV-6B-derived peptides in peripheral blood of healthy carriers. We stimulated PBMCs ex vivo with pools of the 146 octameric or the 153 nonameric peptides in an IFN-γ ELISPOT assay. Responses to HHV-6B peptide pools were much weaker than those to an EBV peptide pool, and generally below 1 / 10 000 PBMCs (Fig 1A).
Therefore, we decided to enrich HHV-6B-specific T cells from peripheral blood by peptide stimulation. PBMCs from four B*08:01-positive healthy HHV-6B carriers were initially stimulated with octamer or nonamer peptide mixes, and then restimulated every week with autologous CD40-activated B cells loaded with the same peptide mixes. Fig 1B shows analyses of such cultures from donor 1 after six to eight weeks of cultivation. Specific reactivity was observed against five subpools of the octameric peptides and at least seven subpools of the nonameric peptides, suggesting the presence of T cells specific for at least twelve HHV-6B peptides in this donor.
After six to eight weeks of cultivation, limiting dilution of the peptide-stimulated cultures was performed to generate T cell clones. Between 6% and 23% of T cell clones were specific for the HHV-6B peptide pool that was used for expansion (Table 2), as demonstrated by their specific IFN-γ secretion in response to peptide-loaded B cells (Fig 1C). Most of these clones could be sufficiently expanded to determine their precise peptide specificity by testing with peptide subpools (Fig 1D) and individual peptides in IFN-γ ELISA assays. Collectively, T cell clones recognized 25 HHV-6B peptides from 19 proteins or open reading frames (Fig 2).
For seven specificities, we verified restriction through HLA-B*08:01 by tests with B*08:01-transfected, peptide-loaded 293T cells and, for comparison, with peptide-loaded B*08:01-matched B cells. The very clear patterns of IFN-γ secretion indicated that all T cell clones tested were restricted through HLA-B*08:01, as shown for four clones in Fig 3.
We analyzed whether specific T cell clones were able to recognize their cognate antigen on HHV-6B-infected cells. Primary CD4 T cells from B*08:01-positive donors were activated with phytohemagglutinin (PHA) and infected with HHV-6B strain HST. Infected cultures were combined with peptide-specific CD8 T cell clones to test for specific IFN-γ secretion. T cell clones with 17 peptide specificities could be tested. For 13 of these, we observed specific recognition of HHV-6B infection at day 6, as shown in Fig 4A and summarized in Fig 2. Since most of the HHV-6B peptides recognized by T cells were fully or closely homologous to corresponding sequences in HHV-6A (Fig 2), we also tested the reactivity of T cell clones with six specificities against HHV-6A-infected cells. All six T cell clones recognized infected cells (Fig 4B). For three epitopes, we could demonstrate presentation by both HHV-6B-infected and HHV-6A-infected cells. Three additional specificities could only be tested against HHV-6A, due to limited cell numbers. Four of the six HHV-6A epitopes were identical to their HHV-6B counterparts, two differed in only one conservatively exchanged amino acid. Overall, these experiments demonstrated that 16 of the 25 candidate peptides against which T cell clones could be established (Fig 2) were bona fide epitopes processed and presented by cells infected with HHV-6B or 6A. Four candidates were not recognized, and five candidates could not be tested because T cell clones did not sufficiently expand and survive. We also tested cytotoxic reactivity against HHV-6B-infected target cells, focusing on CD8 T cells specific for the DFK peptide from U86. Both a DFK-specific CD8 T cell clone and a polyclonal CD8 T cell line, obtained by PBMC stimulation with the peptide DFK, displayed strong cytotoxic activity against HHV-6B-infected cells, but not non-infected cells; HLA-B8 expression of the target cells was required for this activity (Fig 4C–4E).
We proceeded to analyze recognition of target cells over a period of 12 to 18 days of infection with HHV-6B (Fig 5). Some T cell clones reached a maximum of reactivity at three days of infection, others at six days. Timing of maximal recognition did not appear to correlate with the described expression kinetics of HHV-6B antigens. For example, the SPR epitope from immediate-early (IE) antigen U86 was maximally recognized on day 3, but other IE antigens [44] such as U79 and B4 reached a maximum of recognition on day 6. Presumably, time to completion of antigen processing differed between antigens, or potential secondary cycles of virus production and re-infection within the infected CD4 T cell culture augmented the presentation of some antigens to specific CD8 T cells.
We performed time-course T cell recognition assays with T cell clones of additional specificities, including an analysis of recognition of HHV-6A and an additional control condition in the presence of ganciclovir, an inhibitor of HHV-6 replication (Fig 6). Presentation of various IE, E, and L antigens showed distinct peaks of recognition on days 3, 6, or 9. Recognition of all epitopes was partially or fully inhibited by ganciclovir.
Taken together (Fig 2), a majority of the HHV-6B peptide-specific T cell clones tested against infected cells recognized their endogenously processed target epitope in the context of infection. Epitopes from antigens of all kinetic categories (IE, early, late) and of diverse functional roles (regulation, DNA synthesis, virus assembly, structural proteins) were presented by infected cells. No particular class of antigens appeared to be excluded from presentation. At least three epitopes were from proteins with unknown function or putative proteins; our results provide evidence that ORFs including U7, U26, and B4 are translated in infected cells. Multiple epitopes were found in antigens U38 (the DNA polymerase), U41 (the major DNA-binding protein), and U86 (the transcriptional regulator IE2).
A panel of thirteen HLA-B*08:01/peptide multimers (dextramers) was commercially synthesized. For inclusion in this panel, 11 of the 16 epitopes in their HHV-6B variants were arbitrarily chosen. For two of these epitopes, multimers loaded with their variant HHV-6A peptide were also synthesized; these included the HHV-6A variant of EGR (from U79) and the "EFK" variant of DFK (from U86; compare Fig 2). As a positive control, a multimer for the Epstein-Barr virus epitope RAKFKQLL (RAK) from the BZLF1 antigen was synthesized in parallel. All these multimers were used to stain PBMCs from eight healthy donors for analysis in flow cytometry, to determine ex vivo frequencies of HHV-6-specific CD8 T cells (Fig 7). As the examples in Fig 7A show, T cells that bound a multimer DFK/B*08:01, carrying the DFK peptide from U86, often had an elevated frequency and appeared as clearly distinct populations. T cells that bound other HHV-6 multimers were usually much less frequent. Overall, DFK-specific T cells were detectable in 7 of 8 healthy donors, with a median frequency of 0.09% of CD8+ T cells (0.005%– 1.11%). The second most frequent population were T cells specific for the SPR epitope, also from U86 (median 0.025% of CD8+ T cells; 0.007%– 0.07%). Thus, while T cells specific for many HHV-6 epitopes were in most donors not detectable above a standard baseline of 0.01%, we identified an HHV-6B epitope, DFK, that regularly allowed clear ex vivo detection of specific T cells by multimer staining. In contrast, T cells specific for the HHV-6A variant of this epitope, EFK, were of low frequency or absent. DFK-specific T cells displayed a mixed phenotype ex vivo with respect to markers of central memory, effector memory or terminal differentiation (S1 Fig).
The median number per donor of different HHV-6 epitope specificities with a frequency higher than 0.01% was four (Fig 7C), and the highest number was seven. However, this number is likely to underestimate the overall number of specificities including those of lower frequency that are present in a donor, considering that the number of different specificities in donor 1 and 2 that could be obtained as T-cell clones after specific expansion was 17 and 12, respectively (Table 2). There were three HHV-6 epitopes that elicited responses higher than 0.01% in more than half of the donors (Fig 7D). A complete set of FACS plots is provided as supporting information (S2 and S3 Figs).
We analyzed the frequency of HHV-6-specific multimer staining-positive CD8 T cells in peripheral blood of three patients after HLA-B*08:01-positive allo-HSCT from unrelated HLA-matched donors. Patient 1 had detectable HHV-6 in throat swabs at the time of transplantation. In the third to fifth week after transplantation, while in aplasia, this patient underwent an episode of HHV-6 reactivation, detectable in gastric biopsy and throat swabs, with symptoms of skin rash and nausea. Treatment with foscarnet was initiated. At day +54, EBV reactivation was detected, and was treated with cidofovir and rituximab. Probably due to viral infection, engraftment was delayed until day +105. Samples were available for analysis of specific T cells on days +57 and +68, at a time when HHV-6 reactivation had subsided (Fig 8A). DFK-specific T cells were detected at both times at similar levels, while QTR-specific T cells were increasing (Fig 8A and 8B).
Patient 2 showed HHV-6 reactivation at day +29 after allo-HSCT with detection of the virus in bronchoalveolar lavage (BAL), performed due to a CT scan showing pneumonia. A concurrent infection with Aspergillus fumigatus was found, as well as EBV and adenovirus reactivation. Patient 2 developed a histologically proven post-transplant lymphoproliferative disorder (PTLD) at day +94, and treatment with cidofovir and rituximab was performed. HHV-6-specific T cells targeting four of four different epitopes were detected in patient 2 at moderate frequencies in two of two samples after resolution of HHV-6 reactivation (Fig 8C).
Patient 3, who suffered from severe aplastic anemia, was admitted to allo-HSCT with ongoing detection of HHV-6 in throat swabs after immunosuppressive treatment consisting of anti-thymocyte globulin (ATG), corticosteroids, and cyclosporine A. No specific HHV-6-related symptoms were observed. During aplasia, a concurrent enteral adenovirus reactivation occurred, and progressed to disseminated adenovirus disease. The patient was treated with cidofovir and adenovirus-specific T cells, and fully recovered. Viral infection probably contributed to delayed engraftment. HHV-6-specific T cells could be analyzed in an early sample concurrent with ongoing HHV-6 reactivation (day +56) and a late sample (day +1221). Only DFK and the EBV epitope RAK could be studied in the early sample due to a shortage of material. DFK-specific CD8 T cells were absent at the time of reactivation (Fig 8D and 8E), but were well reconstituted at the late time point (Fig 8D, 8F and 8G). In addition, there was evidence for low-frequency establishment of CD8 T cells specific for various other HHV-6 epitopes (Fig 8F and 8G) at the late time point in this patient, who has remained alive and well until now.
Taken together, these data provide tentative evidence that reconstitution of CD8 T cells specific for HLA-B*08:01-restricted HHV-6 epitopes, notably DFK, may be associated with control of viral reactivation in patients after allo-HSCT. However, all patients were treated with cidofovir, which has high activity against HHV-6. Studies in larger patient cohorts will be necessary to establish an association of particular HHV-6 T-cell specificities and control of infection.
Our study identified a total of 16 HLA-B*08:01-restricted epitopes that were presented by infected cells to specific CD8 T cells, based on a set of 299 peptides as epitope candidates. This test set consisted of all HHV-6B peptides that conformed to a simplified HLA-B*08:01 motif (Table 1) defined by the presence of three anchor residues, while any amino acid was allowed in other positions of the peptide. To find out if other internal or flanking sequences were non-randomly enriched for preferred residues or motifs, we aligned our 16 epitopes and their flanking regions in their proteins of origin (Fig 9A) and analyzed their amino acid content in each position, subdividing amino acids into broad categories according to their chemical characteristics (Fig 9B).
In our set of 16 confirmed epitopes, there were nine arginines and seven lysines each in anchor positions N3 and N5, suggesting that there was no strong preference for either of these two. Each of the four permitted aliphatic residues was found in the C-terminal anchor position (C1) of the nonameric epitopes, with a preference for leucine, which may simply mirror the higher frequency of this amino acid in the viral proteome (L, 10.1%; I, 6.4%; V, 6.2%; M, 2.4% in the HHV-6B GenBank reference sequence NC_000898). All six octamers had a leucine in C1. A tendency for leucine to be enriched in N7 was noted. Other than that, there was no strong enrichment of particular amino acids within the peptide other than in the three pre-defined anchor positions, and at least five of the six chemical categories were represented in each internal non-anchor position.
Somewhat more conspicuous patterns were seen in the regions flanking the peptide. N2' was often an uncharged polar amino acid (C, S, T, N, Q) or a basic amino acid (R, K, H), C1' was often serine or another uncharged polar amino acid, and C2' was often a basic amino acid. We calculated the likelihood that such enrichments occurred by chance using Fisher's exact test, comparing the 16 epitopes to the rest of the 299 peptide candidates (Table 3). The lowest probabilities of enrichment by chance were calculated for uncharged polar or basic amino acids in position N2' (p = 0.0010), serines in C1' (p = 0.0016), polar uncharged amino acids in C1' (0.0006), and lysine in C2' (p = 0.0013). Thus, the strongest tendency in HLA-B*08:01-restricted T cell epitopes to follow conserved motifs (apart from the three pre-defined anchor residues) was not found for peptide-internal positions, but for certain flanking positions.
PBMCs from anonymized healthy adult donors were purchased from the Institute for Transfusion Medicine, University of Ulm, Germany. PBMCs from patients after allo-HSCT were obtained at the Department of Internal Medicine III, Hematopoietic Stem Cell Transplantation, Klinikum der Universität München, Munich, Germany, with written informed consent. Anonymized cord blood samples were collected at the Department of Obstetrics and Gynecology (Klinik und Poliklinik für Frauenheilkunde und Geburtshilfe, Klinikum der Universität München, Munich, Germany). The institutional review board (Ethikkommission, Klinikum der Universität München, Munich, Germany) approved these procedures (project no. 071–06–075–06, project no. 17–455).
Standard cell culture medium was RPMI 1640 (Life Technologies/Invitrogen, Karlsruhe, Germany) supplemented with 10% FCS (Biochrom, Berlin, Germany), 100 U/ml penicillin, 100 µg/ml streptomycin (Life Technologies/Invitrogen), and 100 nM sodium selenite (ICN Biochemicals, Aurora, CO). 293T cells were cultivated in DMEM (Invitrogen) with the same supplements. Cells were all cultivated at 37°C and 5% CO2.
PBMCs were obtained by centrifugation on Ficoll/Hypaque (Biochrom). High-resolution HLA typing was performed by PCR-based methods (MVZ, Martinsried, Germany). HHV-6-specific, HLA-B*08:01-restricted T cell lines and clones were derived from four HHV-6 IgG-positive donors. Their full HLA class I types are as follows: donor 1, HLA-A*02:01, A*68:01, B*07:02, B*08:01, Cw*07:01, Cw*07:02; donor 2, HLA-A*01:01, B*08:01, B*15:01, Cw*03:03, Cw*07:01; donor 11, HLA-A*01, A*11, B*08, B*15:01, Cw*03:03, Cw*07:01; donor 12, HLA-A*01:01, A*02:01, B*08:01, B*40:01.
In healthy donors, HHV-6 IgG was determined by immunofluorescence test at Max-von-Pettenkofer Institute, Munich, Germany, with the exception of donors 3, 4, 7, and 8, in whom it was determined using the HHV-6 IgG ELISA kit (Abnova). Cell lines and cultures from these and other HLA-typed donors were used as antigen-presenting cells in T cell assays. Mini-lymphoblastoid cell lines (mLCLs) were generated by infection of PBMC with mini-Epstein-Barr viruses as described [47]. CD40-activated B-cell cultures were established as described [48] and maintained by weekly replating on irradiated (180Gy) LL8 stimulator cells in the presence of 2 ng/ml rIL-4 (R&D Systems). LL8 cells are murine L929 fibroblasts stably transfected with human CD40L [49]. Human embryonic kidney cells 293T (partial HLA type: HLA-A*02:01, B*07:02) were obtained from ATCC (CRL-11268).
HHV-6-specific T cells were analyzed in peripheral blood samples from three adult patients after allo-HSCT. Transplant indication was severe aplastic anemia (SAA) in patients 1 and 3, and acute myeloid leukemia (AML) in patient 2. G-CSF-mobilized peripheral blood stem cells from an HLA-matched unrelated donor were used in patients 1 and 2; bone marrow donated by an HLA-matched unrelated donor was used in patient 3. GvHD prophylaxis consisted of cyclosporine A plus sirolimus (n = 2) or mycophenolate mofetil (n = 1). All patients and donors were HLA-B*08:01-positive and CMV-seronegative. Patients received standard antiviral prophylaxis with acyclovir. Viral infection/reactivation was monitored weekly by quantitative PCR in peripheral blood, including HHV-6. Other specimens like stool, urine and throat swab samples were monitored for virus reactivation weekly on a routine basis as indicated. A detailed overview of the characteristics of patients, donors, and transplant procedures is provided (Supporting S2 Table).
Peptide sequences adhering to the HLA-B*08:01 anchor motif were extracted from the HHV-6B strain Z29 reference sequence (GenBank NC_000898) using the text editor Tex-Edit Plus and a script in the AppleScript language. The 299 peptides of the HLA-B*08:01 candidate library were synthesized by JPT (Berlin, Germany) in a "Research Track" format. Each peptide was analyzed by liquid chromatography–mass spectrometry. Median purity of peptides was 77%. Nineteen peptides had a purity below 50% (minimum 25.5%), none of these was recognized by any T cell clone. Peptides were reconstituted in 100% dimethyl sulfoxide (DMSO) and stored at –20°C. DMSO concentration in all T cell effector assays was kept below 0.1% (vol/vol).
PBMCs from HHV-6-positive donors were enriched for HHV-6B-specific T cells by stimulation with a mix of 146 octameric peptides or 153 nonameric peptides that represent HLA-B*08:01 candidate epitopes from HHV-6B, using a protocol employing autologous CD40-activated B cells. For peptide loading, PBMC (first stimulation) or CD40-activated B cells (all later stimulations) were coincubated with octamer or nonamer peptide pool (1 µg/ml for each peptide) at 37°C for 2 h, and washed three times with PBS.
The T cell stimulation protocol was initiated by peptide loading of PBMC, which were then plated at 5×106 cells in 2 mL per well of a 12-well plate. Ten to 14 days later, cells were pooled, counted using trypan blue staining, and restimulated at 3×106 cells in 2 mL medium per well with freshly irradiated (50 Gy) autologous CD40-activated B cells, previously loaded with peptides, to reach an effector:stimulator ratio of 4:1, in the presence of 25–50 U/mL recombinant IL-2 („Proleukin S“, Novartis). Cells were restimulated every following week with peptide-loaded CD40-activated B cells in the same manner, with the exception that the IL-2 concentration was successively increased to 100 U/mL. Between stimulations, the T cell cultures were expanded using fresh IL-2-containing culture medium as seemed necessary, judging from the visual appearance of the cultures. For cytotoxicity analysis, PBMCs were stimulated with a single peptide (DFK from U86) and autologous CD40-activated B cells in an analogous manner, and tested at day 29 of cultivation.
For single cell cloning of polyclonal T cell cultures, 0.7 or 2.5 T cells/well were seeded into 96-well round-bottom plates, together with 2×104/well irradiated (50 Gy) HLA-B*08:01-positive mini-LCLs loaded with the octameric or nonameric HHV-6B peptide pool, 3×105 cells/well of a mixture of irradiated (50 Gy) allogeneic PBMCs from three donors, and 1000 U/mL IL-2. Outgrowing T cell clones were expanded in 96-well round-bottom plates by restimulating every 2 weeks under equivalent conditions. Later, clones with known peptide specificity were restimulated in an analogous manner but using only the single specific peptide.
HLA/peptide multimers in the form of phycoerythrin-(PE)-labeled HLA-B*08:01/peptide dextramers were purchased from Immudex, Copenhagen, Denmark. Dextramers contained one of thirteen HHV-6 peptides or the peptide RAK (full sequence RAKFKQLL) from the BZLF1 protein from Epstein-Barr virus. Dextramers covered the epitopes EAR, RSK, FEK, QTR, VVK, NVK, MAR, whose peptide sequences are identical in HHV-6A and HHV-6B; the epitopes TNK, EGR-6B and DFK from HHV-6B, which differ from their HHV-6A counterparts in one to three amino acids; the epitopes EGR-6A and EFK from HHV-6A (EFK being the HHV-6A version of DFK); and the HHV-6B epitope SPR, which has no HHV-6A counterpart. The sequences of HHV-6 peptides are provided in Fig 2.
For quantification of antigen-specific CD8+ T cells in peripheral blood from healthy donors using dextramers, a median of 7x105 PBMCs per staining was treated as follows. Cells were stained for 10 minutes at room temperature with 1 μl PE-labeled HLA/peptide dextramer. For negative controls, cells were processed identically, but dextramer was not added. After washing with PBS supplemented with 2% FCS, cells were counterstained on ice for 15 minutes with anti-CD4-FITC (clone RPA-T4), anti-CD3-PE-Cy5 (clone HIT3a), and anti-CD8-APC (clone RPA-T8) antibodies (all BioLegend). Cells were then washed with PBS/FCS and resuspended in 1.6% formaldehyde (Carl Roth) in PBS for fixation, stored at 4°C, and analyzed within one day on a Becton Dickinson FACSCalibur flow cytometer. Data analysis was performed using FlowJo 9.5.3 software (Tree Star): lymphocytes were gated in a forward/sideward scatter dot plot, then CD3+CD4– cells were analyzed for the proportion of multimer-positive cells within CD8+ T cells.
For healthy donors 4, 6, 10 and transplantation patients, a variation of this protocol was used. Instead of anti-CD4-FITC, a "dump channel" mix of FITC-labeled antibodies anti-CD14 (clone TÜK4, Miltenyi Biotec), anti-CD19 (clone LT19, Miltenyi Biotec), and anti-TCR-γδ (clone 5A6.E9, Life Technologies) was used. Viable lymphocytes were gated according to forward/sideward scatter, and FITC-positive cells were excluded. For patient samples, only 3x105 PBMCs were used per staining.
A range of differentiation markers was analyzed on DFK-specific T cells in donor 3 and 6. Staining with dextramer DFK was combined with FITC-labeled anti-CD14 and anti-CD19 antibodies as above (dump channel) and CD3-Alexa Fluor 700 (clone HIT3a, BioLegend); additional antibodies in panel A were CD8-APC (clone RPA-T8), CCR7-PE-Cy7 (clone G043H7), and CD45RA-Pacific Blue (clone HI100; all BioLegend); additional antibodies in panel B were CD8-APC-H7 (clone SK1, BD), CD27-APC (clone O323, BioLegend), CD28-PE-Cy5 (clone CD28.2, BD Pharmingen), and CD57-Pacific Blue (clone HCD57, BioLegend).
Dot plots displaying flow cytometry data in Figs 7 and 8, S2 and S3 Figs span, in both dimensions, a range from 1 to 10000 arbitrary fluorescence units in a logarithmic scale. Data in S1 Fig are presented in a biexponential scale spanning a range from 10−3 to 105 arbitrary units in both dimensions.
To verify the HLA restriction of HHV-6B-specific T cell clones, 293T cells were transfected with a HLA-B*08:01 expression plasmid (kindly provided by Josef Mautner, Munich) by calcium phosphate precipitation. Twenty-four hours later, cells were harvested, washed with PBS, loaded with single peptides at 1 µg/ml, washed three times, and used as targets in T cell assays.
HHV-6B-specific T cell lines and T cell clones were analyzed for antigen-specific IFN-γ secretion in ELISA or ELISPOT assays. Effector cells (104, unless noted otherwise) were cocultivated overnight (16–18 h) with target cells (2x104, unless noted otherwise) in 200 μL per well of a 96 V-well plate at 37°C and 5% CO2. Then supernatants were harvested, and an IFN-γ ELISA was performed according to the manufacturer’s protocol (Mabtech, Nacka, Sweden).
IFN-γ ELISPOT assays were used to determine the frequency of specific T cells in freshly isolated PBMCs and polyclonal T cell lines. They were performed according to the reagent manufacturer’s protocol (Mabtech, Nacka, Sweden) in 96-well MultiScreen-HA plates (Millipore) in 200 μL medium per well, with an overnight incubation period of 16–18 hours at 37°C and 5% CO2. To analyze PBMCs, 250,000 cells were distributed to each well and directly loaded with antigenic peptide. To analyze T cell lines, autologous CD40-stimulated B cells were loaded with antigenic peptides, washed, and co-incubated at 5x104/well together with the T cells at 10,000 cells/well. Spots were developed using the AP Conjugate Substrate Kit from Bio-Rad. Spots were counted in an automated ELISPOT reader (CTL).
To determine the cytotoxic activity of HHV-6B-specific T cells against HHV-6B-infected cells, calcein release assays were performed. HHV-6B-infected target cells (see below) were loaded with Calcein AM (5 μg/ml, Molecular Probes) for 30 min at 37°C in standard medium. Cells were washed three times and resuspended in RPMI medium without phenol red supplemented with 5% FCS. Effector T cells were washed once and resuspended in the same medium. T cells and target cells were combined in V-bottom 96-well plates (200 μl total volume per well), with 5,000 target cells per well and 5,000–80,000 T cells per well (effector: target ratio 1:1 to 16:1), in four replicates of each condition. After 3.5 hours at 37°C and 5% CO2, supernatants (100 μl per well) were collected, and fluorescence was measured (excitation 485 nm, emission 535 nm). Specific lysis was calculated relative to maximal lysis (100%, targets incubated with 0.5% Triton X-100) and minimal lysis (0%, targets incubated in the absence of T cells).
HHV-6A (strain U1102) and HHV-6B (strain HST) were purchased from NCPV, UK, and serially propagated on phytohemagglutinine (PHA)-activated cord blood mononuclear cells. Fresh or cryoconserved cord blood cells at 2x106 cells in 2 ml per well of a 24-well plate were stimulated with 5 μg/ml PHA-M (Calbiochem). Three days later, cells were infected with virus suspension from previous passages (230 μl/well). After 5–7 days, when the cytopathic effect appeared maximal, cell cultures were harvested, cells were pelleted by centrifugation at 300 g for 10 min, and supernatants were stored in aliquots at -80°C.
Peripheral blood cells from adult donors with known HLA types were used to prepare HHV-6A/B-infected target cells for the analysis of T cell recognition. CD4 T cells were positively isolated from PBMC using anti-CD4-coupled paramagnetic beads (Miltenyi Biotec), and 2x106 CD4+ cells were activated in 2 mL per well of a 24-well plate using 5 μg/mL PHA. After 3 days, the cells were pooled, counted, replated at 2x106 cells/well, and infected with 230 μl/well of HHV-6A or HHV-6B virus stocks. Thereafter, infected T cell cultures were resupplied with fresh medium every 3 days on average. At different time points after infection, cells were used as targets for HHV-6-specific T cell clones in cytokine secretion assays. At every time point, infected cells were harvested, washed and counted in Trypan Blue solution immediately before they were combined with HHV-6-specific T cells at constant numbers (104 effector T cells, 2 × 104 infected cells or control targets). In selected experiments as indicated, ganciclovir (20 μg/mL, Roche) was added immediately before infection.
Here we present a cross-sectional analysis of the CD8 T cell response to HHV-6, and an overview of antigens recognized by this response. For one exemplary HLA class I molecule, HLA-B*08:01, we identified candidate epitopes all across the HHV-6B proteome, and tested which of these represent bona fide epitopes. A large set of T cell clones was established to assess and correlate epitope specificity and antiviral reactivity with precision. Frequencies of specific T cells in healthy donors and allogeneic transplant patients were determined by multimer staining. A majority of peptides against which we could raise T cells were presented by infected cells, and epitopes from all classes of viral antigens were presented. Ex vivo frequencies of specific T cells were low for most epitopes. However, U86-specific T cells were readily detectable ex vivo in most donors and patients. U86 is thus a candidate for an immunodominant CD8 T cell antigen of HHV-6. Moreover, we describe the presence of HLA-B*08:01-restricted HHV-6-specific T cells in patients who were able to control episodes of HHV-6 reactivation after stem cell transplantation.
Taken together, the present work provides a cross-sectional overview of the structure of the HHV-6-specific CD8 T cell response at two levels. It shows that multiple viral antigens of different functional and kinetic classes furnish epitopes for T cell recognition; and it describes the quantitative contributions of the different specificities to the T cell repertoire, including identification of a prominent antigen.
Our study extends earlier investigations of the HHV-6-specific CD8 T cell response that were limited to the analysis of responses to five pre-chosen proteins: four virion proteins (U11, U14, U54, U71) and the IE-1 transactivator U90 [18–21]. The motivation to choose those antigens was their correspondence to immunogenic proteins of human CMV. The present work employed a method that was independent of such criteria and targeted CD8 T cell epitopes across the viral proteome. HLA-B*08:01-restricted epitopes were identified in 12 proteins of varied function and from all phases of the viral replication cycle. No epitope was derived from any of the five antigens studied earlier, although 12 candidates from these proteins were included in our analysis. This suggests that immunity to HCMV antigens has limited power to predict the specificity of CD8 T cell responses to HHV-6. We cannot exclude, however, that CD8 T cells that target additional epitopes, including such from the five proteins mentioned, may exist in the T cell repertoire. Of note, U86 attracted the strongest T cell responses among the antigens described here, and its CMV counterpart IE-2/UL122 is a strong CD8 T cell antigen [25]. This suggests that certain commonalities between recognition patterns of CMV and HHV-6 antigens may exist.
However, the overall composition and diversity of the HHV-6-specific CD8 T cell repertoire, as characterized here, appears to stand in marked contrast to the best-studied herpesviruses, CMV (a β-herpesvirus like HHV-6) and EBV (a γ-herpesvirus). Contrary to HHV-6, CMV elicits very large CD8 T cell responses, amounting to an average of 10% of the peripheral CD8 T cell repertoire of healthy carriers [25]. HLA-B*08:01-restricted T cells make a strong contribution to this response [23,50]. EBV-specific CD8 T cells account for a smaller proportion of total CD8 T cells in healthy donors [51], but, for example, the HLA-B*08:01-restricted RAKFKQLL epitope (Table 2) is recognized by a median of about 2% of CD8 T cells, and frequencies above 5% are no rarity [29,52]. CD8 T cell responses to CMV further increase in the elderly [53,54], and EBV-specific CD8 T cells are strongly elevated in patients with symptomatic primary EBV infection [55]. On the other hand, the diversity of the CD8 T cell response to CMV or EBV appears restricted. For example, the database IEDB [56] currently lists only three HLA-B*08:01-restricted epitopes from CMV and four from EBV, counting strain variants as one epitope. In CMV carriers, a median of eight out of 213 ORFs is recognized by CD8 T cells [25], and a majority of EBV antigens appear exempt from CD8 T cell recognition [51]. Thus, it appears that the diversity of epitopes and antigens available for presentation by infected cells is distinctly larger in HHV-6 than in the two paradigmatic human herpesviruses. However, it cannot be excluded that more diverse repertoires of low-frequency CD8 T cell specificities in EBV or CMV have so far escaped detection, possibly because their responses were masked by more dominant CD8 T cell populations.
In contrast, CD8 T cells specific for other herpesviruses such as varicella-zoster virus (VZV) or herpes simplex virus (HSV-1) are maintained at relatively low frequencies in healthy carriers [57–59]. In HSV-1, CD8 T cells appear to target multiple antigens from different phases of infection, whereas IFN-γ responses to individual epitope peptides ex vivo have frequencies of 1 in 104 PBMCs or lower. A large number of potential epitopes from HSV-1 were described, with up to 13 sharing the same HLA class I restriction [57], although it is not clear yet whether a majority of these is presented by infected cells. This structure of the T-cell repertoire appears comparable to the one described here for HHV-6. Less is known about the VZV CD8 epitope repertoire, but available data are compatible with a highly diverse repertoire which is in part shaped by cross-reactivity of CD8 T cells to HSV and VZV [60].
Potential reasons for differentially structured antiviral CD8 T cell repertoires may be sought in the patterns of cellular tropism of these viruses. CMV resides latently in monocytes and myeloid precursors and is reactivated upon their differentiation to dendritic cells [61], whereas EBV infects B cells in diverse activation states [62]. Infection of professional antigen-presenting cells by CMV and EBV may favour competitive clonal expansion and selection of immunodominant T cells into the repertoire [63,64]. In other herpesviruses, tropism for professional antigen-presenting cells is less predominant [5,65]—although HHV-6 was shown to infect monocytes and other antigen-presenting cells in vivo [5]. Also, it appears that the repertoire of viral immunoevasive molecules that directly interfere with steps in the HLA class I presentation pathway is larger for CMV or EBV [49,66,67] than for other herpesviruses [58,68]. Co-expression of many immunoevasive functions in CMV and EBV may limit the number of epitopes that escape such regulatory mechanisms [69,70], and this may lead to competitive advantage and immunodominance of T cells that recognize their epitopes in the context of infection.
Our analysis of T cell epitopes was limited to only one allotype, HLA-B*08:01, and extrapolations to the CD8 T cell repertoire in general must be made with caution. More comprehensive studies on the entire CD8 T cell repertoire to HHV-6 will be necessary to strengthen our present suggestions. However, available information on CD8 T cell responses to other complex viruses indicates that HLA-B*08:01, wherever studied, rarely fails to be an effective presenter of epitopes, as shown by the examples in Table 1.
The groove of MHC class I molecules accomodates peptides for presentation to CD8 T cells [71–73]. Particularly important for stable binding are certain anchor residues [74] whose side chains reach into dedicated pockets in the peptide-binding groove. Allelic variants of MHC class I demand anchor residues required for peptide binding that can differ in their chemical nature and their position in the peptide [27,33,74]. Our identification of HLA-B*08:01-restricted T cell epitopes consisted in a functional screen of all HHV-6B-derived peptides that contained a motif of three required anchor residues [27,75], as depicted in Table 1, while any amino acid was permitted in other positions of the peptide. Application of such a simple anchor-motif based algorithm (SAMBA) is supported by the observation that a majority of well-characterized, independently verified, and potent CD8 T cell epitopes from infectious pathogens perfectly adhere to this motif, whereas amino acid usage in all other positions is more variable (see Table 1 and the references therein). Full conformity to this motif was also shown for abundant self-derived peptides eluted from HLA-B*08:01 in a seminal study [75].
Subsequent studies have, however, increasingly identified B*08:01-binding self peptides that partially deviated from the motif [76,77]. In these cases, peptides were eluted from cells co-expressing several HLA class I molecules, and their HLA-B*08:01 restriction was retrospectively inferred from the peptide sequence. Deviation from the classical motif was also observed when predicted B*08:01-restricted epitopes were validated in ex vivo ELISPOT assays with blood cells loaded with peptide [78]. However, such approaches carry the risk of identifying responses to peptides that are not endogenously processed [26] or not presented by the predicted HLA allotype [69], if those aspects are not independently tested.
Prediction of MHC class I epitopes currently relies on machine-learning algorithms trained on ever increasing datasets of MHC binders or epitopes [79,80]. To the extent that such datasets may contain a growing number of candidates whose HLA restriction and qualities as epitopes have not been verified, further progress in predicting optimal epitopes may be difficult to achieve. Therefore, in our view it is as important as ever to rigorously verify HLA restriction and endogenous presentation, optimally with target cells infected with the pathogen of interest. Reliance on T cell clones increases the accuracy of epitope identification and validation, since this ensures that the very same T cells recognize peptide and infected cells, and minimizes the likelihood of accidental cross-reactivities. Peptide-based functional screens of epitope candidates ex vivo have been successful in identifying CD8 T cell reactivities that were later confirmed to be viral epitopes, for example in CMV [23,54], but such approaches are likely to be less robust when proportions of specific T cells are low, such as for HHV-6.
This study identified sixteen HLA-B*08:01-restricted HHV-6 epitopes–defined as peptides presented by infected cells–out of 299 candidates. We took advantage of this dataset to compare amino acid usage in internal and flanking sequences of epitopes and non-epitopes. In the C-terminal anchor position (C1), Leu appeared to be favoured among admitted aliphatic residues, clearly so in octameric epitopes. Leu was also enriched in the C2 position. Otherwise, no restrictions of amino acid usage in non-anchor positions in HHV-6 epitopes were apparent, which is in line with the idea of distinct functional roles for anchors and non-anchors, and retrospectively supported our use of a SAMBA approach.
However, stronger enrichment of certain amino acids was found in peptide-flanking positions (N2', C1', and C2'). The C terminus of most MHC I ligands is generated by the proteasome [81]. Cut site preferences of human proteasomes have been identified by in vitro digestion of model proteins [82–85], and coincide well with the requirements of many MHC I allotypes (such as HLA-B*08:01) to bind peptides with a bulky hydrophobic residue in the C-terminal position. Downstream of the cut site, amino acid preferences partially diverge between model proteins [82–85]. We found uncharged hydrophilic amino acids, particularly Ser, to be enriched in the C1' position (called P1' in analyses of proteasome function). Ser in this position was also enriched after degradation of HIV Nef by the constitutive proteasome [83] and of prion protein by the immunoproteasome [84]. Increased frequency of Arg [83,85] and depletion of bulky hydrophobic amino acids [82,83,85] also agreed with our findings, whereas enrichment of Ala or Pro [82–85] did not. Although limited in size, our dataset suggests an influence of amino acid identity in C1'/P1' on effective proteasomal processing of HHV-6 epitopes. In the C2'/P2' position, we observed basic amino acids to be enriched; no clear tendency in that regard is found in the literature [84,85].
Formation of the N terminus of MHC I ligands is in many cases a complex multistep process comprising proteasomal degradation, processing by cytosolic aminopeptidases, TAP-mediated transport to the ER, and final trimming by ER aminopeptidases [86]. Nontheless, an N-terminal processing motif of MHC I ligands could be defined [86]. Consistent with our findings, this motif has the basic amino acids Lys and Arg somewhat enriched in the N2' position [86]. Basic amino acids in N-terminal overhangs of MHC I ligand precursors may favour processing by cytosolic or ER aminopeptidases [86], although the ER peptidase ERAP1 does not appear to have this preference [87]. Moreover, basic amino acids close to the N terminus of MHC I ligand precursors may support effective TAP-mediated transport to the ER [88]. Thus, amino acid usage in regions flanking HHV-6 epitope peptides is compatible with some of the described N- and C-terminal MHC I processing preferences. However, such motifs represent tendencies rather than strict criteria, so improving epitope prediction by considering processing motifs remains difficult [79]. Nontheless, we speculate that simplified peptide-flanking motifs may be useful to design screening approaches that prioritize efficiency over completeness.
Identification of multiple CD8 target antigens and epitopes as undertaken here will advance immune monitoring and immunotherapy of HHV-6. Since our study identifies an epitope (DFK from U86) that allowed detection of specific T cells (sometimes at high frequencies of up to 1.1%) in 7/8 healthy carriers and 3/3 patients after allo-HSCT, multimer staining based on this epitope will be a convenient tool for monitoring and monospecific approaches to antiviral T cell therapy [89,90]. HHV-6-specific T cell transfer after allo-HSCT is attractive and feasible. In patients who received allo-HSCT, HHV-6 reactivation and disease is associated with a lack of virus-specific T cells [10,91] and the use of transplantation procedures that lead to imperfect T cell reconstitution [92]. In a first clinical application of HHV-6-specific T cell transfer to allo-HSCT patients, T cells specific for U11, U14, and U90 were part of a protocol that employed multivirus-specific peptide-stimulated T cells derived from the transplant donor [12]. In two patients, HHV-6 reactivation was cleared after transfusion of multivirus-specific T cells, in connection with an emergence of HHV-6-specific T cells in peripheral blood [12]. Partial remissions of HHV-6 infection were also observed in a third-party approach based on similarly prepared T cells [93]. These promising initial results encourage further application and development of HHV-6-specific adoptive immunotherapy. Since multiple epitopes are targeted by HHV-6-specific CD8 T cells, a multiepitope approach [94] may be particularly promising for selection of effective HHV-6-specific T cells for immunotherapy. If TCR-transgenic T cell therapy [95] is considered, HHV-6 antigens from diverse functional classes may be suitable targets.
|
10.1371/journal.pntd.0000620 | Modeling Disease Vector Occurrence when Detection Is Imperfect: Infestation of Amazonian Palm Trees by Triatomine Bugs at Three Spatial Scales | Failure to detect a disease agent or vector where it actually occurs constitutes a serious drawback in epidemiology. In the pervasive situation where no sampling technique is perfect, the explicit analytical treatment of detection failure becomes a key step in the estimation of epidemiological parameters. We illustrate this approach with a study of Attalea palm tree infestation by Rhodnius spp. (Triatominae), the most important vectors of Chagas disease (CD) in northern South America.
The probability of detecting triatomines in infested palms is estimated by repeatedly sampling each palm. This knowledge is used to derive an unbiased estimate of the biologically relevant probability of palm infestation. We combine maximum-likelihood analysis and information-theoretic model selection to test the relationships between environmental covariates and infestation of 298 Amazonian palm trees over three spatial scales: region within Amazonia, landscape, and individual palm. Palm infestation estimates are high (40–60%) across regions, and well above the observed infestation rate (24%). Detection probability is higher (∼0.55 on average) in the richest-soil region than elsewhere (∼0.08). Infestation estimates are similar in forest and rural areas, but lower in urban landscapes. Finally, individual palm covariates (accumulated organic matter and stem height) explain most of infestation rate variation.
Individual palm attributes appear as key drivers of infestation, suggesting that CD surveillance must incorporate local-scale knowledge and that peridomestic palm tree management might help lower transmission risk. Vector populations are probably denser in rich-soil sub-regions, where CD prevalence tends to be higher; this suggests a target for research on broad-scale risk mapping. Landscape-scale effects indicate that palm triatomine populations can endure deforestation in rural areas, but become rarer in heavily disturbed urban settings. Our methodological approach has wide application in infectious disease research; by improving eco-epidemiological parameter estimation, it can also significantly strengthen vector surveillance-control strategies.
| Blood-sucking bugs of the genus Rhodnius are major vectors of Chagas disease. Control and surveillance of Chagas disease transmission critically depend on ascertaining whether households and nearby ecotopes (such as palm trees) are infested by these vectors. However, no bug detection technique works perfectly. Because more sensitive methods are more costly, vector searches face a trade-off between technical prowess and sample size. We compromise by using relatively inexpensive sampling techniques that can be applied multiple times to a large number of palms. With these replicated results, we estimate the probability of failing to detect bugs in a palm that is actually infested. We incorporate this information into our analyses to derive an unbiased estimate of palm infestation, and find it to be about 50% – twice the observed proportion of infested palms. We are then able to model the effects of regional, landscape, and local environmental variables on palm infestation. Individual palm attributes contribute overwhelmingly more than landscape or regional covariates to explaining infestation, suggesting that palm tree management can help mitigate risk locally. Our results illustrate how explicitly accounting for vector, pathogen, or host detection failures can substantially improve epidemiological parameter estimation when perfect detection techniques are unavailable.
| Chagas disease is caused by Trypanosoma cruzi (Kinetoplastida: Trypanosomatidae), a parasitic protozoan transmitted through the feces of infected blood-sucking hemipterans (Reduviidae: Triatominae) [1],[2]. Human infection is endemic throughout Latin America, where it causes loses of more than 650,000 disability-adjusted life years annually [3]. From 1990, burden figures have declined by about 80% [3],[4], reflecting the success of Chagas disease control programs over vast geographical areas [5]. However, the burden of Chagas disease in the Latin American-Caribbean region is still consistently larger than the combined burden of malaria, leprosy, the leishmaniases, lymphatic filariasis, onchocerciasis, schistosomiasis, viral hepatitides B and C, dengue, and the major intestinal nematode infections [6],[7]. Because most transmission is mediated by household-infesting insect vectors, and because no effective treatment or vaccine are available for large-scale use, the elimination of domestic triatomines was defined as one major goal of control programs, together with systematic serological screening of blood donors [8],[9].
The widespread occurrence of native triatomine species that reinvade insecticide-treated households is a major difficulty for the consolidation of Chagas disease control [9]–[12]. Except for a few key vector species (e.g., [13]), the ecological dynamics of reinfestation are still poorly understood, and it is expected that research on sylvatic triatomine populations will help confront the challenge of residual, low-intensity disease transmission mediated by sylvatic vectors. The situation in the Amazon, where enzootic T. cruzi transmission cycles involve a great diversity of vectors and reservoir hosts (e.g., [14],[15]), suitably illustrates these concerns. Adventitious adult triatomines maintain continuous, low-intensity transmission in rural (and some urban) settings; as a result, human infection is hypoendemic in the region, with about 100,000 to 300,000 people chronically carrying T. cruzi [16],[17]. Sylvatic triatomines are also involved in localized disease outbreaks related to oral T. cruzi transmission via contaminated foodstuffs [14],[16], and account for the relatively high infection prevalence (4–5%) reported among extractivist forest workers such as piaçava palm fiber collectors [15],[16]. The vast majority of these transmission events are mediated by triatomines of the genus Rhodnius, which are primarily associated with palm trees [18]–[20]. The widespread occurrence of palm tree-living Rhodnius populations in Amazonia, together with epidemiological evidence suggesting their active role in disease transmission, underscores the importance of obtaining reliable estimates of palm tree infestation rates by these vectors. Such estimates are currently unavailable, and this substantially hinders our understanding of Chagas disease transmission dynamics in the Amazon.
Palms of the genus Attalea (Arecoideae) play a major role as breeding and foraging habitats of sylvatic Rhodnius populations in Amazonia and other Neotropical regions (e.g., [18]–[23]). The strong Attalea-Rhodnius association led to the proposal that the presence of Attalea palms can be used as an ‘ecological indicator’ of areas where enzootic T. cruzi transmission cycles probably occur [23]. Later studies showed that the probabilities of palm infestation by triatomines can differ among sites, landscapes, and palms with varying structural traits [20],[21]. We moved beyond these preliminary proposals, based on limited datasets and crude analytical approaches, and asked under what sets of circumstances is the potential of palms to harbor bug colonies realized; in other words: are all Attalea equally likely to be occupied by Rhodnius bugs? If not, what are the likely causes of variation? In a region as vast as Amazonia, knowledge of the environmental determinants of palm infestation by triatomines may represent a key tool to optimize resource allocation for epidemiological surveillance. Should resources be aimed at intervention in one particular region, in one particular type of landscape, or on certain particular types of palms – regardless of the region and landscape where they are found? Answers to these questions may prove crucial to enhance disease prevention programs [20],[21].
The estimation of palm infestation by triatomines is limited by the inescapable reality of field sampling: the target organisms may be present at a site yet go undetected during the survey. There are two standard solutions to this pervasive problem. One is to develop improved sampling techniques that bring detection close to perfection. The other is to incorporate detection failure explicitly in the analyses; estimates of infestation can thus be derived that statistically compensate for false absences. Near-perfect sampling techniques are expensive and labor-intensive – clearly a problematic option for a vast study area. In this paper, we apply models developed by wildlife biologists to estimate site-occupancy probabilities when detection of the target organism is imperfect [24],[25]. We define palm infestation as site (i.e., palm) occupancy, the probability that a palm is occupied by at least one Rhodnius spp. Our approach leads to strong inferences on Attalea palm occupancy rates by Rhodnius spp. and allows for the comparison of models relating palm occupancy to environmental covariates at three different scales: region, landscape, and individual palm. We aimed at (i) describing palm infestation patterns and the way they vary at different spatial scales; (ii) identifying the most likely causes of such variation; and (iii) incorporating this information into predictive models of palm occupancy that can be useful in the context of disease risk mitigation. More generally, we illustrate a methodological approach that yields reliable estimates of eco-epidemiological parameters out of imperfect data.
Our sample of 298 Attalea palms spanned four regions (totalling 19 localities) in two countries (Fig. 1). The westernmost region was Napo, a white-water river system close to the Ecuadorian Andes. (All model covariates are named in bold typeface on their first appearance in the Methods section.) Moving to the east, we sampled three regions in the Brazilian Amazon: the lower right bank of the black-water Negro river, the left bank of the white-water Amazon river east of Manaus, and the forested part of the northern Branco river basin, an intermediate clear/white-water system. These survey sites spanned areas between ∼120×60 km (Napo) and ∼30×20 km (Negro), and were located, respectively, within each of the following moist forest ecoregions [26]: Napo, Japurá/Solimões-Negro, Uatumã-Trombetas, and Guyanan Highlands/Piedmont. From field observations and available literature [27],[28], we ranked our survey regions in decreasing order of soil fertility as Napo, Amazon, Negro, and Branco. Thus, or sampling is representative of four ecologically distinct sub-regions influenced by the three main Amazonian hydrological systems – white-, black-, and clear-water.
Within each region, we surveyed Attalea palms in three landscape classes: forest, rural, and urban. At each site, a sample of non-adjacent palms was selected haphazardly for the survey. Urban palms where sampled in plots within the street framework of cities, towns, or villages. Rural palms were surrounded by farming land, orchards, or pasture on previously forested sites. Forest palms were located in forested sites, most often medium to large fragments of mature secondary forest. These three landscape classes were easily distinguished in the field, and palms sampled in each of them were at least 50–100 m from the nearest patch of landscape in another class. Our sample included palms of three species (A. maripa, A. speciosa, and A. butyracea); their known distribution is shown in Fig. 1. All three species are large, solitary palms with large inflorescences/infructescences and in which old leaf bases remain adhered to the stem after leaf abscission. Palm identification followed Henderson et al. [29].
Individual palm trees vary considerably with regard to the amounts of epiphytic vegetation and dead organic material (dead fronds, husks, flowers, fruits, fibers, and dead epiphytes) that accumulate on their crowns and stems. We used a pre-established score system [21] to measure the approximate amount of live epiphytic plants and decomposing organic material present on each palm. These epiphyte and organic matter values were first recorded in the field and, for about 85% of palms, cross-checked by another team member by examination of individual palm photographs; we then derived a mean ‘organic score’ value for each palm – ranging from 0 to 4 points, with higher values denoting ‘dirtier’ palms. We measured palm stem height as the linear distance between the ground and the lowest base of a green leaf. Finally, we preliminarily assessed the effects of slash-and-burn farming practices, which are commonplace across the Brazilian Amazon, on palm infestation. We defined two coarse categories to distinguish palms standing on plots that had a fire less than about two years before our survey from palms on plots that were not burnt over a similar period. Fire information was obtained from landowners and complemented by recording fire scars on palms and nearby trees and the presence and size of fire-adapted pioneer trees in each survey plot.
We sampled each individual palm with a combination of mouse-baited adhesive traps [30],[31] and manual bug searches [32] (Fig. 2). Traps were set in the afternoon and checked the following morning, after approximately 15 hours of operation. We placed traps among organic debris or epiphytes in the palm crown, around the upper end of the stem, or directly in the angle between palm fronds. Most palms (234, or 78.5%) were sampled with four traps, with a minimum of one trap in eight palms and a maximum of nine in one palm. The total trapping effort was 1,098 trap-nights. Manual searches were performed on the organic matter of the palm crown after trap removal. We searched either directly in the palm crown or by collecting organic material in a 50-liter plastic bag and later checking bag contents on a white canvas. Both sampling techniques were used in 255 palms (85.6%), only manual searches in nine, and only traps in 34. Each individual trap or manual search was treated as a sampling event yielding a binary result of either “1” for bug detection or “0” for no bugs detected. Thus, a typical palm tree was sampled five times – four traps and one manual search. Each detection history is represented by a row of “1”s and “0”s. For instance, “1100-----0” represents a palm with two positive traps, two negative traps, and a negative manual search (the last “0”); the five dashes indicate that only four traps, up to a maximum of nine, were operated in this particular palm. The raw dataset is provided as Supporting Information (Dataset S1).
We combine two different but interconnected procedures: parameter estimation and model selection. All our models have a biological process component that expresses the probability that a palm is occupied by bugs (ψ), and a sampling process component that expresses the probability that we detect bugs in a palm where they actually occur (p). This hierarchical approach makes it possible to estimate the probability that animals are present in places where they are not seen, accommodating an explicit treatment of imperfect detection [24],[25],[33],[34]. We fit models using the software PRESENCE [35], which provides maximum-likelihood estimates of parameters and their standard errors (SE) in user-defined models that can contain covariates of occupancy and/or detection. Before performing the analyses, we built a set of 23 models (below) each expressing an a priori hypothesis of palm occupancy and bug detection. Model selection followed the Akaike Information Criterion (AIC), which combines information and maximum-likelihood theories to find models with the best compromise between model fit and complexity [36]. We use model selection as a tool for hypothesis testing: each model represents one hypothesis, and hypotheses represented by models with lower AIC values are better supported by the data.
We treat palms as independent sites with regard to occupancy by bugs of the genus Rhodnius; to ensure independence, several sites were surveyed within each locality, and neighboring palms were rarely sampled. Live-bait traps and manual searches are treated as replicate sampling events with an average probability of detecting bugs, conditioned on palm occupancy. Field observations and exploratory analyses motivated us to compare the performance of manual searches and traps in detecting bugs; furthermore, we observed relatively high numbers of triatomines per palm in the Napo region, suggesting that bug presence might be easier to detect in Napo palms than elsewhere. Accordingly, we modeled detection always as an additive logistic function of two binary covariates: sampling technique and region, with the latter specifying only whether sampling took place in Napo or elsewhere. Since we aimed at understanding which spatial scale contributes most to explaining observed variation in palm occupancy, we built models that include different palm, landscape, and regional covariates of occupancy. Our a priori set of 23 models includes six regional models, four landscape models, six local (palm) models, six models with different combinations of covariates from different scales, and one null model without covariates of occupancy. Some of the combined models include interactions between covariates at different scales. In particular, considering the more fertile soils of the Napo region, we model an interaction between Napo and the rural landscape, as well as between Napo and the forest landscape. These models represent hypotheses stating that the relationship between landscape and occupancy differs between Napo and the remaining regions. For ease of presentation, we will report modeling results grouped by spatial scale, concluding with a comparison of the best models across scales.
We first estimated detection probability with a simple model that has no covariates of palm occupancy. We designate this model with the notation ‘ψ(.), p(manual+Napo)’, where the ‘.’ denotes no covariates on the occupancy part of the model and ‘manual’ and ‘Napo’ designate the technique and regional covariates of detection, respectively. Under this null model of no predictable variation in palm occupancy rates, the probability of detecting bugs where they actually occur ranges from 0.05 (SE = 0.01) with traps in the Brazilian Amazon to 0.82 (SE = 0.05) with manual searches in Napo, Ecuador. Both covariates increase detection probabilities; the Napo effect estimate is 3.01 (SE = 0.3). Had we not taken detection failure into account, we would report a proportion of 0.24 palms occupied by bugs – the number of palms where we detected bugs divided by the total number of palms sampled, which when expressed as a percentage is the classical ‘infestation index’ [9] (Table 1). When we consider that the probability of detection may be less than one, our null model estimate of occupancy is 0.59 (CI95% 0.42–0.75).
We found little evidence of regional variation in occupancy, as shown by the small differences in AIC values between the null model and models with regional covariates (Table 2). When we constrain models to only one regional covariate, the region that contributes most to explaining the data is Napo. All the models that estimate occupancy in the Napo region separately from other regions set that value at 0.68 (CI95% 0.50–0.83), almost twice the average occupancy estimated for Brazilian regions (0.37; CI95% 0.22–0.54). The second model in Table 2 includes regional covariates for the two hypothetical extremes of occupancy, Napo and Branco. Despite our prior expectation, based on published soil richness information, this model does not explain the data any better than the single-covariate Napo model. Thus, even if the Napo region appears to have higher palm occupancy rates, the data do not provide strong evidence of variation in occupancy across regions, and in particular among regions within Brazil.
Estimated palm occupancy is highest in rural and lowest in urban settings, without striking differences between estimates for different landscapes (Table 3). The models with interaction terms (Napo*forest and Napo*rural) do not explain the data particularly better than models without those terms. Among models with only one landscape covariate, the best model estimates a negative effect of urban landscapes on occupancy and lumps rural and forest areas into one landscape class. Estimated palm infestation rates are 0.33 (CI95% 0.15–0.57) for urban and 0.63 (CI95% 0.45–0.78) for forest/rural landscapes. Despite these broad patterns, there is no strong evidence of landscape-level effects: AIC values vary within less than 10 units for all models, and there is overlap of 95% CIs for estimates of occupancy in different landscapes.
All the models that include the ‘organic score’ palm attribute perform substantially better than the null model (Table 4). We modeled the effects of organic score, height, and recent fire separately and in two additive combinations (all effects and the combination of height and organic score) after preliminary analyses suggested that recent fire was the least important of the three covariates. AIC variation across models indicates that height and organic score are indeed most useful to explain the data. A model with all covariates does not rank any better than the model with height and organic score alone. When the three covariates are modeled separately, organic score ranks better than height, which, in turn, ranks better than fire. The strength of these relationships between infestation and individual palm traits is at odds with expectations under random bug migration among palms within a given site, indicating that the assumption of palm independence with regard to occupancy holds.
Tables 2 and 3 show how regional and landscape models fall within less than 10 AIC units of the null model, suggesting that they do not improve our ability to explain the data when compared with a model lacking occupancy covariates. Conversely, Tables 4 and 5 show strong support for local-scale models that use palm attributes as covariates of occupancy. Models that include regional and/or landscape covariates jointly with palm attributes also perform substantially better than the null model. However, these multi-scale models do not explain the data any better than a simple local model of occupancy as a function of organic score and palm height – the first model of Tables 4 and 5, where both effects are positive and significantly larger than zero (1.41, SE = 0.41; and 0.43, SE = 0.13, respectively). Figure 3 shows occupancy estimates according to this best-performing model. Short and ‘clean’ Attalea palms have the lowest probability of infestation, whereas tall palms (∼10 m) with plenty of accumulated organic debris are predicted to be almost certainly infested. According to these point estimates of occupancy by Rhodnius spp., a ‘clean’ palm would have, at most, a 0.3 probability of infestation; this probability would rise to over 0.5 in a palm with an organic score close to 4. Parameter estimates for the best-ranking models are provided as Supporting Information (Table S1).
A coherent view of the epidemiology of Chagas disease in Amazonia is currently emerging; discrete foci of relatively intense transmission, related to large-scale harvesting or consumption of forest products, seem to punctuate a widespread background pattern of low-intensity, vector-borne transmission [15]–[17]. Faced with the logistical impossibility of full geographical coverage, surveillance systems rely on a combination of two strategies: (i) detection of acute, febrile cases of the disease through existing health services (malaria posts and the regular health care network), and (ii) identification of higher-risk areas or situations that can be targeted through localized control and prevention efforts [37]. The first strategy is limited by the low sensitivity of clinical diagnosis [2],[38]; the detection of T. cruzi in malaria blood smears depends on the levels of parasitemia and requires skilled technicians. The second approach demands a clear understanding of the environmental circumstances that signal a higher risk of disease transmission. We focus on this second option, using the quantification of vector occurrence as a proxy for epidemiological risk and modeling palm occupancy by vectors as a function of environmental covariates over three spatial scales. To the best of our knowledge, this is the first attempt to develop quantitative models relating environmental factors to the occurrence of triatomine vectors in Amazonia.
Had we measured palm infestation as the percentage of palms where bugs were detected [9], we would report an infestation index of 24.2% (72 out of 298 palms; Table 1). Instead, we explicitly considered the possibility that bug detection fails in some palms that are actually infested, and derived an unbiased estimate of palm occupancy that is twice as high as the classical infestation index. This hierarchical strategy of modeling occupancy and detection as separate but inter-related processes stems from methods developed for estimating animal population parameters under imperfect detection [25],[33],[34], and is particularly useful when target organisms are of small size, dull-colored, and secretive (see Box 1). Many human disease vectors match this description, and most triatomine species surely do. Vector population studies that disregard the imperfections of the sampling process are likely to yield biased conclusions that may result in flawed recommendations for disease control and surveillance [see 39,40].
It must be noted that environmental constraints not included in our analyses could also modify palm occupancy. For instance, bug populations are under the influence of seasonality, predation pressure, and host availability. The efficacy of live-bait traps may vary with the nutritional status of the bugs, their aggressiveness or the performance of adhesive tapes under different weather conditions. Thus, while our models provide a simple and informative explanation of the data at hand, a more detailed assessment of triatomine population ecology and T. cruzi transmission dynamics in Amazonia will require the measurement and analysis of additional covariates.
Our data contain indirect information on vector abundance that is reflected in the estimates of detection probability [41]. The high estimates of detection probabilities in the Napo region (∼0.55 vs. ∼0.08 elsewhere) match our field observation of relatively larger numbers of bugs per occupied palm (9.04 vs. 2.24 in Brazil); this suggests a possible relation between soil fertility and bug density, perhaps mediated by higher primary productivity in rich-soil ecosystems. Whether this relation holds and has any public health relevance in other Amazonian fertile-soil regions is still an open question. It must be noted, however, that the prevalence of human T. cruzi infection in the Ecuadorian Amazon, including our Napo survey area, is substantially higher (2.4%) than the overall estimate (∼1%) for the whole Amazon basin [16],[17]. Such difference warns against using palm occupancy as the sole metric of transmission risk, and calls for further research to test the soil fertility-vector abundance hypothesis. Studies of vector abundance should also investigate how the number of bugs in a palm relates to the probability that adult specimens fly into a nearby house [42]. There is evidence that in denser triatomine colonies each individual has less access to bloodmeals, and that adult bugs are more likely to start dispersive flights when starved [43],[44], but the data are still inconclusive for sylvatic Rhodnius populations.
The effects of anthropogenic habitat disturbance on triatomine bug populations have been discussed extensively (e.g., [9],[20],[42],[45]); however, the evidence to support the claim that habitat disturbance triggers house invasion or colonization by triatomines is still weak. Our results show similar palm occupancy rates in forest and rural areas, but lower occupancy in urban settings. This suggests that palm tree Rhodnius populations can endure moderate habitat degradation, including slash-and-burn farming, in deforested rural areas, but tend to become rarer in heavily disturbed urban landscapes. Such endurance may sustain the risk of vector-human contact in rural sites, particularly when selective deforestation respects large palm trees near houses – a common practice across the Neotropics. We caution that our observations about urban landscapes may not apply directly to large urban forest fragments or to the contact zones between forests and expanding urban settlements; triatomines are known to occur in these environments, and may regularly enter houses near forest edges (e.g., [19],[46]).
Our data provide substantial support to previous observations suggesting that individual palm tree attributes have a strong influence on infestation probabilities [20],[21]. The mechanisms underlying this phenomenon have not been thoroughly investigated; we hypothesize that larger and ‘dirtier’ palms constitute better micro-environments for the bugs in terms of both structural traits and host availability. High organic score values translate into higher architectural complexity, resulting in more hiding and oviposition sites, and probably help maintain stable and buffered microclimate conditions [cf. 20]. The number of potential vertebrate hosts available as bloodmeal sources for the bugs can also be expected to be higher in larger palms with higher organic score values [47], where more hiding/nesting sites, and often also fruits and seeds, are available. Our hypothesis predicts that a Rhodnius population infesting a large, dirty palm tree has less chances of going extinct than a population infesting a small, clean palm. This hypothesis may be tested with a patch occupancy dynamics study [48].
This paper highlights the importance of accounting for imperfect detection in the study of vector ecology; in addition, our assessment of the explanatory power of regional, landscape, and local environmental covariates aimed at identifying those that hold more promise for improving vector surveillance and control strategies [49],[50].
Our results are relatively discouraging with regard to broad-scale risk mapping; the use of soil richness datasets seems attractive, but prior validation studies are necessary. On the other hand, local-scale covariates are overwhelmingly more useful than regional or landscape features in explaining variations in palm occupancy. This suggests that the assessment of potential disease risk situations will require detailed knowledge of local, site-specific conditions. The participation of decentralized vector control teams linked to local malaria control services [16],[37] may therefore be key to the advancement of Chagas disease prevention in Amazonia. Our results also suggest that peridomestic palm tree management could lower palm infestation rates and, therefore, might help reduce transmission risk [21]. Model-predicted effects of removing organic debris from palms range from halving to reducing palm infestation probability by more than 70% (Fig. 3). This result indicates correlation, not necessarily causation, but provides a clear-cut working hypothesis that can be put to test in the context of environmental management research.
Imperfect detection of the target organism is a real and pervasive problem both in wildlife management and in epidemiology. Wildlife biologists often use sampling strategies (e.g., [51]) and analytical tools [52],[53] that yield unbiased parameter estimates under imperfect detection. Latent class analysis and capture-recapture approaches are used to formally account for detection failure in epidemiological studies; they allow estimation of prevalence or incidence rates when a diagnostic gold standard is unavailable or undercount of disease events is likely (e.g., [54]–[58]). Even if the contribution of these and similar approaches is growing, we still find that many epidemiological and most vector ecology studies simply overlook the problem of imperfect detection.
Here we show how replicate sampling of vector ecotopes with a practical, yet imperfect field methodology can be used to (i) derive unbiased statistical estimates of eco-epidemiological parameters and (ii) test hypotheses about the effects of environmental covariates on such parameters. As long as model assumptions (e.g., population closure or independent detection histories) hold reasonably and study design is adequate, this strategy can help enhance research on vectors, pathogens, and hosts (see Box 1). For instance, replicate malaria blood smears could be used to measure between-slide variation in Plasmodium spp. detection. The same reasoning applies to vector surveillance schemes with replicate sampling, e.g., of Aedes aegypti [59], or when pathogen diagnosis involves serial testing, e.g., for intestinal parasites [60]. The generality of our methodological proposal is particularly compelling in the case of vector-borne zoonotic diseases, which are those more likely to become emerging public health threats [61], but the formal treatment of imperfect detection can significantly strengthen other areas of eco-epidemiological research.
A. Paucar, C. Carpio, R. Perry, and technicians of Fiocruz and the Ecuadorian and Brazilian vector control services participated in fieldwork. We thank T.V. Barrett (INPA, Brazil), C.J. Schofield (LSHTM and ECLAT, UK), F. Noireau (IRD, Bolivia), and S.L.B. Luz (ILMD-Fiocruz, Brazil) for helpful discussion and suggestions. The Brazilian Instituto Nacional de Colonização e Reforma Agrária provided logistic support for several field trips. This paper is contribution number 9 of the Research Program on Infectious Disease Ecology in the Amazon (RP-IDEA) of the Instituto Leônidas e Maria Deane.
|
10.1371/journal.pntd.0006081 | Sustaining visceral leishmaniasis elimination in Bangladesh – Could a policy brief help? | Bangladesh has made significant progress towards elimination of visceral leishmaniasis, and is on track to achieve its target of less than one case per 10,000 inhabitants in each subdistrict in 2017. As the incidence of disease falls, it is likely that the political capital and financial resources dedicated towards the elimination of visceral leishmaniasis may decrease, raising the prospect of disease resurgence. Policy memos may play a crucial role during the transition of the elimination plan from the ‘attack’ to the ‘consolidation’ and ‘maintenance’ phases, highlighting key stakeholders and areas where ongoing investment is crucial. An example of a policy brief is outlined in this paper. The background to the current elimination efforts is highlighted, with emphasis on remaining uncertainties including the impact of disease reservoirs and sustainable surveillance strategies. A stakeholder map is provided outlining the current and projected future activities of key bodies. Identification of key stakeholders subsequently frames the discussion of three key policy recommendations in the Bangladeshi context for the transition to the consolidation and maintenance phases of the elimination program. Recommendations include determining optimal vector control and surveillance strategies, shifting the emphasis towards horizontal integration of disease programs, and prioritising remaining research questions with a focus on operational and technical capacity. Achieving elimination is as much a political as a scientific question. Integrating the discussion of key stakeholders with policy priorities and the research agenda provides a novel insight into potential pathways forwards in the elimination of visceral leishmaniasis in Bangladesh and in the rest of the Indian subcontinent.
| In this article, we examine the challenges of eliminating visceral leishmaniasis in Bangladesh from a policy perspective. Visceral leishmaniasis is a neglected tropical disease which causes significant morbidity and mortality, but recent efforts in Bangladesh, India and Nepal (the countries that used to have two-thirds of all cases in the world) have made significant headway in reducing the extent of the disease. However, ongoing concerted effort is needed to ensure that elimination of the disease is maintained in the long-term. This will require engagement and coordinated effort from a variety of stakeholders, including national and international policy makers, funders and implementers. We present the identified challenges, stakeholder map and potential solutions in the form of a policy brief to distill the key actors and actions that are necessary to maintain momentum towards elimination of visceral leishmaniasis. Such actions include determining optimal long-term vector control and surveillance strategies, and attempting to integrate the visceral leishmaniasis program into existing health systems. The research agenda should also be prioritised, with a focus on addressing operational and technical questions.
| The conclusion of 2017 marks the deadline for the achievement of elimination of visceral leishmaniasis (VL; kala-azar) as a public health problem in five South Asian states. Bangladesh, India and Nepal signed a Memorandum of Understanding (MOU) in 2005, seeking to achieve elimination by 2015. This deadline was extended in 2014 to the end of 2017 with the inclusion of Bhutan and Thailand in the MOU [1]. In the context of Bangladesh, elimination as a public health problem is defined as less than one case per 10,000 inhabitants in each upazila (subdistrict) [2]. Of the 489 upazilas in Bangladesh, 100 were reported to be initially endemic for kala-azar [3]. By April 2017 the incidence of VL has dropped below the target in all upazilas, and attainment of the target will hopefully be confirmed at the end of the year [4]. As the elimination efforts shift from achieving the target to consolidating and maintaining progress, it is timely to consider where further action is needed. This paper seeks to present the challenge of consolidating progress towards elimination within the context of a policy framework, analyzing key players and issues where further action is necessary. Box 1 summarises the role and structure of the policy brief.
Visceral leishmaniasis, or kala-azar, is caused by protozoan parasites from the genus Leishmania which are spread by the female sandfly Phlebotomus argentipes, and is classified among the neglected tropical diseases [5]. It is characterized by fever, fatigue, weight loss, hepatosplenomegaly, lymphadenopathy and anaemia, with a near universal case fatality without treatment [6,7]. However, the majority of infected subjects are asymptomatic. [6] Among those treated, a proportion will develop post-kala-azar dermal leishmaniasis (PKDL), with the emergence of skin lesions usually six months to three years after the original condition has been cured [6]. In some cases; however, presentation may be delayed or not associated with previous clinical kala-azar.
Modern-day Bangladesh was the site of the first reported outbreak of visceral leishmaniasis, which accounted for 75,000 deaths between 1824 and 1827 [5]. Over the years, a number of attempts have been made to control the burden of visceral leishmaniasis, which disproportionately affects the Indian subcontinent [8]. The Malaria Eradication Program in the 1950s had a positive impact on the control of visceral leishmaniasis in Bangladesh, with only 59 cases reported between 1968 and 1980 [5]. However, subsequent disease resurgence with cessation of widespread spraying of dichlorodiphenyltrichloroethane (DDT) has prompted new and wide-ranging efforts to curb the disease, beginning in 2005 with the signing of the MOU.
A number of factors suggest that long-term elimination of kala-azar may be achievable. These include that in the Indian subcontinent the disease is only hosted by humans, with a single vector and close geographic clustering of cases [9]. The recent availability of a diagnostic test, based on the rk39 antibody, has facilitated improved diagnostic testing in the field. Similarly, the advent of liposomal amphotericin B as a single-dose treatment has provided access to readily available and tolerable treatment [9]. Access to liposomal amphotericin B has been facilitated initially by a 2007 pricing agreement and subsequently in 2011 by a donation agreement between the manufacturer and WHO [9–11]. These factors have supported the development and implementation of the elimination plan currently in place in Bangladesh.
Much as in the other countries of the Indian subcontinent, the national kala-azar elimination program (NKEP) in Bangladesh consists of four discrete phases: a preparatory phase, attack phase, consolidation phase and maintenance phase [12]. The attack phase is defined by widespread indoor-residual spraying, the implementation of integrated vector management, early diagnosis and treatment and close case and vector surveillance [12]. The consolidation phase has been proposed to be distinguished from the attack phase by limited indoor residual spraying around affected cases (the index case approach), and a shift in focus to treatment of co-infection such as HIV and treatment adherence. The maintenance phase has been described as a state of close surveillance with intervention in small regions where outbreaks occur [12]. However, the optimal approach for transitioning from the attack to the consolidation phases remain contested and the subject of inquiry.
After signing the MOU, Bangladesh introduced a national elimination program in 2008 [13]. Key tenets of the program included training health staff practicing in endemic areas, the introduction of the rapid rk39 diagnostic test and free provision of oral miltefosine therapy [13]. Miltefosine is an oral drug taken as a 28-day course to treat visceral leishmaniasis, which has faced deployment issues and is now surpassed by single-dose liposomal amphotericin B [14]. In 2012, the programme also introduced a 12-week oral miltefosine therapy protocol for PKDL patients, replacing the poorly tolerated regimen with sodium stibogluconate [13]. Blanket coverage with indoor residual spraying with deltamethrin and time-limited provision of long-lasting insecticide treated bednets was also implemented in 2011 in the areas of highest endemicity [13]. Implementation research conducted in conjunction with the Special Programme for Research and Training in Tropical Diseases from the World Health Organization (WHO/TDR) informed key strategies in terms of case identification, case management and curtailing transmission. The NKEP has deployed indoor residual spraying with insecticide and case search through house-to-house visits in 60 households around recently reported VL/PKDL cases (the so-called index case search), known as no kala-azar transmission activity (NKTA), although the systematic deployment of this intervention has been conditional to the availability of adequate human and financial resources.
As the attack phase shifts into the consolidation and later into the maintenance phases in Bangladesh, a number of key uncertainties remain which need to be addressed to ensure the continued progress of the elimination program. The current target of <1 per 10,000 cases per upazila was inherited from historical leprosy campaigns and its epidemiological significance for visceral leishmaniasis is uncertain [15]. Modelling studies have sought to delineate whether existing efforts are sufficient to maintain disease control after this target is reached. Seven modelling papers were identified in a recent literature review [8], and more work is underway. However, modelers are struggling with assumptions about the drivers of transmission (symptomatic visceral leishmaniasis patients, asymptomatic carriers, PKDL patients, patients in whom disease reactivation occurs, which is higher among HIV-coinfected patients, and the ‘window of infectivity’ during which a subject is infectious to others), as well as the vectorial capacity of sandflies to transmit the infection [7,16–18]. The relative contribution of these reservoirs remains unclear, and significantly impacts projections regarding the success of control and elimination strategies, especially in the consolidation and maintenance phases.
Detection of asymptomatic cases is currently limited in the field, and treatment is not recommended. The available rk39 test is antibody-based, and therefore unable to distinguish between current and previous infections; in order to diagnose active visceral leishmaniasis it is used as part of an algorithm that requires also presence of fever and enlarged spleen [6]. The availability of cost-effective and field suitable antigenic tests, such as nucleic acid detection tests, would represent a significant advancement [18]. Finally, the optimal methods for disease surveillance and case detection in the consolidation and maintenance phase remain to be determined. The attack and consolidation phases required a targeted, vertical approach (including the ‘fever camp’ and ‘index case’ strategy), but approaches that are more integrated in the general healthcare system are required to make visceral leishmaniasis case detection sustainable and cost-effective in the long-term [19].
Most pressingly, as the incidence and mortality from visceral leishmaniasis falls, donors and local decision-makers may seek to redirect funds elsewhere into more apparently urgent concerns. However, as the experience post cessation of the DDT spraying programme in the 1960s has highlighted, disease resurgence is a real threat which would lay existing efforts to waste. This policy memo therefore seeks to highlight key priorities for decision-makers in managing the transition from the attack to the consolidation phase, including investment in remaining areas of scientific uncertainty which may threaten elimination efforts.
In this context, it is crucial to highlight the existing stakeholders working towards the elimination of visceral leishmaniasis in Bangladesh. Table 1 classifies the key identified stakeholders as national state actors, intergovernmental organizations and non-state actors, highlights their role during the attack phase and suggests potential activities during the consolidation phase. In Table 2, the stakeholders are mapped according to their primary role in the elimination of the visceral leishmaniasis, demonstrating pertinent interconnections between stakeholders and areas of potential collaboration.
As Bangladesh transitions from the attack to the consolidation phase of the visceral leishmaniasis elimination program, the operational plan will also need to adjust to reflect the lower disease burden. The onus for advancing the elimination plan lies with national bodies as some international stakeholders wind down their contribution; where possible, donors must be engaged to continue their involvement to prevent disease resurgence. Importantly, international stakeholders and national bodies should collaborate and coordinate to ensure that adequate capacities remain in the country when international aid ceases. National bodies should emphasise the opportunity cost of disease resurgence as a result of reduced or diverted efforts. Three key policy priorities are identified and explored as central to the transition from the attack to the consolidation phase of the elimination program.
As the incidence of disease falls, cost-effective strategies for long-term transmission control and disease surveillance must be established. Blanket provision of indoor residual spraying is unsuited as a long-term control strategy in view of significant cost and resource requirements [29]. Implementation research is contributing new findings suggesting that alternative long-term vector control strategies are feasible. Studies in Bangladesh and elsewhere examine and compare the effects of insecticide treated bednets, wall painting and wall lining, in addition to environmental management. Bednets appear to confer protection, whether insecticide-treated or not [29–31]. A multi-centre cluster randomised controlled trial across Bangladesh, India and Nepal found that insecticide impregnated wall linings had the most sustained impact on sandfly density, surpassing insecticide treated bednets and environmental management; these effects were shown to extend for at least two years in a cluster-randomised study in Bangladesh in terms of sandfly reduction and mortality [31, 32]. Within three months after the intervention, another cluster randomized study found that wall painting was more effective in reducing sandfly density than insecticide spraying, wall lining and bednets [33]. Combination approaches are likely to prove effective in the long-term, with a recent study identifying that long-lasting insecticide treated bednets in combination with outdoor spraying of breeding areas yielded the highest reduction in sandfly density compared with alternative combined modalities [34]. With this information and other studies underway, Bangladesh must now transition from the indoor residual spraying-led approach of the attack phase to one which more sustainably utilises these novel control strategies for the consolidation phase, while evaluating long-term effects and costs.
Bangladesh to date has also effectively utilised active case detection strategies in the attack phase, searching houses for suspected cases during indoor residual spraying in preselected hyperendemic villages (‘blanket search approach’), or in the fifty houses surrounding any case who presents to the health centre (‘index case approach’) [20]. The former yields higher numbers of new cases (during the attack phase) but at greater effort and cost than the index search approach [19, 35]. The latter is likely to represent the most appropriate search strategy as disease incidence falls, provided cases are identified soon enough to limit the number of secondary infections generated [19]. However, a disturbing finding is that the median time from onset of symptoms to treatment during the attack phase was 78 days in Bangladesh, essentially due to patient’s healer shopping practices and healers failing to diagnose and refer suspected cases, and had not improved from the 58 days estimated in a previous cross-sectional study in 2007 [36, 37]. With declining incidence and awareness these delays are even likely to increase.
The ‘fever camp approach’ has also been evaluated as an active detection strategy in the Indian subcontinent. In initial studies, the fever camp demonstrated higher sensitivity in detecting clinical cases than alternative approaches at a greater overall cost but lower cost per case in the Bangladeshi context [19, 35]. However, as the incidence of cases falls, the experience in India has demonstrated that the cost per case will rise markedly [35]. In order to improve cost-effectiveness, a combined fever camp approach assessing for a number of infectious diseases has been proposed. One study evaluated combined camps assessing for visceral leishmaniasis, tuberculosis, malaria and leprosy in Bangladesh, India and Nepal. While this small pilot study could only detect a small number of new cases, the approach was acceptable to the community, with a cost of USD232 per case detected, and provided an opportunity for additional interventions, such as bednet impregnation and community education [38].
Case detection and transmission control activities can also be combined. Various types of index case-based intervention packages are being tested in Bangladesh. These include fever camps at the village of the index case looking for cases of visceral leishmaniasis and PKDL as well as other febrile illnesses, combined with installation of insecticide-impregnated wall lining or impregnation of existing bednets [39].
The assessment of multiple diseases in the combined camp approach highlights the need for Bangladesh to horizontally integrate its visceral leishmaniasis elimination program with other health activities. This may be achieved at the level of the health service, but also through integration with other private entities operating in Bangladesh. By way of example, the Bangladesh Rural Advancement Committee currently offers pathology services for malaria and HIV testing, promotes the use of bednets and utilises community health volunteers to administer health interventions, and could be readily collaborated with to expand access to services for visceral leishmaniasis [21]. The web-based surveillance system DHIS2 by GIZ has the potential to be expanded to other infectious diseases and presents a powerful tool in the management of disease outbreaks. It has significantly improved disease surveillance, making routine data available through public servers and providing real-time information from public health facilities to the Ministry of Health and Family Welfare [40]. The system leverages the government’s existing web-based platform for maternal and child health services to maintain clinic information on patients with visceral leishmaniasis and improve follow-up [25]. Horizontal integration with the treatment of other infectious diseases will improve service delivery and may also engage donors and international stakeholders in refining models of care. This will necessitate health system strengthening including ongoing educational activities by icddr,b and the Directorate of General Health Services. The provision of technical advice by partners regarding health system strengthening may also be of benefit.
As highlighted earlier in the paper, there are a number of outstanding questions that require continued investments in research to ensure lasting visceral leishmaniasis elimination. This is indeed a hard sell, at a time when the feeling is that the goal is reached, and decision-makers start planning for divesting. However, without filling critical knowledge gaps and innovative tools and approaches, the remarkable achievements of the attack phase are vulnerable to the disease returning in the medium to long-run.
Research has provided the interventions that are being used now and made it possible to eliminate visceral leishmaniasis. This includes the development of tools for diagnosing (the rK39 rapid diagnostic test) and treating the infection (miltefosine, liposomal amphotericin B, paromomycin), as well as implementation research that identified effective ways of conducting active case identification and reducing transmission through vector control and community participation. The contribution of many actors, both in the countries and internationally, should be acknowledged.
However, as pointed out above, some of these tools and approaches may not work or may not be sustainable in the consolidation and maintenance phases. In particular, the WHO Special Programme for Research and Training in Tropical Disease (WHO/TDR) has supported for more than a decade the joint work of disease control managers and researchers from Bangladesh, India and Nepal to identify research questions and conduct the research to inform policy decisions in terms of transmission reduction, vector control and case identification and management. This group highlighted knowledge gaps and research investment needs ranging from research and development of new tools to intervention and implementation research [41,42]. While a deeper understanding of the epidemiology and transmission of visceral leishmaniasis is needed, improvement in diagnosis, treatment and prevention is also pressing in the transition between attack and consolidation phase. Importantly, this group emphasizes the need to reconsider a target based on reducing the incidence (rather than the prevalence) of visceral leishmaniasis, and maintaining zero transmission in the areas that have achieved the elimination target [42]. These analyses should inform decisions by international and local actors, and be coordinated across the concerned countries of the Indian subcontinent.
Achieving elimination of visceral leishmaniasis is as much a political as a scientific question. As Bangladesh moves towards the consolidation phase of its elimination program, national stakeholders must continue to engage donors and international bodies in coordination with the other concerned countries of the Indian subcontinent to ensure that the long-term goal of elimination is achieved, and resources are not prematurely redirected. The key focus will be on evolving the operational plan towards more sustainable surveillance and vector control strategies, horizontal integration of health resources, and addressing remaining development and implementation research questions to assist in achieving long-term elimination of visceral leishmaniasis.
|
10.1371/journal.pntd.0004229 | Quality of Sterile Male Tsetse after Long Distance Transport as Chilled, Irradiated Pupae | Tsetse flies transmit trypanosomes that cause human and African animal trypanosomosis, a debilitating disease of humans (sleeping sickness) and livestock (nagana). An area-wide integrated pest management campaign against Glossina palpalis gambiensis has been implemented in Senegal since 2010 that includes a sterile insect technique (SIT) component. The SIT can only be successful when the sterile males that are destined for release have a flight ability, survival and competitiveness that are as close as possible to that of their wild male counterparts.
Tests were developed to assess the quality of G. p. gambiensis males that emerged from pupae that were produced and irradiated in Burkina Faso and Slovakia (irradiation done in Seibersdorf, Austria) and transported weekly under chilled conditions to Dakar, Senegal. For each consignment a sample of 50 pupae was used for a quality control test (QC group). To assess flight ability, the pupae were put in a cylinder filtering emerged flies that were able to escape the cylinder. The survival of these flyers was thereafter monitored under stress conditions (without feeding). Remaining pupae were emerged and released in the target area of the eradication programme (RF group). The following parameter values were obtained for the QC flies: average emergence rate more than 69%, median survival of 6 days, and average flight ability of more than 35%. The quality protocol was a good proxy of fly quality, explaining a large part of the variances of the examined parameters.
The quality protocol described here will allow the accurate monitoring of the quality of shipped sterile male tsetse used in operational eradication programmes in the framework of the Pan-African Tsetse and Trypanosomosis Eradication Campaign.
| An area-wide integrated pest management campaign against Glossina palpalis gambiensis has been implemented in Senegal since 2010 that includes a sterile insect technique component. The sterile males used for the releases emerged from pupae that were produced and irradiated in Burkina Faso and Slovakia (irradiation done in Seibersdorf, Austria) and transported weekly under chilled conditions to Dakar, Senegal. Tests were developed to assess the quality (flight ability and survival) of sterile males. To assess flight ability, for each consignment a sample of 50 pupae (QC flies) was put in a cylinder filtering emerged flies that were able to escape the cylinder. The survival of these flyers was monitored under stress conditions. Remaining pupae (RF flies) were emerged and released in the target area of the eradication programme. The quality assessment of the QC flies was a good proxy of the quality of the RF flies. The quality protocol described here will allow the accurate monitoring of the quality of shipped sterile tsetse males used in operational eradication programmes.
| Tsetse flies are hematophageous insects found in sub-Saharan Africa and are the main vectors of trypanosomes, the causative agents of African Animal Trypanosomosis (AAT) and Human African Trypanosomosis (HAT) [1]. The debilitating disease AAT limits the exploitation of fertile land for agricultural activities in an area of 10 million km2 [2] and is considered the main constraint to more productive livestock systems in sub-Saharan Africa [3,4]. The direct annual production losses of cattle in terms of decreased meat and milk production, abortions, etc. are estimated at USD 600–1200 million [5] and the overall annual losses in livestock and crop production have been estimated as high as USD 4750 million [6].
To suppress or eradicate these disease vectors, four methods that are environmentally and economically acceptable can be used in a context of area-wide integrated pest management (AW-IPM) approaches [4,7,8] i.e. the sequential aerosol technique (SAT) [9,10], the deployment of insecticide-impregnated traps/targets [11], the live-bait technology [12] and the sterile insect technique (SIT) [13,14]. The SIT is used throughout the world to suppress, eradicate, contain or prevent the introduction of several insect pests such as fruit flies [15], moths [16], screwworm flies [17–19], mosquitoes [20] and tsetse flies [14]. The effectiveness of the SIT to eradicate tsetse fly populations was demonstrated in Nigeria with Glossina palpalis palpalis [21], in Burkina Faso with G. palpalis gambiensis, G. tachinoides and G. morsitans submorsitans [13,22] and on Unguja Island, Zanzibar with G. austeni [14]. In Senegal, a programme is underway to eradicate a G. p. gambiensis population from the Niayes area [23–27]. This campaign is part of the Pan-African Tsetse and Trypanosomosis Eradication Campaign (PATTEC), an initiative of the African Heads of State and Government to ensure increased food security through better management of the tsetse fly and trypanosomosis problem [28].
The data of the feasibility study (2007–2010) indicated the potential to create a sustainable zone free of G. p. gambiensis in the Niayes [24,29], and therefore, the Government of Senegal opted for an AW-IPM approach that included an SIT component. An agreement was made between the Government of Senegal and the Centre International de Recherche-Développement sur l’Elevage en zone Subhumide (CIRDES) in Bobo-Dioulasso, Burkina Faso and the Slovak Academy of Sciences (SAS) in Bratislava, Slovakia to produce the sterile flies for the eradication campaign in Senegal. The male flies were transported as chilled pupae to Dakar where they could emerged under standard conditions [27].
In AW-IPM programmes that have an SIT component, the quality of the released sterile males remains one of the most crucial prerequisites for success, as flies of low quality (i.e. low survival rate and/or deformed wings) can’t compete with wild males in the field to mate with females and induce sterility in the native population [8,30,31]. Therefore, routine quality control procedures are crucial for the SIT component to identify weaknesses in production or handling procedures that result in low quality of the sterile males which may lead to potential failure of these programmes [32]. In past tsetse eradication campaigns [13,14,21,22] the rearing facility and target area were not far apart, so there was no need for pupal shipments. As an example, in the eradication programme on Unguja Island, Zanzibar, the sterile male flies were produced in Tanga, mainland Tanzania and the sterile adult flies were collected twice a week with light aircraft and released from the air in biodegradable cartons. A quality control system was implemented that consisted of taking one release carton before loading the aircraft in Tanga and one carton during the release flights. Both in Tanga and Unguja, the flies were released in a specially designed release arena and the following quality parameters assessed: number of flies in the box, mortality, number of non-flyers, sexing error, and feeding status [33].
In this study, we developed and validated a quality control protocol to assess the quality of male G. p. gambiensis that were irradiated and shipped as pupae under chilled conditions. Four biological parameters were measured: i) adult emergence, ii) percentage of flies with deformed wings, iii) flight ability of the sterile flies in the insectary and in the field and iv) survival of the flyers (those that were capable of flying out of the cylinder in the insectary) under stress conditions. These parameters were used to assess the reliability of this quality protocol to 1) predict field performance of the flies, 2) monitor and compare the performance of flies from two locations with different treatment protocols, and 3) develop quality criteria for use in feedback mechanisms to improve rearing systems.
The study was carried out at the Institut Sénégalais de Recherches Agricoles / Laboratoire National de l’Elevage et de Recherches Vétérinaires, Service Bio-Ecologie et de Pathologies Parasitaires (ISRA/LNERV/BEPP) in Dakar. Insectary conditions were 24–25°C, 75 ± 5% RH and 12:12 light:dark photoperiod for emergence and the monitoring of the flies.
Male G. p. gambiensis pupae from colonies kept at Burkina Faso and Slovakia were irradiated under chilled (4–6°C) conditions to lower their metabolic rate to prevent emergence [27,34]. The SAS pupae were irradiated in a Gammacell 220 (MDS Nordion, Ottawa, Canada) (dose rate of 3.11 Gy.sec-1 on 1 May 2012 and 2.19 Gy.sec-1 on 1 January 2015) or in an X-ray irradiator (Radsource 2400) (dose rate of 14.30 Gy.min-1) located at the FAO/IAEA Insect Pest Control Laboratory, Seibersdorf, Austria. The CIRDES pupae were irradiated in a 137Cs source for 24 minutes 30 seconds to give a dose of 110 Gy. The male pupae were packaged in cartons (for SAS) and in petri dishes (CIRDES) that were placed in insulated transport boxes containing phase change material packs (S8) (PCM Phase Change Material Products Limited, Cambridgeshire, United Kingdom) to maintain the temperature at 8–10°C and shipped to Dakar by commercial aircraft [27].
The study was implemented from May 2012 to January 2015. A shipment of CIRDES and SAS pupae was received every week at the ISRA in Dakar. Each consignment contained two batches (1 and 2) of pupae that had a different larviposition date and consequently had been exposed to a different chilling period before shipping, i.e. batch 1 was chilled at 8°C for one day longer than batch 2 in the source insectary before transport. Each batch contained an average of 2500 pupae. A total of 50 pupae were sampled from each batch for the quality control test (QC) and the remaining pupae were emerged to be released in the operational eradication programme, i.e. the flies destined for release in the programme (RF). The pupae of the QC and RF groups were kept under the same environmental conditions (24–25°C, 75 ± 5% RH and a photoperiod of L:D 12:12 h). The 50 pupae for the QC group were selected to assess whether a small sample of each received pupae consignment was adequate to predict the quality of the shipped and released flies (RF).
Pupae from the RF group were placed in Petri dishes under ~1cm of sand mixed with a fluorescent dye (DayGlo) (0.5g dye/200g of sand), to mimic the natural emergence conditions in the soil (Fig 1A) and to allow discrimination from wild flies in the monitoring traps as these sterile male flies were released in the operational programme. Emerged flies were sorted and classed as “normal” (flies with no apparent morphological deficiencies) and “abnormal flies” (i.e. with deformed wings). Normal flies were offered at least three bovine blood meals (originating from slaughterhouse of Dakar, with the consent from the slaughterhouse to obtain the blood samples from livestock) containing 10 mg of the trypanocidal drug isometamidium per litre of blood using the in vitro silicon membrane feeding system before being transported to the field for release. The trypanocidal drug prevents the cyclical development of trypanosomes in the released sterile males [35–37]. Irradiated and marked males were transported by car to the release sites (~ 1 hour for Diacksao Peulh and Kayar and 10 minutes for the Parc de Hann [24]) in Roubaud-type cages (4.5 x 13 x 8 cm) that were covered with netting with a mesh size of 1 mm x 1 mm, each containing on average 120 sterile males. Cages were kept in climate controlled containers (temperature and humidity of 24–26°C, 75 ± 5% respectively) during the transport and temperature and humidity were recorded every 5 minutes with a Hobo data logger. Flies were released every Friday afternoon between 16:00 and 18:00 h over a white cloth (2 x 1.5 m). Males remaining on the cloth after 5 minutes were counted and considered as non-flyers. Ground releases of these flies took place from May 2012 to March 2013. Thereafter, all sterile male flies were released by air.
The pupae of the QC group were kept under the same conditions as the RF group but the Petri dishes with the pupae were put in a flight cylinder, i.e. a PVC tube 10 cm high and 8.4 cm in diameter (Fig 1B). The inner wall of the cylinder was coated with unscented talcum powder to prevent the flies from crawling out. This method was initially developed for routine quality assessments of sterile fruit flies [38,39] and moths [40], and adapted here to tsetse flies. This protocol gave an indication of the propensity of the sterile male flies to fly out of the cylinder and only those flies that managed to escape the flight cylinder after emergence were considered as “available for the SIT”. Flies with deformed wings and those with normal wings but unable to escape the flight tube were counted, as well as the number of pupae that did not emerge.
The survival of the sterile males of the QC group that escaped the flight cylinder was assessed under stress conditions (no food). Every morning (except Sundays), the emerged flies were collected and transferred to standard fly holding (10.3 cm diameter and 4.5 cm high) cages (Fig 1C). The flies emerged on a given day were pooled in one cage. Dead flies were counted daily and removed from the cages.
The data sets (both QC and RF groups) on percentage emergence, flies with deformed wings and flight ability were each divided into training and test sets. The training set was used to build the model and the test set to measure its performance [41]. For the data on emergence and percentage of flies with deformed wings, 60% of the entire data set (n = 364 rows), selected at random, was used as a training set and the remaining as the test set. For the flight ability, 75% of the entire data set (n = 80 rows) was used for the training set and 25% for the test set. The difference in the proportion of data used for the training set in the first and second cases was related to sample size.
A binomial linear mixed effect model was used to analyze emergence rates. The emergence rate measured within the QC group, the origin of the pupae (CIRDES and SAS), the batches (1 and 2) and their second and third order interactions were used as explanatory variables and the emergence rate of the RF group as the response variable. The shipment date was considered as a random effect. The best model was selected on the basis of the lowest corrected Akaike information criterion (AICc), and the significance of fixed effects was tested using the likelyhood ratio test [42,43]. The R2 (coefficient of determination) was used to describe the proportion of variance explained by the model for the training and test data sets [44,45].
The same analysis was used for the percentage of flies with deformed wings and the percentage of flyers.
Flight ability was analyzed between QC and RF groups using only the CIRDES data sets because field data were not available for the SAS shipments. Flight ability was compared among years (2012, 2013 and 2014) using the same binomial model.
The survival of the sterile males of the QC group that had escaped from the flight cylinder and kept under starvation was analyzed using Kaplan-Meier survival curves. Survival curves were compared between origins (CIRDES and SAS), batches (1 and 2) and years using the coxph model [46]. The median survival was considered to be the average probable survival of the studied flies. The R Software (version 3.1.0) was used for all statistical analyzes [47].
The complete data sets are available in S1 and S2 and S3 Tables.
The study was conducted in the framework of the tsetse eradication campaign in Senegal, led by the Directorate of Veterinary Services, Ministry of Livestock and the ISRA/LNERV, Ministry of Agriculture and Rural Equipment. This project received official approval from the Ministry of Environment of Senegal, under the permit N°0874/MEPN/DE/DEIE/mbf.
A total of 1,581,366 irradiated pupae were used for this study of which 1,271,121 (80.4%) originated from the CIRDES insectary (123 shipments) and 310,245 (19.6%) pupae originated from the SAS insectary (53 shipments).
The emergence rate of pupae of the RF group was significantly greater for shipments originating from CIRDES than those from SAS (P < 10−3; Table 1), as well as for batch 2 pupae than batch 1 pupae regardless of the origin of pupae (P < 10−3; Table 1). The percentage of flies with deformed wings was significantly lower for the flies that originated from the CIRDES than the SAS flies and for batch 2 pupae than batch 1 pupae regardless of the origin (P < 10−3; Table 1). The flight ability of the CIRDES flies in the field was significantly better for flies derived from batch 2 pupae than batch 1 pupae (P < 10−3; Table 1).
Adult emergence of pupae of the QC group was similar between origins (P = 0.8) but differed between batches regardless of the origin (P < 10−3; Table 1). The percentage of flies with deformed wings that emerged from the SAS pupae was significantly lower than that for the CIRDES pupae (P < 10−3; Table 1). It was similar between batches for CIRDES and different for SAS (P < 10−3; Table 1). The flight ability was similar between batches (P > 0.05; Table 1).
The comparison of the different parameters between the QC and the RF groups showed that the emergence rates were similar for the CIRDES flies while they were significantly greater in the QC group of the SAS flies (P < 10−3; Table 1). The percentage of flies with deformed wings was lower in the RF group as compared with the QC group for the CIRDES pupae, whereas it was the opposite for the SAS pupae (P < 10−3; Table 1). The percentage of the CIRDES flies escaping the flight cylinder in the insectary was significantly lower than the percentage of flies taking off in the field after the release (P < 10−3) i.e. 34.9 ± 17.8% and 55.1 ± 13.5%, respectively. The flight ability of batch 2 flies of the RF group was significantly greater than for batch 1 flies (P < 10−3) whereas no difference was observed between batches of the QC group (P = 0.6; Table 1). The predicted probabilities of occurrence using the QC data allowed us to predict the results observed in the RF group with good accuracy: the emergence rates, percentage of flies with deformed wings and flight ability were strongly correlated to predictions of the training data set (P < 10−3; R2 of 0.90, 0.94 and 0.95 respectively; Fig 2). For the test data set, the model predicted 55%, 53% and 45% of the variances respectively (P < 10−3, Fig 2).
Survival curves of QC flies kept under starvation are presented by batch and origin in Fig 3. Flies from batch 2 pupae survived significantly longer than those from batch 1 pupae for CIRDES (P = 0.01) whereas for SAS, batch 2 flies survived marginally longer than from batch 1 (P = 0.09) shipments. The CIRDES flies survived marginally longer than the SAS flies (P = 0.06). The median survival was 6 days regardless of the batch and origin of pupae (Fig 3). The maximum survival observed was 12 days after emergence for the CIRDES (batch 2) and 10 days for the SAS (batch 2) flies.
From 2012 to 2014, the percentage of QC flies escaping the cylinder gradually increased regardless of the origin of pupae (P < 10−3; Table 2). Flies lived significantly longer in the survival tests in 2013 and 2014 as compared with 2012 for both CIRDES and SAS flies (P < 10−3). Thus, the quality of sterile male flies (flight ability and survival) was significantly improved among years and these improvements were more prominent for flies from SAS (Table 2).
The quality protocol implemented in this study was developed for a programme that required long distance transport of chilled male tsetse pupae and was shown to be a good proxy for the insectary rearing output. Indeed, the emergence rates, percentage of flies with deformed wings and flyers from the QC and RF groups were highly correlated. Overall, these results highlight that the quality protocol procedures had no negative impact on adult emergence and predicted well the amount of sterile males available for the SIT component. In eradication programmes such as the one implemented in the Niayes of Senegal, thousands of sterile male flies need to be processed weekly for release requiring many preliminary activities in the insectary (to separate and to count normal and abnormal flies after emergence, assess mortality rate and percentage of non-flyers after release in the field). With the results obtained from the QC group, it was shown that all these parameters predicted well the biological quality of the sterile male flies, which will reduce considerably the work load. More importantly, multiple handling of flies (generally at 2–4°C for the sorting) generates stress which reduces their quality which can be avoided using a sample for the quality control test [27].
Quality control protocols for SIT programs were initially developed for fruit flies, especially the Mediterranean fruit fly Ceratitis capitata and has more recently been extended to Anastrepha and Bactrocera fruit fly species [38,39]. For these insects, the average flight ability after irradiation and transport of pupae was 65% for C. capitata, 75% for Anastrepha suspensa and 55% for Bactrocera oleae [38,39,48]. The flight ability obtained in the present study with G. p. gambiensis, BKF strain was on average 35.8 ± 18.4%. Although caution is required when comparing data from different species and when pupae were shipped under different conditions, it provides an indication that our results with G. p. gambiensis were rather low. This low propensity to fly could be due to mechanical shocks and vibrations that were absorbed by the pupae during transport or possibly different handling procedures in the different insectaries. In addition, the length of the cooling period of the pupae seems to be an important quality reducing factor, especially in terms of emergence rate. The impact of these different variables on emergence of adults was shown before [27]. Adults emergence may also be affected by excessive temperatures or inappropriate relative humidity during the rearing process [30]. In addition, it is well established that irradiation could potentially lower the quality of the produced flies especially when the irradiation dose that is required to obtain 95–100% sterility is high and therefore results in severe somatic damage [30,49–51].
The released sterile males must be active to find a blood meal, shelter and to compete with wild males for mating with wild females and successfully transfer the sterile sperm, and they must survive long enough to be able to find the virgin females [30]. Data of the QC group indicated that about 20% of the flies that emerged were “normal-looking” flies that had their wings deployed but did not escape the flight cylinder. What is measured here is the propensity of the flies to fly i.e. some of those flies that stayed in the cylinder probably can fly, but for one or the other reason they don’t. This was confirmed by the data from the field in that most of these flies were able to take off from the release cloth; however, they still might be poor flyers (but this was not assessed in this study). Indeed, after the preliminary sorting at the insectary, all normal-looking flies (i.e those that were mobile and had deployed wings) were transported to the field and released using the ground release protocol where the flies were released on a cloth (2 x 1.5 m) and checked after 5 minutes.
These observations confirm the necessity to implement a quality control protocol for sterile males to make eradication campaigns more effective. Weekly data on the percentage of released sterile males as compared to the number of shipped pupae allows for crucial feedback information to the rearing facility and to better plan the operational phase of the SIT component of AW-IPM programmes [4]. For example, by improving the packaging and transport protocols (such as the use of cotton for the CIRDES pupae and sawdust/vermiculite for the SAS pupae to cushion the mechanical shocks) flight ability was increased significantly reaching 55% in 2014.
There was no difference between the survival of the sterile males that emerged from the CIRDES and the SAS pupae. This indicates that the quality of the blood diet and the performance of the females in the colonies of the two rearing facilities were equivalent. The sterile males did not receive any blood meals during the survival experiment, and hence, their survival depended only on the fat reserves acquired during larval development. As tsetse reproduce by adenotrophic viviparity [4], these fat reserves are closely linked to the quality of the blood meals that are taken by the female parents. Under these conditions, the median survival of sterile males was 6 days regardless of the origin of the pupae with more than 80% and 10% surviving until 4 and 8 days after emergence, respectively. These results were similar to thoses observed with G. pallidipes, i.e. 90% of G. pallidipes males that emerged from pupae that had been exposed to a low temperature of 15°C survived unfed until 4 days but less than 10% survived after 8 days [52]. In order to simulate the proposed use of the chilled adult release system for area-wide tsetse SIT, the tenerale male flies of the same tsetse species exposed to a temperature of 7°C for 48 and 72 hours followed by 6 hours at 4°C and monitored without being offered a blood meal showed a median survival of 4 days [52].
The flies that emerged from batch 2 pupae survived on average longer than those emerging from batch 1 pupae indicating that the duration of the chilling at 8°C had a negative impact on fly quality. The median survival of the sterile males of 6 days without food as observed under laboratory conditions is encouraging, as the sterile males that are destined for release are being offered a blood meal at least three times before being released. This will undoubtedly increase their fat reserves thus enhancing their survival until they have found a host and hence, their competitiveness.
In conclusion, although the quality protocol data indicated that the percentage of flyers was less than 40%, the quality of the transported sterile pupae improved with time. More importantly, the data from the field indicate that the competitiveness of those male flies that were released was very good [53] resulting in excellent progress in the eradication campaign [54–56]. Research continues to improve the transport conditions of the pupae to potentially further increase the proportion of flyers.
|
10.1371/journal.pgen.1002404 | Acute Multiple Organ Failure in Adult Mice Deleted for the Developmental Regulator Wt1 | There is much interest in the mechanisms that regulate adult tissue homeostasis and their relationship to processes governing foetal development. Mice deleted for the Wilms' tumour gene, Wt1, lack kidneys, gonads, and spleen and die at mid-gestation due to defective coronary vasculature. Wt1 is vital for maintaining the mesenchymal–epithelial balance in these tissues and is required for the epithelial-to-mesenchyme transition (EMT) that generates coronary vascular progenitors. Although Wt1 is only expressed in rare cell populations in adults including glomerular podocytes, 1% of bone marrow cells, and mesothelium, we hypothesised that this might be important for homeostasis of adult tissues; hence, we deleted the gene ubiquitously in young and adult mice. Within just a few days, the mice suffered glomerulosclerosis, atrophy of the exocrine pancreas and spleen, severe reduction in bone and fat, and failure of erythropoiesis. FACS and culture experiments showed that Wt1 has an intrinsic role in both haematopoietic and mesenchymal stem cell lineages and suggest that defects within these contribute to the phenotypes we observe. We propose that glomerulosclerosis arises in part through down regulation of nephrin, a known Wt1 target gene. Protein profiling in mutant serum showed that there was no systemic inflammatory or nutritional response in the mutant mice. However, there was a dramatic reduction in circulating IGF-1 levels, which is likely to contribute to the bone and fat phenotypes. The reduction of IGF-1 did not result from a decrease in circulating GH, and there is no apparent pathology of the pituitary and adrenal glands. These findings 1) suggest that Wt1 is a major regulator of the homeostasis of some adult tissues, through both local and systemic actions; 2) highlight the differences between foetal and adult tissue regulation; 3) point to the importance of adult mesenchyme in tissue turnover.
| It is important to understand the cellular and molecular pathways that regulate the maintenance and turnover of adult tissues. These processes often go awry in diseases and are likely to deteriorate with ageing. Here we show that removal of a single gene, the Wilms' Tumour gene, Wt1, in the adult mouse leads to the extremely rapid deterioration of multiple tissues. Within 7–9 days after gene removal kidneys fail, the pancreas and spleen suffer severe atrophy, there is widespread loss of bone and body fat, and red blood cells are no longer produced. Our findings reveal the vulnerability of adult tissues, while opening up avenues for dissecting the pathways controlling tissue turnover. Further experiments showed that the tissue failure we observed is due both to local defects of stem/progenitor cell activities and to significant changes in the serum levels of some key master regulators. In particular there is a dramatic reduction in the levels of IGF-1, a key regulator of homeostasis and aging. Our studies also show that the control of adult tissue turnover may be different from that during foetal development. These findings have important implications for understanding and treating common human diseases.
| Although much is known about the mechanisms that govern cellular differentiation during development, we know less about the processes that regulate cell turnover and homeostasis in the adult. Perhaps the exceptions to this rule are rapidly turning over tissues such as intestine, skin and haematopoietic tissue. Recently it has been shown that genes required for regulating differentiation during foetal development may not be used in regulating turnover of the same tissues in the adult [1], [2].
Mutation of the Wilms tumour gene, WT1, in humans may lead to the eponymous paediatric kidney cancer, glomerulosclerosis of the kidney and gonadal dysgenesis, which can manifest as male to female sex reversal [3]. During foetal development, Wt1 is expressed in the kidney, gonads, spleen, the mesothelium which surrounds most organs as well as ill-defined body mesenchyme. Knockout mice lack kidneys, gonads, and spleen and the animals die at mid-gestation through the lack of coronary vasculature formation [4]. There are no apparent defects of the skeletal, haematopoietic, digestive, or metabolic systems.
Recently we have shown that Wt1 is a key regulator of the balance between the epithelial and mesenchymal states in a number of developing organs. Whereas it is required for the mesenchymal to epithelial transition (MET) underlying the formation of kidney nephrons, in the heart it is essential for the reverse process, the epithelial to mesenchyme transition (EMT) required for the production of proliferating cardiovascular progenitors from the epicardium (a mesothelium) [5]. In a similar vein Wt1 expressing mesothelial cells in the intestine and lung produce mesenchymal progenitors for vascular smooth muscle [6], [7]. Furthermore, very recent evidence proves that, in the developing liver, Wt1 expressing mesothelial cells provide the precursors for stellate cells [8], [9], [10]. Stellate cells in the liver and the pancreas have aroused much interest through their ability to regulate tissue fibrosis, via the production of cytokines [11], [12]. They are also important for the progression of pancreatic cancer [13].
In the adult, Wt1 is expressed in very few tissues in a small percent of cells. These include the mesothelium surrounding a number of visceral organs [14], the glomerular podocyte cells of the kidney, Sertoli/granulosa cells in the testes/ovaries [15], [16], [17] and 1% of bone marrow (BM) cells (with properties of restricted haematopoietic progenitors) [18]. Nothern Blot analysis has shown that Wt1 is also expressed in a variety of epithelial cells including spleen, lung and heart. Our own data, including those provided in this paper suggest that this mainly reflects expression in the mesothelial lining of these tissues. We speculated that the expression of Wt1 in these rare sites in the adult could have functional significance, for the following reasons. Firstly, given the importance of the mesothelium as a source of progenitor cells, requiring Wt1 function during development, we hypothesised that mesothelia might perform a similar function in the adult and this might require Wt1.
Secondly, Wt1 is essential for the formation and maturation of podocytes [19]. We hypothesised that continued expression of Wt1 in the adult would reflect a role in kidney maintenance.
Thirdly, WT1 is mutated or overexpressed in acute myeloid leukaemia (AML) [20]. However, Wt1 is not required for foetal haematopoiesis [21]. Given Wt1 expression
in adult bone marrow and association with leukaemia, we surmised that Wt1 might play a role in adult haematopoiesis.
Finally, WT1 is expressed at high levels in most adult cancers studied [22], though expression has not been detected in the normal tissue counterparts. It has been proposed that WT1 might be an oncogene in adult cancer in contrast to
its function as a tumour suppressor in paediatric kidney cancer [3]. As a prelude to testing this, it was necessary first to determine whether the gene is essential for normal development or maintenance of the epithelia from which these tumours arise.
To address these propositions, we deleted the Wt1 gene ubiquitously in adult mice. While our findings inform on these issues, the results far exceeded our expectations. The range, severity, and rapidity of
the phenotypes observed were dramatic and unexpected and raise major questions about adult tissue homeostasis.
To enable inducible deletion of Wt1 in the adult, we generated tamoxifen inducible Wt1 KOs by crossing CAGG promoter driven Cre-ER™ mice with our homozygous Wt1 conditional mice, where the first exon of Wt1 is flanked
by loxP sites [5]. Successful Wt1 deletion was demonstrated by recombination PCR and the depletion of Wt1 expression in mesothelia (Figure S1 and Figure S2). Deletion of Wt1 in the mesothelium did not affect the integrity of the tissue (Figure S3). The health status of the mutant animals deteriorated quickly and all the mice had to be culled by 10 days post-induction (p.i.). Prior to death, the mutant mice presented dramatic
phenotypes; they were less active and oedemic. Upon dissection, fluid was sometimes found in the abdominal cavity and in the subcutaneous tissues. Detailed gravimetric analysis showed that there was a reduction in the spleen to body weight ratio as well as in the heart to body weight ratio (Table 1). Subsequent histological analysis of internal organs revealed pale kidneys, severe spleen and pancreas atrophy, and deficiency of fat tissues. For most tissues, mice treated at 3, 10, or
13 weeks of age developed the same phenotypes. The only exception to this involved fat, as we discuss in more detail later. Before considering each phenotype, it is important to emphasise that not all tissues showed overt
signs of damage. For example, we observed no obvious macroscopic changes to the lung, liver or intestine- three tissues often involved in systemic inflammatory responses. Furthermore, although there was a 30% reduction in the heart/body weight ratio there was no obvious cardiovascular pathology (Table 1).
Wt1 is crucial for kidney development as the conventional Wt1-null embryos suffer from renal agenesis [4]. Upon induction of Wt1 deletion in our model, expression of Wt1 in the podocytes was completely depleted (Figure 1B) and the mutant mice were shown to have severe proteinuria (Table 2). H&E staining showed that the tubules were filled with protein casts (Figure 1A, arrow). The mutant kidneys had well developed glomerulopathy with cytopathic changes in podocytes and parietal epithelium. There was almost complete loss of synaptopodin and nephrin expression in the podocytes in the mutant kidneys (Figure 1C and 1D). EM studies showed that the foot processes of the podocytes were completely lost in the mutant kidneys (Figure 1E, day 10 post-injection). The development of the kidney phenotype in our model was extremely rapid. Five days post-tamoxifen injection, H&E stained kidney sections showed normal histology while podocyte effacement started to appear (Figure 1F). At day 7 post-injection, protein casts in the tubules were already present and the glomeruli started showing signs of degeneration (Figure S4a). Finally, plasma levels of urea and creatinine were normal at day 5 p.i., started to rise at day 7 p.i., and were significantly elevated at day 10 p.i (Table 2). In our model, mice that were heterozygous for the Wt1 conditional allele (CAGG-CreER™; Wt1loxP/+) did not exhibit any kidney abnormalities after tamoxifen-mediated deletion of Wt1. In addition, tamoxifen treated mice that were only positive for the CAGG-CreER™ allele and wild type for the Wt1 loxP sites (i.e. CAGG-CreER™ positive; Wt1+/+) were also included as controls and did not demonstrate any phenotypes. The kidney phenotype in our model is similar to other nephrotic syndrome mice where podocytes are damaged [23], [24], [25], [26]. However, none of these other mouse models presented any of the other phenotypes we describe below apart from the kidney defects. Most importantly, Wt1 has been deleted specifically in adult podocytes. These animals develop glomeruloscelosis similar to that described here but did not develop the other phenotypes we report below. Furthermore,
the mice survived well beyond the timeframe reported here [27].
Wt1 is expressed at E9 in the urogenital ridge and subsequently in the sex cords of the genital ridge in mice and it is a crucial factor for gonad development and sex determination [28]. In adult mice, Wt1 is expressed in Sertoli cells in the testes and granulosa cells in the ovaries [15]. We observed a reduction in the size of the testes and ovaries; however the difference was not statistically significant (Table 1). None of the testis markers studied showed any difference in expression patterns (Figure S5).
Asplenia in the conventional Wt1-null mice correlates with enhanced apoptosis in the primordial spleen cells [29]. In the adult Wt1 KO model, the mutant spleen was much paler and smaller in size compared with the control spleen (Figure 2A, arrow). There was a reduction in the number of proliferating cells in the mutant spleen; however the number of cells expressing an apoptotic marker (active caspase 3) remained unchanged (Figure S8A–S8D). The spleen to body weight ratio was reduced by 60% in the mutants of both the young (Figure 2D, 3 week old, p-value = 0.003; 8 controls and 5 mutants were analysed) and mature groups (Figure 2D, p-value = 0.000, 9 controls and 12 mutants were analysed).
The mutant mice had diminished extramedullary haematopoiesis within the red pulp compartment while white pulp remained largely unaffected (Figure 2B, 2C). FACS analysis showed an almost complete absence of erythrocytes (Ter-119 positive) in the mutant spleens (Figure 2E, 0.69±0.17% in the mutant c.f. 55.7±3.9% in the control spleen, p-value = 0.024; five controls and three mutants were analysed) and in Wt1-mutant bone marrow (Figure 2E, 7.3±3.1% in the mutant c.f. 30.3±4.0% in the control bone marrow, p-value = 0.025; five controls and three mutants were analysed).
Maturation of red blood cells requires erythropoietin (EPO) [30], which is synthesised mainly in the kidney. Furthermore, Wt1 has been shown to transcriptionally activate the EPO gene [31]. To determine whether the defect in erythropoiesis is intrinsic to the haematopoietic system, we cultured the mutant bone marrow cells in a methylcellulose-based system where a complete set of factors for supporting haematopoietic differentiation is provided in the medium. After two weeks in culture, despite the presence of all the required growth factors, the Wt1-mutant bone marrow cells failed to differentiate into the erythrocyte lineage, while the control bone marrow cells, as expected, did form red blood cells (Figure 2F, 5.0%±1.87%
in the mutant compared with 31.3%±9.6% in the control; five controls and three mutants were analysed, p-value = 0.05).
To address whether this defect in erythropoiesis reflects a cell autonomous role for Wt1 in haematopoiesis, we set out to characterise the 1% of bone marrow cells that express Wt1. Using the Wt1-GFP knockin mouse (Wt1GFP/+), we FACS sorted GFP positive cells from the bone marrow of Wt1GFP/+ mice and cultured them in a methylcellulose-based system. It has been shown previously that some Wt1-expressing cells in the bone marrow express markers characteristic of short-term haematopoietic stem cells (Ter119−CD45+Mac-1loc-kit+Sca-1+) [18] but the differentiation potential of these cells was not investigated. Hence we investigated the potential of these Wt1-GFP cells to differentiate to different haematopoietic lineages in culture. First we stained the GFP positive BM cells with a set of haematopoietic stem cell markers (CD150, CD48, and CD244) [32] and showed that approximately 50% of GFP positive BM cells were in the population of oligolineage-restricted progenitors (CD150−CD48+CD244−). Before culturing, no GFP-positive cells were positive for Ter-119 or Cd11b and only a few percent of the cells expressed CD45. After two weeks in culture, the GFP-positive cells were able to form Ter119 (red blood cells), CD45 (white blood cells), and CD11b (granulocytes) positive cells (Figure 2G). From this we can conclude that the Wt1-expressing cells are oligolineage-restricted progenitors.
We then set out to test if the reduction of erythrocytes reflected a decrease in the number of erythrocyte progenitors (Pre CFU-E) using the high resolution myeloerythroid progenitor cell staging method described by Pronk et al [33]. Representative flow cytometric profiles are shown in Figure 3. We saw a significant reduction in the % of Pre CFU-E in the mutant spleen (Figure 3, 0.27±0.06 in the controls and 0.03±0.008 in the mutants, p-value = 0.001; 7 control and 8 mutant mice were analysed). Erythrocyte progenitor cells branch from megakaryocyte-erythrocyte progenitors (PreMegE). Another progenitor that branches from PreMegE is the Megakaryocyte progenitor (MkP) which produces platelets. Both MkP and Pre MegE were reduced significantly in the mutant spleen (Figure 3) However, the number of platelets in the circulation was not affected (control platelet number is 880.5±89.9 K/µL and mutant platelet number is 817.5±164 K/µL). Mutant mice did not show any obvious bleeding tendencies. The half life of platelets is about 35 hours [34]. Platelet deficiency may have developed if the mice had survived longer.
We observed abnormalities of the growth plate in both the tibias and femurs of Wt1-mutant mice. The vascular invasion zones were irregular and anaemic (Figure 4A, indicated by arrow). The proliferative zone chondrocytes of the mutant mice were irregular with less surrounding territorial matrix than control mice (Figure 4A). The inner (marrow) surface of the long bone from the mutant mice was ragged compared with control mice (Figure 4B, arrows), suggesting increased bone resorption. We then analysed the bone architecture of femurs, tibias, and spine 9 days after induction of Wt1 deletion using μCT (Figure 4C). The 3D movie of the trabecular bone loss is shown in Videos S1 and S2. Trabecular bone volume was reduced by 30% in the mutants (Figure 4D), mostly due to a reduction in trabecular number and a small reduction in trabecular thickness. Furthermore, trabecular connectivity was also reduced. Taken together, these changes in bone architecture would be expected to lead to a substantial reduction in bone strength (Figure 4D). The bone loss observed could be due to either reduced bone growth or increased bone absorption. However, bone formation is a relatively slow process, and in view of the rapidity of the phenotype observed here it seemed that increased bone resorption was the more likely cause. We therefore stained sections of the long bones for the osteoclast marker TRAcP and observed dramatically increased numbers of osteoclasts on the bone surface of the Wt1-mutant bones (Figure 4E). To test if these bone phenotypes might reflect an intrinsic role for Wt1 in the osteoclast and osteoblast lineages, we harvested fresh bone marrow cells from the mutant mice, induced Wt1 deletion by treating the bone marrow cells with 4-OH tamoxifen for three days and cultured the cells in media supplemented with M-CSF and RANKL to induce osteoclast differentiation. Surprisingly and in contrast to the in vivo study, mutant bone marrow cells in which Wt1 had been deleted by tamoxifen treatment were less capable of forming osteoclasts in vitro (Figure 4F, p-value = 0.05 and 0.029 at 10 and 30 µg/ml of RANKL respectively; three separate experiments were performed each using bone marrow pooled from 2–3 control or mutant mice). When we used a similar in vitro approach using culture medium inducing osteoblastic differentiation, we observed that Wt1-mutant osteoblasts had reduced bone differentiation ability as levels of the osteoblast marker enzyme alkaline phosphatase were reduced (Figure 4G, p-value = 0.037; three separate experiments were performed). These results suggest that Wt1 plays an intrinsic role in both osteoclast and osteoblast differentiation, and that the loss of Wt1 is likely to disturb bone homeostasis.
The Wt1-mutant mice also displayed reduction in the size of fat pads. In addition to the abdominal fat pads which mainly comprise white adipocytes, interscapular brown adipocytes were also atrophic and had fewer lipid cytoplasmic vacuoles than controls (Figure 5A–5J). Although the trend of fat loss was consistent in mutant mice, the reduction of fat pad size seemed to be more variable in the older group of animals (13 weeks, Figure 5K, arrows). In some mutant animals, the reduction in the size of fat pads was observed in both the interscapular and abdominal fat pads, while in other mutants the lipid vacuole size reduction was seen in the abdominal fat pads but not in the interscapular fat pads. The weight of fat pads in the mutant mice did not reflect their actual size because of the oedema (data not shown), and we therefore analysed fat pad volume using whole body μCT scans. Mice were scanned at the start (before tamoxifen injection) and the end of the experiment (9 days after induction). Results from the μCT scan confirmed the substantial fat loss in the mutants (Figure S7, arrows). There was no difference in the number of apoptotic and proliferating cells in the fat pads between mutant and control mice. Histological analysis of the adipose tissues showed that the reduction in the size of fat pads reflected a decrease in the vacuole size of the adipose tissues, as seen in the abdominal fat pads (Figure 5L–5M, p<0.05; three controls and three mutant mice were analysed). Consistent with this loss of fat, there was a significant reduction in the level of AP2 expression in mutant abdominal fat pads (Figure 5N, p-value = 0.05; three controls and three mutants were analysed).
Wt1 expression in fat has not been reported previously. However, here we show that Wt1 is expressed in the mesentery, epididymal, and retroperitoneal fat pads, but not at detectable levels in the abdominal fat pad nor in the interscapular brown adipose tissue (Figure 5O, 5P). Given the fact that adipocytes and osteoblasts have a common origin in the bone marrow, we examined whether there was any alteration in the number of adipocytes in the bone marrow. Labelling adipocytes using AdipoRed, we found a reduction in the number of adipocytes in the mutant bone marrow (Figure 5Q, p-value = 0.021; four controls and four mutants were analysed).
As adipocytes and osteoblasts arise from the stromal mesenchymal population in the bone marrow, we speculated that Wt1 loss might lead to a disturbance in this population which can be quantified using an antibody to Stro-1. We did in fact find a significant (five fold) increase in this population of cells following Wt1 loss (Figure 4H, p-value = 0.02; four controls and four mutants were analysed).
Figure 6G–6J (arrows) shows the successful depletion of Wt1 expression in the pancreatic mesothelium. The pancreas from the mutant mice was severely atrophied. H&E staining demonstrated that there was a substantial amount of cell loss in the exocrine tissues while the endocrine pancreas remained largely unaffected (Figure 6A, 6B). Acini in the mutant pancreas were loosely packed and acinar cells appeared atrophied and presented less eosinophilic cytoplasmic staining, suggesting a reduced zymogen content. Residual acinar epithelial cells were rounded and less cohesive with neighbouring cells. Similar aberrant histology started to appear at day 7 after Cre activation (Figure S4B). We saw an increase in the number of apoptotic cells in the mutant pancreas (Figure 6C, 6D) while the number of proliferating cells remained unchanged (Figure 6E, 6F). Although the pathology of our model shares many similarities to pancreatitis mouse models, there was no elevation of serum amylase level in the Wt1-mutant mice (Table 2). Given the severity of the pancreas phenotype, it is surprising to see the lack of any elevation of serum amylase. However, this probably reflects the short space of time between the onset of the phenotype and death of the mice. Pancreatitis involves inflammation of the pancreatic tissues and in Wt1-mutant mice we observed a low-grade inflammation in much of the pancreas and scattered foci of more severe active inflammation. In the Wt1-mutant pancreas, the presence of infiltrating macrophages was confirmed by staining with macrophage marker F4/80 (Figure S6E, S6F); however, staining of CD11b, Gr1, and CD3 were absent (data not shown). Both insulin and amylase expression were normal in the mutant pancreas sections (Figure S6A–S6D).
To try to gain more insight into the origin of the pancreatic phenotype, we examined more closely the cell types that express Wt1 in the exocrine pancreas. Pancreatic stellate cells (PSCs) have been implicated in pancreatitis and pancreatic cancer. We show Wt1 is expressed in the mesothelial lining of the pancreas as well as in PSCs. Desmin is a marker for PSCs [35]. The interstitial cells that express Wt1 also express desmin, and this was demonstrated in sectioned pancreata (Figure 6K–6M) and in cultured PSCs (Figure 6N).
One possible explanation for the dramatic and acute nature of the phenotypes observed in these mice is a systemic inflammatory response, even though analysis of the diseased pancreas did not suggest this. Furthermore, even though the animals appeared to show no signs of distress and their stomachs were full at 9–10 days, it is possible that the bone and fat defects were due to nutritional deprivation. To assess these possibilities, we carried out quantitative analysis of 40 cytokines and 38 adipokines in mutant versus wildtype serum using antibody arrays. Perhaps surprisingly, given the severity of the phenotypes there was no statistically significant change in the levels of any inflammatory cytokines (Figure 7A; three controls and three mutants were analysed), arguing that the phenotypes were not due to a systemic inflammatory response. As a positive control to test that the arrays were working, we treated the mice with LPS and then assayed cytokine levels. There was a 23 fold induction in MIP-2, an 11 fold induction of JE, a 6 fold induction of KC, and a 3 fold induction of TNFα (Figure 7B). These findings demonstrate that the assays work and are able to measure an acute systemic inflammatory response. Similarly, there was no indication of nutritional deprivation. Following calorific restriction, there is reported to be a 60–80% reduction in serum leptin levels [36], [37], a 65% reduction in TNFα [38], a 100% increase in AgRP/FIAF [39], and a 75% increase in the levels of adiponectin [40]. We saw no significant changes in any of these molecules (Table 3 and Figure 7C), supporting the idea that the mice were not suffering nutritional deprivation and, in turn, this was not causing any of the phenotypes. However, we did observe a dramatic 85% reduction in the levels of IGF-1 and 3.5 fold increase in the levels of FGF21 (Figure 7C). This could in part account for the bone and fat phenotypes respectively. To investigate if the reduction of IGF-1 levels could due to global growth hormone deficiency, we measured circulating growth hormone (GH) using ELISA. We observed a slight elevation of GH levels in the mutant serum (Figure 7J; three controls and five mutants were analysed, p-value = 0.025). Histology analysis showed absence of any pathological abnormalities in the pituitary and adrenal glands (Figure 7D–7I).
The multiple organ disturbance observed in adult mice deleted for Wt1 is striking, and, we believe, unprecedented in terms of severity and rapidity of onset. There is perhaps no need to point out that most of these phenotypes have relevance for diseases common in adults, even though our starting point was a gene more or less defined for its role in the development of several organs. Our study shows that Wt1 plays a key role in regulating the production or turnover of red blood cells, bone and fat in the adult. Despite intensive
analysis of Wt1-null foetuses, including those surviving to 18 days gestation, no developmental defects in these tissues were found previously [4], [29]. Thus our study contributes to the growing body of evidence that adult tissues may employ different or additional players compared to foetal development. Wt1 is among a list of genes whose methylation increases with age in a genome-wide CpG island methylation profiling study [41]. Therefore Wt1 expression levels may well decrease with age. It will be important to determine whether Wt1 levels in these key cell populations reduce during aging or under different environmental influences. If so, this could contribute to disease-related phenotypes described here.
Although there is much future work needed to elucidate the mechanisms underlying these phenotypes, there are several conclusions we can draw at present. Perhaps, surprisingly, we could detect no significant changes in serum cytokine levels, arguing that the phenotypes we observe are unlikely to be due to a systemic inflammatory response, even though this is often associated with damage to the tissues that are affected in the Wt1 mutant mice. As we argue below, the phenotypes involving the kidney and erythrocytes reflect an intrinsic function of Wt1 in these tissues or their progenitors. On the other hand, we believe loss of fat and bone is likely to be a combination of systemic and local factors.
The phenotypes involving the haematopoietic system and bone, have their origins wholly or partly within the bone marrow itself. Wt1 is expressed in a restricted haematopoietic progenitor population and its loss leads to disturbance in red blood cell and osteoclast production. This is consistent with the previous finding that Wt1 expression is upregulated during early myeloid differentiation (particularly in the common myeloid progenitors and megakaryocyte-erythroid progenitors) [18]. In keeping with this, we found the levels of PreMegE, MkP, and Pre CFU-E were significantly decreased in mutant spleen, Given the association of Wt1 with AML, we might have expected an imbalance in the myeloid compartment. Preliminary analysis has not demonstrated a reduction in the absolute number of monocytes and granulocytes in the circulation of Wt1 mutant mice. However, this may have only become evident if the mice had survived longer.
The bone loss in most part is likely to result from the increase in osteoclasts that we observed in the bone marrow. Paradoxically, mutant mice showed a reduction in osteoclast formation ability in vitro. The bone marrow compartment in which we saw an increase in the number of osteoclasts consists of a mixed population of cells. The mesenchymal stromal cells and haemaatopoietic stem cells are in close proximity in the bone marrow and there is known to be crosstalk between these cell types [42], [43]. Our in vitro osteoclast formation cell culture system started with a restricted population of cells (bone marrow stromal cells). The in vivo and in vitro difference could be due to factor(s) that are present in the bone marrow but absent in the in vitro culturing system.
However, we also found that Wt1 is required for osteoblast synthesis in bone marrow culture pointing to a role in the mesenchymal lineage. Consistent with this, our preliminary experiments have shown that non-haematopoietic Wt1-GFP positive cells from the bone marrow stroma are able to differentiate to bone and fat (unpublished observations). Furthermore, we show here that Wt1 loss also leads to an increase in Stro1 positive stromal mesenchymal stem cells, which may explain partly the disturbance in adipocyte and osteoblast production in the bone marrow. Our serum protein analysis showed a dramatic reduction of IGF-1 levels and this might be expected to contribute to the bone loss phenotype. Interestingly, deletion of IGF-1 specifically in the
liver, the major source of synthesis, only leads to a 75% reduction in circulating IGF-1 levels and there is no apparent phenotype [44]. However, mice that are double homozygous mutant for IGF-1 and the binding protein acid labile subunit (ALS) [45] show an 85% reduction in IGF1- levels and a similar degree of bone thinning to that seen in our Wt1 adult knockout mice. Hence, it seems reasonable to conclude that the 85% reduction of IGF-1 levels in our mutant mice is a major factor behind the bone phenotype. In the Wt1 mutant mice, the IGF-1 levels are much lower than those observed when IGF-1 is deleted specifically in the liver, so either Wt1 is required for IGF-1 expression in non-hepatocytes, or for factors that stabilise IGF-1 in the serum. Growth hormone, produced by the pituitary gland, is a major regulator of IGF-1 levels. One possibility was that the reduction in IGF-1 level was due to defects in the pituitary axis and downregulation of GH. However, we detected no pathological abnormalities in the pituitary and adrenal glands, and if anything GH levels were increased.
Obesity is a major health problem and there is considerable topical interest in the factors that regulate fat levels. Loss of Wt1 not only leads to reduced adipocyte production in the bone marrow but also to rapid systemic loss of fat, with dramatically reduced vacuole size. There are several reasons why we believe this fat loss is not due to under-nourishment. Fat vacuole reduction was already apparent 7 days after tamoxifen injection, at which time the health status of the animals was normal. Nine days post-injection, the mutant mice still actively sought food and their stomachs were full on autopsy. Importantly, there was no change in the levels of circulating leptin, adiponectin, TNFα, and AgRP/FIAF, all of which would be expected to change dramatically after one or two days of calorific restriction. There was a reduction in the level of lipocalin 2 in mutant serum (Figure 7C). Lipocalin 2 is abundantly produced from adipocytes [46], [47]. The reduction of lipocalin 2 could be caused by the reduced volume of adipose tissues in mutant mice. Taken together our findings provide evidence that Wt1 may influence both the formation and maintenance of adipocytes. The fat loss is extremely rapid and given that Wt1 only appears to be expressed in a proportion of fat pads affected, it seems likely that systemic factors might be involved. We found that the levels of circulating FGF21 increased by 3.5 fold in the mutant animals and this would be expected to induce some fat loss [48].
The RT-PCR result showed that Wt1 expression was detected in fat pads (Figure 5O). In preliminary experiments to address whether this reflects expression in mature adipocytes or the stromal vascular compartments, we digested and fractioned fat tissues from the Wt1-GFP knockin mice into the floating mature adipocyte layer and the stromal vascular fraction. The majority of the GFP signal was seen in the stromal vascular fraction (unpublished data). This supports the idea that systemic or local paracrine factors dependent on Wt1 are regulating adipocyte homeostasis.
The effect of Wt1 loss on bone and fat turnover is interesting in the context of Wilms' tumours. We and others have shown that the 15–20% subset of Wilms' tumours arising through WT1 loss are more likely to be stromal (mesenchymal) predominant and often contain ectopic tissues, including bone, fat, cartilage and muscle [49], [50], [51]. Taken together this and our new findings underline the key role of the mesenchyme and Wt1 in tissue turnover and maintenance.
With regard to the pancreatic atrophy, this does not appear to be typical pancreatitis as there was no increase in serum α-amylase. However, as discussed above, amylase level may have increased if the mice had lived longer. Serum cytokine profiles showed that there was no systemic inflammatory response in the mutant mice. In line with this there was no observable pathology in liver, lung and intestine, all tissues susceptible to inflammation. It remains to be seen whether the severe pancreatic atrophy is due to loss of Wt1 function within the tissue itself. We can exclude an effect through loss of Wt1 function in the islet or acinar cells as deletion of the gene specifically in these cell types using PDX1-Cre did not lead to overt pathology in the pancreas or elsewhere (P. Hohenstein, V. Brunton, M. Frame, O. Samson and N. Hastie unpublished observations). One possibility is that the pancreatic atrophy arises through activation of the sub-population of stellate cells that express Wt1 although further study is required to investigate this hypothesis. Activated stellate cells produce cytokines [52] and we speculate that these may be responsible for destroying the acinar cells. Given the published data on foetal liver [8], the parallel between pancreatic and hepatic stellate cells, and the role of Wt1 in generating vascular progenitors from the epicardium by EMT [5], we hypothesise that a proportion of pancreatic stellate cells arise from the mesothelium, via an EMT, once more pointing to the role of this tissue as a source of mesenchymal progenitor cells.
Despite the accumulating knowledge about the importance of Wt1 at multiple stages of kidney development, the function of Wt1 in the podocytes of mature glomeruli has remained the subject of some speculation. Even though children and adult mice with Wt1 mutations characteristic of Denys-Drash and Frasier syndrome develop glomerulosclerosis, it was always possible that the damage had its origin in utero, rather than reflecting a continued function for Wt1 in the maintenance of the adult kidney. Our results provide the first evidence that Wt1 is crucial for maintaining the integrity of mature podocytes. Our model allowed us to test whether the glomerulosclerosis we observed arises through abnormalities of cell proliferation or the differentiation state of the mature podocytes. We did not see major changes in proliferation or apoptosis in the mutant glomeruli deleted for Wt1 (using proliferation marker anti-phosph-histone H3 and apoptosis marker active caspase 3, Figure S8E–S8H). However, we showed that loss of Wt1 expression resulted in damage to the foot processes of the podocytes therefore causing a morphological alteration. Nephrin is necessary for the renal filtration barrier and is also a known downstream target of Wt1 during kidney development [53]. Consistent with this we found that nephrin expression levels reduce dramatically after Wt1 deletion, indicating that the transcriptional regulation of nephrin by Wt1 continues into adult life. Here we show that Wt1, known to be a key regulator of nephrogenesis, is also vital for the maintenance of adult glomerular structure and function, something that has been the subject of speculation but not proven until now.
Clearly these findings should be followed up using tissue specific Cre lines. However, at present suitable Cre lines are not available for several of the crucial lineages we wished to investigate, including the mesothelium and mesenchymal stem cells. In the mean time, we have been able to use cultures to show that several of the phenotypes we observed are intrinsic to the bone marrow.
The results presented in this study open new avenues of research into mesenchymal cell function in adult tissues. The cell types that express Wt1 in adult tissues e.g. the hepatic and pancreatic stellate cells and bone marrow progenitors are mesenchymal. The other major cell types expressing Wt1, namely the podocytes and mesothelia are considered epithelial, but are unusual in expressing high levels of mesenchymal markers, such as vimentin. Given our findings, it is interesting to speculate on the possible relationships between the cell types expressing and requiring Wt1 in these different tissues. Different reports have shown that stellate cells may arise from the mesothelium and bone marrow [10], [54]. Our studies suggest that Wt1 may have a function in both stellate cells and bone marrow mesenchymal stem cells. Stellate cells, like the epicardially-derived cells requiring Wt1, synthesise retinoic acid. One of the striking features of stellate cells is the presence of vitamin A (retinoid) droplets and this becomes lost upon stellate cell activation. In the epicardium we have shown that RALDH2 levels and RA are reduced when Wt1 is deleted and that RALDH2 is a direct transcriptional target of Wt1 [9]. We have shown that Wt1 is required for the EMT that generates RA-synthesising coronary vascular progenitors from the epicardium and it is interesting that an EMT is required for activation of stellate cells. It is also notable that stellate cells synthesise high levels of fat and it will be interesting to see if the Wt1 expressing cells in fat have similarities to stellate cells and mesenchymal bone marrow cells.
Finally our findings may also have implications for cancer therapy. There is a growing number of studies developing anti-WT1 immune therapy for common cancers predicated on the belief that WT1 is expressed at high levels in cancers [20], [55], but very low levels in the normal adult. Our findings raise questions about this approach as damage to these normal Wt1-expressing tissues might have adverse effects.
Mice were housed and bred in animal facilities at the MRC HGU and the University of Edinburgh. Animals were kept in compliance with Home Office regulations. The Wt1-conditional line was made in our group [5]. To obtain [CAGG-CreER™ positive, Wt1loxP/loxP] and [CAGG-CreER™ negative, Wt1loxP/loxP] transgenic mice, [CAGG-CreER™ positive] males were mated with Wt1loxP/loxP females, and the resulting offspring intercrossed. Wt1-GFP knockin mice used in this study were kindly provided by Professor H Sugiyama [18].
Cre recombinase was induced by intraperitoneal administration of tamoxifen (4 mg/40 g body weight for 5 days; Sigma). All animal work was carried out under the permission of license. To delete Wt1 in vitro, cells were treated with 4-OH tamoxifen (1 µM, Sigma) for three days.
Full methods are described in Text S1. Antibodies and primers are listed in Tables S1 and S2.
|
10.1371/journal.pntd.0004675 | Dermatophytosis among Schoolchildren in Three Eco-climatic Zones of Mali | Dermatophytosis, and particularly the subtype tinea capitis, is common among African children; however, the risk factors associated with this condition are poorly understood. To describe the epidemiology of dermatophytosis in distinct eco-climatic zones, three cross-sectional surveys were conducted in public primary schools located in the Sahelian, Sudanian and Sudano-Guinean eco-climatic zones in Mali.
Among 590 children (average age 9.7 years) the overall clinical prevalence of tinea capitis was 39.3%. Tinea capitis prevalence was 59.5% in the Sudano-Guinean zone, 41.6% in the Sudanian zone and 17% in the Sahelian eco-climatic zone. Microsporum audouinii was isolated primarily from large and/or microsporic lesions. Trichophyton soudanense was primarily isolated from trichophytic lesions. Based on the multivariate analysis, tinea capitis was independently associated with male gender (OR = 2.51, 95%CI [1.74–3.61], P<10−4) and residing in the Sudano-Guinean eco-climatic zone (OR = 7.45, 95%CI [4.63–11.99], P<10−4). Two anthropophilic dermatophytes species, Trichophyton soudanense and Microsporum audouinii, were the most frequent species associated with tinea capitis among primary schoolchildren in Mali.
Tinea capitis risk increased with increasing climate humidity in this relatively homogenous schoolchild population in Mali, which suggests a significant role of climatic factors in the epidemiology of dermatophytosis.
| Dermatophytosis, and particularly the subtype tinea capitis, is common among African children; however, the risk factors associated with this condition are poorly understood. To describe the epidemiology of dermatophytosis in distinct eco-climatic zones, three cross-sectional surveys were conducted in public primary schools located in the Sahelian, Sudanian and Sudano-Guinean eco-climatic zones in Mali. Among 590 children (average age 9.7 years) the overall clinical prevalence of tinea capitis was 39.3%. Tinea capitis prevalence was 59.5% in the Sudano-Guinean zone, 41.6% in the Sudanian zone and 17% in the Sahelian eco-climatic zone. Microsporum audouinii was isolated primarily from large and/or microsporic lesions. Trichophyton soudanense was primarily isolated from trichophytic lesions. Based on the multivariate analysis, tinea capitis was independently associated with male gender (OR = 2.51, 95%CI [1.74–3.61], P<10−4) and residing in the Sudano-Guinean eco-climatic zone (OR = 7.45, 95%CI [4.63–11.99], P<10−4). Two anthropophilic dermatophytes species, Trichophyton soudanense and Microsporum audouinii, were the most frequent species associated with tinea capitis among primary schoolchildren in Mali. Tinea capitis risk increased with increasing climate humidity in this relatively homogenous schoolchild population, which strongly suggests that climatic factors play a significant role in the epidemiology of dermatophytosis.
| Dermatophytosis represents one of the most common infectious diseases worldwide and causes serious chronic morbidity [1]. The condition is caused by dermatophytes, which are fungi that require keratin for growth. An increase in the incidence of such infections has been noted worldwide, especially in developing countries [2,3]. In particular, tinea capitis represents a major public health issue among children in developing countries. This dermatophytosis of the scalp and hair shafts is almost exclusively a childhood disease, and evidence suggests that it occurs more often in children of African or Caribbean origin [1]. Many factors including gender, age, urban/rural environment, socio-economic level and cultural habits have been shown to significantly impact the development of dermatophytosis worldwide, especially throughout the African continent [4–9]. It has been hypothesized that climate also plays an important role in the heterogeneity of dermatophytosis epidemiology in Africa [10]. Many studies have described dermatophytosis epidemiology in various geographical settings throughout the African continent. However, heterogeneous study design limits the assessment of numerous potential confounding factors and the striking differences observed cannot reliably be attributed to variations in climate. In fact, no study has been established to specifically address the impact of climate on dermatophytosis presentation. In Mali, the climate ranges from subtropical in the south to arid in the north. Therefore, the current study aimed to assess the prevalence, risk factors and etiological agents of tinea capitis, the most frequent dermatophytosis subtype, among primary schoolchildren in three eco-climatic zones in Mali.
Three cross-sectional surveys were carried out during the dry season, in December 2009, December 2010 and February 2012, in three public primary schools in (i) Sirakoro-Meguetana, a semi-urban community located in the suburbs of Bamako, the capital of Mali, in the Sudanian eco-climatic zone; (ii) Bandiagara, a urban community in the Sahelian eco-climatic zone and (iii) Bougoula-Hameau, a semi-urban community located in the suburbs of Sikasso in the Sudano-Guinean eco-climatic zone. The study sites and eco-climatic zones previously defined [11] are shown on the map in Fig 1.
The characteristics of each eco-climatic zone, from north to south, are detailed below. The Sahelian zone is arid with annual rainfall levels between 250 to 550 mm. In Sévaré, located 55 km from Bandiagara, the average rainfall levels range from 0 mm in January to 161 mm in August, with total annual precipitation of approximately 485 mm. The mean temperatures in Sévaré range from 23.2°C in January to 33.0°C in May, with an average annual temperature of 28.3°C. The Sudanian zone is semi-arid to sub-humid with annual rainfall ranging between 550 and 1100 mm. In Bamako, the average rainfall ranges from 0 mm in January to 291 mm in August, with total annual precipitation of approximately 955 mm. The mean temperatures in Bamako range from 25.0°C in December to 31.5°C. The Sudano-Guinean zone is sub-humid with average annual rainfall greater than 1100 mm. In Sikasso, the average rainfall levels range from 1 mm in January to 298 mm in August, with total annual precipitation of approximately 1125 mm. The mean temperatures in Sikasso range from 23.8°C in January to 30.2°C in April, with an average annual temperature of 28.3°C.
Pupils, aged 6 to 15 years, were randomly selected in each primary school using a block randomization design adjusted on the number of pupils in each classroom. Oral informed consent was obtained from the children and their parents or guardians. The exclusion criterion was a history of antifungal treatment (oral or topical, conventional or traditional) within two weeks. Medical history and information concerning exposure to potential dermatophytosis risk factors were recorded, including contact with animals and specific hair grooming habits, and a complete physical examination of the skin and appendages, including fingernails and hair, was performed on all children by one of the investigators. The data were recorded on a standardized clinical report form.
The study protocol was reviewed and approved by the Faculty of Medicine’s Institutional Review Board at the University of Bamako, Mali. The study protocol was also approved by the Local Education Authorities at each study site. Parent or guardian provided written informed consent on behalf of the participating children.
Samples were collected from each lesion that was compatible with dermatophytosis. Skin samples were isolated from the peripheral erythematous border of the lesion. Scalp lesions were collected by scraping the area with a sterile curette, and broken and lusterless hairs were selected and plucked using sterile tweezers (Fig 2A). One portion of the each sample was used for direct examination via microscopy, while the second portion was inoculated directly onto Sabouraud Dextrose Agar (SDA) (bioMérieux, Marcy l’Etoile, France) with antibiotics and cycloheximide for mycological examination. In the Sahelian and Sudano-Guinean zones, an additional sample was collected directly from the same lesion via sterile gauze. The samples inoculated on SDA were incubated at room temperature before being transferred to the Parasitology-Mycology Laboratory at the University Hospital of Marseilles, where they were incubated at 27°C for a of 4-6-week period. The sterile gauze samples collected in the Sahelian and Sudano-Guinean zones were stored at ambient temperature in individually sealed plastic bags before being inoculated on SDA agar (BioMérieux) and subsequently incubated for 4–6 weeks at 27°C at the Parasitology-Mycology Laboratory in Marseilles. Dermatophyte colonies were identified based on examination of the macro- and micro-morphological features of the fungus, and those with atypical morphological features were further identified via rDNA internal transcribed spacer 2 (ITS2) sequence analysis as previously described [12].
A sample size of 200 children in each study site was calculated to estimate a 12% dermatophytosis prevalence rate with a 4.5% precision at α = 5%. The data were analyzed using SAS 9.2 for Windows (SAS Institute Inc., Cary, NC, USA). Continuous variables were expressed as the mean (SD), while categorical variables were expressed as proportions and percentages. Continuous variables were compared using ANOVA. Categorical variables were compared using the Chi square or Fisher's exact tests as required. All statistical tests were two-sided with a P<0.05 significance level. Univariate and multivariate unconditional logistic regression analyses were performed to estimate odds ratios (ORs) with a 95% confidence interval (CI). All covariates with a P<0.20 significance level in the univariate analysis were included in the multivariate logistic regression model. A stepwise selection was performed to retain the most parsimonious model including the covariates that displayed an independent statistically significant (P<0.05) effect on tinea capitis risk.
Of the 590 randomly selected schoolchildren, 286 males and 304 females participated in this study, including 190 from Sirakoro-Meguetana, 200 from Bandiagara and 200 from Sikasso (Fig 1). Although the age distribution of the participating children in each eco-climatic zone differed significantly (P = 0.001), the mean age of each zone, which ranged from 9.3 to 10.2 years, was quite similar (Table 1). The sex ratio of the participating schoolchildren in each eco-climatic zone did not significantly differ (P = 0.151). As expected, the ethnic group distribution of each eco-climatic zone differed significantly (P<0.001) (Table 1).
As detailed in Table 1, 312 children presented with clinical dermatophytosis lesions, thereby yielding an overall dermatophytosis prevalence of 52.9% (95% CI [48.7–57.0]) among the entire study population. The most frequent (232/590) clinical presentation was tinea capitis with 39.3% 95% CI [35.4–43.4] prevalence. Therefore, tinea capitis was separately detailed in the study, and all other clinical presentations of dermatophytosis were categorized as non-tinea capitis, which was found in 80 (13.6%, 95% CI [10.9–16.6]) of the participating children. Among the 80 non-tinea capitis dermatophytosis cases, the most common clinical presentation was tinea corporis (81.3%), followed by tinea cruris (8.7%), tinea pedis (7.5%) and tinea unguium (2.5%), irrespective of the geographic area.
Overall, 39.3% (95% CI [35.4–43.4]) of the schoolchildren presented with clinical tinea capitis lesions. The prevalence of tinea capitis significantly (P<0.001) differed depending on the geographic area, with prevalence rates of 17.0% (95% CI [12.1–22.9]), 41.6% (95% CI [34.5–48.9]) and 59.5% (95% CI [52.4–66.4]) recorded in the Sahelian, Sudanian and Sudano-Guinean eco-climatic zones, respectively. The characteristics of the 232 schoolchildren presenting with clinical tinea capitis lesions are detailed in Table 1. The tinea capitis lesions were described as diffuse primarily in the Sudano-Guinean (55.5%) and Sudanian (51.9%) zones, while diffuse lesions were observed only in 26.5% of cases in the Sahelian zone (P = 0.012). Interestingly, this individual marker of infection intensity correlated with prevalence. Inflammatory and suppurative forms of tinea capitis were rarely observed. Tinea capitis presentation in the Sudano-Guinean zone was characterized by the predominance of microsporosis (Fig 2B), i.e. large lesions (>2 cm) involving Microsporum audouinii, in contrast to the Sahelian and Sudanian zones, which were characterized by the predominance of trichophytosis (Fig 2A), i.e. multiple (n>2) lesions involving Trichophyton soudanense (Table 1). As expected, the majority of microsporosis lesions (64.5%) and a particularly high proportion (80.3%) of large (>2 cm) tinea capitis lesions, displayed a typical green fluorescence upon Wood’s lamp examination, while most trichophytosis lesions (95.5%) did not display this feature. Of note, 30.3% of the diffuse scalp lesions tested positive upon Wood’s lamp examination.
Eighty children (13.6%) presented with non-tinea capitis clinical dermatophytosis (Table 2), of which tinea corporis was the most frequent (81.3%) clinical presentation. The prevalence of tinea corporis varied depending on the eco-climatic zone, ranging from 88.4% in the Sudanian to 71.4% in the Sudano-Guinean zone. Overall, six children (7.5%) presented with athlete’s foot, of which four (14.3%) originated from the Sudano-Guinean zone and two (4.7%) were from the Sudanian zone.
The majority of tinea capitis lesions (81.5%) tested positive upon direct microscopic examination. The distribution of endothrix or endo-ectothrix parasitism type did not significantly differ between the geographic areas (Table 1). In contrast, only 3.8% of the non-tinea capitis dermatophytosis samples tested positive upon direct examination (Table 3).
A dermatophyte was isolated in 58 of the 80 samples collected from children with tinea corporis, tinea cruris, athlete foot or onychomycosis, thereby yielding a global prevalence of 9.8% (95%CI [7.6–12.5]) of mycologically confirmed non-tinea capitis dermatophytosis cases (Table 3). Two dermatophyte species were predominant, namely T. soudanense (63.8%) and T. mentagrophytes (5%). The combination of both T. soudanense and T. mentagrophytes was found in 3.8% of positive cultures. M. audouinii was not isolated from non-tinea capitis lesions in this study. Twenty-two (27.4%) non-dermatophyte filamentous fungi were recovered, primarily in the Sudanian eco-climatic zone (Table 3).
The global and gender-specific distributions of the assessed tinea capitis risk factors are tabulated in Table 1. The distribution of the potential risk factors and habits associated with tinea capitis among schoolchildren according to their tinea capitis status is shown in Suppl Table 1. In the univariate analysis, several hairdressing habits were found significantly associated with tinea capitis; however, the associations were non-significant when gender was considered, which acted as a notable confounding factor. Proximity to cattle was associated with a significant decrease in tinea capitis risk among females but not among males. In contrast, the presence of a dog in the household was associated with increased tinea capitis risk among males but not among females. None of these risk factors were found statistically significant in the multivariate analysis. The effects of the most significant tinea capitis risk factors are detailed in Table 4. In the univariate analysis, tinea capitis was significantly associated with male gender (OR = 7.85, 95%CI [5.22–11.81], P<0.001) and residing in Bougoula-Hameau (OR = 7.17, 95%CI [4.51–11.4], P<0.001). Notably, age had no significant effect on this study population. In the multivariate analysis, both male gender (OR = 2.51, 95%CI [1.74–3.61], P<0.001) and residing in the Sudano-Guinean eco-climatic zone (OR = 7.45, 95%CI [4.63–11.99], P<0.001) were statistically significant independent tinea capitis risk factors (Table 4).
Overall, our study highlights three major findings: a dramatic disparity in tinea capitis epidemiology between distinct eco-climatic zones within Mali, evidence of an increased tinea capitis risk among male children and the dermatophyte species distribution. The dermatophytosis prevalence variations observed in the current study might be associated with varied exposure of the surveyed schoolchildren in each area to several risk factors, including eco-climatic, socio-economic factors and genetic or ethno-cultural elements. Notably, with the exception of ethno-cultural characteristics, the schoolchildren surveyed in each area were homogenous, especially in regards to age, and each survey was performed at the same period during the dry season. Therefore, the observed geographical differences in tinea capitis epidemiology may be attributed to variations in the local environment, and more specifically eco-climatic differences. Indeed, the highest prevalence of dermatophytosis was recorded in Bougoula-Hameau (59.5%), where the Sudano-Guinean climate is characterized by relatively higher levels of humidity; followed by Sirakoro-Meguetana (41.6%), where the Sudanian climate is characterized by intermediate humidity levels; and Bandiagara (17%), where the Sahelian climate is characterized by relatively lower humidity levels. This association trend of tinea capitis risk that increases with the humidity level of the climate correlates with data from Togo, where tinea capitis prevalence rates were 11% in a dry region in the North of the country and 20% in a humid area in the South of the country [13]. These geographical discrepancies might be associated with hot and humid climates, which favor the growth and spread of fungi and may predispose populations to various skin diseases including tinea capitis [14].
Tinea capitis primarily affects children in developing countries, while tinea pedis and tinea unguium pose the greatest burden to adults and the elderly in developed countries [15]. The high prevalence dermatophytosis rate (52.9%) found in this study is similar with those reported in many studies concerning African schoolchildren [16–19]. In correlation with our findings, tinea capitis was also the most frequent dermatophytosis presentation among children [4–6,20]. Its 39.3% prevalence rate was similar to those observed among outpatients attending the dermatology consultation at the Marchoux Dermatology Institute in Bamako (38.5%) [21] and higher than those reported in previous surveys of schoolchildren in Bamako (7% and 12.5%) [22,23].
In agreement with the majority of African studies, our findings highlight male gender as a significant tinea capitis risk factor among schoolchildren [6,7,16,19,22–24]. For example, in Mali, the prevalence of tinea capitis was 4.4% and 2.1% (P< 10−6) in schoolboys and schoolgirls, respectively [22]. Meanwhile, the prevalence was 3 and 5 times higher in boys than in girls in Abidjan (Côte d’Ivoire) and Central Nigeria, respectively [6,7].
Many potential risk factors for tinea capitis have been proposed [25]. Although our study did not address genetic susceptibility to dermatophytosis, we considered ethno-cultural risk factors, which are likely to play a significant role in dermatophytosis epidemiology. It has been reported that dermatophytosis prevalence is influenced by the hairdressing mode, extracurricular activity and cultural habits, rather than population density [4,6]. In the present study population, the following factors were associated with tinea capitis in the univariate but not in the multivariate analysis: contact with dogs (P = 0.002), public hairdressing practices (P = 0.022), home hairdressing practices (P<0.001), traditional braiding practices (P<0.001) and head shaving practices (P<0.001). These differences between crude and adjusted risk estimates in this study are caused by multicollinearity, nested effects (i.e. contact with animals and contact with dogs …) or non-independence with quasi-complete separation of data points (i.e. type of hairdressing mode (braiding or head shaving) according to the gender) among the predictors in the multivariate analysis. In Egypt, El-Khalawany et al. [25] have shown that contact with animals was a common predisposing factor for tinea capitis in rural areas, whereas transmission from other family members was more common among individuals residing in urban areas. However, increased dermatophyte risk due to contact with dogs was unexpected in our study, as we did not isolate Microsporum canis, the dermatophyte species usually associated with canines. Further studies are required to assess whether dogs might be involved in anthropophilic dermatophyte species transmission.
One limitation of the study is the relatively high false-negative dermatophyte culture rate in the first study compared with the two other study sites, which is associated with the subsequent introduction of sterile gauze in the sampling procedure. This limitation was taken into account by applying a clinical definition of tinea capitis in the risk factor study. We can reasonably presume that the false-negative culture results occurred at random in Sirakoro-Meguetana and thus did not alter the dermatophyte species distribution evaluation. As observed in other studies [5,7] the most common Trichophyton species was T. soudanense (36.6%), which was the most common species associated with tinea capitis in the Sahelian climate zone (64.7%). T. soudanense is one of the most common clinical dermatophyte species in West and Central Africa, where infections of this anthropophilic species are spread via direct contact between people. Two other species of the Trichophyton genus were isolated in this study: T. mentagrophytes (exclusively in the Sudanian zone in 6.6% of cases) and T. violaceum (only in the Sudano-Guinean zone in 5.9% of cases). The anthropophilic species T. violaceum has been shown associated with tinea capitis in Conakry-Guinea, where a 56.7% prevalence of this species was reported [26]. However, in Côte d’Ivoire, a low prevalence (2.3%) of this species has also been associated with tinea capitis [5]. It should be noted that T. soudanense and T. violaceum are phylogenetically very similar and that some experts consider these two taxa to be synonyms [27]. M. audouinii was the only species of the Microsporum genus isolated in this study. It was the second most common species following T. soudanense, which correlates with previous studies of schoolchildren in Bamako [22,23].
In conclusion, tinea capitis was diagnosed in 39.3% of a representative population of Malian schoolchildren. Two anthropophilic dermatophyte species, T. soudanense and M. audouinii were isolated in the majority of cases. Male gender and residing in the tropical North Guinea climatic zone of Mali were identified as independent tinea capitis risk factors. Tinea capitis risk increased with increasing humidity among the relatively homogenous populations located in distinct climatic geographic areas, thereby indicating that climatic factors may play a significant role in dermatophytosis epidemiology. Further epidemiological studies are required to elucidate the respective role that climatic and ethno-cultural factors play in dermatophytosis distribution.
|
10.1371/journal.pntd.0001682 | Identification of Mimotopes with Diagnostic Potential for Trypanosoma brucei gambiense Variant Surface Glycoproteins Using Human Antibody Fractions | At present, screening of the population at risk for gambiense human African trypanosomiasis (HAT) is based on detection of antibodies against native variant surface glycoproteins (VSGs) of Trypanosoma brucei (T.b.) gambiense. Drawbacks of these native VSGs include culture of infective T.b. gambiense trypanosomes in laboratory rodents, necessary for production, and the exposure of non-specific epitopes that may cause cross-reactions. We therefore aimed at identifying peptides that mimic epitopes, hence called “mimotopes,” specific to T.b. gambiense VSGs and that may replace the native proteins in antibody detection tests.
A Ph.D.-12 peptide phage display library was screened with polyclonal antibodies from patient sera, previously affinity purified on VSG LiTat 1.3 or LiTat 1.5. The peptide sequences were derived from the DNA sequence of the selected phages and synthesised as biotinylated peptides. Respectively, eighteen and twenty different mimotopes were identified for VSG LiTat 1.3 and LiTat 1.5, of which six and five were retained for assessment of their diagnostic performance. Based on alignment of the peptide sequences on the original protein sequence of VSG LiTat 1.3 and 1.5, three additional peptides were synthesised. We evaluated the diagnostic performance of the synthetic peptides in indirect ELISA with 102 sera from HAT patients and 102 endemic negative controls. All mimotopes had areas under the curve (AUCs) of ≥0.85, indicating their diagnostic potential. One peptide corresponding to the VSG LiTat 1.3 protein sequence also had an AUC of ≥0.85, while the peptide based on the sequence of VSG LiTat 1.5 had an AUC of only 0.79.
We delivered the proof of principle that mimotopes for T.b. gambiense VSGs, with diagnostic potential, can be selected by phage display using polyclonal human antibodies.
| Control of the chronic form of sleeping sickness or gambiense human African trypanosomiasis (HAT) consists of accurate diagnosis followed by treatment. We aim to replace the native variant surface glycoprotein (VSG) parasite antigens that are presently used in most antibody detection tests with peptides that can be synthesised in vitro. Antibodies recognising VSG were purified from HAT patient sera and were used to select phage-expressed peptides that mimic VSG epitopes from a Ph.D.-12 phage display library. The diagnostic potential of the corresponding synthetic peptides was demonstrated in indirect ELISA with sera from HAT patients and endemic negative controls. We proved that diagnostic mimotopes for T.b. gambiense VSGs can be selected by phage display technology, using polyclonal human antibodies.
| The chronic form of sleeping sickness or human African trypanosomiasis (HAT) in West and Central Africa is caused by the protozoan parasite Trypanosoma brucei (T.b.) gambiense while T.b. rhodesiense causes a more fulminant, acute form in East and Southern Africa. Both subspecies of T. brucei are cyclically transmitted by tsetse flies of the genus Glossina and mainly affect poor, rural populations. The true burden of this disease is unknown as many cases remain undiagnosed or unreported [1], [2].
Since untreated HAT is almost always fatal and no inexpensive, safe and easily administered drugs are available, accurate case detection is crucial. Parasite detection is laborious and insensitive, and remains therefore limited to disease suspects. In the absence of reliable clinical symptoms or antigen detection tests, HAT suspects are identified through screening of the population at risk for presence of trypanosome specific antibodies. The commonly used antibody detection tests, card agglutination test for trypanosomiasis (CATT) [3], LATEX/T.b. gambiense and ELISA/T.b. gambiense [4], [5] detect antibodies against the highly immunogenic variant surface glycoproteins (VSGs) of T.b. gambiense. Even though the genome of T. brucei contains >1000 VSG genes, only one variable antigen type (VAT) is expressed at a time. Stochastic switching of VSG allows the trypanosome to evade the specific antibody responses that were raised against earlier VATs [6]–[10]. Some VATs, such as LiTat 1.3 and 1.5, are recognised by almost all gambiense HAT patients and therefore called predominant. The dense VSG monolayer on the living trypanosome shields all non-specific epitopes. The hypervariable N-terminal VSG domain (300–400 residues) is exposed to the immune system and comprises the VAT-specific epitopes, while the relatively conserved C-terminal domain (40–80 residues) is hidden by the intact VSG coat [6], [9], [11], [12].
Disadvantages of the present antibody detection tests include the occurrence of non-specific reactions. This might be explained by exposure of non-HAT-specific epitopes that are normally shielded on the living trypanosome [12], [13]. In addition, diagnostic test production actually requires culture of infective T.b. gambiense in large numbers of laboratory rodents and poses an important risk of infection to the manufacturing staff [14].
These drawbacks can be circumvented through the use of synthetic peptides that mimic HAT-specific VSG epitopes (mimotopes) and can be produced in a standardised way [15]. One way to identify such mimotopes is by peptide phage display. This technique is based on DNA recombination resulting in foreign peptides with random sequences that are displayed fused to the pIII surface protein of the M13 phage. After an in vitro selection process based on binding affinity and several rounds of enrichment (panning), the encoded peptide insert sequence of the selected phage is deduced from the phage DNA. We previously reported successful identification of mimotopes for VSG LiTat 1.3 and LiTat 1.5 by performing phage display with three monoclonal antibodies [16]. However, by the use of only three monoclonal antibodies, representing only a fraction of the VSG-specific antibody response, some mimotopes with diagnostic potential might have been missed. Additionally, the mouse and human immune system may recognise different B cell epitopes. The use of polyclonal human antibodies might therefore increase chances of selecting diagnostic mimotopes [17]. Polyclonal antibodies from human sera have been previously used for selection of mimotopes with diagnostic potential for e.g. hepatitis C [15], typhoid fever [18] and Epstein Barr virus [17]. Some mimotopes have been patented for incorporation in commercially available tests, e.g. for neurocysticercosis [19].
In this manuscript we describe the identification of mimotopes for VSG LiTat 1.3 and LiTat 1.5 through phage display, using sera from HAT patients and endemic negative persons.
Sera from HAT patients and endemic controls were collected within different diagnostic studies [5], [20]. All individuals gave their written informed consent before providing blood. Permission for these studies was obtained from the national ethical committee of the Democratic Republic of the Congo (DR Congo) and from the Institute of Tropical Medicine Antwerp (ITMA) ethical committee, reference number 03 07 1 413 and 04 44 1 472. Forty additional endemic negative control specimens were obtained from the archived specimen bank of the Parasite Diagnostics Unit at ITMA. All specimens were anonymised.
Variant surface glycoproteins were purified from cloned populations of T.b. gambiense Variable Antigen Type (VAT) LiTat 1.3 and 1.5 [4]. VSG LiTat 1.3 or LiTat 1.5 were coated onto magnetic particles (MP, Estapor, 10% suspension, 1.04 µm, 9 µeq/g COOH). A volume of 250 µL of MP suspension was washed twice with 1 mL of buffer A (10 mmol/L NaH2PO4, pH 6.0). The MP were activated with 2.5 ml of buffer A containing 25 mg of 1-ethyl-3-(3-dimethylaminopropyl) carbodiimide (Pierce) and 15 mg of N-hydroxysuccinimide (Sigma). The MP were rotated for 15 minutes at room temperature (rT) and washed with 1 mL of buffer B (2 mmol/L HCl) where after 350 µg of VSG (LiTat 1.3 or LiTat 1.5) in 1.5 mL of buffer C (20 mmol/L NaH2PO4/Na2HPO4, pH 7.5) was added to the pellet of MP. After rotating for 2 h at rT, the MP were washed three times with buffer C and resuspended to a final concentration of 8% in buffer C containing 100 mmol/L glycin, 1% bovine serum albumin (BSA) and 0.1% NaN3. Successful coating of the MP was evaluated by agglutination with a HAT positive serum diluted 1/4 in phosphate buffered saline (PBS, 0.01 mol/L phosphate, 0.14 mol/L NaCl, pH 7.4) containing 0.02% w/v NaN3.
Antibodies specific to VSG LiTat 1.3 or LiTat 1.5 were purified from nine HAT positive sera originating from the DR Congo [20]. One mL of LiTat 1.3 or LiTat 1.5 coated MP was rotated for 2 h at rT with 125 µL of HAT positive serum. After five washes with 800 µL of PBS, the specific antibodies were eluted from the MP by adding 700 µL of 0.2 mol/L glycine/HCl (pH 2.2) followed by magnetic separation after five minutes. The eluates, corresponding to the affinity purified antibody fractions, were neutralised with 100 µL of 1 mol/L Tris/HCl pH 9.1.
Indirect ELISA was used to screen the affinity purified antibody fractions and all human serum samples on reactivity with VSG LiTat 1.3 and LiTat 1.5. ELISA plates (Nunc MaxiSorp™) were coated overnight (ON) at 4°C with 100 µL/well of 2 µg/mL of each VSG separately in phosphate buffer (PB, 0.01 mol/L phosphate, pH 6.5) or with 1.7 1011 particle/mL of wild type phage (WTP) in PBS. One plate was left empty as antigen negative control (Ag0). The plates were tapped dry, saturated with 350 µL/well of PBS-Blotto (0.01 mol/L phosphate, 0.2 mol/L NaCl, 1% w/v skimmed milk powder, 0.05% w/v NaN3) during 1 h at rT and washed three times with 0.05% v/v Tween-20 in PBS (PBST) (ELx50, Bio-Tek ELISA washer). The purified antibody fractions were diluted 1/25 and human serum samples 1/150 in PBS-Blotto. One hundred µL/well of each dilution was applied in duplicate and incubated for 1 h at rT. After three washes with PBST 100 µL/well of horse radish peroxidase (PO)-conjugated goat anti-human IgG (H+L) (Jackson), 1/40000 diluted in PBST, was added. An hour and five washes later, wells were incubated for 1 h at rT with 100 µL/well of 2.2′-azino-bis-(3-ethylbenzthiazoline-6-sulfonic acid) (ABTS) chromogen/substrate solution (50 mg tablet/100 mL of ABTS buffer, Roche). The plate was shaken for ten seconds and the optical density (OD) was read at 414 nm (Labsystems Multiskan RC 351). The measured OD was corrected (ODc) with the corresponding OD in the Ag0 wells.
Three LiTat 1.3 positive pools, each consisting of three different VSG LiTat 1.3-specific antibody fractions, three LiTat 1.5 positive pools, each consisting of three different VSG LiTat 1.5-specific antibody fractions and one negative pool of four endemic negative sera were prepared. For each pool the antibodies were coated onto anti-human IgG (H+L) functionalised magnetic particles (MP) (1% w/v, 1.05 µm, Estapor/Merck) according to the guidelines of the manufacturer.
The panning was performed with the Ph.D.-12 (12-mer) phage display library (New England Biolabs, NEB) [21] through two rounds consisting of 1) a positive selection with anti-VSG (LiTat 1.3 or 1.5, respectively) antibodies coated on MP, 2) a negative selection with endemic negative serum antibodies coated on MP and 3) phage amplification [22]. Each positive selection was followed by phage titration and sandwich ELISA. After these two rounds a third positive selection was performed.
Positive selection was performed as previously described [16]. Bound phages were eluted for ten minutes with 600 µL of 0.2 mol/L glycine-HCl containing 1 mg/mL BSA (pH 2.2) and neutralised with 90 µL of Tris-HCl (1 mol/L, pH 9.1).
Six hundred µL of the elution from the positive selection was rotated ON at 4°C with 1 mg of MP coated with endemic negative serum antibodies, in a total volume of 1 mL of PBSG.
The unbound phages in 900 µL of the supernatant of the negative selection were amplified, in a culture of Escherichia (E.) coli (strain ER2738, NEB) at early log (0.01–0.05 A600), and purified with PEG-NaCl as previously described [16], [21].
Phages from the first, second and third positive selection were diluted in PBS 101 to 104, 102 to 105, 104 to 107, respectively. Ten µL of these dilutions were incubated for five minutes with 200 µL of an E. coli culture in mid-log (0.4–0.5 A600). The mixture was then pipetted into 4 mL of Top-Agar (50°C) and poured onto agar plates containing 1 mL/L IPTG/X-gal (1.25 g isopropyl β-D-thiogalactoside, 1 g 5-bromo-4-chloro-3-indolyl-β-D-galactoside, 25 mL dimethylformamide); ninety-four blue clones were picked and each clone was inoculated in 200 µL of lysogeny broth (LB) in a sterile culture plate (BD Falcon™ Clear 96-well Microtest™ Plate) [21]. This plate was shaken overnight at 30°C, and then the bacteria were pelleted by 5 min centrifugation at 1312 g. The supernatant was tested in a sandwich ELISA.
ELISA plates were coated ON at 4°C with 100 µL/well of VSG LiTat 1.3- or LiTat 1.5-specific positive antibody pools (5 µg/mL in PBS) or a 1/10000 dilution in PBS of the negative serum pool. The ELISA was performed as previously described [16]. Briefly, the wells were incubated for 1 h at rT with 100 µL of phage dilution in PBS-Blotto (1/3 for culture plate supernatant or 1/20 for PEG-NaCl purified phage). PO-anti-M13 pVIII mAb (GE Healthcare), diluted 1/2000 in PBST was added to the wells for 1 h at rT. The wells were then incubated for 1 h at rT with ABTS and read at 414 nm.
Phage clones were withheld after the first two positive selections if 1) the OD with the corresponding positive pool (ODpos)>average ODpos+2*standard deviation (sdpos) and 2) the OD with the negative pool (ODneg)<average ODneg.
After the third positive selection, phages were sequenced if 1) ODpos>average ODpos+1* sdpos with at least one of the positive pools, 2) ODpos with the 3rd positive pool >0.150 or 0.200 for phages selected for VSG LiTat 1.3 or LiTat 1.5 respectively and 3) ODneg<average ODneg. Withheld phage clones were sequenced and tested in a similar sandwich ELISA with as capture antibody the nine individual affinity purified antibody fractions, diluted 1/70 in PBS.
Purification of phage DNA was performed according to the NEB manual [21]. Sequence determination was performed as described before [16]. The obtained sequence chromatograms were read with Chromas 2.33 (Technelysium Pty Ltd). Sequence alignment was performed manually and with RELIC software [23]. A protein data base (pdb) model of the N-terminal domain of VSG LiTat 1.5, was created using SWISS-MODEL [24], [25]. Modelling was based on the known structure of VSG MITat 1.2 (pdb 1vsgA), previously derived by X-ray crystallography [26]. For VSG LiTat 1.3 however the server could not find a template with sufficient sequence homology, hence the pdb was created by Thomas Juetteman from the PyMol helpdesk (PyMOL Molecular Graphics System, Schrödinger, LLC). In order to identify possible conformational epitopes, the 3D-Epitope-Explorer (3DEX) [27] was used to find structural homology between the mimotope sequences and the respective VSG protein sequence.
Molecular graphics images were produced using the UCSF Chimera package from the Resource for Biocomputing, Visualization, and Informatics at the University of California, San Francisco (supported by NIH P41 RR001081) (http://www.cgl.ucsf.edu/chimera).
The peptides were synthesised at >85% purity (Peptide 2.0, Chantilly, VA, U.S.). The GGGS-spacer, separating the library insert and the pIII phage protein, was added to the C-terminus of the peptides that were selected by phage display [16], [21]. All peptides were C-terminally elongated with an additional lysine-biotin and amidated (-CONH2), mimicking the uncharged peptide bond in a protein. All synthetic peptides were reconstituted in sterile deionised H2O to a concentration of 2 mg/mL.
First, the reactivity of all biotinylated synthetic peptides was evaluated with the nine sera used for affinity antibody purification and nine endemic negative controls. Second, the diagnostic performance of the synthetic peptides was evaluated with human serum samples that were previously screened (indirect ELISA on VSG, serum dilution 1/100) on reactivity with VSG LiTat 1.3 and 1.5. All 102 serum samples from gambiense HAT patients originated from DR Congo [20]. Of the 102 endemic gambiense HAT negative serum samples, 71 originated from the DR Congo and 31 from Benin. The indirect ELISA on biotinylated peptides was performed as the indirect ELISA on VSG but 150 µL/well was applied in all but the saturation and washing steps. ELISA plates were coated with 10 µg/mL streptavidin (NEB) in carbonate buffer (0.1 mol/L, pH 9.2) or with 2 µg/mL VSG LiTat 1.3 and LiTat 1.5 in PB, or wells were left empty (Ag0). After saturation with PBS-Blotto, the peptides were added at a concentration of 2 µg/mL in PBS to the wells containing streptavidin. The peptide-free wells received only PBS. To the VSG-containing and Ag0 wells PBS-5% w/v sucrose was added. After incubation of 1 h at rT the plates were tapped dry, sealed and frozen at −80°C. The serum samples were centrifuged for 5 min at 15700 g and diluted 1/100 in PBS-Blotto. After thawing of the plates and three washes with PBST, the serum dilutions were applied in duplicate. After one hour, we added PO-conjugated goat anti-human IgG (H+L), 1/40000 diluted in PBST. ABTS was used as chromogen/substrate solution and the OD was read as described above. The measured OD was corrected by subtracting the corresponding OD in the peptide-free or Ag0 wells and the average of the duplicate corrected ODs was taken (ODc).
The accuracy of the synthetic peptides to detect VSG-specific antibodies for diagnosis of sleeping sickness was assessed by the area under the receiver operator characteristics (ROC) curve (AUC) [28]. Confidence intervals were calculated according to DeLong [29]. For the whole range of cut-offs the Youden index was determined (Youden index = sensitivity+specificity−1) [30] and the cut-off with maximal Youden index was retained.
In indirect ELISA, all affinity purified antibody fractions reacted specifically with their corresponding VSG and not with WTP. The antibody fractions that were purified with VSG LiTat 1.3, had an average ODc of 0.533±0.319 with VSG LiTat 1.3, and average ODcs of only 0.032±0.032 with VSG LiTat 1.5 and −0.008±0.015 with WTP. The antibody fractions purified with VSG LiTat 1.5 had an average ODc of 1.406±0.487 with VSG LiTat 1.5, and average ODcs of only 0.037±0.064 with VSG LiTat 1.3 and −0.017±0.018 with WTP.
The negative serum samples did not react with VSG LiTat 1.3 (ODc 0.034±0.078), nor with VSG LiTat 1.5 (ODc 0.028±0.069), nor with WTP (ODc 0.013±0.029).
During the selection process, none of 94 phage clones of the first positive selection, eight of 188 phage clones of the second positive selection and 11 of 188 phage clones of the third positive selection reacted in the sandwich ELISA and were sequenced, resulting in 18 sequences (table 1).
The alignment results of VSG LiTat 1.3 [GenBank AJ304413] and the eighteen peptide sequences displayed by the phage clones are presented in figure 1. All peptides could be aligned within amino acid stretch (AA) 72 to 116 of the N-terminal domain of VSG LiTat 1.3 (alignment 1). The common motive (F/W)ExDxK(A/V/L)x(A/V/L) was repeated twice in this VSG AA stretch. Therefore twelve sequences could be aligned twice within this region. The peptide displayed by phage 3-3-F6, ETDNMKPLHLRQ, could even be aligned three times within this region of VSG LiTat 1.3, having ETD, DNxKP and ExD identical within amino acids 78 to 80, 87 to 91 and 102 to 104 of the protein sequence. The peptide sequence displayed by phage 3-2-D10 had the highest identity within AA 72 to 116 of the VSG LiTat 1.3 sequence (7/16 AA, 44%). The reverse sequence of the peptides displayed by phage clones 3-2-C5 and 3-3-E3 showed respectively 31 and 13% identity within AA 180 to 196 (alignment 2). Within the C-terminal domain (alignment 3), the peptide expressed by phage clone 3-2-C5 and 3-3-E3 had 25% identity (4/16 AA) within AA 404 to 443 of VSG LiTat 1.3. The peptide expressed by phage clones 3-2-G10, 3-2-G5 and 3-2-B12 were respectively 31%, 19% and 19% identical within AA 404 to 443 of VSG LiTat 1.3.
All selected phage clones were tested in a sandwich ELISA with the individual purified antibody fractions. The peptides displayed by the seven phage clones with the highest average ODc were withheld for synthesis as biotinylated peptides (table 1). The peptide displayed by phage clone 3-2-B12 was not withheld, since it was a homologue of 3-2-G10 and 3-2-G5 but had a lower average ODc. Based on the alignment results, AA stretch 78 to 110 and AA stretch 424 to 439 of the protein sequence of VSG LiTat 1.3 were also synthesised as biotinylated peptides (respectively peptide 1.3/78-110 and peptide 1.3/424-439).
The reactivity of all nine biotinylated synthetic peptides was evaluated in indirect ELISA with the nine HAT positive sera used for affinity antibody purification, and with nine endemic negative controls. Peptide 1.3/78-110 was the best performing peptide with ODc 1.469. Peptide 1.3/424-439 gave a lower average ODc (0.246) than peptides 3-2-G10 and 3-2-G5 (0.564 and 0.920), sharing the same common motive, and was not withheld for further testing. Peptide 3-2-E2, a homologue of peptide 3-2-D10, also gave a lower average ODc (0.541 versus 0.763) and was also not withheld for testing on diagnostic performance.
During the selection process, one of 94 phage clones of the first positive selection, two of 94 phage clones of the second positive selection and 20 of 188 phage clones of the third positive selection reacted in the sandwich ELISA and were sequenced, resulting in 20 sequences (table 2).
The alignment results of VSG LiTat 1.5 [GenBank HQ662603] and the 20 peptide sequences displayed by the phage clones are presented in figure 2.
The peptide expressed by phage clone 5-1-F9 (19% identity) could be aligned within AA 33 to 47 (alignment 1).
Within the N-terminal domain, 18 phage peptides could be aligned with minimum 6% identity within AA 81 to 119 (alignment 2), by analogy with alignment 1 for VSG LiTat 1.3. The peptides expressed by clones 5-3-C7, 5-3-A8 and 5-3-B5 respectively had 0, 0 and 6% identity within this AA stretch, but shared the common motive (W/F)Y with the peptide expressed by phage 5-3-B8 (19% identity). The peptide of phage clone 5-3-G6 had only 1/16 AA (6%) identity with VSG LiTat 1.5 but two more AA were homologous within this region. The reverse sequences of phage clone peptides 5-3-F7 (13% identity), 5-1-F9 (19% identity) and 5-3-D5 (13% identity) could also be aligned within this VSG LiTat 1.5 region. The peptide expressed by clone 5-3-B9 had the highest identity within the AA 81 to 119 stretch (5/16 AA, 31% identity, if a gap of 1 AA was allowed).
Peptides expressed by phage clones 5-3-C1, 5-3-A4 and 5-3-A6, with common motive “KLANP”, could also be aligned between AA 145 to 166 of the VSG LiTat 1.5 protein sequence (alignment 3) with respectively 25, 13 and 13% identity. Within the VSG LiTat 1.5 AA stretch 245 to 281, the peptide expressed by phage clone 5-3-A6, showed 19% identity and the reverse peptide sequence of phages 5-3-B9, 5-3-A3 and 5-3-A4 showed respectively 19, 19 and 13% identity (alignment 4). Within the boundary with the C-terminal domain of VSG LiTat 1.5, showed the peptides expressed by phage clones 5-3-A8, 5-3-C1 and 5-3-B6 respectively 19, 31 and 31% identity within AA stretch 341 to 368, if a gap of three AA was allowed for peptide 5-3-C1 (alignment 5). Within the C-terminal domain of VSG LiTat 1.5, phage clone peptide 5-3-B9 and 5-3-A4 had respectively 31 and 13% identity between AA 468 to 489 (alignment 6).
All selected phage clones were tested in a sandwich ELISA with the individual purified antibody fractions (table 2). The peptides displayed by the seven phage clones with the highest average ODc, were chosen for synthesis as biotinylated peptides, except for the peptide displayed by phage clone 5-3-A6, which was similar to 5-3-C1 but had a lower average ODc. Based on the alignment results and by analogy with VSG LiTat 1.3, the AA stretch 81 to 109 of the protein sequence of VSG LiTat 1.5 was synthesised as biotinylated peptide (peptide 1.5/81-109).
The reactivity of all eight biotinylated synthetic peptides was evaluated in indirect ELISA with the nine sera used for affinity antibody purification and with nine endemic negative controls. Peptide 5-3-A4 and 5-3-G6 had the lowest average ODcs (0.145 and 0.109) and shared a common motive with peptide 5-3-C1 with a higher average ODc (0.289) and were therefore not withheld for testing on diagnostic performance. Peptide 1.5/81-109 had an ODc of 0.382 and was withheld.
The accuracy of the biotinylated peptides to detect VSG-specific antibodies was assessed with sera from 102 gambiense HAT patients and 102 endemic negative controls (table 3). Among the mimotopes for VSG LiTat 1.3, the highest AUC was obtained with peptide 3-2-G5 (0.93) and peptide 3-2-G10 (0.95). Sensitivities and specificities at the cut-off with the highest Youden index were respectively 0.85 and 0.94 for peptide 3-2-G5, and 0.90 and 0.93 for peptide 3-2-G10. Of the mimotopes for VSG LiTat 1.5 the highest AUC was obtained with peptide 5-1-F9 (0.95) and 5-2-D3 (0.94) with respective sensitivities and specificities of 0.94 and 0.95 for peptide 5-1-F9 and 0.92 and 0.89 for peptide 5-2-D3. With peptide 1.3/78-110, an AUC of 0.95 was observed, with a sensitivity of 0.96 and a specificity of 0.85. With peptide 1.5/81-109, an AUC of 0.79, a sensitivity of 0.81 and a specificity of 0.75 were obtained.
VSG LiTat 1.3 and 1.5 obtained an area under the curve of respectively 1.000 and 0.997. The sensitivity and specificity were both 1.000 at cut-off 1.318 for VSG LiTat 1.3 and 1.000 and 0.990 at cut-off 1.182 for VSG LiTat 1.5.
By using 3DEX software and setting the number of hits at a minimum of 5 AA, none of the VSG LiTat 1.3 mimotopes with AUC>0.90 could be mapped as a conformational epitope on the protein model of the VSG LiTat 1.3. In contrast, among the VSG LiTat 1.5 mimotopes with AUC>0.90, peptide 5-2-D3 could be mapped with 8/12 AA (E 168|N 164|D 152|G 153|T 150|K 146|L 144|A 141) on the three-dimensional VSG LiTat 1.5 protein model (figure 3).
In this manuscript we describe how mimotopes and regions that take part in epitope formation for VSGs LiTat 1.3 and LiTat 1.5 of T.b. gambiense were identified by screening of a Ph.D.-12 phage display library with polyclonal antibodies that were purified from sera of sleeping sickness patients.
As sera from sleeping sickness patients contain an important fraction of trypanosome unrelated antibodies [31], the risk of selecting mimotopes that are unrelated to sleeping sickness by using human sera for the screening was considerable.
We identified a linear region between amino acid 72 and 114 of the protein sequence of both VSG LiTat 1.3 and LiTat 1.5 wherein most of the peptide sequences could be aligned with the VSG protein sequence. This region is localised in the hypervariable N-terminal domain of the VSG and was for both VSGs synthesised as a linear biotinylated peptide and tested in indirect ELISA with a panel of 102 HAT positive and 102 endemic negative sera.
Peptide 1.3/78-110, corresponding to AA stretch 78 to 110 of VSG LiTat 1.3, had an AUC of 0.95, indicating diagnostic potential. The epitope of VSG LiTat 1.3, recognised by the human serum antibodies used for screening of the peptide library, therefore seems to be linear and located within AA stretch 78 to 110. The peptide sequences that were selected for VSG LiTat 1.3 had in average 3/16 amino acids in common within AA 72 to 114 with a maximum of 7/16 (44%) identical amino acids. Interestingly, a common motive of the peptide sequences was repeated twice within AA 72 to 114 of VSG LiTat 1.3: (F/W)ExDxK(A/L/V)x(A/L/V), from AA 77 to 85 and 101 to 109.
The two mimotopes of VSG LiTat 1.3 with the highest AUC, peptide 3-2-G5 and 3-2-G10, seemed to share a common epitope (correlation coefficient of ODcs with human sera in ELISA 0.71, data not shown), their motive WxxDxK reoccurred twice within AA 72 to 114. Their motive, (I/V/A)(T/S)DSK, could also be aligned within the C-terminal domain (AA 424 to 439). This AA stretch was synthesised as a biotinylated peptide as well, but had a low average ODc upon a first screening with nine HAT sera and was discarded for further evaluation of diagnostic performance. We think it unlikely that peptide 3-2-G5 and 3-2-G10 are mimotopes for a linear epitope localised in the C-terminal domain of VSG LiTat 1.3. Additionally, epitopes localised in the relatively conserved C-terminal domain are more likely to react with non-VSG-specific antibodies.
As for VSG LiTat 1.3, a repetitive motive was present within AA 81 to 114 of VSG LiTat 1.5: (Y/F/W)(x or xx)(A/L/I/V)A(A/I/L)(D or K) (A/L)xxxxE, from AA 83 to 94 and AA 99 to 111. The peptide sequences selected for VSG LiTat 1.5 had in average 2/16 AA in common within AA 82 to 114 of the protein sequence, with a maximum of 5/16 (31%) identical AA. Contrary to the results for peptide 1.3/78-110, peptide 1.5/81-109, corresponding to AA stretch 81 to 109 of VSG LiTat 1.5, had an AUC of only 0.79, while the AUC of all of the individual peptides aligned in this region was >0.85. Motive AYSxxxIKL of peptide 5-3-C1 (AUC 0.87), corresponded to LYSxxxAKL (AA 99 to 106) of the VSG LiTat 1.5 protein sequence. Peptide 5-3-C1 seems therefore to mimic an epitope that is, albeit partly, localised in this region. The similar peptide 5-2-D3, with motive F(x)xxxxKL, performed better in ELISA (AUC 0.94). It is possible that peptide 5-2-D3 and 5-3-C1 are mimotopes for a discontinuous epitope as they share the common motive (A/I/L) (Y/F) xxxxxKLANPG with four other peptides, while ANPG was not found in the VSG protein sequence. We therefore suspect the epitope of VSG LiTat 1.5, recognised by the human serum antibodies used for screening, to be discontinuous and to be at least partly localised within this region. This might explain the weaker performance of the linear peptide 1.5/81-109 compared to the mimotope peptides. This finding was supported by the results of the 3DEX analysis of the mimotopes that had an AUC>90, locating peptide 5-2-D3 with 8/12 AA on the three-dimensional VSG LiTat 1.5 protein model (E 168|N 164|D 152|G 153|T 150|K 146|L 144|A 141).
In a previous study [16] we were able to identify mimotope peptides for the native trypanosomal variant surface glycoproteins by screening of peptide phage display libraries with monoclonal antibodies. Through phage display with polyclonal human antibodies we now identified different mimotopes and regions taking part in epitope formation. Because the three monoclonal antibodies used in the first study represent only a fraction of the VSG-specific antibody response, some mimotopes with diagnostic potential might have been missed. Additionally, the mouse and human immune system may recognise different B cell epitopes. Other factors may have contributed to finding different motives using the two approaches. As a result of a short infection period of two weeks, the mouse monoclonals do not recognise all VSG-epitopes. They were selected for strict VAT-specificity with purified VSGs and identified mimotopes, not necessarily dominant, that were located near the surface of the VSG N-terminal domain. The polyclonal human antibodies result from a long infection and recognise also less exposed VSG-epitopes. It may be that by affinity purification on purified VSG an antibody fraction that recognises non-surface epitopes was mainly retained, as the mimotopes of VSG LiTat 1.3 seem to be located in this region. Another explanation may lie in the presence of self-reactive VSG-specific antibodies in sera from uninfected individuals, as has been demonstrated by Müller et al. [32]. Thus the negative selection with human antibodies from control sera may have eliminated the phages expressing the mimotopes for the VSG-specific epitopes also recognised by the mAbs. Both panning strategies thus seem complementary, in contrast to what has been described by Tang et al. [18], who selected a greater number of different 12-mer sequences with polyclonal serum for Salmonella enterica, but some of the common motives were also selected by panning of a monoclonal antibody. In the manuscript of Casey et al. [17] the mimotopes for Epstein-Barr (EBV) virus, selected with polyclonal EBV immune rabbit and patient sera were also not recognised by the monoclonal antibodies used for mimotope selection in a previous study.
Diagnostic evaluation of individual mimotopes and combinations [patent application GB1202460.0] indicates that screening of peptide phage display libraries with patient's antibodies resulted in a more efficient selection of diagnostic peptides than with monoclonal antibodies. We therefore prefer screening with patient's antibodies. As an alternative approach to phage display linear epitopes may be replaced by synthetic peptides identified by scanning of overlapping peptides spanning the native protein sequence [33]. Furthermore, there are alternative in vitro methods for phage display such as yeast cell or bacterial display or, non-cellular, ribosome or mRNA display [34].
Our study has nevertheless some limitations. First, considering the broad antibody spectrum in HAT sera due to polyclonal B cell activation [35], we opted to use antibody fractions that were affinity purified for VSG LiTat 1.3 and 1.5. Thus, mimotopes for other predominant VSGs or other trypanosome antigens with diagnostic potential have not been selected. An alternative approach to identify additional diagnostic mimotopes may consist of screening peptide phage libraries with patient antibodies against other candidate diagnostic proteins [36]. Examples are the T.b. gambiense-specific glycoprotein TgsGP [37], the T.b. rhodesiense-specific serum resistance associated (SRA) protein [38] and Trypanozoon-specific trypanosome antigens such as invariant surface glycoprotein (ISG) 65 and ISG 75 [39], microtubule associated repetitive protein 1 (MARP1) and GM6 [40]. Some of them have already been tested for their diagnostic potential in the form of recombinant fusion proteins expressed in E. coli but none are yet used in diagnostic tests for HAT. Second, even with affinity purified antibodies there is a risk that non-specific mimotopes are selected with antibodies against VSG epitopes that are normally hidden in the intact VSG coat. Usually, most of the phage particles display a consensus binding sequence after two or three rounds of enrichment [21]. We therefore performed three rounds of positive selection and two selections with negative sera. Remarkably, the mimotopes with the highest AUC for VSG LiTat 1.3 and 1.5 were selected after only two or even one round of panning. Third, no affinity measurements e.g. via surface plasmon resonance, have been performed. Considering the polyclonal character of antibodies in patients' sera and the inherent differences in antibody response between individual patients, we opted to assess only the diagnostic potential of the selected peptides by means of ELISA.
Before the native T.b. gambiense VSGs LiTat 1.3 and LiTat 1.5 in the currently existing diagnostic formats can be replaced by synthetic peptides, further improvements should be considered. It is possible to define critical residues, essential for binding with the antibody, by e.g. alanine scanning mutagenesis [41]. Thus the epitope of the human serum antibodies might be recreated as has recently been done for a linear epitope on the VP1 protein of foot-and mouth disease virus [42]. Other, non-essential, parts of the peptides can then be eliminated in order to increase specificity. Phage clones that express peptides with a higher binding affinity might be selected by increasing the number of selection rounds and/or the stringency of the washing steps.
In conclusion, with this study we demonstrate that mimotopes of T.b. gambiense VSG LiTat 1.3 and 1.5 can be selected from a phage display library and that these mimotopes and corresponding amino acid stretches within the VSGs have diagnostic potential.
|
10.1371/journal.pgen.1003674 | The C. elegans DSB-2 Protein Reveals a Regulatory Network that Controls Competence for Meiotic DSB Formation and Promotes Crossover Assurance | For most organisms, chromosome segregation during meiosis relies on deliberate induction of DNA double-strand breaks (DSBs) and repair of a subset of these DSBs as inter-homolog crossovers (COs). However, timing and levels of DSB formation must be tightly controlled to avoid jeopardizing genome integrity. Here we identify the DSB-2 protein, which is required for efficient DSB formation during C. elegans meiosis but is dispensable for later steps of meiotic recombination. DSB-2 localizes to chromatin during the time of DSB formation, and its disappearance coincides with a decline in RAD-51 foci marking early recombination intermediates and precedes appearance of COSA-1 foci marking CO-designated sites. These and other data suggest that DSB-2 and its paralog DSB-1 promote competence for DSB formation. Further, immunofluorescence analyses of wild-type gonads and various meiotic mutants reveal that association of DSB-2 with chromatin is coordinated with multiple distinct aspects of the meiotic program, including the phosphorylation state of nuclear envelope protein SUN-1 and dependence on RAD-50 to load the RAD-51 recombinase at DSB sites. Moreover, association of DSB-2 with chromatin is prolonged in mutants impaired for either DSB formation or formation of downstream CO intermediates. These and other data suggest that association of DSB-2 with chromatin is an indicator of competence for DSB formation, and that cells respond to a deficit of CO-competent recombination intermediates by prolonging the DSB-competent state. In the context of this model, we propose that formation of sufficient CO-competent intermediates engages a negative feedback response that leads to cessation of DSB formation as part of a major coordinated transition in meiotic prophase progression. The proposed negative feedback regulation of DSB formation simultaneously (1) ensures that sufficient DSBs are made to guarantee CO formation and (2) prevents excessive DSB levels that could have deleterious effects.
| Formation of haploid gametes during meiosis relies on deliberate induction of DNA double-strand breaks (DSBs), followed by repair of a subset of DSBs as crossovers between homologous chromosomes. Crossovers form the basis of connections that enable homologs to segregate toward opposite spindle poles at meiosis I, thereby reducing ploidy. Thus, germ cells must generate enough DSBs to guarantee a crossover for every chromosome pair while avoiding an excessive number of DSBs that might endanger their genomes. Here, we provide insight into how this crucial balance is achieved. We identify C. elegans DSB-2 as a key regulator of DSB formation, and we propose that its association with chromatin is an indicator of DSB competence. Disappearance of DSB-2 is part of a coordinated transition affecting multiple distinct aspects of the meiotic program, and failure to form crossover-eligible recombination intermediates elicits a delay in DSB-2 removal and other transition events. Our data are consistent with a model in which meiotic DSB formation is governed by a negative feedback network wherein cells detect the presence of downstream crossover intermediates and respond by shutting down DSB formation, thereby ensuring that sufficient DSBs are made to guarantee crossovers while simultaneously minimizing the threat to genomic integrity.
| For most diploid organisms, the formation of haploid gametes relies on crossover (CO) recombination between homologous chromosomes for accurate chromosome segregation. Recombination is initiated during meiotic prophase by the programmed induction of DNA double strand breaks (DSBs), catalyzed by the evolutionarily conserved topoisomerase-like protein Spo11 [1]. A subset of these DSBs are repaired by a specialized meiotic DSB repair pathway that uses the homolog as a recombination partner and generates intermediates that can be resolved as COs. This specialized repair is completed during the pachytene stage of meiotic prophase, in the context of meiosis-specific chromosome organization in which homologs are paired and connected along their axes by a structure known as the synaptonemal complex (SC). By the last stage of meiotic prophase (diakinesis), the SC has disassembled, and chromosomes have further condensed and reorganized to reveal CO-dependent structures called chiasmata, which connect homologous chromosomes and allow them to orient and segregate to opposite poles at the meiosis I division [2].
DSB formation must be tightly regulated to ensure successful meiosis: cells must both turn on DSB formation to achieve inter-homolog COs, but also turn off DSB formation to allow repair and subsequent chromosome re-organization in preparation for the meiotic divisions. Thus, DSB formation and repair must be coordinated with other aspects of meiotic chromosome dynamics. In addition, cells must make enough DSBs to guarantee one CO per chromosome pair, but too many DSBs could lead to unrepaired DNA damage and compromise genomic integrity.
While Spo11 catalyzes DSB formation, little is known about how Spo11 activity is regulated and how the timing and number of DSBs are controlled. Several proteins besides Spo11 are required for meiotic DSB formation in various systems, although their mode(s) of action are not well understood [3], [4], [5]. The highly conserved Rad50/Mre11 complex is required for DSB formation in some systems but not in others, and even in an organism where it is normally required (C. elegans), Spo11-dependent DSBs can form independently of Rad50/Mre11 in some contexts [6], [7]. Further, many of the known DSB-promoting proteins are not well conserved at the sequence level, showing rapid divergence even among closely related species [4]. In C. elegans, the chromatin-associated proteins HIM-17, XND-1, and HIM-5 have been implicated in promoting normal levels and/or timing of DSB formation, particularly on the X chromosomes [8], [9], [10]. These proteins localize to chromatin throughout the germ line and are proposed to exert their effects by modulating the chromatin environment to affect accessibility of the DSB machinery. However, the localization of these proteins is not limited to the time of DSB formation, suggesting that other factors must control when the DSB machinery is active.
In the current work, we identify the C. elegans DSB-2 protein (encoded by dsb-2, member of new gene class dsb for DNA double-strand break factor) as a novel factor required specifically to promote the DSB step of meiotic recombination. We show that DSB-2 localizes to chromatin in meiotic prophase germ cells, and that the timing of its appearance and disappearance corresponds to the time window during which DSBs are formed. These and other data implicate DSB-2 in regulating the timing of competence for DSB formation by SPO-11. Further, we find that the presence of DSB-2 on chromatin is regulated coordinately with multiple distinct aspects of the meiotic program, including specialized meiotic DSB repair features and the phosphorylation state of nuclear envelope protein SUN-1. Thus, we propose that disappearance of DSB-2 reflects loss of competence for DSB formation, which occurs as part of a major coordinated transition in meiotic prophase progression. Moreover, our data suggest the existence of a regulatory network wherein germ cells can detect the presence or absence of downstream CO-eligible recombination intermediates. In the context of this model, successful formation of monitored intermediates would trigger removal of DSB-2 (and other factors) from chromatin and consequent shut-down of DSB formation, whereas a deficit of relevant intermediates would elicit a delay in DSB-2 removal (and in other aspects of meiotic progression). We propose that the negative feedback property inherent in such a regulatory network provides a means to ensure that sufficient DSBs are made to guarantee CO formation, while at the same time protecting the chromosomes against formation of excessive levels of DSBs that could jeopardize genomic integrity.
The dsb-2(me96) allele was isolated following EMS mutagenesis in a screen for meiotic abnormalities visible in oocytes at diakinesis, the last stage of meiotic prophase (see Materials and Methods). Whereas WT oocyte nuclei consistently exhibit 6 pairs of homologous chromosomes attached by chiasmata (bivalents), oocyte nuclei in the dsb-2(me96) mutant exhibit a variable number of unattached (achiasmate) chromosomes (univalents), indicating a defect in chiasma formation (Figure 1A) resulting from an underlying defect in CO formation (below).
The me96 mutation was mapped to a 126 kb interval on chromosome II, and RNAi against F26H11.6 (a gene in the candidate interval) phenocopied the me96 mutant (Materials and Methods). Sequencing revealed a T-to-A transversion in F26H11.6 in the me96 mutant; as this mutation results in an early stop at codon 14 (TTA = >TAA) of F26H11.6 (280 codons total), me96 is presumed to be a null allele. A second, independently-isolated, dsb-2 allele (me97) contains a premature stop at codon 168 in the same gene, further confirming the identity of F26H11.6 as dsb-2.
Homology searches revealed an F26H11.6 paralog in C. elegans (F08G5.1), and genes encoding orthologs of both proteins were found in other Caenorhabditis species (briggsae, remanei and japonica) but were not identified in other organisms (Figures S1 and S2). Based on independent analyses implicating both genes in meiotic double-strand break formation [11], F08G5.1 and F26H11.6 were designated as dsb-1 and dsb-2, respectively. Multiple sequence alignment of this highly diverged protein family shows two readily alignable regions corresponding to residues 1–103 and 195–251 of the C. elegans F26H11.6/DSB-2 protein, each containing several conserved sites (Figure S1). These two regions are connected by a variable segment that in each protein contains an S/T Q cluster domain [12], a feature suggesting that these proteins are potential targets for phosphorylation by the ATM/ATR family of protein kinases.
The number of achiasmate chromosomes detected in diakinesis-stage oocytes of dsb-2 mutant hermaphrodites increases with maternal age, indicating a worsening of phenotype over time (Figure 1B). Whereas adult dsb-2(me96) hermaphrodites fixed one day after the L4 larval stage (24 hours post-L4) had an average of 8.5 DAPI-stained bodies at diakinesis (reflecting a mixture of bivalents and univalents), 48 hour post-L4 hermaphrodites had an average of 11.1 DAPI bodies (indicating that nearly all chromosome pairs lacked chiasmata). Further, as lack of chiasmata connecting homologs results in mis-segregation of chromosomes, both the frequency of inviable embryos (reflecting autosomal aneuploidy) and the frequency of males (XO, reflecting X chromosome mis-segregation) produced by dsb-2(me96) hermaphrodites likewise increased with maternal age (Figure 1C): frequencies rose from 27% dead embryos and 6% males on day 1 of egg-laying to 89% dead embryos and 29% males on day 3.
Age dependence of the dsb-2 mutant phenotype was also observed for the dsb-2(me97) allele, using GFP::COSA-1 as a cytological marker of crossover (CO) sites (Figure 1D). During wild-type meiosis, GFP::COSA-1 localizes to 6 foci per nucleus during the late pachytene and diplotene stages, marking the single CO/emerging chiasma on each homolog pair [13]. Whereas 6 GFP::COSA-1 foci were consistently observed in late pachytene nuclei of control worms regardless of maternal age, the number of GFP::COSA-1 foci was substantially reduced in dsb-2(me97) worms at 24 hours post-L4 and further declined by 48 hours post-L4 .
The age effect in dsb-2 mutants is not caused by persistence of maternal gene product in the germ line, as it was observed in homozygous mutant worms derived from either heterozygous parents or homozygous mutant parents (where no maternal product should be present). In addition, the age effect is evident at both standard (20°C), and elevated (25°C) growth temperatures.
Together, our data indicate that the function of DSB-2 is required throughout reproductive life to generate normal levels of COs and chiasmata, and becomes increasingly important for meiotic success in germ cell nuclei that enter the meiotic program at progressively later times. This implies that changes must occur as the worms age that render crossing over and chiasma formation increasingly sensitive to the loss of DSB-2 protein.
Successful chiasma formation requires pairing of homologous chromosomes, assembly of the synaptonemal complex (SC), and CO recombination between the homologs. Homolog pairing and SC assembly are not dependent on initiation or progression of recombination during C. elegans meiosis [14], facilitating investigation of potential involvement of DSB-2 in these events. To this end, we conducted immunofluorescence analyses on germ lines dissected from dsb-2 worms at 48 hours post-L4, when the CO/chiasma deficit is severe. Several lines of evidence indicate that the lack of chiasmata in dsb-2 mutants is due to a defect in the initiation of meiotic recombination.
First, dsb-2 mutant worms are proficient for pairing of the X chromosomes, as immunofluorescence of pachytene nuclei showed a single focus of HIM-8, a protein that binds a specific region of the X chromosome known as the pairing center [15], [16] (Figure 2A). Second, dsb-2 mutants are proficient for assembly of the SC, as immunostaining revealed proper loading of HIM-3 (an SC lateral element component) and SYP-1 (an SC central region component) [17], [18] along the lengths of aligned homologs (Figure 2B). Proficiency for pairing and synapsis suggests that dsb-2 mutants are deficient in the process of meiotic recombination per se.
Meiotic recombination is initiated by formation of DNA double-strand breaks (DSBs) by the SPO-11 protein [14], [19], followed by processing of these DSBs to enable loading of the DNA-strand exchange protein RAD-51, which can be detected as foci from zygotene to mid-pachytene stages in WT germlines [20], [21]. dsb-2 germ lines display greatly reduced levels of RAD-51 foci, with most nuclei having no foci (Figure 2C,D), suggesting either that fewer DSBs are made or that loading of RAD-51 is impaired. However, the dsb-2 mutant is proficient for loading of RAD-51 when DSBs are induced by gamma-irradiation, as seen by the presence of RAD-51 foci in germline nuclei fixed 1 hour post-irradiation (Figure 2C).
Furthermore, irradiation bypasses the requirement for DSB-2 and restores chiasma formation (Figure 2E). It was previously shown that in C. elegans, providing DSBs by irradiation rescues chiasma formation in the spo-11 mutant, which lacks the enzyme responsible for making programmed DSBs [14], [22]. The same effect was seen upon irradiation of dsb-2 mutant worms, demonstrating that the chiasma defect in dsb-2 worms is a result of a defect in SPO-11-induced DSB formation. In both dsb-2 worms and age-matched spo-11 worms, 1 kRad of irradiation resulted in efficient restoration of chiasmata in diakinesis-stage oocytes examined 18 hours post-irradiation (Figure 2E). Thus DSB-2 is a novel protein required for robust meiotic DSB formation.
Immunofluorescence experiments using an antibody against the DSB-2 protein (Materials and Methods) showed that DSB-2 localizes to chromatin in germ cell nuclei from meiotic entry to mid-pachytene (Figure 3A). DSB-2 staining is first detected in the transition zone (TZ; corresponds to leptotene and zygotene stages, when pairing and SC assembly occur), and is strongest overall in early pachytene, where it localizes to chromatin in an uneven pattern, showing a few bright patches per nucleus as well as fainter stretches/foci associated with most of the chromatin. Towards mid-pachytene, the bright patches diminish and the chromatin signal fades and then disappears from most nuclei. However, in a subset of nuclei in the mid/late pachytene region, DSB-2 staining becomes brighter, with bright stretches/foci along most of the chromatin; a few of these “outlier” brightly-staining nuclei are present in later pachytene and likely represent nuclei destined for apoptosis (see Discussion).
Apart from the outlier nuclei, the “DSB-2-positive” region of the germ line corresponds to nuclei at the stages in which DSB formation is presumed to occur [8], [21]. Indeed, co-immunostaining experiments showed that RAD-51 foci (marking DSB-dependent recombination intermediates) appear in nuclei shortly after DSB-2 staining appears on chromatin upon meiotic entry, and the RAD-51 foci disappear shortly after DSB-2 is no longer present on chromatin in mid-pachytene nuclei (Figure 3C). Previous work has demonstrated that germ cell nuclei at later stages of meiotic prophase are proficient to load RAD-51 when DSBs are introduced by irradiation [6]. Thus, the disappearance of RAD-51 foci in the endogenous case likely indicates that DSBs are no longer being formed and that existing DSBs have progressed to subsequent stages of repair. Indeed, we observe that COSA-1 foci marking designated CO intermediates appear in nuclei only after the removal of DSB-2 from chromatin (Figure 3D). In addition, the “outlier” bright-staining DSB-2 nuclei in late pachytene contain high levels of RAD-51 foci and lack COSA-1 foci, suggesting these nuclei are arrested in their progression and may be triggering a checkpoint response. Thus, the close correspondence between the zone where DSB-2 localizes on chromatin and the zone where RAD-51 foci are detected is not only consistent with the demonstrated role for DSB-2 in promoting DSB formation, but further suggests that loss of DSB-2 coincides with loss of competence for DSB formation and progression to a subsequent stage of DSB repair.
We used immunofluorescence analyses to investigate the relationships between DSB-2 and other meiotic factors that act at the DSB formation step. Figure 4 shows the relationship between DSB-2 and its paralog DSB-1, which was independently implicated in DSB formation [11]. Nuclear localization of DSB-1 and DSB-2 is detected in the same region of the gonad, and their staining patterns on chromatin have a similar appearance (Figure 4A). However, the relative intensity patterns of the two proteins differ during meiotic progression. Within the gonad, DSB-1 signal is detected on nuclei slightly before DSB-2 and has a stronger intensity early on, which then declines as nuclei progress through pachytene (except for the outlier nuclei); DSB-2 signal is weaker early on and peaks in intensity later than DSB-1 before eventually declining. Both proteins disappear from nuclei at the same time, and both localize to the same outlier nuclei. Within each nucleus, the intensity patterns on chromatin are also different, such that the DSB-1 and DSB-2 signals partially overlap but do not match each other (Fig. 4A inset).
Whereas DSB-2 localization is abolished in dsb-1 mutant germ lines [11], some DSB-1 protein is present on chromatin in the dsb-2 mutant (Figure 4B,C). DSB-1 staining in dsb-2 young adult germ lines (12 hours post-L4) appears comparable to age-matched wild-type controls despite that fact that RAD-51 foci are already substantially diminished by this stage; this indicates that the presence of DSB-1 on chromatin is not sufficient to promote efficient DSB formation in the absence of DSB-2. Further, the association of residual DSB-1 protein in the dsb-2 mutant appears to change with age, as DSB-1 staining in older dsb-2 germ lines (48 hours post-L4) is typically fainter and declines and disappears sooner than in WT. Together these data suggest that DSB-2 may be required to augment the DSB-promoting activity of DSB-1, possibly by affecting the nature of its association with chromatin, and that the reliance on DSB-2 for this augmentation becomes more acute with increasing age.
We further showed that DSB-2 localizes to chromatin independently of DSB formation, indicating that DSB-2 localization is not a consequence of DSB formation. Specifically, in spo-11 mutants, which lack endogenous DSBs, DSB-2 is detected on chromosomes in transition zone and pachytene nuclei, and the overall appearance of the staining within nuclei is similar to that in WT nuclei. However, DSB-2 association with chromatin extends further into late pachytene, suggesting that endogenous DSB formation affects timing of DSB-2 removal (Figure 5A; see below).
Finally, we assessed DSB-2 localization in germ lines lacking HIM-17, a THAP-domain containing protein that associates with germline chromatin and is required for normal levels of meiotic DSB formation [8]. In him-17 mutant germ lines, DSB-2 is detected on chromatin in nuclei from transition zone to late pachytene (Figure 5B), but the DSB-2 signal has an altered appearance within the nuclei: the bright patches characteristic of DSB-2 localization in WT germ cells are not observed, and DSB-2 instead displays only the fainter, more uniform distribution (Figure 5C). Thus, improper localization of DSB-2 may contribute to the observed defect in DSB formation in him-17 mutants. Taken together, these data suggest that association of DSB-2 and DSB-1 with chromatin is required to regulate competence for DSB formation by SPO-11.
The distribution of DSB-2 positive nuclei within the germ line is similar to that reported for nuclei exhibiting phosphorylation of serine-8 of nuclear envelope (NE) protein SUN-1[23]. SUN-1 is a part of a conserved protein complex that spans the NE and mediates attachment of the chromosomes to the cytoskeletal motility machinery [24], [25]. Although the SUN-1 protein is present throughout the germ line, SUN-1 S8P is detected only in a subset of nuclei during meiotic prophase [23]: SUN-1 S8P appears abruptly at the onset of meiotic prophase, with TZ nuclei exhibiting both bright SUN-1 S8P patches, corresponding to the chromosome attachment points that mediate chromosome movement, and a diffuse SUN-1 S8P staining throughout the NE; in early pachytene, the patches dissipate (except for one), but the diffuse NE staining persists, weakening until it disappears around mid-pachytene; however, a few outlier nuclei maintain SUN-1 S8P staining in later pachytene (Figure 3A and B) [26].
Co-staining experiments revealed that DSB-2 and SUN-1 S8P tend to be detected in the same nuclei (Figure 3A). The relative intensity patterns are different, with SUN-1 S8P exhibiting a much stronger signal in the TZ, and showing generally weaker signal towards mid-pachytene when compared with DSB-2 (Figure 3A). Most outlier nuclei are bright for both marks, but some are bright only for one of the marks and weak for the other. Nevertheless, the correlation is striking, suggesting that these two features (presence of DSB-2 on chromatin and of SUN-1 S8P on the NE) may be co-regulated.
In support of this hypothesis, we found that DSB-2 localization depends on the CHK-2 protein kinase. CHK-2 was previously shown to be required for several early prophase events including DSB formation, homolog pairing and synapsis, reorganization of chromosomes within the nucleus, chromosome movement, and associated phosphorylation of SUN-1 [23], [24], [27], [28]. We found that both DSB-2 staining and SUN-1 S8P (in early meiotic prophase) were severely reduced or absent in chk-2 mutant gonads (Figure 6B), indicating that CHK-2 represents a common regulator of these two distinct features of the meiotic program. Although chromatin-associated DSB-2 staining was not observed by immunofluorescence, Western blot analysis indicated that the DSB-2 protein is expressed in the chk-2 mutant (Figure 6B).
Whereas localization of DSB-2 on chromatin and Ser-8 phosphorylation of SUN-1 at the NE in meiotic prophase nuclei tend to be correlated, they do not depend on each other. SUN-1 S8P immunostaining is present on meiotic prophase nuclei in dsb-2 mutant worms, and the zone of SUN-1 S8P-positive nuclei is extended into later pachytene (Figure 6A, see below). Conversely, DSB-2 is able to load on chromatin in nuclei in sun-1(gk199) null mutant germ lines despite severe defects in germline organization and abnormal chromosome morphology (data not shown). Thus, these two features appear to be independent downstream readouts of CHK-2 activity in meiosis. Together, our data suggest that CHK-2 coordinates the meiotic program by acting as a common upstream regulator of two parallel pathways, thereby linking competence for DSB formation (mediated through DSB-2) with chromosome and NE dynamics (mediated through SUN-1 S8P).
The correlation between DSB-2 and SUN-1 S8P was also tested in him-19 mutants, which show an age-dependent pleiotropic phenotype that includes multiple defects (in DSB formation, chromosome clustering and movement in TZ, pairing and synapsis) that are hypothesized to result from mis-regulation of CHK-2 activity [29]. In 2-day old him-19 worms, SUN-1 S8P is missing from most of the TZ and early pachytene regions, but is present on a few scattered nuclei [23] that are also positive for DSB-2 (Figure 6C), consistent with these two features being controlled by common factors including CHK-2.
The removal of DSB-2 and SUN-1 S8P at mid-pachytene during WT meiosis, concurrent with the timing of disappearance of RAD-51 foci, led us to hypothesize the existence of a coordinated regulatory mechanism that simultaneously shuts down competence for DSB formation and changes other properties of the nucleus as it enters another stage of meiotic progression. In spo-11 and him-17 mutants, the zone of DSB-2 and SUN-1 S8P marked nuclei was extended beyond what was seen in WT (Figure 5A and B, Figure 7); extension of the SUN-1 S8P-positive zone in the spo-11 mutant was also reported by Woglar et al.[26]. In addition, in dsb-2 mutants, the zone of SUN-1 S8P staining was also prolonged (Figures 6A, 7). All of these mutants have defective DSB formation, and thus lack or have a deficit of downstream recombination intermediates and COs. We hypothesized that the deficit of appropriate recombination intermediates prolonged the zone of nuclei marked by DSB-2 and SUN-1 S8P. To test this hypothesis, we analyzed DSB-2 and SUN-1 S8P staining in several classes of meiotic mutants.
We tested mutants lacking proteins involved in early steps of DSB processing and repair: the rad50 mutant, which lacks the RAD-50 protein that has been implicated in meiotic DSB formation, DSB resection and RAD51 loading [6], [30]; the rad51 mutant, which lacks the RAD-51 recombinase that catalyzes strand exchange [20]; and the rad54 mutant, in which unloading of RAD-51 and progression of DSB repair are disrupted [31]. We found that in all of these mutants, DSB-2 and SUN-1 S8P staining are extended over most of the pachytene region (which also tends to be smaller than in WT gonads) (Figures 8, 7). This prolonged staining in mutants defective in DSB formation, processing, and repair suggests that such mutants lack the signals that would normally trigger removal of DSB-2 and SUN-1 S8P.
We next assessed zhp-3, msh-5, and cosa-1 mutants, which have a specific defect in CO formation. These mutants are proficient for homolog pairing and synapsis and can initiate and repair DSBs, but not as COs [13], [21], [22], [32]. All of these mutants showed an extended zone of DSB-2 and SUN-1 S8P staining (Figure 9 B, C, D), thus suggesting that lack of the CO-eligible recombination intermediates that depend on ZHP-3, MSH-5 and COSA-1 will prolong DSB-2 localization to chromatin and phosphorylation of SUN-1 S8.
Finally, we tested whether meiosis-specific chromosome structures are required to mediate the persistence of DSB-2 and SUN-1 S8P when CO-eligible inter-homolog recombination intermediates are reduced or lacking. We first examined the syp-1 mutant, which loads chromosome axis proteins but lacks a key structural component of the central region of the synaptonemal complex, and thus cannot establish synapsis between homologs [18]. In this mutant, DSB-dependent RAD-51 foci form and persist at elevated levels before disappearing at the very end of pachytene, and COs do not form [18], [21]; in addition, chromosome clustering, chromosome movement and SUN-1 phosphorylation are all greatly prolonged [18], [26], [28], [33]. We found that DSB-2 and SUN-1 S8P staining were both extended to the end of the pachytene region in the syp-1 mutant (Figure 9A). Thus, lack of SYP proteins leads to both lack of inter-homolog COs and prolonged DSB-2 and SUN-1 S8P staining.
In contrast, lack of HORMA domain chromosome axis proteins HTP-1 or HTP-3 does not lead to extended DSB-2 or SUN-1 S8P staining in the respective mutant gonads, despite a lack or severe deficit of inter-homolog COs (Figure 10). htp-1 mutants are defective in pairing of autosomes and assemble SCs between nonhomologous chromosomes, and they exhibit reduced RAD-51 foci reflecting reduced DSB formation and/or altered kinetics of repair [34], [35]; htp-3 mutants are defective in pairing and SC formation for all chromosomes and appear to lack DSBs [36], [37]. We find that despite the deficit or lack of COs in the htp-1 and htp-3 mutants, the zone of DSB-2 and SUN-1 S8P-positive nuclei was not extended (Figures 10, 7). This finding suggests that HTP-1 and HTP-3, or features of axis organization that are dependent on these proteins, are needed for DSB-2 and SUN-1 S8P to persist when CO recombination intermediates are absent.
In addition to acquiring and subsequently losing competence to form DSBs during meiotic prophase progression, C. elegans germ cells also switch on, then subsequently switch off, a specialized meiotic mode of DSB repair [6], [13], [38], [39]. Whereas switching on this meiotic DSB repair mode enables formation of inter-homolog intermediates capable of yielding COs, switching off this repair mode is proposed to facilitate repair of any remaining DSBs in order to guarantee restoration of genome integrity prior to cell division. One notable feature of this specialized meiotic DSB repair mode is a requirement for RAD-50 to load RAD-51 on DSBs induced by gamma-irradiation: whereas essentially all germ cells in wild-type gonads rapidly acquire RAD-51 foci following gamma-irradiation, formation of irradiation-induced RAD-51 foci is strongly inhibited in a specific subset of rad-50 mutant germ cells, from meiotic prophase onset until after the transition to late pachytene [6]. Thus, dependence on RAD-50 for RAD-51 loading at DSBs provides a means to visualize germ cells in which the meiotic DSB repair mode is engaged.
We used this feature to test the hypothesis that the presence of DSB-2 on chromatin correlates with engagement of the meiotic mode of DSB repair. By co-staining for DSB-2 and RAD-51 following irradiation of rad-50 mutant gonads, we found a striking correspondence between the nuclei in which DSB-2 was present on chromatin and the nuclei in which RAD-51 loading was inhibited (Figure 11A). Further, we similarly observed strong correspondence between the presence of DSB-2 and inhibition of RAD-51 loading in htp-1; rad-50 double mutant gonads, in which both features are restricted to a smaller region of the germ line than in the rad-50 single mutant [6]; Figure 11B). Moreover, in both rad-50 and htp-1; rad-50 gonads, nuclei exhibited this inverse correlation between DSB-2 and RAD-51 staining even when neighboring nuclei were in a different mode. In the context of a model in which association of DSB-2 with chromatin is a marker for a DSB-competent state, these results suggest that competence for DSB formation and utilization of the meiotic DSB repair mode are coordinately turned on and shut off, and that coordination of these processes occurs at the level of individual nuclei.
In this work, we identify DSB-2 as a protein that is required for efficient meiotic DSB formation and that localizes to chromatin during the stages of meiotic prophase when DSBs are thought to form. DSB-2 localizes to chromatin independently of SPO-11 (and thus of DSB formation) and is restricted to the region of the gonad where RAD-51 foci mark processed DSBs (from TZ to mid-pachytene). Further, the fact that exogenous DSBs induced by irradiation rescue the chiasma defect in dsb-2 mutant germ cells indicates that the downstream DNA processing and CO formation machinery are functional in the mutant. Moreover, the timing of disappearance of DSB-2 coincides with the cessation of DSB formation (implied by the disappearance of RAD-51 foci), suggesting a model in which removal of DSB-2 (and presumably other factors) results in shutting down of DSB formation. Based on these data, we propose that DSB-2 regulates competence for SPO-11-dependent DSB formation during C. elegans meiosis.
Several properties distinguish DSB-2 from other previously identified chromatin-associated proteins (HIM-17, XND-1 and HIM-5) that influence DSB formation in C. elegans. Whereas HIM-17, XND-1 and HIM-5 proteins localize to chromatin in nuclei throughout the germ line [8], [9], [10], the presence of DSB-2 on chromatin correlates with the timing of DSB formation. Further, while him-17 and xnd-1 mutants display pleiotropic phenotypes indicating that HIM-17 and XND-1 have additional roles regulating germ line proliferation and/or organization [9], [40], dsb-2 mutants are specifically defective in meiotic DSB formation. In addition, whereas XND-1 and HIM-5 affect DSB formation predominantly on the X chromosomes, DSB-2 is required for efficient DSB formation on all chromosomes. Together these data suggest that DSB-2 has a more direct role in promoting DSB formation than do HIM-17, XND-1 or HIM-5.
We interpret the region of the germ line where nuclei are positive for DSB-2 localization to represent the zone in which nuclei are competent to undergo DSB formation. Consistent with this interpretation, in meiotic mutants in which the DSB-2-positive zone is extended (and that are capable of making DSBs and loading RAD-51), RAD-51 foci are higher in number and persist beyond mid-pachytene [20], [21], [31], [32]. In principle, persistence of RAD-51 foci could be due to excess/prolonged DSB formation, delayed RAD-51 removal, or both. Thus, caution is warranted when using such mutants to estimate numbers of DSBs. We suggest that in mutants with an extended DSB-2 positive zone (in which the DSB machinery is functional) germ cells may continue to make additional DSBs for a prolonged period, whether or not they are ultimately competent to repair them.
How might DSB-2 control DSB competence? Given its broad yet uneven localization on chromatin, it might act by altering chromatin structure to create an environment that is permissive for the activity of SPO-11 and the DSB machinery. It might also act directly upon SPO-11 and the DSB machinery, by recruiting and/or activating it at certain locations depending upon the underlying chromatin structure. It is intriguing that DSB-2 localizes to a few bright patches/foci in addition to its broader chromatin staining. The fact that these bright patches are absent in him-17 mutants, which are defective in DSB formation, suggests that the patches may have functional significance.
Immunofluorescence analyses of DSB-2 in both wild type and meiotic mutants were highly informative regarding how DSB formation is coordinated with multiple distinct aspects of the meiotic program. We found that presence of DSB-2 on chromosomes and the presence of SUN-1 S8P are highly correlated, despite the fact that neither feature is required for the other. Further, we identified CHK-2 as a common upstream regulator of these two features, and we suggest that CHK-2 links acquisition of competence for DSB formation (promoted by DSB-2) with nuclear/chromosomal processes required for successful pairing and synapsis of homologous chromosomes (mediated by SUN-1 at the NE). Moreover, the correlated removal of both DSB-2 and SUN-1 S8P at mid-pachytene, at the same time that RAD-51 foci disappear, further suggests the existence of coordinated regulatory mechanisms that shut down competence for DSB formation and change other properties of the nucleus as germ cells transition to a later stage of meiotic progression.
As seen in multiple experimental systems, DSB formation is restricted to a specific time window in early prophase, indicating that cells must have a means to shut down the meiotic DSB machinery [3]. However, little is known about what controls this transition. Recent evidence from Drosphila, mice and budding yeast suggests that ATM, a protein kinase involved in DNA damage response, may play a role in limiting meiotic DSB formation [41], [42], [43]. It was suggested that ATM is activated by meiotic DSBs and inhibits further DSB formation at the local level by triggering a negative feedback loop. Based on the current work, we propose that additional negative feedback regulation operates at the nucleus-wide level to mediate shutdown of DSB formation during C. elegans meiosis.
Our evidence that germ cells have the capacity to monitor and respond to the presence or absence of DSB-dependent CO-eligible recombination intermediates is based on the analysis of DSB-2 localization in various meiotic mutants. We found that DSB-2 persists in mutants with defects in DSB formation (spo-11, him-17, rad-50), in mutants with defects in early steps of DSB processing (rad-50, rad-51, rad-54), as well as in mutants that can make DSBs but repair them by pathways that do not yield inter-homolog COs (zhp-3, msh-5, cosa-1). Although we cannot exclude the possibility that different defects in these mutants elicit the same response, the parsimonious explanation is that DSB-2 persistence reflects a response to the common deficit shared by all of these mutants, i.e., the inability to generate CO recombination intermediates. Thus, we infer that CO-eligible recombination intermediates are required for removal of DSB-2 with WT timing. We propose a model in which the appearance of CO-eligible recombination intermediates results in a signal (or quenching of an inhibitory signal) that is necessary to trigger the shutdown of DSB formation, in part by removal of DSB-2 (Figure 12). We suggest that this change occurs at the nucleus-wide level when cells sense that sufficient CO-eligible intermediates have been formed to guarantee one CO per chromosome pair. Once this requirement is met, cells are permitted to enter a different state of meiotic progression; if this condition is not met, cells experience a delay in this transition. This type of coupling can be viewed as analogous to checkpoint mechanisms that make cell cycle progression contingent upon fulfillment of a requirement to complete a monitored event. However, it is also appropriate to consider such a coupling as reflecting operation of a negative feedback circuit wherein the formation of threshold levels of a downstream product (i.e. CO-eligible recombination intermediates) feeds back to inhibit an earlier step in the pathway (i.e. DSB formation). Thus, we envision a regulatory network governing DSB formation that involves negative feedback operating on (at least) two levels, one that inhibits DSB formation locally (in a region where a DSB has already formed [41], [42], [43]), and one that inhibits DSB formation nucleus-wide once sufficient CO-eligible recombination intermediates are established. This regulatory network would ensure that sufficient DSBs are made to guarantee that every chromosome pair undergoes a CO [13], [31], [39], while protecting against excessive DSB levels or local concentration of DSBs that could have deleterious effects.
We further propose that multiple aspects of the meiotic recombination program undergo a coordinated transition that in wild type germ cells is marked by disappearance of DSB-2 and SUN-1 S8P (Figure 12). We proposed in a previous study that access to the homologous chromosome as a repair partner is shut down once sufficient CO-eligible recombination intermediates are formed [39]. We suggested that this transition occurs around mid-pachytene in WT germ lines, and we showed that inter-homolog access is prolonged in msh-5 mutants [39]. In light of the current results, an attractive possibility is that the appearance of sufficient CO-eligible recombination intermediates simultaneously signals both shut-down of DSB formation and shut down of inter-homolog access for DSB repair. Moreover, we found that another specialized aspect of the meiotic DSB repair program, namely the dependence on RAD-50 for rapid loading of RAD-51 on IR-induced DSBs, is restricted to nuclei positive for DSB-2. This finding further strengthens the case that cessation of programmed DSB formation is coordinated with a major transition in the mode of DSB repair.
It is notable that in mutants defective for HORMA domain axis proteins HTP-1 and HTP-3, the DSB-2/SUN-1 S8P - positive zone is not extended despite the absence of CO-eligible recombination intermediates on most or all chromosomes. This finding raises the possibility that this family of proteins, which was previously implicated in the operation of checkpoint-like coupling mechanisms that coordinate early prophase chromosome movement, homolog recognition and SC assembly [34], may also be required for operation of checkpoint-like mechanisms that make later events in meiotic progression contingent upon the formation of CO-eligible recombination intermediates.
We speculate that the regulatory network that coordinates this meiotic transition (i.e. the shutdown of DSB formation and accompanying changes) likely involves the activities of one or more protein kinases. As the CHK-2 protein kinase is required to promote the acquisition of both DSB-2 and SUN-1 S8P, it is likely that the disappearance of DSB-2 and SUN-1 S8P requires inactivation of CHK-2, suggesting that CHK-2 may be a key target of feedback regulation. Further, DSB-2 contains several potential phosphorylation sites both for CHK-2 and for the ATM/ATR protein kinases [12], [44]. Future work will investigate the significance of these for DSB-2 function and regulation.
The fact that DSB-2 and SUN-1 S8P are coordinately removed in wild-type meiosis (and coordinately prolonged in mutants) implies that the NE also responds to signaling from CO-eligible recombination intermediates. Our findings confirm and extend the recent report of Woglar et al., who similarly showed that the SUN-1 phosphorylation is prolonged in spo-11 and rad-51 mutants and concluded that establishment of CO intermediates is necessary for exit from early pachytene (as defined by loss of phospho-SUN-1) [26]. The change in SUN-1 phosphorylation status at this transition may be indicative of global changes in properties of the nucleus that occur as it enters a different stage of meiotic progression; e.g., the fluidity of the nuclear membrane, which is modified upon entry into meiotic prophase [28], may revert to a more constrained state similar to that of non-meiotic germ cells. Such a change would be analogous to the observed reversion to the non-meiotic mode of DSB repair that occurs at this same transition.
While DSB-2 and SUN-1 S8P immunofluorescence signals become dimmer and disappear from most nuclei by the time they reach the mid-pachytene region of the germ line, a few “outlier” nuclei show bright DSB-2 and SUN-1 S8P staining later in the pachytene region. Sometimes the chromatin in these nuclei has a clustered organization reminiscent of zygotene or early pachytene stages, but in contrast to earlier nuclei, these outlier nuclei have brighter DSB-2 staining covering most of the chromatin as well as high levels of RAD-51 foci. This difference suggests that these nuclei are arrested in their progression and may have triggered a checkpoint response. This response could be due to failure to make appropriate CO-eligible recombination intermediates and/or to the presence of excess or persistent DNA breaks. These processes may be inter-related: if the failure to make CO-eligible recombination intermediates keeps DSB formation active, this could increase the chance of accumulating levels of DNA damage that challenge the capacity for repair. Accumulation of high levels of DSB-2 and SUN-1 S8P may indicate that these nuclei are triggering the recombination/DNA damage checkpoint and will be targeted for future apoptosis. While these outlier nuclei may be destined for apoptosis, however, they likely have not yet engaged the cell death program, as outlier nuclei are still detected in mutants lacking the pro-apoptotic factors CED-3 or CED-4 [11], [26].
An intriguing aspect of the dsb-2 mutant phenotype is that the defect in meiotic recombination worsens with age. This implies that the DSB-1 protein retains some residual DSB-promoting activity in the absence of its paralog, but also indicates that the requirement for DSB-2 becomes more acute in older germ cells. Interestingly, CO distribution has also been found to differ between young and old WT C. elegans oocytes [45]. This suggests that meiotic recombination processes such as DSB formation and CO distribution are sensitive to changes in the germline environment that occur as worms age. However, the ability to achieve accurate and reliable meiosis in the context of a changing environment is advantageous for the reproductive success of the organism. The C. elegans reproductive system has substantial plasticity in this regard, as the duration of progression through meiotic prophase varies markedly with both sex and age and can be modulated dramatically in the female germ line by the availability of sperm [46]. The operation of feedback networks such as that demonstrated here provides a means to regulate and coordinate key events and transitions in a manner that buffers the system against a varying environment, thereby promoting reproductive success.
Strains were maintained at 20°C under standard conditions. Experiments were performed at 20°C unless otherwise noted. Strains used in this study:
AV334 unc-119 III; ruIs32 [Ppie-1::GFP-his-11; unc-119(+)] III mnT12 (IV;X)
AZ212 unc-119 III; ruIs32 [Ppie-1::GFP-his-11; unc-119(+)] III
AV477 dsb-2(me96) II 4X outcrossed
AV501 rol-1(e91) dsb-2(me96) II
AV511 rol-1(e91) dsb-2(me96) unc-52(e998) II
AV539 rol-1(e91) dsb-2(me96)/mnC1 [dpy-10(e128) unc-52(e444)] II
AV727 meIs8[unc-119(+) pie-1promoter::gfp::cosa-1] II ;; ItIs37[unc-119(+)pie-1::mcherry::histoneH2B]; ltIs38[pAA1;pie-1 promoter::GFP::PH::unc-119(+)]
AV758 dsb-2(me97) meIs8[unc-119(+) pie-1promoter::gfp::cosa-1] II ;; ItIs37[unc-119(+)pie-1::mcherry::histoneH2B]; ltIs38[pAA1;pie-1 promoter::GFP::PH::unc-119(+)]
AV630 meIs8[unc-119(+) pie-1promoter::gfp::cosa-1] II
AV645 spo-11(ok79)/nT1 IV; +/nT1[qIs51] V
AV146 chk-2(me64) rol-9(sc148)/unc-57(e369) rol-9(sc148) V
AV660 chk-2(me64) rol-9(sc148)/sC4(s2172)[dpy-21(e428)] V
VC292 +/nT1 IV; sun-1(gk199)/nT1 V
VC255 +/nT1 IV, him-17(ok424)/nT1 V
AV158 +/nT1 IV; rad-50(ok197)/nT1 [unc-?(n754) let-? qIs50] V
TG9 dpy-13(e184) rad-51(lg8701) IV/nT1[let-?(m435)] (IV;V)
VC531 rad-54 and tag-157(ok615) I/hT2[gli-4(e937) let(9782) qIs48] I; III
AV449 zhp-3(me95)/hT2 [bli-4(e937) let-? (q782) qIs48] I
AV603 msh-5(me23)/nT1 IV; +/nT1[qIs51] V
AV596 cosa-1(tm3298)/qC1[qIs26] III
AV307 +/nT1 IV; syp-1(me17)/nT1 V
AV393 htp-1(gk174) IV/nT1[unc-?(n754) let-? qIs50] (IV;V)
TY4986 htp-3(y428) ccIs4251 I/hT2[bli-4(e937)let-?(q782) qIs48] (I,III).
AV473 +/nT1 IV; rad-50(ok197)/nT1[qIs51] V
AV443 htp-1(gk174)/nT1[ unc-?(n754) let-? qIs50] IV; rad-50 (ok197)/nT1 [qIs51] V
Bristol (N2) wild type
CB4856 Hawaiian wild type
The dsb-2(me96) allele was isolated in a genetic screen for meiotic mutants exhibiting defects in chiasma formation or chromosome organization in diakinesis-stage oocytes, conducted in collaboration with M. Hayashi [47]. The AV334 strain used for this screen, which allows visualization of chromosomes using a germline-expressed GFP::histone H2B fusion protein, also contains a fusion of chromosomes IV and X. Parental (P0) L4 hermaphrodites were treated with ethyl methanosulfonate (EMS) as in [48] and were plated individually. F1 progeny were picked to individual plates to produce progeny, and pools of F2 progeny worms from each F1 plate were mounted on multi-well slides in anesthetic (0.1% tricaine and 0.01% tetramisole in M9 buffer) and their germ lines were visualized for meiotic defects. Two mutations affecting meiotic recombination, me95 and me96, were identified based on the presence of univalents at diakinesis in a the subset of F2s (from independent F1s) and were recovered by plating of siblings; repeated outcrossing (with N2) and selecting for the mutant phenotype removed the chromosome fusion as well as the GFP::H2B transgene. Mapping, complementation testing and sequencing revealed that me95 is an allele of zhp-3, containing a C-to-T transition that results in a premature stop at codon 348 of the predicted 387 amino acid coding sequence of K02B12.8a. Mapping of the me96 mutation (below) indicated that it identified a new component of the meiotic machinery.
Initial SNP mapping based on the methods of [49] and [50] placed me96 near genetic map position 20 on the right side of chromosome II. To select for informative COs near this region, rol-1 me96 unc-52 worms were crossed with CB4856 males, and Rol non-Unc F2 progeny were selected and genotyped for SNP pkP2117 at genetic map position 17.9 to select for COs occurring between this marker and unc-52 (genetic map position 23). Informative recombinants were assessed for me96 phenotype and typed using additional SNP markers in the region, narrowing down the position of me96 to a 165 kb interval (between SNP markers uCE2-2315 and uCE2-2332) comprising 36 candidate genes. RNAi of candidate genes was performed using bacterial clones from the RNAi feeding library [51], [52] as in [53]. Worms used were AZ212 worms, which contain GFP::histone, as this genotype was shown to be more sensitive to RNAi [54]. RNAi against candidate gene F26H11.6 (at 15°C, but not at 20°C) phenocopied the me96 mutation, eliciting a mixture of bivalents and univalents at diakinesis. Sequencing identified a T-to-A transversion generating an early stop at codon 14 (TTA = >TAA) of the predicted F26H11.6 coding sequence (280 codons total) in the me96 mutant.
A second dsb-2 allele (me97) was isolated independently in a screen for mutants with altered numbers of GFP::COSA-1 foci (which mark CO sites in late pachytene); me97 fails to complement me96 and contains a premature stop at codon 168, further confirming the identity of F26H11.6 as the dsb-2 gene. Except where otherwise noted, all analyses were conducted using the me96 allele.
L4 hermaphrodites were picked to individual plates, allowed to lay eggs, and transferred to fresh plates every 24 hours for three days. Hermaphrodites start laying eggs after they transition from L4 to adult, and lay most eggs in the first three days. Inviable embryos that do not hatch are indicative of autosomal mis-segregation, while male progeny indicate X-chromosome mis-segregation. Eggs from eight dsb-2(me96); rol-1(e91) hermaphrodites (where rol-1 is a marker with no meiotic defects) were counted. The number of eggs counted for each day is: 573 (day 1), 879 (day 2) and 488 (day 3), with an average of 243 eggs per hermaphrodite over the three day interval.
Numbers of DNA bodies present in diakinesis oocytes were assessed in intact adult hermaphrodites of indicated age, fixed in ethanol and stained with 4′,6-diamidino-2-phenylindole (DAPI) as in [40]. This method underestimates the frequency of achiasmate chromosomes, as some univalents lie too close to each other to be resolved unambiguously. Numbers of nuclei scored were: 88 and 106 for dsb-2 worms, 1 day and 2 day post L4 respectively; 43 and 57 for WT worms, 1 day and 2 day post L4 respectively.
Numbers of GFP::COSA-1 foci in late pachytene nuclei of live anesthetized worms (0.1% tricaine and 0.01% tetramisole in M9 buffer ) were quantified by taking 3D image stacks on a DeltaVision microscope. GFP foci were counted in the last five rows of pachytene nuclei; only nuclei completely contained within the stack were scored, and nuclei with features indicative of apoptosis (compact and bright mCherry::histoneH2B signal) were excluded. 24 h control data were from fixed immunofluorescence images [13]. Numbers of nuclei scored: dsb-2 24 h, n = 127; dsb-2 48 h, n = 101; WT 24 h, n = 76; WT 48 h, n = 78. (Note: Whereas most nuclei in a mutant that lacks DSBs and COs have zero COSA-1 foci, 20% have one or two foci presumably reflecting non-specific aggregation of CO proteins when a suitable substrate is absent [13]. Thus, a subset of dsb-2 nuclei with one or two COSA-1 foci may similarly lack COs, especially at 48 h post L4 where nuclei with zero foci are frequent.)
Immunofluorescence was conducted as in [55] with minor modifications. Unless otherwise noted, all experiments were performed at 40–48 hours post L4. Worms were cut at the vulva to dissect the gonads (in egg buffer with 0.1% Tween-20) and fixed with 1% paraformaldehyde (in egg buffer) for 5 minutes. Slides (Superfost Plus) were covered with a coverslip and frozen in liquid nitrogen. The coverslip was removed, and slides were immersed in cold (−20°C) methanol for 1 minute. Slides were washed three times for 8–10 minutes in phosphate-buffered saline containing 0.1% Tween-20 (PBST) and then blocked for one hour with 0.5% bovine serum albumin (BSA) diluted in PBST. Primary antibody solution was added (50 µl) on top of the dissected gonads and covered with a parafilm square. Slides were incubated overnight in a humid chamber at room temperature, then washed three times for 8–10 minutes in PBST. Secondary antibody solution was added (50 µl) and slides were incubated with parafilm cover for 2 hours at room temperature in the dark. Slides were washed three times with PBST and incubated for 5 minutes with 2 µg/ml DAPI solution in the dark, followed by two more washes. Slides were mounted with Vectashield and the coverslip was sealed with nail polish.
The following primary antibodies were used at the indicated dilutions in PBST with 0.5% BSA: guinea pig anti-HIM-8 (1∶500) [16], rabbit anti-HIM-3 (1∶200) [17], guinea pig anti-SYP-1 (1∶200) [18], rabbit anti-RAD-51 (1∶500) [21] , guinea pig anti-SUN1 S8P (1∶1000) [23], rabbit anti-DSB-2 (1∶5000), rat anti-RAD-51 (1∶250), guinea pig anti-DSB-1 (1∶500) [11].
An affinity-purified rabbit polyclonal antibody against DSB-2 was generated by SDIX (Newark, DE) using the C-terminal 100 amino acids of F26H11.6 as the immunogen. Specificity of the antibody was demonstrated both by the lack of chromatin staining in immunofluorescence analysis of dsb-2 mutant gonads (Figure 6A) and by Western blot analysis (Figure 6B).
Rat anti-RAD-51 antibody was generated using a His-tagged fusion protein expressed from plasmid pET28a containing the entire RAD-51S coding sequence [56]; immunizations and bleeds were performed by SDIX. Rat anti -RAD-51 was affinity purified against membrane-bound protein as described in [57] with the following modifications: nitrocellulose membrane was blocked in 5% milk in 1×TBST; and, eluates containing rat anti -RAD-51 were further purified by dialysis with 12–14 kDa dialysis tubing (Spectrum) in 1×TBST for 1 hour and overnight at 4°C. Specificity was demonstrated by showing that rat anti-RAD-51 foci colocalize with rabbit anti-RAD-51 foci [21] by immunofluorescence and that these recombination-dependent foci are eliminated in spo-11(me44) gonads.
All secondary antibodies were Alexa Fluor goat from Invitrogen used at 1∶200 dilution in PBST with 0.5% BSA.
Immunofluorescence images were acquired using the DeltaVision microscopy system (Applied Precision) and deconvolved using softWoRx software. Images shown are maximum-intensity projections of Z-stacks acquired at 0.3 µm intervals.
For each wild-type germ line evaluated, RAD-51 foci were quantified in 8 contiguous rows of pachytene nuclei from the region where foci were most abundant. The average distance (in rows of nuclei) between the position of this peak and the end of the transition zone was calculated for wild-type germ lines, and this distance was used to define the corresponding regions to be scored in dsb-2(me96) mutant germ lines (in which the abundance of RAD-51 foci was low throughout). Quantitation was carried out on deconvolved 3D image stacks using SoftWoRx software; only nuclei that were completely contained with in the image stack were scored. Occasional atypical nuclei with condensed, bright DAPI signals were excluded. Numbers of nuclei scored: WT, n = 335; dsb-2, n = 196.
For each genotype, sixty adult worms (24 hours post-L4) were picked into M9+0.05% Tween 20, washed gently three times, then lysed by resuspension in 2× Laemmli Sample Buffer (Bio-Rad). Gel electrophoresis was performed on a 4–15% Criteriot TGX gradient gel (Bio-Rad), followed by transfer of proteins to a PVDF membrane. Blots were probed with rabbit anti-DSB-2 (1∶1000 in 5% milk) for 2 hours, followed by HRP-conjugated secondary antibody and detection by ECL.
Worms were exposed to 1 kRad (10Gy) of gamma-irradiation using a Cs-137 source. The 1 kRad dose was chosen based on its sufficiency to restore chiasmata to 95% of chromosome pairs in affected nuclei of the spo-11(ok79) mutant [6]. Worms were irradiated at 36 hours post L4, and the number of DNA bodies at diakinesis was assessed in worms fixed at 18 hours post-irradiation, for both dsb-2 and age-matched spo-11 mutants. The dsb-2 worms also carried the rol-1 marker, which does not affect meiosis. This assay tends to underestimate the incidence of achiasmate chromosomes, as some lie too close together to be resolved. Numbers of nuclei scored were: 71 and 76 for dsb-2 worms, untreated and irradiated respectively; 76 and 45 for spo-11 worms, untreated and irradiated respectively.
Worms were exposed to 5 kRad (50Gy) of gamma-irradiation using a Cs-137 source. Formation of RAD-51 foci was assessed by immunofluorescence in gonads dissected and fixed 1 hour after irradiation. Germ lines from rad-50 and htp-1; rad-50 mutants were irradiated at 24 hours post-L4, and stained with DAPI, RAD-51 antibody and DSB-2 antibody. Germ lines from dsb-2 mutants were irradiated at 48 hours post-L4 and stained with DAPI and RAD-51 antibody.
|
10.1371/journal.pntd.0002973 | Auranofin Is Highly Efficacious against Toxoplasma gondii In Vitro and in an In Vivo Experimental Model of Acute Toxoplasmosis | The mainstay of toxoplasmosis treatment targets the folate biosynthetic pathways and has not changed for the last 50 years. The activity of these chemotherapeutic agents is restricted to one lifecycle stage of Toxoplasma gondii, they have significant toxicity, and the impending threat of emerging resistance to these agents makes the discovery of new therapies a priority. We now demonstrate that auranofin, an orally administered gold containing compound that was FDA approved for treatment of rheumatoid arthritis, has activity against Toxoplasma gondii in vitro (IC50 = 0.28 µM) and in vivo (1 mg/kg).
Replication within human foreskin fibroblasts of RH tachyzoites was inhibited by auranofin. At 0.4 µM, auranofin inhibited replication, as measured by percent infected fibroblasts at 24 hrs, (10.94% vs. 24.66% of controls; p = 0.0003) with no effect on parasite invasion (16.95% vs. 12.91% p = 0.4331). After 18 hrs, 62% of extracellular parasites treated with auranofin were non-viable compared to control using an ATP viability assay (p = 0.0003). In vivo, a previously standardized chicken embryo model of acute toxoplasmosis was used. Fourteen day old chicken embryos were injected through the chorioallantoic vein with 1×104 tachyzoites of the virulent RH strain. The treatment group received one dose of auranofin at the time of inoculation (1 mg/kg estimated body weight). On day 5, auranofin-treated chicken embryos were 100% protected against death (p = 0.0002) and had a significantly reduced parasite load as determined by histopathology, immunohistochemistry and by the number of parasites quantified by real-time PCR.
These results reveal in vitro and in vivo activity of auranofin against T. gondii, suggesting that it may be an effective alternative treatment for toxoplasmosis.
| Toxoplasma gondii is a protozoan parasite that infects at least two thirds of the world human population. Once it infects the human host, it has great predilection for the brain and the retina of the eye. It remains latent until the host's immune system weakens, and then causes organ tissue damage. There are very few treatments available that are active against this parasite, and they all fail to eradicate it from the human body. Hence, there is always a risk for recurrence and/or disabling long-term complications such as blindness or neurological abnormalities. Despite this fact, it has been over fifty years since most anti-Toxoplasma agents were initially described. Most recently, in an attempt to expedite the process of drug discovery, older drugs are making a comeback by being re-purposed for new diseases. Auranofin, which was originally designed to treat rheumatoid arthritis, has consistently shown antiparasitic activity against multiple organisms, including parasites of great public health importance such as Plasmodium, Schistosoma and Leishmania, although most of these reports are based on in vitro assays. Herein, we present our studies that demonstrate that auranofin is active against Toxoplasma gondii in vitro and in an animal model of acute Toxoplasma infection, suggesting that auranofin has great potential to become a new anti-Toxoplasma agent.
| Toxoplasma gondii is the second leading cause of hospitalizations (8%) and deaths (24%) among foodborne pathogens in the US. People typically become infected by three principal routes of transmission: foodborne, animal-to-human (zoonotic) and mother-to-child (congenital), and rarely as post-solid organ transplant, blood transfusion or work related injuries. The number of primarily infected individuals varies widely worldwide: 22.5% of the American population is infected with this parasite [1], while in other parts of the world, the infection prevalence can be as high as 95%. These individuals are at risk of developing disease which usually follows after congenital transmission or reactivation of T. gondii latent forms (bradyzoites) in immunocompromised hosts [2]. Unfortunately, current available therapies have significant toxicity and are only active against one lifecycle stage of the parasite, the tachyzoite, and have no effect over the bradyzoite form [3], [4]. Furthermore, the impending threat of emergence of resistance to these therapies makes the discovery of new therapeutic targets a priority.
One promising re-profiled drug, auranofin, a gold containing compound that is FDA approved for the treatment of rheumatoid arthritis, has recently shown broad antiparasitic activity against Plasmodium falciparum [5], Leishmania infantum [6], Schistosoma mansoni [7] and Entamoeba histolytica [8] among others. Auranofin's anti-parasitic activity seems to stem from its gold molecule that readily dissociates and targets thioredoxin reductase, which we have recently demonstrated in our work with Entamoeba histolytica trophozoites [8] and cysts of Entamoeba invadens (manuscript in preparation). Given that thioredoxin reductase is a highly conserved enzyme in protozoan parasites [9] and based on our preliminary data, we hypothesized that auranofin has activity against T. gondii.
Per Public Health Services (PHS) Policy, the Institutional Animal Care and Use Committee (IACUC) oversight is not required for egg model of toxoplasmosis using unhatched eggs. PHS Policy is applicable to proposed activities that involve live vertebrate animals. While embryonal stages of avian species develop vertebrae at a stage in their development prior to hatching, Office for Protection from Research Risks (OPRR) has interpreted "live vertebrate animal" to apply to avians (e.g., chick embryos) only after hatching (http://www.gpo.gov/fdsys/pkg/CFR-2009-title9-vol1/xml/CFR-2009-title9-vol1-chapI-subchapA.xml).
Primary human foreskin fibroblasts (HFF) were initially cultured in Dulbecco's modified Eagle's medium supplemented with 10% fetal bovine serum (Gibco, Life Technologies, Carlsbad, Calif.), penicillin and streptomycin (50 µg/ml each) and maintained subsequently in the same medium with 2% fetal bovine serum (FBS). T. gondii RH tachyzoites (National Institutes of Health AIDS Reference and Reagent Repository, Bethesda, MD) and RH tachyzoites expressing cytoplasmic yellow fluorescent protein (YFP, kindly provided by M.J.Gubbels, Boston College, Boston, Massachusetts) [10] were maintained by serial passage in HFF monolayers at 37°C in a humid 5% CO2 atmosphere.
Auranofin (Enzo Life Sciences), was dissolved in 100% ethanol as a stock solution (4 mg/mL) and then diluted in complete tissue culture medium (DMEM +2% FBS) for final concentrations of 0.1 to 19 µM. For in vivo experiments, the auranofin concentration used was 1 mg/kg of estimated body weight.
Pyrimethamine (Sigma Aldrich) was dissolved in 100% ethanol in a stock concentration of 5 mg/mL. Three final dilutions in complete tissue culture medium were examined: 0.02, 0.1 and 0.2 µM. Sulfadiazine (Sigma Aldrich) was dissolved in complete medium at a final stock concentration of 5 mg/mL. Three doses in complete medium were evaluated: 0.2, 1 and 2 µM. Testing of pyrimethamine/sulfadiazine combinations by checkerboard method (using abovementioned concentrations) was carried out in triplicates and in three independent experiments.
Black, 96-well, tissue culture-treated plates with optical clear bottoms were purchased from Greiner bioOne (Germany). HFF cells were added to wells in a final volume of 200 µl and grown to confluence. Freshly syringed, lysed parasites were filtered through a 5- µm polycarbonate filter (sterile Millex SV low binding durapore PVDF syringe filters), centrifuged, and resuspended in parasite culture medium without phenol red (Gibco, Life Technologies, Carlsbad, Calif.). HFF host cells were infected with 0.5×103 YFP- expressing RH tachyzoites in the presence of different dilutions of auranofin, sulfadiazine, pyrimethamine, or control (ethanol alone) in triplicate. Plates were kept in a humidified incubator with 5% CO2 at 37°C for five days. After 5 days of incubation, plates were washed, fixed with 4% paraformaldehyde and read with a Synergy Mx BioTek (Vermont, US) multimode microplate reader Gen5 Software (excitation 510 nm; emission 540 nm). For excitation, a single flash from a UV Xenon lamp was used for each well, and emission signals were recorded with a sensitivity setting of 100. The values are presented as percentages of growth inhibition relative to the untreated controls (defined as 100% survival).
YFP-expressing RH tachyzoites, grown in DMEM 2% FBS without phenol red, syringed, lysed and filtered as described above, were used to infect 6-well plates containing fresh, confluent HFF monolayers. Each well was inoculated with 0.5×103 YFP-expressing RH tachyzoites. The treatment group was treated with auranofin (0.4 µM), and the parasites were allowed to grow for seven days before fixation with 4% paraformaldehyde. Plaques were defined as independent foci of green fluorescence that correspond to a cluster of YFP tachyzoites infecting multiple adjacent HFFs. These plaques were visualized and counted per low power field (20x) with an inverted fluorescence microscope.
CellTiter 96 Non-Radioactive Cell Proliferation Assay was performed according to the manufacturer's instructions (Promega). Confluent monolayers of HFF host cells (approximately 1×104) plated in clear bottom 96-well plates were treated with multiple dilutions of auranofin (0.1–19.2 µM). After 5 days incubation, 15 µL of dye solution (tetrazolium salt) was added to each well and the plates were incubated at 37°C for 4 hr. The Solubilization Solution/Stop Mix was then added to the culture wells to solubilize the formazan product, and the absorbance at 570 nm was recorded using a 96-well plate reader (Synergy Mx BioTek (Vermont, US) multimode microplate reader Gen5 Software). The 570 nm absorbance reading is directly proportional to the number of cells normally used in proliferation assays. The values are presented as percentages relative to the untreated controls (defined as 100% survival).
In preparation for invasion and replication assays, RH wild type tachyzoites were grown for 72 hrs. Intracellular parasites were collected as described above. Monolayers of confluent HFF cells were grown in 8-well chamber slides. Three independent experiments were conducted with triplicates, and at least 100 cells were counted. The number of infected cells and the number of tachyzoites per vacuole was determined per each high power field by light microscopy.
The effect of auranofin on tachyzoites of RH wild type T. gondii was assessed by ATP assays. Extracellular parasites (0.25×106 per experimental group) collected as previously described, were kept in suspension with complete medium (DMEM +2%FBS) with or without auranofin (0.4 µM) for 0, 2, 4, 6, 8 hrs and 18 hrs (overnight). After incubation, experimental groups were sonicated, aliquots were spun down at 4000 rpm for 5 min, and supernatants extracted for ATP assays. CellTiter-Glo Luminescent Cell Viability Assay was performed according to manufacturer's instructions (Promega). Fifty microliters of supernatant from each experimental group were aliquoted into wells in a 96 opaque-wells plate (Nunc) in triplicate. Subsequently, an equal volume of CellTiter-Glo Reagent was added to each well. Stabilization of the luminescence was accomplished by incubating the plate for 10 mins at room temperature. Readings were performed with a GloMax Luminometer (Promega). The readings are presented as relative luminescence units (RLU).
We have previously standardized the chicken embryo model [11]. Briefly, twelve day old pathogen-free fertilized chicken eggs (McIntyre Farms, Hemet, CA) were incubated at 37°C in a humid incubator. At 14-days old, a small window was cut through the shell with a hand drill directly over the blood vessel in each egg, and the vein was visualized with 1 drop of sterile mineral oil on the exposed membrane. Tachyzoites (1×104) in Dulbecco's modified Eagle's medium with or without auranofin (1 mg/kg of predicted weight for age) were injected directly into the chorioallantoic vein [11] with a 28-gauge needle without pre-incubation. The windows were sealed with tape, and the embryos were incubated in a 37°C incubator. The eggs were candled once a day daily to assess viability. Livers and brains were harvested from the embryos either at the time of their death or 5 (or 8) days post-infection, whichever occurred first. One half of each collected organ was fixed in 4% paraformaldehyde for histopathology (hematoxylin and eosin staining and for immunohistochemistry staining for T. gondii with anti-T. gondii HRP antibody). The second half of each organ was frozen at −70°C for subsequent quantitative PCR analysis.
To quantify tachyzoites in vivo, a standard curve was constructed by adding 107 tachyzoites to brain or liver samples (100 mg) from 19-day-old chicken embryos and homogenizing the preparation with a cordless homogenizer (VWR) in cell lysis solution. Total genomic DNA was extracted from 25 mg tissue aliquots with the Qiagen DNeasy Blood & Tissue Kit per manufacturer's instructions (Qiagen, Alameda, CA). DNA was eluted in 200 µl of DNA elution buffer, and then serially diluted to create a standard curve [11]. Tissue from experimental embryos was similarly harvested and genomic DNA extracted as above. Using 2 µl aliquots of eluted genomic DNA as template, quantitative PCR amplification was performed to determine the relative amount of T. gondii surface antigen (SAG1) gene, a constitutively produced gene. Quantitative PCR was performed in duplicate using primers 5′-GTC ATT GTA GTG GGT CCT TCC-3′ and 5′-GCC TCA TCG GTC GTC AAT AA-3′ and PrimeTime probe 5′-TCC TAC GGT GCA AAC AGC ACT CTT-3′ (IDT), and the cycling conditions were 95°C for 10 min, followed by 40 cycles of 95°C for 15 s, and 60°C for 1 min. The relative amount of product generated was measured by determining the threshold cycle when the level of specific PCR product as measured by probe fluorescence that exponentially increased and crossed the threshold of a passive reference dye (ROX) in each sample. The standard curves (with known numbers of tachyzoites added to uninfected liver or brain) were used to extrapolate the numbers of tachyzoites present in unknown samples. Results are presented as relative log10 of relative numbers of tachyzoites per organ.
Results were analyzed using GraphPad Prism software 6.0. All the in vitro experiments and the qPCR quantification results were analyzed with two-tailed, non paired, non-parametric tests to determine statistically significant differences (p<0.05; CI 95%) between control and treatment groups. A Kaplan Meier survival curve was calculated comparing control vs. chicken embryos treated with a single dose of auranofin.
Response data measurements were fit to a sigmoid Emax model using the computer program NONMEM ver 7.2 (ICON, Dublin, Ireland). A naïve-pooled approach was employed incorporating all individual experiment results in the analysis.
For IC50 determination experiments, 96-well plates (clear bottoms) with confluent monolayers of HFF cells were infected with 0.5×103 YFP-RH tachyzoites for 4 hrs. At the end of infection period, extracellular parasites were removed and complete medium (DMEM +2% FBS without phenol) was added back with twofold serial dilutions of auranofin yielding a concentration range of 0.15–4.8 µM. Fluorescence measurements at day 5 post infection showed that auranofin inhibited growth by 50% at a concentration of 0.28 µM (IC50) (Figure 1A, Hill coefficient = 1.94; maximum response or Emax: 82%). In comparison, all different combinations of pyrimethamine and sulfadiazine generated a maximum response of 80% (Emax: 80%; Figure 1C).
For host cell toxicity assays, 96-well plates with confluent monolayers of HFFs host cells were treated with twofold serial dilutions of auranofin yielding a concentration range of 0.3–19.2 µM. Triplicates per experimental group were read at 120 hrs (Day 5). By measuring absorbance at day 5 post-infection, auranofin caused cell cytotoxicity in 50% of the cells at a concentration of 8.21 µM (TD50) (Figure 1B, Hill coefficient 3.89).
Fourteen day old chicken embryos were injected through the chorioallantoic vein with 1×104 tachyzoites of the virulent RH strain. The treatment group received auranofin at the time of inoculation at a dose of 1 mg/kg (estimated body weight). While all control embryos died by day 4, auranofin-treated chicken embryos were 100% protected against death by day 5 (p = 0.0002) (Figure 3A) and had a significantly reduced parasite load as determined by histopathology and by the number of parasites quantified by real-time PCR (expressed as a log10) from their brains (5.27 vs 2.98; p = 0.002) and livers (6.705 vs 3.11; p = 0.0003) (Figure 3B) and histopathology and immunohistochemistry (Figure 4). Of note, the amount of tissue decomposition found in the control chicken embryos suggested that they died one day before their documented death (on day 3).
We demonstrate for the first time, that auranofin has significant activity against T. gondii. In vitro, auranofin reduced parasite replication, while in vivo, it reduced the parasite load and most remarkably, only one dose of auranofin prevented death in a model of acute toxoplasmosis. Our study of the anti- T. gondii effects of auranofin in vitro showed that it affects parasite viability, reducing its replication ability without affecting its capacity to invade the host cell in the absence of host cell toxicity. These results are compelling since auranofin is a FDA approved drug with a known safety profile, which can expedite its use in clinical trials.
Auranofin is active against T. gondii similarly to the activity described for other protozoans of great public health importance such as Plasmodium falciparum [5] and Leishmania infantum [6]. In our study, our Emax modeling of the treatment response to auranofin demonstrated a Hill coefficient greater than 1, suggesting positive co-operativity or independent binding of auranofin to its target. Additionally, auranofin's maximum inhibition of T. gondii growth was 82%, which is equivalent to the effect observed with current standard of therapy for toxoplasmosis (all sulfadiazine and pyrimethamine combinations, across the board, generated a calculated maximum effect of 80%) (Figure 1C).
Auranofin had a TD50 for HFF host cells that was 29 fold higher than the IC50 of 0.28 µM suggesting a high therapeutic index. This therapeutic index supports its safety as a potential alternative treatment for acute T. gondii infection.
The independent IC50 of sulfadiazine (26.05 µM) and pyrimethamine IC50 (0.402 µM), as demonstrated by Meneceur et al [12] are higher than that of auranofin (0.28 µM). However, the combination of lower doses of pyrimethamine and sulfadiazine demonstrated a maximum inhibitory effect over the growth of T. gondii of 80% consistently throughout multiple combinations (Figure 1C). Auranofin's maximum inhibitory effect was equivalent to that of sulfadiazine-pyrimethamine (Figure 1A and 1C). This suggests that auranofin could be an alternative candidate for the treatment of acute toxoplasmosis, although further studies are needed before consideration of clinical trials.
Auranofin's anti-toxoplasmic mechanism of action is not known. However, from our in vitro studies we can surmise that auranofin affects replication while it does not exert any effect during the T. gondii dynamic invasion process. We challenged HFFs with T. gondii tachyzoites for 5, 15 and 30 min and found no differences in invasion (data not shown). Even when the invasion time was prolonged to 1 hr, we did not detect any difference in the rate of invasion whether cells were treated or not with auranofin. Contrary to these findings, we observed statistically significant differences in the percentage of infected cells after overnight incubation post-infection in the presence of auranofin as compared to controls. This latter observation strongly suggests that auranofin affects replication of the parasite by inhibiting its growth.
On the other hand, the molecular target of auranofin, as an antiparasitic agent, is strongly suggested by the current literature. Angelucci et al [7], demonstrated that auranofin inhibited Schistosoma mansoni glutathione-thioredoxin reductase, which the parasite solely relies on for antioxidant protection. Similarly, we recently reported that Entamoeba histolytica was rendered vulnerable to auranofin's antiparasitic effect because it inhibits its sole thioredoxin reductase [8]. As an intracellular parasite, T. gondii needs to circumvent host cell-mediated oxidant attacks during its invasion and replication; therefore, it is conceivable that T. gondii thioredoxin reductase might be the target for auranofin. However, T. gondii possesses multiple anti-oxidant enzymes that might be directly or indirectly affected by auranofin: thioredoxin reductase, glutathione reductase and thioredoxin-dependent peroxidases [13]. Other targets within the dynamic parasitophorous vacuole (which results from the direct interaction between the parasite and the host cell) are also unknown. We performed assays to differentiate the load of reactive oxygen species (ROS) with dichlorodihydrofluorescein between auranofin-treated and control cells. Although we observed differences between control and auranofin-treated infected cells per fluorescence microscopy, we failed to demonstrate quantifiable differences (data not shown). Further studies are underway to determine the exact molecular target and mechanism of action involved in auranofin's anti-Toxoplasma activity.
The most striking results came from our in vivo chicken embryo model of acute toxoplasmosis. All chicken embryos treated with a single dose of auranofin survived to the end of the experiment (after 5 days post infection), while all the control subjects succumbed to overwhelming infection no later than day 3 post infection (Figure 3A). Similar effects were observed if the embryos were injected on day 12 (instead of day 14) and were allowed to incubate for 8 days post infection (data included in the survival curve). Hence, a single dose of auranofin provided 100% protection from death in all acutely infected embryos, while it reduced the parasite load in organs such as the brain and the liver with almost no inflammatory reaction associated with the lower parasite load (see hematoxylin & eosin histology in Figure 4). Although, parasites were not completely absent in the auranofin-treated group, the observed 100% in vivo survival was achieved with only one dose of auranofin. We cannot demonstrate the viability of the parasites detected in the treatment group since we used DNA detection by qPCR. Further studies with mouse animal models with a standard daily treatment regimen are part of our immediate future studies.
One of the limitations of this chicken embryo model is its short course. The chicken embryo is not allowed to hatch, hence we are not able to prolong incubation beyond 21 days (5–8 days post infection).Chicken embryos inoculations with T. gondii tachyzoites at stages earlier than 12 days old are technically challenging given the fragility of their blood vessels, which is also the reason why repeated doses of auranofin are not possible.
In contrast, the standard mouse animal model of acute toxoplasmosis requires at least 10 days of daily therapy and subsequent post-treatment follow up in order to determine the efficacy of the study drugs to eradicate parasite load and ensure survival of the mice. We are planning further experiments in this standard model in the future.
Given its effect on both the parasite and the host, auranofin stands out as a unique anti-parasitic agent: it can protect sanctuary organs such as the brain, where the host‘s own protective inflammatory responses might cause further organ damage. Additional pharmacokinetic studies for auranofin in the CNS are necessary in order to establish its bioavailability in the setting of an abnormally permeable blood brain barrier during a CNS infection. This is particularly important, since most pharmacokinetic studies on auranofin were performed in the early ‘80 s [14]–[16] in uninfected animals with normal blood brain barriers.
In summary, these results reveal significant in vitro and in vivo activity of auranofin against T. gondii, suggesting that it may be an effective alternative treatment for acute toxoplasmosis in the future.
|
10.1371/journal.ppat.0030179 | Receptor-Binding and Oncogenic Properties of Polyoma Viruses Isolated from Feral Mice | Laboratory strains of the mouse polyoma virus differ markedly in their abilities to replicate and induce tumors in newborn mice. Major determinants of pathogenicity lie in the sialic binding pocket of the major capsid protein Vp1 and dictate receptor-binding properties of the virus. Substitutions at two sites in Vp1 define three prototype strains, which vary greatly in pathogenicity. These strains replicate in a limited fashion and induce few or no tumors, cause a disseminated infection leading to the development of multiple solid tumors, or replicate and spread acutely causing early death. This investigation was undertaken to determine the Vp1 type(s) of new virus isolates from naturally infected mice. Compared with laboratory strains, truly wild-type viruses are constrained with respect to their selectivity and avidity of binding to cell receptors. Fifteen of 15 new isolates carried the Vp1 type identical to that of highly tumorigenic laboratory strains. Upon injection into newborn laboratory mice, the new isolates induced a broad spectrum of tumors, including ones of epithelial as well as mesenchymal origin. Though invariant in their Vp1 coding sequences, these isolates showed considerable variation in their regulatory sequences. The common Vp1 type has two essential features: 1) failure to recognize “pseudoreceptors” with branched chain sialic acids binding to which would attenuate virus spread, and 2) maintenance of a hydrophobic contact with true receptors bearing a single sialic acid, which retards virus spread and avoids acute and potentially lethal infection of the host. Conservation of these receptor-binding properties under natural selection preserves the oncogenic potential of the virus. These findings emphasize the importance of immune protection of neonates under conditions of natural transmission.
| Strains of the mouse polyoma virus adapted to growth in cell culture vary greatly in their abilities to cause disease. Pathogenicities of these laboratory strains range from “attenuated” to “highly virulent” when tested in animals. The biological differences are based in large part on variations in the outer capsid protein, which dictate the manner in which the virus recognizes and binds to cell receptors. In contrast, strains of virus newly isolated from wild mice are uniform in their receptor-binding properties. Naturally occurring strains avoid binding to pseudoreceptors, which would severely limit their ability to spread. At the same time, their avidity of binding to true receptors is sufficiently strong to avoid rapid dissociation and potentially lethal spread. They are therefore neither attenuated nor virulent. The new isolates do, however, retain the ability to induce a broad spectrum of tumors in the laboratory. These findings emphasize the importance of neonatal and maternal immune responses in allowing a potentially highly oncogenic virus to disseminate without causing disease.
| Interest in the murine polyoma virus (Py) grew out of its discovery as a tumor-inducing agent in its natural host under experimental conditions [1]. The original virus isolates were obtained from tissues of inbred laboratory mice [2]. Plaque purification and molecular cloning have been used in different laboratories to establish independent “wild-type” virus strains, which have been propagated in cell culture over many years. Studies of these strains have provided a detailed understanding of the replicative and transforming functions of the virus [3], as well as their tumorigenic properties [4]. Laboratory strains have also been highly purified and used for structural studies of the virus and its interaction with receptors [5].
Cell transformation based on focus formation or growth in soft agar has generally been accepted as the in vitro counterpart of tumor induction in the animal. However, assays in cell culture do not give a complete account of functions required by the virus to replicate or induce tumors in mice. Py mutants altered in tumor (T) antigen functions essential for transformation may retain the ability to induce tumors [6,7]. Moreover, different “wild-type” virus strains commonly used in the laboratory, while equally efficient in transforming cells in vitro, may differ greatly in their abilities to replicate and induce tumors in the intact host [8]. These differences are related not to T antigen functions but rather to receptor-binding properties of the major capsid protein Vp1.
Structural, genetic, and biological studies have come together to provide an understanding of how Vp1 polymorphisms affect virus binding to sialic acid moieties as an essential component of cell receptors [4,5,9,10]. Single amino acid substitutions at two positions in Vp1 alter the selectivity and avidity of virus binding to sialic acid [5,11,12]. These polymorphisms have little or no effect on the behavior of the virus in cell culture [13,14] but have profound effects in the animal [10,14,15].
Despite its pathogenic potential in the laboratory, Py establishes a silent infection in nature. Virus shed from persistently infected mothers is transmitted to neonates which maintain the virus for life without developing tumors [2,16,17]. In contrast, when inoculated into newborn mice from a virus-free colony, as little as a few plaque-forming units of an appropriate virus strain is sufficient to induce the development of multiple tumors ([8] and unpublished data). The absence of pathogenic effects in naturally infected mice suggests that virus strains found in nature might correspond to non– (or low) tumor-inducing strains. It is also possible that more aggressively replicating strains with potential to induce tumors or to kill the host outright may be found in nature, their pathogenic effects offset by protective antibody delivered through the milk and by rapidly maturing immune responses of the pups. Differences in host genetic backgrounds between inbred laboratory mice and feral mice may also contribute to the different outcomes.
Because laboratory strains of virus have been propagated and manipulated in culture over many years, it is unclear how their properties relate to those of naturally occurring virus strains. This investigation was undertaken to determine the Vp1 types and pathogenic properties of truly wild-type strains of Py. To this end, we trapped feral mice at multiple locations and screened them serologically for Py. Independent isolates of the virus were established from seropositive animals. Vp1 coding sequences and biological properties of these isolates have been compared with prototype laboratory strains.
Properties of three prototype “wild-type” laboratory strains of Py are summarized in Table 1. The abilities of these strains and of several mutants derived from them to replicate in newborn mice were previously compared by whole mouse section hybridization (Figure 1). The small plaque strain RA shows limited replication and induces few or no tumors. Tumors that do arise are strictly of mesenchymal origin and develop after a long latency approaching one year [8]. The large plaque strain PTA gives rise to a disseminated infection followed by development of multiple tumors that are of epithelial as well as mesenchymal origin. Tumors may be detected grossly as early as 5–6 wk and grow rapidly, leading to a moribund condition usually within 2–3 mo in essentially 100% of the animals [8]. LID, a large plaque strain derived from PTA, is virulent [18], killing neonates within a few weeks due to rapid dissemination and destruction of vital tissues [9,19]. The LID strain has acquired virulence without losing tumorigenicity, because mice inoculated with low doses may survive and develop tumors [9].
The different levels of pathogenicity of these three strains are governed to a large extent by their Vp1 type [9,14]. Amino acid substitutions at positions 91 and 296 are critical in determining receptor recognition properties of the virus [5,9]. The common receptor(s) for all Py strains contain a terminal sialic acid (NeuNAc) in an α-2,3 linkage to galactose. Certain gangliosides carrying this linkage have been shown to be functional receptors, binding and internalizing the virus and conveying it to the endoplasmic reticulum for disassembly and translocation [20–23]. The sialic acid binds in a shallow groove of complementary shape on the surface of the virus with no apparent conformational changes induced in Vp1 upon binding. Co-crystal structures of whole virus or recombinant Vp1 pentamers together with sialyloligosaccharides have been determined at high resolution. These reveal electrostatic, hydrogen bond, and van der Waals interactions between side chains of Vp1 and the oligosaccharide receptor. Substitutions of glycine for glutamic acid at position 91 and alanine for valine at position 296 have profound effects, the former in allowing the virus to discriminate between true receptors and “pseudoreceptors” and the latter in determining binding affinity and ease of dissociation [5,11,12].
The effects of these substitutions can be rationalized as follows. The carbohydrate moiety of the receptor is commonly found in a trisaccharide [NeuNAc-(α2,3)-Gal-(β1,3)-GalNAc] in which the GalNAc in the third position may be linked via an α2,6 linkage to another sialic acid, forming a branched di-sialic acid chain [NeuNAc-(α2,3)-Gal-(β1,3)-(NeuNAc-(α2,6)-GalNAc]. Small plaque strains such as RA bind to both the branched and straight chain oligosaccharides, while large plaque strains bind only to the straight chain [24,25]. The crystal structures indicate that electrostatic repulsion and steric interference between the glutamic acid at position 91 of Vp1 and an α2,6-linked sialic acid would prevent virus binding to the branched chain oligosaccharide. For this reason, large plaque strains such as PTA and LID with glutamic acid at 91 are constrained to bind only the straight chain receptor. Substitution of glycine for glutamic acid at this position opens up another pocket able to accommodate the second (α2,6-linked) NeuNAc. RA and other small plaque strains are thus able to bind both the branched and the straight chain oligosaccharides. Because small plaque strains have broader receptor binding capability and at the same time are inhibited in replication in the mouse (compare RA and PTA in Figure 1), the branched chain oligosaccharide is considered to be a pseudoreceptor, i.e., one that binds virus without leading to infection and thus retarding virus spread [5,10]. The polymorphism at position 91 also dictates plaque size, presumably by affecting release from cell debri and relative ease of spread in cell monolayers. It also affects interaction with erythrocyte receptors, altering the pH- and temperature-dependence of hemagglutination [13].
The valine to alanine substitution at position 296 introduces a subtle alteration in a hydrophobic surface forming part of the sialic acid binding groove. Valine at this position (PTA) makes a van der Waals contact with the sialic acid ring, while the shorter chain alanine (LID) does not. The alanine substitution thus weakens the hydrophobic interaction with the receptor [9]. Loss of this interaction is expected to facilitate dissociation of the virus from cell debri following a lytic infection. The ability of LID to spread rapidly and destroy vital tissues underlies its virulent behavior. Site-directed mutagenesis has confirmed the importance of these substitutions (Figure 1 and [10]). Thus, PTA-V296A spreads more rapidly and extensively than PTA and to a degree similar to LID. It also induces early death in neonatal mice [10]. Studies of site-directed mutants have also shown that recognition of branched chain pseudoreceptors exerts a dominant effect biologically. Introduction of 91G into virulent or tumorigenic strains results in attenuation of spread. This is seen by comparing PTA-E91G, PTAV296A-E91G, and LID-E91G with their respective parental strains (Figure 1). PTA-E91G has been shown to be less tumorigenic than PTA, consistent with its reduced ability to spread [10].
To determine the Vp1 type(s) of Py under natural conditions, feral mice were trapped and screened serologically for ones that carried the virus. A total of 71 mice were trapped alive from 12 locations in and around Boston and as far away as Cuttyhunk Island, MA, and New York City. Mice were humanely killed, sera collected, and kidneys frozen for attempts at virus isolation. Screening for Py was carried out on the serum from each mouse using a hemagglutination inhibition (HA-I) assay. Results are shown in Table 2. Using a cutoff of 160 HA-I for non-specific inhibition, mice fell into two groups based on their HA-I titers and the location where they were trapped. Mice from four locations were uniformly negative, with HA-I titers ranging from 20 to 160. Mice trapped at the other eight locations were all positive, with HA-I titers ranging from 640 to 10,240. Eight locations yielded multiple mice. Mice from a given location were either all positive or all negative indicating that whenever present the virus spreads efficiently within each breeding population. No tumors were noted in either the HA-I–positive or –negative mice, with one exception, a fibroma, which upon histological examination appeared not to be typical of Py tumors. These findings of efficient spread with no discernible pathology are consistent with earlier reports on the epidemiology of natural infections by Py [16,26,27].
Py was isolated and amplified from kidney homogenates of HA-I–positive mice. Primary kidney epithelial cells prepared from virus-free baby mice were used for amplification. Cytopathic effects typical of Py developed over a period of 7–12 d in every inoculated culture. First-round lysates gave plaque titers in the range of 106–107 pfu/ml. These titers were lower by 1–2 orders of magnitude compared with those of laboratory-adapted strains. Several isolates were tested and shown to be neutralized by anti-Vp1 antibody or by sera from laboratory mice infected by Py (unpublished data). The feral mouse virus isolates were further amplified and analyzed by DNA sequencing of selected regions of the viral genome. Six isolates from mice trapped at three locations were tested for their abilities to induce tumors in newborn mice (see further below).
One virus isolate from each of the eight locations was chosen initially for sequence analysis (Figure 2). To determine the Vp1 types in these isolates, virus was concentrated from first or second passage lysates and analyzed by PCR and sequencing. A 1083-bp fragment covering 94% of Vp1 coding sequences was amplified and purified from each of the eight viruses. Internal primers were used to determine sequences flanking the 91 and 296 coding regions. All eight isolates encoded glutamic acid at position 91 and valine at 296. An additional seven viruses were analyzed from different mice taken from each of the sites where multiple animals were trapped. These seven isolates also encoded glutamic acid at 91 and valine at 296 in their Vp1. Thus, 15 of 15 isolates from mice trapped at eight different locations carried the polymorphisms found in standard large plaque tumorigenic laboratory strains represented by PTA as the prototype. The entire Vp1 coding regions of the original isolates were sequenced and confirmed to be identical to that of PTA.
Regulatory sequences flanking the replication origins of the prototype wild-type strains are variable compared with their highly conserved coding regions [28]. It was therefore of interest to determine the degree of variation in regulatory sequences of the new isolates. It should be noted that all three prototype strains were being propagated in our laboratory during the course of this investigation with the RA strain being most widely used. This raises the possibility of contamination in attempts to amplify and sequence the new virus isolates. To address this concern and to learn more about the new isolates, crude lysates were concentrated and used for analysis. The entire non-coding regions between the ATG of the T antigens on the early side of the replication origin and the ATG of Vp2 on the late side (ca. 450 bp) were amplified and sequenced in one isolate from each of the eight sites. Results are presented schematically in Figure 2 and the sequences are given in Figure S1.
Insertions, deletions, and single base changes distinguish the different isolates. The S and Z2 isolates were identical to each other but different from all three prototype strains. Differences in the new isolates were found dispersed throughout the non-coding regions in both the “A” and “B” enhancers and on the early side of the origin. PTA carries two copies of a 40-bp sequence around the Bgl1 site, while the related LID strain carries a single copy [9]. Two of the isolates (C and H) carried the duplication as found in PTA, and the other five (MB, Z1, Z2, D, and S) carried a single copy. Origin regions of several additional isolates derived from mice trapped at the same location were sequenced. Sequences for independent isolates from each of three sites examined were found to be the same as shown in Figure 2. Of the 15 isolates examined, only 1 (C) was found to be identical to PTA throughout the origin region. None matched either RA or LID. These results rule against contamination and support the conclusion that the isolates represent new strains.
PTA, RA, and LID have been compared for their abilities to induce tumors following inoculation into newborn mice of the highly susceptible C3H/BiDa strain [8,9]. PTA induces multiple tumors in 100% of these animals. Tumors are of both epithelial and mesenchymal types. RA induces only mesenchymal tumors and at a much lower frequency and with longer average latency. The absence of epithelial tumors with RA is an important feature distinguishing RA from PTA [8]. LID at doses of approximately 105 pfu/animal or higher kills C3H/BiDa within roughly two weeks. At much lower doses, animals survive and develop a tumor profile similar to that induced by PTA.
The finding of a uniform Vp1 type identical to PTA in the feral mouse isolates makes several predictions about the biological behavior of the new strains. First, they should induce tumors at high frequency and these should include epithelial as well as mesenchymal types. Second, because the isolates lack the virulence determinant of LID (296A), they should not cause early deaths (≤ 3 wk). To test these predictions, two isolates from each of three locations were inoculated into newborn C3H/BiDa mice. The results clearly support the predictions based on the importance of Vp1 type. All six isolates induced epithelial as well as mesenchymal tumors (Table 3). There were no early deaths. Latencies of tumor development, measured as the average age to necropsy, varied considerably but were longer than with PTA. The longer latencies are attributable to the lower titers of the new isolates. Latencies are known to increase as the titers decrease [8].
Py establishes a silent persistent infection in natural populations of mice, yet behaves in the laboratory as a powerful oncogenic agent (reviewed in [4]). This discordance could be explained if naturally occurring virus strains were attenuated in some manner compared with highly oncogenic strains used in the laboratory. Variations in the major capsid protein of Py dictate a wide range of biological properties upon inoculation into newborn mice, ranging from non-pathogenic to tumorigenic to virulent. The purpose of the present investigation was to determine if these variations, presently known only in laboratory strains of virus, are also found in nature. Fifteen new isolates of virus were derived from feral mice trapped at eight different locations. Only a single type of Vp1 was found, and it matches that of the large plaque highly oncogenic laboratory strain PTA [8,14]. The finding of PTA-like Vp1 sequences in new isolates is not due to rapid selection during passage in culture, as both LID and RA are routinely and faithfully passaged under the same conditions without change in their Vp1s. Six of six of the new isolates tested proved to be highly tumorigenic with a tissue tropism similar to that of PTA. Natural selection therefore results in virus that retains its pathogenic potential. Natural variants of the virus have adopted a “middle ground” with respect to their common Vp1 type and ability to spread in the natural host. Their oncogenic potential is effectively offset by protective maternal antibody and possibly other host or environmental factors that accompany natural transmission (reviewed in [4]).
Glutamic acid at position 91 and valine at position 296 define critical features of this common Vp1. Glutamic acid–91 is critical for discriminating between true receptors carrying a terminal unbranched sialic acid and pseudoreceptors with branched chain sialic acids [5,11,12]. A glycine at this position allows binding to pseudoreceptors, leading to significant attenuation of virus spread [10]. No naturally occurring strains of Py with glycine-91 have been found. The failure to find such strains implies that binding to pseudoreceptors leads to an inhibition of virus spread to a degree that is incompatible with natural transmission. Therefore, binding to branched chain sialic acids should not be viewed as a natural means for attenuating the pathogenic potential of the virus.
Valine at position 296 establishes a hydrophobic contact with the receptor, decreasing the rate of dissociation and spread of the virus. Substitution of alanine at this position results in loss of the hydrophobic contact and a decreased overall avidity of binding [9]. The magnitude of the change in avidity has not been determined precisely. However, given the engagement of multiple capsomeres (each a pentamer of Vp1) at the cell surface, even a modest decrease in affinity would be expected to confer a significantly lower overall avidity of virus binding to the cell. The alanine substitution clearly results in more rapid and extensive virus spread and results in a lethal infection of newborn mice [9,10]. No variants with this substitution have been found in nature. This suggests the importance of control of the rate of virus dissociation from infected cell debri and subsequent spread within the newly infected host. Given the importance of glutamic acid at 91 in allowing efficient spread without interference from pseudoreceptors, the coupling with valine as opposed to alanine at 296 would appear to be essential because the latter would confer a degree of virulence potentially incompatible with host survival.
Some viruses promote their release and spread through mechanisms that effectively destroy host cell receptors. Prominent examples are the neuraminidase of influenza viruses and the Vpu and nef proteins of HIV and simian immunodeficiency virus which act directly or indirectly to destroy cell receptors. For Py and perhaps other viruses that lack the means to destroy their own receptors, it appears to be critical that they regulate their receptor-binding properties within narrow limits. Affinity of binding must be high enough to promote efficient cell attachment and entry, yet not so high as to inhibit dissociation from cell debris and prevent virus spread. At the same time, receptor affinity must not be so low as to allow rapid dissociation and spread, which could endanger the life of the host. The properties of Py strains RA and LID, lying outside these limits, serve to illustrate the constraints of natural selection.
While Vp1-coding sequences were entirely conserved, regulatory sequences showed considerable variability among the new virus isolates. This variability was evident comparing isolates from different locations. In a limited sampling of multiple virus isolates from the same location, regulatory sequences were conserved, indicating relative stability of virus circulating at a given time within the same host breeding population. This finding is consistent with the epidemiology and natural history of infection in wild mice ([16,26,27] and reviewed in [4]).
The origin and biological significance of the variability in regulatory sequences of the new isolates are not clear. Sequences at the replication origin per se as well as the binding sites for the large T antigen are conserved among the feral mouse isolates and laboratory strains of virus, although the number of large T binding sites varies. Signaling pathways from the middle and small T antigens converge on the enhancer regions at different sites via a number of cellular factors affecting viral DNA replication as well as transcription [29,30]. Different enhancer sequences play roles in the host range properties of the virus in cell culture [30–35] and in the animal [36–38]. Particular features of the tumor profile are controlled by a 40-bp duplication upstream of the early promoter [39]. Studies of the related pneumotropic polyoma virus of mice have provided evidence for incorporation of host sequences into the viral enhancers [40,41]. Further studies of the new Py isolates and their molecularly cloned derivatives will be required to determine whether and to what extent different enhancer sequences may dictate differences in tissue tropism for replication and induction of tumors.
Feral mice (Mus musculus) from 12 locations in and around Boston and as far away as Cuttyhunk Island, MA, and New York City were trapped and recovered alive using Sherman Traps and mouse chow coated with peanut butter as bait. Mice were humanely killed, sera collected for serological testing, and kidneys frozen at −80 °C. One kidney from each animal was sent to the Laboratory of Comparative Medicine at Yale University School of Medicine for testing of common mouse pathogens before attempting virus isolation in our laboratory. Results on all mice were negative for all agents tested, including: hantavirus; mouse parvovirus; lymphocytic choriomeningitis virus; Theiler's mouse encephalomyelitis virus; Sendai virus; minute virus of mice; mouse hepatitis virus; ectromeila; reovirus; and mycoplasma.
Serological testing for Py was performed in our laboratory using a standard HA-I assay in V-shaped 96-well microtiter plates. Serial 2-fold dilutions of mouse sera were incubated with an equal volume of 2–4 HA units of wild-type polyoma (PTA strain) for 30 min at 37 °C, followed by addition of a suspension of guinea pig erythrocytes (2 × 107/ml). Further incubation was at 4 °C for a minimum of 6 h. HA-I titer was scored as the reciprocal of the highest dilution of serum that prevented agglutination.
Mice with sera showing HA-I titers of ≥ 160 were judged to be positive and were considered likely carriers of the virus. Frozen kidneys from these HA-I–positive mice were thawed and homogenized in 2 ml serum-free Dulbecco's Modified Eagle's Medium. Aliquots of the homogenates were used to inoculate primary baby mouse kidney epithelial cells prepared from Py-free baby ICR mice. Virus lysates from these cultures were titered by plaque assay on NIH3T3 cells, concentrated by pelleting, and resuspended for further analysis.
Viral DNA was extracted from crude or concentrated virus. Selected regions of the viral genome were amplified by PCR and sequenced. For Vp1, a 1083-bp region covering most of the Vp1 coding sequence was first amplified [5′ GAATATAGCTGAATACACAG3′ (sense) and 5′ AGGTGCTGGACCTTGTGACAGGG3′ (anti-sense)] and then sequenced using primers 5′ AACAGTGAGCCAGAGCCCACCACC3′ for the 91 region and 5′ GAGGGAGGCCATGGGATAGGG3′ for the 296 region. Additional sequencing was carried out to confirm and extend coverage of the entire Vp1. The origin regions were amplified with primers 5′ GTACCGCTGTATTCCTAG3′ and 5′ CATTCTCAGATTGTATACTTCAG3′ giving a fragment of 1061 bp encompassing all of the non-coding sequences between the ATG of Vp2 on the late side and the ATG for the T antigens on the early side of the origin. Sequencing of this amplicon was carried out with primers 5′ CTCATTACACCCTCCAAAGTC3′ and 5′ CATTCTCAGATTGTATACTTCAG3′. Amplicons were sequenced using cycle sequencing with BigDye v3.1 dideoxy fluorescent terminators and a 3730xL DNA Analyzer from Applied Biosystems.
Tumor induction studies were carried out with six of the 15 virus isolates following previously described procedures [8]. Briefly, newborn C3H/BiDa mice (< 24 h old) were inoculated with virus intraperitoneally at approximately 105 pfu/animal. Six to 12 animals were used for each virus. Animals were followed for tumor development grossly and were necropsied when moribund. Tumors were confirmed by histological examination and grouped as epithelial or mesenchymal. The former includes tumors arising from mammary glands, thymus, skin, and salivary glands and the latter includes fibrosarcoma, osteosarcoma, and renal sarcoma. Results with the six new isolates are compared with previously published results for PTA and RA [8].
GenBank (http://www.ncbi.nlm.nih.gov/Genbank/index.html) accession numbers for sequences of laboratory strains of polyoma virus are as follows: A2 strain (JO2288), PTA strain (U27812), LID strain (U27813), and RA strain Vp1 (M34958). Additional sequences for non-coding and coding regions of RA may be found in [14] and [28]. |
10.1371/journal.ppat.1007828 | Genetic dissection of a Leishmania flagellar proteome demonstrates requirement for directional motility in sand fly infections | The protozoan parasite Leishmania possesses a single flagellum, which is remodelled during the parasite’s life cycle from a long motile flagellum in promastigote forms in the sand fly to a short immotile flagellum in amastigotes residing in mammalian phagocytes. This study examined the protein composition and in vivo function of the promastigote flagellum. Protein mass spectrometry and label free protein enrichment testing of isolated flagella and deflagellated cell bodies defined a flagellar proteome for L. mexicana promastigote forms (available via ProteomeXchange with identifier PXD011057). This information was used to generate a CRISPR-Cas9 knockout library of 100 mutants to screen for flagellar defects. This first large-scale knockout screen in a Leishmania sp. identified 56 mutants with altered swimming speed (52 reduced and 4 increased) and defined distinct mutant categories (faster swimmers, slower swimmers, slow uncoordinated swimmers and paralysed cells, including aflagellate promastigotes and cells with curled flagella and disruptions of the paraflagellar rod). Each mutant was tagged with a unique 17-nt barcode, providing a simple barcode sequencing (bar-seq) method for measuring the relative fitness of L. mexicana mutants in vivo. In mixed infections of the permissive sand fly vector Lutzomyia longipalpis, paralysed promastigotes and uncoordinated swimmers were severely diminished in the fly after defecation of the bloodmeal. Subsequent examination of flies infected with a single paralysed mutant lacking the central pair protein PF16 or an uncoordinated swimmer lacking the axonemal protein MBO2 showed that these promastigotes did not reach anterior regions of the fly alimentary tract. These data show that L. mexicana need directional motility for successful colonisation of sand flies.
| Leishmania are protozoan parasites, transmitted between mammals by the bite of phlebotomine sand flies. Promastigote forms in the sand fly have a long flagellum, which is motile and used for anchoring the parasites to prevent clearance with the digested blood meal remnants. To dissect flagellar functions and their importance in life cycle progression, we generated here a comprehensive list of >300 flagellar proteins and produced a CRISPR-Cas9 gene knockout library of 100 mutant Leishmania. We studied their behaviour in vitro before examining their fate in the sand fly Lutzomyia longipalpis. Measuring mutant swimming speeds showed that about half behaved differently compared to the wild type: a few swam faster, many slower and some were completely paralysed. We also found a group of uncoordinated swimmers. To test whether flagellar motility is required for parasite migration from the fly midgut to the foregut from where they reach the next host, we infected sand flies with a mixed mutant population. Each mutant carried a unique tag and tracking these tags up to nine days after infection showed that paralysed and uncoordinated Leishmania were rapidly lost from flies. These data indicate that directional swimming is important for successful colonisation of sand flies.
| Eukaryotic flagella / cilia are complex multifunctional organelles conserved from protists to humans [1]. Protists use flagella for swimming, feeding, cell-to-cell communication, adherence to substrates and morphogenesis [2]. Single-celled organisms, most prominently among them the green algae Chlamydomonas reinhardtii, have served as important model organisms to study molecular mechanisms of ciliogenesis and ciliary function [3], spurred on by the recognition that ciliary defects cause human genetic disorders collectively termed “ciliopathies” [4]. The eukaryotic flagellum is a complex, highly structured organelle and dissection of the molecular mechanisms underpinning its diverse functions requires detailed knowledge of its component parts. Proteomic studies of isolated flagella or axonemes from diverse species typically identified at least 300 distinct proteins [5–9] and phylogenetic profiling identified a set of 274 evolutionarily conserved ciliary genes [10]. All of these datasets comprise many “hypothetical” proteins still awaiting functional characterisation in addition to well-characterised core components of the microtubule axoneme, associated motor proteins and regulatory complexes.
Insights into conserved ciliary biology have helped elucidation of flagellar function in eukaryotic microbes, with a particular focus on human pathogens [11,12]. Among these, flagella have been most extensively studied in the causative agent of African trypanosomiasis, Trypanosoma brucei [13], which uses flagellar motility for locomotion and immune evasion [14] and exhibits close spatio-temporal coordination between flagellum assembly and cell morphogenesis during division [15]. The T. brucei bloodstream form is particularly sensitive to the loss of flagellar function [6,16], highlighting a potential Achilles’ heel that might be exploitable for new anti-parasitic treatments.
The Leishmania flagellum is also a multi-functional organelle, which undergoes striking structural changes during the parasite’s life cycle [17–19]. Amastigote forms proliferating in mammalian macrophages possess a short sensory-type 9+0 microtubule axoneme, which is remodelled to a canonical long motile 9+2 axoneme during differentiation to promastigote forms, which live in blood-feeding phlebotomine sand flies (Diptera: Psychodidae). In the fly, nectomonad promastigote forms attach via their flagella to the microvilli of the posterior midgut [20] to protect the parasites from being cleared during defecation of remnants of the blood meal. In the oesophageal valve, broad haptomonad forms attach to the cuticular lining via their flagellar tips, forming hemidesmosomes [20]. These life cycle descriptions (S1 Fig) [21,22] imply that periods of attachment must be followed by migration to more anterior regions of the alimentary tract and the propulsive function of the Leishmania flagellum is presumed to drive this forward migration but this has not been directly tested.
To enable a detailed genetic dissection of flagellar functions and mechanisms in Leishmania, we defined here a flagellar proteome for motile L. mexicana promastigotes. We used new CRISPR-Cas9 genome editing methods [23] to generate a Leishmania knockout library of 100 mutants, over half of which showed altered swimming speed. We also developed a barcode sequencing (bar-seq) protocol to test the fitness of mutants in the permissive sand fly vector Lutzomyia longipalpis. This study identified new genes required for flagellar motility and shows that whilst culture-form promastigotes tolerated loss of the flagellum, paralysed mutants and uncoordinated swimmers failed to colonise sand flies indicating that flagellum movement is required for completion of the parasite’s life cycle. Furthermore this flagellum movement must be able to give effective translocation and if cells cannot undergo directional motility then they cannot be transmitted through the fly.
To enable a systematic genetic dissection of flagellar functions we sought to isolate L. mexicana promastigote flagella comprising the axoneme, extra-axonemal structures and the surrounding membrane for subsequent analysis by protein mass spectrometry (MS). Mechanical shearing in the presence of 75 mM Ca2+ successfully separated cells into flagella (F) and deflagellated cell bodies (CB) (Fig 1A and 1B). Subsequent centrifugation on sucrose gradients allowed isolation of F and CB fractions with little cross-contamination: the CB fraction contained only 2.03% (±0.69%) isolated flagella and the F fractions contained 0.56% (±0.15%) deflagellated cell bodies (S2A Fig). Isolated flagella still retained their membrane: First, examination of F fractions by transmission electron microscopy (TEM) confirmed that most axonemes were bounded by a membrane (S2B–S2E Fig) and second, tracking an abundant promastigote flagellar membrane protein, the small myristoylated protein 1 (SMP-1, [24]) tagged with enhanced green fluorescent protein (eGFP) showed that it remained associated with isolated flagella (Fig 1; 75% of flagella retained SMP-1::eGFP signal, N = 906). Analysis of the SMP-1::eGFP signal also facilitated flagellar length measurements in whole cells, F and CB fractions, which showed that flagella were separated from the cell body near the exit point from the flagellar pocket. The average break point was 2.7 μm distal to the base of the flagellum. The length of the isolated flagella was similar to those on intact cells, indicating that isolated flagella remained in one piece, with little fragmentation (S3 Fig). Two independently prepared sets of F and CB fractions were separated into detergent soluble (S) and insoluble fractions (I), yielding four fractions, FS, FI, CBS and CBI (Fig 1C). All four fractions for both replicates were analysed by liquid chromatography tandem mass spectrometry (MS), which detected a total of 2711 distinct proteins (Fig 1D). Enrichment of detected proteins between biological replicates correlated well (Pearson’s r > 0.72, Spearman’s rs > 0.83, S4 Fig). To discover proteins enriched in each of the four fractions, we used a label-free normalized spectral index quantitation method (SINQ, [25]; S1, S2 and S3 Tables) to generate a SINQ enrichment plot (Fig 2A). The promastigote flagellar proteome, defined as proteins enriched in F vs. CB fractions consisted of 701 unique proteins detected in at least one MS run; 352 of these were enriched in F vs. CB fractions in both MS runs.
To validate the data, we mapped well-characterised flagellar proteins onto the enrichment data plot (Fig 2A). Axonemal, paraflagellar rod (PFR), flagellar tip and flagellar membrane proteins mapped to the FI and FS quadrants. Basal body, FAZ and tripartite attachment complex (TAC) proteins mapped to the CBI and CBS quadrants because F fractions contained exclusively the cell-external portion of the flagellum. Intraflagellar transport (IFT) proteins clustered around the midpoint of the plot, indicating their abundance was similar in the F and CB fractions, which is consistent with their known dynamic association with the flagellar basal body and axoneme.
We also found substantial overlaps between L. mexicana proteins in the FI quadrant and proteins detected in previously published flagellar proteomes of L. donovani and T. brucei (S5A–S5D Fig). However, L. mexicana proteins in the FS quadrant showed only a moderate overlap with reported soluble T. brucei flagellar proteins (S5C Fig). We designed a website (www.leishgedit.net/leishgedit_db) for interactive browsing of proteins in the enrichment plots shown in Figs 2 and S5.
Prediction of lipid modification sites identified 15 proteins in the Fs fraction with an MGXXXS/T N-terminal myristoylation site indicating possible association with the flagellar membrane. Proteins with predicted trans-membrane domains (TMD; annotation from TritrypDB.org) were predominantly detected in the detergent soluble fractions (S5F Fig). Overall, TMD proteins were however underrepresented in the proteome compared to their frequency predicted from the genome (Chi-squared test, p < 0.0001), as were proteins smaller than 10 kDa (S6 Fig). Underrepresentation of small and hydrophobic proteins could be due to technical limitations of the sample preparation and MS protocol [26], for example through loss of proteins at the gel fractionation stage due to their size or their propensity to aggregate, or use of a suboptimal protease for proteolytic cleavage.
Although ribosomal proteins were detected in individual F fractions, the enrichment plot clustered them around the midpoint, with many enriched in the cell body fractions (S5E Fig). Our simple strategy of testing for enrichment thus successfully filtered out likely contaminating proteins from the promastigote flagellar proteome, as recently observed for enrichment of other cytoskeleton structures in T. brucei [27,28].
Interestingly, a comparison of these proteomics data with L. mexicana RNA-seq data from promastigotes and amastigotes [29] showed that proteins enriched in the flagellar fractions were significantly more likely to have higher RNA abundance in promastigotes vs. amastigotes, compared to proteins detected in the cell body fraction (Fig 2B; Chi-squared test, p < 0.0001). This is consistent with the disassembly of the motile axoneme during differentiation from promastigotes to amastigotes [17]. Whilst on a global scale transcript levels correlate poorly with protein abundance in Leishmania spp. [30] these data indicate that modulation of mRNA levels is a key regulatory step in Leishmania flagellar biogenesis and differentiation from a 9+2 to a 9+0 flagellum.
Many of the proteins detected in the F fractions had orthologs in previously defined flagellar and ciliary proteomes yet lacked any functional characterisation. Arguably, endowing cells with motility is the primary function of the promastigote flagellum and we took advantage of our high-throughput CRISPR-Cas9 toolkit [23] to identify proteins required for motility and subsequently study the phenotypes of the mutant Leishmania. In our knockout (KO) library (S4 Table) we included 19 highly conserved axonemal proteins known to be involved in the regulation of flagellar beating, three intraflagellar transport (IFT) proteins, 60 flagellar proteins with transcript enrichment in promastigotes [29] and eight additional soluble and four insoluble flagellar proteins. Twenty of the selected proteins were detected in the promastigote flagellar proteome but have to our knowledge not been linked to flagella before. We also made deletion mutants for two genes implicated in membrane protein trafficking, BBS2 and Kharon1. Finally, deletion mutants for four glycoconjugate synthesis genes encoding phosphomannose isomerase (PMI), phosphomannomutase (PMM) and GDP-mannose pyrophosphorylase (GDP-MP) were produced as control cell lines for sand fly infection experiments. Flagellar localisation of a subset of proteins was independently examined by generating cell lines expressing proteins tagged with a fluorescent protein at the N- and/or C-terminus (S7 Fig). For 35 proteins, both N- or C-terminally tagged fusion proteins were examined and 28 showed consistent localisations. For CFAP44 and CMF10, the C-terminal tag gave a clear flagellar localisation whereas the N-terminal tag resulted in flagellar and cell body signal. For six proteins (LmxM.17.0800, LmxM.29.3360, LmxM.08_29.1000, LmxM.27.0670, PKAC1 and CD047) a clear flagellar localisation was observed with N-terminal tags but the C-terminally tagged proteins were exclusively seen in the cell body. Addition of a fusion protein can in some cases result in protein mis-localisation, and further analysis of these discrepancies may reveal sequence features controlling flagellar targeting in these proteins.
Orthofinder [31] was used to generate genome-wide orthologous protein sequence families using genome sequences of 33 ciliated and 15 non-ciliated species from across eukaryotic life, including L. mexicana and T. brucei (S5 Table). Twenty-two proteins were only found in kinetoplastids (L. mexicana and T. brucei), 30 were conserved specifically in ciliated organisms and 23 widely conserved across eukaryotes whilst the remainder showed no clear pattern. In the following, we refer to genes of unknown function by their GeneID from TriTrypDB.org [32] and where we identified named orthologs we used the corresponding gene names.
The target genes were then deleted as described previously [23]. To facilitate high-throughput generation of knockout (KO) cell lines, PCR reactions and transfections were performed in 96-well plates. Analysis of drug-resistant transfectants by PCR confirmed loss of the target ORF and integration of the drug-resistance gene in 94 of 98 cell lines (S8 Fig). This 96% success rate highlights the power of our gene deletion strategy. The reason for the presence of the target ORF in the remaining four cell lines was not further investigated, but was confirmed by diagnostic PCR of two independently isolated samples of genomic DNA from the relevant mutants.
The flagellar mutants generated in this study, the previously generated paralysed cell line ΔPF16 [23], the parental line L. mex Cas9 T7, and wild type promastigotes were subjected to motility assays using dark field microscopy to track the swimming behaviour of cells and measure swimming speed and directionality as previously described [33]. Parental cells immobilised though formaldehyde fixation were also measured. Wild type L. mexicana promastigotes use a tip-to-base flagellar beat for propulsive motility, interrupted by episodes of base-to-tip ciliary beats [34] and their swimming trajectories follow curving paths, with occasional changes in direction. The majority of cells achieve a large displacement from their starting position over time; this is directional motility of the cell, albeit in a random direction in a homogenous culture environment [33]. More than half of all mutant lines showed a significant deviation from the normal average swimming speed measured for the parental cell line and wild type controls (Fig 3A and 3B): 52 (53.6%) mutants showed a significant reduction in speed and 4 (4.1%) swam faster (Student’s t-test, p<0.005; Fig 3A, S4 Table). We used the ratio of velocity to speed per cell as a measure of swimming path directionality per cell, this is equivalent to the ratio of displacement achieved to the distance travelled to reach that point. Plotting mean swimming speed against mean directionality shows broad groups of mutants (Fig 3B): Those which are paralysed, slower swimmers, slow uncoordinated swimmers, faster swimmers and a single mutant that had faster and more directional swimming (ΔLmxM.36.3620). The mechanistic contribution to swimming behaviour remains to be clarified for many proteins in this set. Loss of flagellar waveform modulators would cause altered motility patterns, and this is exemplified by two mutants in this set: the ΔdDC2 mutant, which lacks the outer dynein arm docking complex protein dDC2 and can perform a ciliary beat but no flagellar beat [35] clusters with the uncoordinated group. By contrast, ΔLC4-like, which lacks a distal regulator of outer dynein arms and spends more time doing a flagellar beat at a higher beat frequency [35], was among the faster swimmers.
The most severe loss of motility was observed in three cell lines that had no visible external flagellum (Fig 4); all of these were deletions of conserved intraflagellar transport (IFT) proteins (ΔIFT122B, ΔIFT139 and ΔIFT88). Ablation of the central pair (CP) protein hydin also resulted in almost complete paralysis, comparable to the deletion of the CP protein PF16 [23].
In a subset of paralysed or slow-swimming uncoordinated mutants (Fig 3C) we noted that the flagella tended to be in a curled rather than straight conformation. Δhydin mutants had the highest proportion of curled-up flagella (62.6%, Fig 4 and S9 Fig) while fewer than 1% of flagella were curled-up in the parental cell line and many other slow swimming mutants (S9 Fig). A high proportion (>10%) of curled-up flagella was also found in four paralysed KO lines (inner dynein arm intermediate chain protein mutant ΔIC140, 57%; ΔPF16, 14%; tether and tether head complex protein mutants ΔCFAP44, 15% and ΔCFAP43, 19%) and three uncoordinated KO lines (ΔMBO2, 26%; nexin-dynein regulatory complex protein mutant ΔDRC4, 13%; ΔLmxM.33.0560, 12%). The curls were observed in aldehyde fixed cells as well as in live cells in culture, indicating they were not an artefact of microscopy sample preparation. This novel phenotype might be caused by disrupted dynein regulation and warrants further investigation.
We generated 13 add-back cell lines to rescue mutant phenotypes by transfecting episomes containing the deleted ORF. Four complemented mutants fully recovered parental swimming speed (complemented ΔIFT88, ΔLmxM.14.1220, ΔLmxM.18.1090 and ΔLmxM.08_29.2440; Fig 3) and complemented ΔCFAP44 and ΔMBO2 lines showed fewer curled flagella (S9 Fig). Complementation of the other 7 slow swimming mutants resulted in a significant increase in swimming speed close to parental levels (Fig 3) and reduction of curling compared to the KO lines (S9 Fig).
Null mutants for the major PFR protein PFR2, lacking the paracrystalline PFR lattice structure, are known to have impaired motility [36]. To compare motility of a ΔPFR2 mutant with other mutants generated in this study, we used CRISPR-Cas9 to delete both allelic copies of the PFR2 array (PFR2A, PFR2B and PFR2C) and confirmed loss of PFR2 expression by western blot (S10 Fig). This ΔPFR2 line had slower and less directional swimming compared to the parental cells, clustering with other slow swimming mutants defined in Fig 3B. To test whether gene deletion in other slow swimming mutants had a major disruptive effect on the PFR, which might explain their motility defect, we expressed PFR2::mNG in KO lines and looked for changes to PFR length or loss of PFR integrity (defined as gaps in the PFR2::mNG signal) (Figs 4 and 5; S8 Table). Three mutants had shorter flagella compared to the parental cell line, but the PFR remained proportional to the overall flagellar length and was uninterrupted (ΔARL-3A, ΔCFAP44, and ΔFLAM2). Six mutants had PFR-specific defects (Fig 5B): a shorter flagellum with a disproportionately shorter PFR (ΔLmxM.27.0860; ΔTTC29; ΔLmxM.14.1220), a normal-length flagellum with a shorter PFR (ΔFM458) or a shorter PFR with gaps (ΔLmxM.21.1110, 25.3% of all flagella; ΔMBO2 only 4.1% of all flagella). Interestingly, these comparatively subtle alterations to PFR length and integrity reduced swimming speed to similar levels as PFR2 deletion (Fig 3C).
Thus, our screen readily identified promastigote mutants with impaired motility and even the most severe phenotype, ablation of flagellar assembly caused by loss of IFT components, was compatible with promastigote survival in vitro, in line with earlier reports [37], [38], [39].
Whilst flagellar motility is generally believed to be required for development in sand flies, enabling Leishmania migration from the midgut to the mouthparts [40–42], this has not been directly tested. To interrogate the phenotypes of larger cohorts of Leishmania mutants in parallel, we developed a multiplexed bar-seq strategy inspired by pioneering phenotyping screens in yeast [43] and the malaria parasite Plasmodium berghei [44,45]. We pooled mutant L. mexicana lines that were each tagged with a unique 17 bp barcode. This enabled us to measure the relative abundance of each line at different time points after sand fly infection (S11 Fig). Seventeen were flagellar mutants described above and five were parental control cell lines tagged with unique barcodes in their small subunit (SSU) ribosomal RNA locus (S11 Fig). The flagellar mutants were chosen to represent different phenotypes which may impact in different ways on their persistence and migration in the fly: aflagellate parasites, parasites with a short flagellum, paralysed parasites with a flagellum of normal length, slow swimming parasites with more (“uncoordinated”) or less severe defects in directionality and parasites lacking proteins implicated in flagellar protein trafficking. We also generated a barcoded ΔLPG1 KO mutant, which is only defective in LPG synthesis [23,46] and three barcoded mutants defective in the pathway leading to mannose activation for synthesis of LPG and other glycoconjugates: KOs of phosphomannose isomerase [47] (ΔPMI), phosphomannomutase [48] (ΔPMM) and GDP-mannose pyrophosphorylase [49] (ΔGDP-MP). These mutants were included as control lines expected to be outcompeted by the parental cell lines based on prior demonstration that loss of the LPG coat is detrimental to parasite development in the fly [50,51].
The barcoded cell lines were pooled in equal proportions and first we determined their relative growth rates in culture. Over the 96h observation period, five cell lines became depleted: ΔIFT88, ΔLPG1, ΔPMI, ΔPMM and ΔGDP-MP (S12 Fig). These showed also the longest doubling times when measured in individual cultures (S12 Fig).
To generate pools to infect L. longipalpis, the cell lines were divided into four sub-pools according to their in vitro growth rates and grown for 48 hours until they reached late log phase and then these were pooled in equal proportions just before the infection. The relative abundance of each line was determined by sequencing DNA isolated from the mixed promastigote pool and from flies at two, six and nine days after infection. The results show progressively diminishing proportions for the control mutants defective in LPG synthesis (ΔLPG1) or a broader range of glycoconjugates including LPG (ΔPMI, ΔPMM and ΔGDP-MP) (Fig 6, S9 Table) indicating that parasites lacking these molecules were at a competitive disadvantage in these infections. This effect was apparent as early as two days after infection, consistent with a protective role for PG-containing glycoconjugates in the digesting bloodmeal [51] and a role for LPG in L. mexicana attachment to L. longipalpis [50].
Paralysed and uncoordinated mutants also became noticeably scarcer as the infection progressed (Fig 6, S9 Table, S13 Fig). The aflagellate ΔIFT88 mutant showed the most severe phenotype and a significant decrease over time was also measured for ΔPF16, ΔCFAP43, ΔCFAP44, ΔIC140, ΔdDC2 and ΔRSP4/6. By contrast, mutants with a mild swimming defect (slower swimmers ΔLmxM.21.1110, ΔFM458 and ΔLmxM.18.1090 and faster ΔLC4-like) (Fig 3D, S13 Fig) remained as abundant as the normal swimmers throughout the infection (Fig 6, S9 Table). The exceptions were the slower swimmers ΔKharon1 (Fig 3D, S13 Fig), which is also defective in the transport of a flagellar glucose transporter [52], and ΔARL-3A, which has a short flagellum (Fig 5). Both of these were rarer in the fly compared to the starting pool.
To gain anatomical resolution and determine whether an immotile mutant fails to migrate to anterior portions of the fly gut, we infected separate batches of L. longipalpis with motile parasite lines and complemented KO lines as controls, with the motile ΔBBS2 mutant, which lacks a component of the BBSome complex [53] which is expected to play a role in flagellar membrane trafficking, and with the paralysed ΔPF16 mutant (Fig 7). The ΔPF16 mutants are among the least motile cells that retain a long flagellum (Fig 5), while having only moderate levels of flagellar curling (S9 Fig). The axonemal defect resulting in paralysis is a well-characterised disruption of the central pair in kinetoplastids (Fig 3B and [23,54,55]) and is similar to the defect of the pf16 Chlamydomonas reinhardtii mutant [56] indicating it is a well-conserved core axoneme component. Two days post blood-meal (PBM), the L. mexicana wild type and L. mex Cas9 T7 [23] control cell lines and the ΔBBS2 mutant developed well, with infection rates above 70%; the ΔPF16 mutant produced the lowest infection rate (below 50%). The introduction of an add-back copy of PF16 into the ΔPF16 line restored infection levels (Fig 7A). In all lines, promastigotes were localized in the abdominal midgut, within the bloodmeal enclosed in the peritrophic matrix (Fig 7B). After defecation (day 6 PBM), all control lines and the ΔBBS2 mutant replicated well and developed late-stage infections with colonisation of the whole mesenteron including the stomodeal valve (Fig 7B) which is a prerequisite for successful transmission. Their infection rates ranged from 56% to 83%. By contrast, ΔPF16 Leishmania failed to develop; the infection rate was less than 2% (a single positive fly out of 62 dissected (Fig 7A), with parasites restricted to the abdominal midgut (Fig 7B)), indicating that ΔPF16 parasites were lost during defecation and were unable to develop late stage infections in L. longipalpis. Since the pooled data (Fig 6) showed that uncoordinated swimmers were also progressively lost during an infection, we tested whether the uncoordinated swimmer ΔMBO2 would also fail to reach the stomodeal valve. Dissection of flies at 2 and 6 days after infection with the ΔMBO2 mutant line or a complemented ΔMBO2 line showed that ΔMBO2 Leishmania failed to thrive. At day 6 the infection rate was 7.5% (4 positive flies out of 53 dissected (Fig 7A), with parasites restricted to the abdominal midgut (Fig 7B)), similar to the ΔPF16 mutants. This defect was rescued by restored expression of MBO2 (Fig 7). Our data provide strong evidence that flagellum-driven directional motility is an essential requirement for successful Leishmania development in sand flies and, by implication, parasite transmission.
This study demonstrates the power of high-throughput CRISPR-Cas9 knockout screens to discover mutant phenotypes in Leishmania. We first defined a flagellar proteome by pursuing a flagellar isolation protocol yielding a defined section of intact flagella and comparing both the flagella and the deflagellated cell body fractions to define a relative enrichment score for each protein. The SINQ method [25,27,28] eliminated from our analysis abundant cell body proteins that were likely cross-contaminants in the flagellar fractions. The flagellar proteins (Fig 2) defined by this method showed similarities in numbers and types of proteins to other analyses of eukaryotic flagella and cilia (S5 Fig, S4 Table, [5]). We then used these high-confidence flagellar proteome data in conjunction with transcriptomics data and prior knowledge of conserved axonemal proteins to demonstrate a role in motility for >50 genes from a set of one hundred. We also show the importance of directional flagellar motility in the colonisation of sand flies. The data from the pooled mutant population show a progressive loss of paralysed or uncoordinated swimmers over nine days from infection. Because these data report total abundance of each genotype in the whole fly without discriminating between regions of the gut, we probed this question further in infections with the ΔPF16 mutant, which is essentially paralysed and incapable of sustained directional motility due to a defined defect in the central pair complex of the axoneme [23]. The results show that ΔPF16 Leishmania were rapidly lost from most of the dissected flies, consistent with the depletion of this mutant from the mixed pool, and additionally shows that none of the few remaining parasites reached anterior parts of the alimentary tract. A similarly severe defect in colonisation was observed in the ΔMBO2 mutant. MBO2 is an evolutionarily conserved axonemal protein [57] and derives its name from Chlamydomonas mutants that move backwards only because the algal flagella remain locked in a flagellar beat and cannot readily switch to a ciliary beat [58]. Whilst the precise function of MBO2 remains unknown, it is likely that the uncoordinated swimming behaviour of Leishmania ΔMBO2 mutants (Fig 3B, S13 Fig) is also the result of defective waveform control.
Taken together, these findings show that parasite motility is required for completion of the Leishmania life cycle, in line with the essential role of motility in other vector-transmitted protists. For example, Rotureau et al., [59] showed that loss of forward motility, caused by ablation of outer dynein arms though KO of DNAI1, rendered T. brucei unable to reach the tsetse fly foregut. It seems likely that loss of motility also contributed to the inability of L. amazonensis to progress beyond the abdominal midgut of L. longipalpis when the parasites overexpressed GTP-locked ADP-ribosylation factor-like protein 3A (Arl-3A) and as a result grew only short flagella [60].
The interesting question remains to what extent flagellar motility and attachment via the flagellum are linked. Observations of attached Leishmania in dissected sand flies show adhesion specifically via the flagellum but the precise molecular interactions between flagellum and the microvillar gut lining remain to be clarified. The dominant cell surface glycoconjugate LPG which covers the entire parasite surface including the flagellum is known to be important in Leishmania attachment to sand fly guts [61] and our results support the view that LPG plays an important role in L. mexicana infection of L. longipalpis [50]. The proportion of ΔLPG1 mutants had decreased by two days after infection and reduced further as infection progressed. The observed loss of fitness of the ΔPMM, ΔGDP-MP and ΔPMI mutants is likely the cumulative effect of the loss of LPG and a broader range of mannose-containing glycoconjugates which were shown to protect Leishmania in the digesting bloodmeal [51]. The consistency of the pooled mutant data with the reported phenotypes of individual glycoconjugate-deficient mutants demonstrates the power of this new rapid method for mutant phenotyping in Leishmania. However, whilst a role for LPG in L. mexicana attachment to the fly is well established, the possible contribution of flagellum-specific surface molecules [62] has not yet been conclusively resolved. Zauli et al., [38] reported isolation of L. braziliensis from a patient’s skin lesion which differentiated to promastigotes with an “atypical” morphology. These cells had a short flagellum barely protruding from the flagellar pocket, with an amorphous tip suggestive of a defect in flagellum elongation. In experimental infections of L. longipalpis, these parasites persisted in the fly following defecation of the blood meal, suggesting that they remained sufficiently anchored without a long flagellum. It would be interesting to follow up the subsequent development of this mutant in the fly. Interestingly, here only 1.6% of flies infected with the paralysed ΔPF16 mutant and 7.5% infected with ΔMBO2 were still positive 6 days post infection, compared to 65% of dissected flies infected with the parental cell line. It is possible that loss of directional motility impedes traversal of the peritrophic matrix and it would be informative to look for differences between mutants in the subsequent colonisation of the microvillar lining.
Several lines of evidence suggest a role for the trypanosomatid flagellum in environmental sensing [42,63–65]. Evidence for specific signal transduction pathways aiding promastigote navigation through the sand fly is however limited. Cyclic nucleotide signal transduction pathways may have important roles in coupling environmental sensing with regulation of flagellar beat patterns [66,67] and have been shown to be involved in the migration of T. brucei in the tsetse fly [68]. In our flagellar proteome we identified several adenylate cyclases (ACs), cAMP-specific phosphodiesterases (PDEs) and PKA subunits and mapped their localisations to distinct flagellar subdomains by protein tagging (S7 Fig). The motility assays showed that deletion of PKA subunits (ΔLmxM.34.4010 (partial KO only) and ΔFM458) reduced swimming speed, whereas deletion of two different PDEs (ΔLmxM.18.1090 and ΔLmxM.08_29.2440) increased it, pointing to an activating role for cAMP in Leishmania motility. Knockout of receptor-type adenylate cyclase a-like protein LmxM.36.3180 had no effect on swimming speed in our motility assay but given the possible redundancy with other flagellar ACs, this preliminary finding should be followed up by examination of other AC mutants individually and in combinations. In our pooled KO screen in sand flies, KOs of PDE LmxM.18.1090 and PKA RSU (FM458) remained as abundant as the controls, indicating that the mild motility phenotypes measured in vitro did not significantly impair colonisation of flies.
Perturbation of the flagellar membrane might be expected to interfere with sensory functions mediated through the flagellum. Ablation of membrane proteins LmxM.17.0870 and LmxM.23.1020 (S7 Fig) did not significantly enhance or reduce the relative abundance of the respective mutants in sand flies over the nine-day observation period. BBS2 is an integral part of the core BBSome complex which is highly conserved across ciliated eukaryotes [69] and functions as a cargo adaptor for ciliary membrane protein trafficking in Chlamydomonas flagella and metazoan cilia [70]. Our pooled mutant data and infections with the BBS2 deletion mutant alone found that loss of this gene had no discernible detrimental effect on survival in sand flies and the parasites’ ability to reach the anterior gut. By contrast, KO of Kharon1, a protein shown to be required for trafficking of the glucose transporter LmGT1, and perhaps other proteins, to the promastigote flagellum [52] led to slightly reduced fitness in the flies from the earliest time point. The ΔArl-3A mutants were also less abundant compared to the controls. This is reminiscent of the previously published abortive phenotype of L. amazonensis overexpressing the constitutively GTP bound LdARL-3A-Q70L [60]. This mutant formed only a short flagellum, similar to the ΔArl-3A mutant generated in the present study (Fig 5). Failed attachment as a result of the shortened flagellum was thought to be a likely cause for the rapid clearance of LdARL-3A-Q70L-expressing parasites but it was noted that an inability to migrate at later stages of development would also lead to the disappearance of the mutants [60]. In our study the phenotype of the ΔArl-3A mutants was however mild compared to the aflagellate (ΔIFT88) or paralysed mutants. Arl-3A acts as guanine nucleotide exchange factor in the transport of lipidated proteins to the flagellar membrane [71] and protein mis-targeting could contribute to the phenotype in addition to flagellar shortening. Further insights into the contribution of flagellar membrane proteins to attachment or directional swimming behaviour may be uncovered by further biochemical studies into flagellar membrane composition and subjecting different mutants (with or without overt motility phenotypes in culture) to chemotaxis assays and fly infections. Flagella isolated by the method used in this study provide suitable starting material for further targeted experiments to identify integral membrane proteins. This could be achieved by using for example carbonate fractionation, as used for the enrichment of membrane proteins in olfactory cilia [72], or combining surface labelling with subsequent affinity purification prior to mass spectrometry [73]
In contrast to the absolute requirement of motility for movement through the sand fly vector, flagellar motility is dispensable for promastigote proliferation in culture. Promastigotes are viable and able to divide even if they fail to assemble a flagellum at all, as demonstrated originally by the deletion of cytoplasmic dynein-2 heavy chain gene LmxDHC2.2 [37] and IFT140 [39] and the phenotypes of knockouts of anterograde and retrograde IFT components in the present study. The ensuing prediction that most gene deletions affecting flagellar function are expected to yield viable promastigotes in the laboratory is borne out by our high success rate of obtaining 96% of attempted knockouts. Thus, in Leishmania, flagellar mutant phenotypes can be observed in replication-competent cells over many cell cycles and our mutant library enables detailed systematic studies of KO phenotypes to probe protein functions in flagellum assembly, motility and signal transduction.
A fruitful area for further studies will be dissection of PFR function and assembly mechanisms. This extra-axonemal structure is required for motility as demonstrated through deletion of the major structural PFR components, PFR1 and PFR2 in Leishmania [36,74] and ablation of PFR2 by RNAi in T. brucei procyclic forms [75] but its precise role remains unclear. The PFR comprises more than 40 proteins, some with structural roles, others with roles in adenine nucleotide homeostasis, cAMP signalling, calcium signalling and many uncharacterised components [76,77] and it may anchor metabolic and regulatory proteins as well as influencing the mechanical properties of the flagellum. Our results showed that fragmentation of the PFR caused by loss of LmxM.21.1110 reduced swimming speed to levels similar to the structurally more severe PFR2 KO. Whether LmxM.21.1110 is required for correct PFR assembly or stabilisation is currently unknown.
Motility mutants analysed in our screen also included deletions of genes with human orthologs linked to ciliopathies (such as hydin) or male infertility (CFAP43 and CFAP44) [78]. Leishmania offers a genetically tractable system to gain further mechanistic insight into their functions. The hydin mutant has been extensively characterised in other species: in mammals, mutations in the hydin gene cause early-onset hydrocephalus [79] and subsequent studies on C. reinhardtii, T. brucei and mice showed that hydin localises to the C2 projection of the central pair complex [80], and that loss of hydin function causes mispositioning and loss of the CP [81] and motility defects [80–82]. The motility phenotype in the L. mexicana Δhydin mutant was consistent with these existing data and we made the new observation that the mutant flagella show extensive curling (Fig 4, S9 Fig). Interestingly, hydin knockdown in C. reinhardtii caused flagella to arrest at the switch point between effective and recovery stroke, leaving cells with one flagellum pointing up and the other down, prompting speculation that this may indicate a role for hydin in signal transmission to dynein arms [80]. Consistent with this hypothesis, cilia of hy3/hy3 mouse mutants frequently stalled at the transition point between the effective and recovery stroke [82]. Curling may represent the failure of flagellum bending to reverse during progression of the normal flagellum waveform down the flagellum, leaving the flagellum locked at one extreme of bending, analogous to the ciliary beat hydin phenotype. In L. mexicana, the Δhydin mutant presented the most severe manifestation of the curling phenotype, which was also observed in a lower proportion of other mutants (S9 Fig). This phenotype may be a consequence of mis-regulated dyneins and the set of mutants exhibiting curling will facilitate further experiments to establish the mechanistic basis for flagellar curling.
Genetic, biochemical and structural studies have provided elegant and detailed models for the mechanisms of flagellar motility [83,84]. Phylogenetic profiling and comparative proteomics studies have yielded insights into the evolutionary history, core conserved structures and lineage-specific adaptations of eukaryotic flagella. Our CRISPR-Cas9 KO method enables rapid targeted gene deletion and characterisation of loss-of-function phenotypes for large cohorts of Leishmania genes in vitro and in vivo and hence new opportunities to interrogate the functions of hitherto poorly characterised flagellar proteins in motility regulation, environmental sensing and axoneme remodelling from 9+2 to 9+0. The bar-seq strategy for phenotyping of mutants can also be used to probe parasite-host interactions in mammals.
Promastigote-form L. mexicana (WHO strain MNYC/BZ/62/M379) were grown at 28°C in M199 medium (Life Technologies) supplemented with 2.2 g/L NaHCO3, 0.005% haemin, 40 mM 4-(2-Hydroxyethyl)piperazine-1-ethanesulfonic acid (HEPES) pH 7.4 and 10% FCS. The modified cell line L. mexicana SMP1:TYGFPTY [17] was cultured in supplemented M199 with the addition of 40 μg/ml G-418 Disulfate. L. mex Cas9 T7 [23] was cultured in supplemented M199 with the addition of 50 μg/ml Nourseothricin Sulphate and 32 μg/ml Hygromycin B.
To avoid proteolytic degradation, all procedures were performed on ice. 2·109 L. mexicana SMP1:TYGFPTY cells were collected at 800g for 15 min at 4°C, washed once in 20 ml phosphate buffered saline (PBS) and resuspended in 5 ml 10 mM PIPES [10 mM NaCl, 10 mM piperazine-N,N′-bis(2-ethanesulfonic acid, 1 mM CaCl2, 1 mM MgCl2, 0.32 M sucrose, adjusted to pH 7.2]. 0.375 ml of 1 M Ca2+ solution (final conc. 0.075 M) and a protease inhibitor cocktail [final concentration, 5 μM E-64, 50 μM Leupeptin hydrochloride, 7.5 μM Pepstatin A and 500 μM Phenylmethylsulfonyl fluoride (PMSF)] were added to the cell suspension. Cells were deflagellated by passing them 100 times through a 200 μl gel loading pipette tip (Starlab) attached to a 10 ml syringe. Flagella and cell bodies were separated through density gradient centrifugation, using a modified version of the protocol in [85]. The sample was loaded on top of the first sucrose-bed containing three layers of 10 mM PIPES with 33% (upper), 53% (middle) and 63% (bottom) w/v sucrose [10 mM NaCl, 10 mM piperazine-N,N′-bis(2-ethanesulfonic acid, 1 mM CaCl2, 1 mM MgCl2, adjusted to pH 7.2 with either 0.96M, 1.55M or 1.84M sucrose] and centrifuged at 800g for 15 min at 4°C. The pellet in the 63% sucrose layer was diluted with 10 ml 10 mM PIPES and centrifuged at 800g for 15 min at 4°C. The supernatant was discarded and the pellet resuspended in 40 μl 10 mM PIPES. This was the cell body fraction. The top layer of the first sucrose-bed, containing flagella, was collected and sucrose sedimentation was repeated with a second sucrose-bed containing only one layer of 10 mM PIPES with 33% w/v sucrose. The resulting top layer of the second sucrose bed was transferred to an ultra-centrifugation tube (Beckmann tubes) and collected by ultra-centrifugation at 100,000g for 1 h at 4°C (Beckman Coulter). The supernatant was discarded and the pellet resuspended in 40 μl 10 mM PIPES. This was the flagellar fraction. All other sucrose layers contained a mixture of flagella and cell bodies and were discarded. 1 μl of flagellar and cell body fractions was used for counting and imaging and 36 μl of each fraction were used for proteomic analysis.
Cell body and flagellar fractions were supplemented with 4 μl protease inhibitor cocktail (see above) and 10 μl octylglycoside (1% (w/v) final conc.), incubated for 20 min on ice and centrifuged at 18,500g for 1 h at 4°C to separate soluble (supernatant) from insoluble (pellet) proteins. 50 μl ice cold reducing 2x Laemmli buffer was added to the resulting supernatant. Pellets were dissolved in 100 μl 1x Laemmli buffer. To avoid aggregation of hydrophobic proteins, fractions were not boiled prior to SDS-PAGE [86]. 20 μl of flagella fractions and 10 μl of cell body fractions (~5–20 μg protein) were pre-fractionated on a 10% polyacrylamide gel, stained overnight with SYPRO Ruby Protein Gel Stain (Molecular Probes) and destained for 30 min in 10% (v/v) Methanol / 7% (v/v) acetic acid. Sample preparation in the following was carried out as described in [87]. Briefly, gel pieces were destained with 50% acetonitrile, reduced with 10mM TCEP (Tris(2-carboxyethyl)phosphine hydrochloride) for 30 minutes at RT, followed by alkylation with 55 mM Iodoacetamide for 60 minutes in the dark at RT. Samples were deglycosylated with PNGase F over two days at RT and digested overnight at 37°C with 100 ng trypsin. Samples were acidified to pH 3.0 using 0.1% trifluoroacetic acid and desalted by reversed phase liquid chromatography. Samples were analysed on an Ultimate 3000 RSLCnano HPLC (Dionex, Camberley, UK) system run in direct injection mode coupled to a QExactive Orbitrap mass spectrometer (Thermo Electron, Hemel Hempstead, UK).
MS-data were converted from .RAW to .MGF file using ProteoWizard (S6 Table) and uploaded to the Central Proteomics Facilities Pipeline (CPFP [88]). Protein lists were generated by using CPFP meta-searches (S6 Table) against the predicted L. mexicana proteome (gene models based on [29], followed by label-free SINQ quantification (S1 and S6 Tables). For SINQ enrichment plots detected GeneIDs were filtered (p ≥ 0.95, ≥ 2 peptides) and plotted using normalized spectral indices. For missing indices pseudo spectral indices of 10−10 were inserted. The mass spectrometry proteomics data have been deposited to the ProteomeXchange Consortium via the PRIDE [89] partner repository with the dataset identifier PXD011057.
Tagging was achieved by insertion of a drug-selectable marker cassette and fluorescent protein gene into the endogenous gene to produce an in-frame gene fusion. The fusion PCR method described in Dean et al., [90] was used for tagging with eYFP, using pJ1170 (pLENT-YB) as the template plasmid for PCR and selection with 5 μg/ml Blasticidin-S deaminase. The CRISPR-Cas9 method described in Beneke et al., [23] was used for tagging with mNeonGreen. The online primer design tool www.LeishGEdit.net was used to design primers for amplification of the 5’ or 3’ sgRNA template and primers for amplification of donor DNA from pPLOTv1 blast-mNeonGreen-blast or pPLOTv1 puro-mNeonGreen-puro. Transfectants were selected with either 5 μg/ml Blasticidin-S deaminase or 20 μg/ml Puromycin Dihydrochloride.
Gene deletions were essentially done as described in Beneke et al., [23]. The online primer design tool www.LeishGEdit.net was used to design primers for amplification of the 5’ and 3’ sgRNA templates and for amplification of donor DNA from pTBlast and either pTPuro or pTNeo. Primers for deletion of PFR2A-C were designed with the EuPaGDT primer design tool [91] using the PFR2 array sequence from L. mex Cas9 T7. For amplification of the sgRNA template DNA, primers:
For amplification of a pTNeo donor DNA fragment with 80 bp homology arms, primers:
For transfections on 96-well plates the protocol was modified as follows (similar to descriptions in [92]): 52 x 107 late log phase L. mex Cas9 T7 cells (1 x 107 cells per reaction) were collected by centrifugation at 800g for 15 min. A transfection buffer was prepared by mixing 2 ml 1.5 mM CaCl2, 6.5 ml modified 3x Tb-BSF buffer (22.3 mM Na2HPO4, 7.67 mM NaH2PO4, 45 mM KCl, 450 mM sucrose, 75 mM HEPES pH 7.4) and 1.5 ml ddH2O. The cells were re-suspended in 3 ml of this transfection buffer and centrifuged again as above. The heat-sterilised sgRNA and donor DNA PCR products were placed into 48 wells of a 96-well disposable electroporation plate (4 mm gap, 250 μl, BTX) such that a given well combined all of the donor DNAs and the sgRNA templates for a given target gene. After centrifugation, cells were re-suspended in 5.2 ml transfection buffer and 100 μl of the cell suspension dispensed into each of the 48 wells containing the PCR products. Plates were sealed with foil and placed into the HT-200 plate handler of a BTX ECM 830 Electroporation System. Transfection used the following settings: 1500 V, 24 pulses, 2 counted pulses, 500 ms interval, unipolar, 100 μs. After transfection cells were recovered in 1 ml supplemented M199 in 24-well plates and incubated for 8–16 h at 28°C / 5% CO2 before addition of the relevant selection drugs by adding 1 ml of supplemented M199 with double the concentration of the desired drugs. Mutants were selected with 5 μg/ml Blasticidin-S deaminase in combination with either 20 μg/ml Puromycin Dihydrochloride or 40 μg/ml G-418 Disulfate and further incubated. Drug resistant populations typically emerged after one week.
Drug-selected populations were passaged at least twice (one with at least a 1:100 dilution) before extraction of genomic DNA using the protocol described in [93]. A diagnostic PCR was done to test for the presence of the target gene ORF in putative KO lines and the parental cell line (S8 Fig). Primer3 [94] was used to design, for the entire L. mexicana genome (gene models based on [29]), primers to amplify a short 100–300 bp unique fragment of the target gene ORF (S7 Table). In a second reaction, primers 518F: 5’-CACCCTCATTGAAAGAGCAAC-3’ and 518R: 5’-CACTATCGCTTTGATCCCAGGA-3’ were used to amplify the blasticidin resistance gene from the same genomic DNA samples. For ΔPFR2 additional primers were used to confirm deletions (S10 Fig;
Leishmania cells expressing fluorescent fusion proteins were imaged live. Samples were prepared as described in [17]. Cells were immediately imaged with a Zeiss Axioimager.Z2 microscope with a 63× numerical aperture (NA) 1.40 oil immersion objective and a Hamamatsu ORCA-Flash4.0 camera or a 63× NA 1.4 objective lens on a DM5500 B microscope (Leica Microsystems) with a Neo sCMOS camera (Andor Technology) at the ambient temperature of 25–28°C.
For transmission electron microscopy, deflagellated cell bodies and isolated flagella were prepared with a modified version of the chemical fixation protocol described in Höög et al., [95]. Pellets of cell fractions were fixed with 2.5% glutaraldehyde in 10 mM PIPES (buffer as described above) overnight at 4°C. Pellets were washed four times for 15 min in 10 mM PIPES and overlaid with 10 mM PIPES, containing 1% osmium tetraoxide and incubated at 4°C for 1 h in darkness, then washed five times with ddH2O for 5 min each time and stained with 500 μl of 0.5% uranyl acetate in darkness at 4°C overnight. Samples were dehydrated, embedded in resin, sectioned and on-section stained as described previously [95]. Electron micrographs were captured on a Tecnai 12 TEM (FEI) with an Ultrascan 1000 CCD camera (Gatan).
Micrographs were processed using Fiji [96]. To enable comparisons between the parental and tagged cell lines, the same display settings for the green fluorescence channel were used. PFR length (defined by reporter signal) and flagellar length (distance between stained kinetoplast DNA and flagellar tip) was measured with the ROI manager in Fiji [96].
Mutant and parental cell lines were tracked using the previously described method in [33] with three modifications. Firstly, the scripts were modified for batch analysis of multiple files. Secondly, prior to calculation of swimming statistics any ‘drift’ due to bulk fluid flow in the sample was subtracted. As swimming direction of each cell in the population is in a random direction any drift is visible as a mean population movement between successive frames. We treated drift as an apparent translation and scaling of cell positions between successive video frames, which was then subtracted. Finally, the primary statistics we considered were mean speed (using the path at 200 ms resolution) and directionality (mean velocity as a fraction of mean speed). Swimming behaviour was measured in triplicates at approximately 6·106 cells/ml and analysed from 5 μl of cell culture on a glass slide in a 250-μm deep chamber covered with a 1.5 mm cover slip using darkfield illumination with a 10× NA 0.3 objective and a Hamamatsu ORCA-Flash4.0 camera on a Zeiss Axioimager.Z2 microscope at the ambient temperature of 25–28°C. The sample was stored inverted prior to use, then turned upright immediately prior to imaging to ensure consistent motion of immotile cells through sedimentation between samples. A sample of the parental cell line killed with a final concentration of 1% formaldehyde was used as a reference for motion of completely paralysed cells through sedimentation and Brownian motion alone.
Sand flies were either infected with pooled mutant populations of L. mexicana or individually with L. mexicana MNYC/BZ/62/M379 wild type (WT), parental line L. mex Cas9 T7, knockout cell line ΔPF16, its add-back (PF16AB) [23], knockout cell line ΔBBS2, its add-back (BBS2AB), knockout cell line ΔMBO2 and its add-back (MBO2AB). For pooling, parasites were pooled into four sub-pools with different starting densities, depending on the mutant growth rates, so that the pools would reach late log phase at the same time. Sub-pools were seeded at 5·106 cells/ml, 3·105 cells/ml, 1·105 cells/ml or 8·104 cells/ml, respectively, and grown for 48 hours. The sub-pools were mixed in equal proportions just before the infection. All parasites were cultivated at 23°C in M199 medium supplemented with 20% foetal calf serum (Gibco), 1% BME vitamins (Sigma), 2% sterile urine and 250 μg/ml amikin (Amikin, Bristol-Myers Squibb). Before experimental infections, logarithmic growing parasites were washed three times in PBS and resuspended in defibrinated heat-inactivated rabbit blood at a concentration of 106 promastigotes/ml. Lutzomyia longipalpis (from Jacobina, Brazil) were maintained at 26°C and high humidity on 50% sucrose solution and a 12h light/12h dark photoperiod. Sand fly females, 3–5 days old, were fed through a chick skin membrane as described previously [97]. Fully-engorged females were separated and maintained at 26° C with free access to 50% sucrose solution. They were dissected on days 2 or 6 post blood-meal (PBM) and the guts were checked for localisation and intensity of infection by light microscopy. Parasite load was graded as described previously by [98] into four categories: zero, weak (<100 parasites/gut), moderate (100–1000 parasites/gut) and heavy (>1000 parasites/gut).
Mutant Leishmania lines were grown separately or in sub-pools as described above to late log phase and mixed in equal proportions (1·107 cells per cell line). This pool was divided equally into three aliquots. DNA was extracted using the Roche High Pure Nucleic Acid Kit or Qiagen DNeasy Blood & Tissue Kit according to the manufacturers instructions and eluted in 20 μl elution buffer. Each aliquot was then either kept in promastigote culture over 96 hours, where DNA was extracted every 24 hours from approximately 1·107 cells as above, or used to infect three separate batches of 150 sand flies. The three batches were kept separate and DNA was extracted from 50 whole sand flies two, six and nine days post blood meal, using the same kit as follows: Sand flies were placed in two 1.5 ml microcentrifuge tubes (25 flies per tube), overlaid with 100 μl tissue lysis buffer and frozen at -20°C. The dead flies were defrosted and after addition of 100 μl tissue lysis buffer and 40 μl proteinase K, flies were disrupted with a disposable plastic pestle (Bel-Art) and DNA purified according to the kit manufacturer’s instructions. DNA was eluted with 50 μl elution buffer and eluates from the same timepoint were combined, yielding 100 μl in total.
For each sample, 600 ng isolated DNA was incubated with exonuclease VII (NEB) overnight at 37°C, purified using SPRI magnetic beads and amplified using PAGE purified custom designed p5 and p7 primers (Life Technologies), containing indexes for multiplexing and adapters for Illumina sequencing. Amplicons were again bead purified and multiplexed by pooling them in equal proportions. The final sequencing pool was once again bead purified and quantified by qPCR using NEB Library Quant Kit. Library size was determined using Agilent High Sensitivity DNA Kit on a 2100 Bioanalyzer instrument. The pool was diluted to 4 nM and spiked with 30% single indexed Leishmania genomic DNA, prepared using Illumina TruSeq Nano DNA Library kit according to the manufacturers instructions. The final library was spiked with 1% PhiX DNA and the Illumina sequencer was loaded with 8 pM to allow low cluster density (~800 K/mm2). Sequencing was performed using MiSeq v3 150 cycles kit following the manufactures instructions with paired-end sequencing (2x75 cycles, 6 and 8 cycles index read).
MiSeq raw files were de-multiplexed using bcl2fastq (Illumina). De-multiplexed samples were subjected to barcode counting using a bash script. Each gene in the Leishmania genome was assigned a unique barcode—the number of times each of these barcodes appeared in the sequencing output was recorded (S9 and S10 Tables). The criteria for barcode counting was a 100% match to the 17 nt total length. Counts for each barcode were normalized for each sample by calculating their abundance relative to all 25 barcodes. One of the mutants selected for the pooled screen was excluded from the analysis because sequencing of the cell line showed eight nucleotide mismatches in the p5 sequence (likely introduced during oligonucleotide synthesis) which precluded amplification of the barcode region. To calculate “fitness” normalized barcode counts in the pooled population before feeding were divided by normalized counts at the relevant time point post blood meal.
Orthofinder 1.1.2 was used to generate orthogroups for predicted protein coding genes from 48 eukaryotic genomes (32 ciliated species and 16 non-ciliated species): The 45 previously used by Hodges et al. [69] (with Leishmania major replaced with Leishmania mexicana) and supplemented with Aspergillus nidulans, Plasmodium berghei and Schistosoma mansoni.
|
10.1371/journal.ppat.1006431 | Generality of toxins in defensive symbiosis: Ribosome-inactivating proteins and defense against parasitic wasps in Drosophila | While it has become increasingly clear that multicellular organisms often harbor microbial symbionts that protect their hosts against natural enemies, the mechanistic underpinnings underlying most defensive symbioses are largely unknown. Spiroplasma bacteria are widespread associates of terrestrial arthropods, and include strains that protect diverse Drosophila flies against parasitic wasps and nematodes. Recent work implicated a ribosome-inactivating protein (RIP) encoded by Spiroplasma, and related to Shiga-like toxins in enterohemorrhagic Escherichia coli, in defense against a virulent parasitic nematode in the woodland fly, Drosophila neotestacea. Here we test the generality of RIP-mediated protection by examining whether Spiroplasma RIPs also play a role in wasp protection, in D. melanogaster and D. neotestacea. We find strong evidence for a major role of RIPs, with ribosomal RNA (rRNA) from the larval endoparasitic wasps, Leptopilina heterotoma and Leptopilina boulardi, exhibiting the hallmarks of RIP activity. In Spiroplasma-containing hosts, parasitic wasp ribosomes show abundant site-specific depurination in the α-sarcin/ricin loop of the 28S rRNA, with depurination occurring soon after wasp eggs hatch inside fly larvae. Interestingly, we found that the pupal ectoparasitic wasp, Pachycrepoideus vindemmiae, escapes protection by Spiroplasma, and its ribosomes do not show high levels of depurination. We also show that fly ribosomes show little evidence of targeting by RIPs. Finally, we find that the genome of D. neotestacea’s defensive Spiroplasma encodes a diverse repertoire of RIP genes, which are differ in abundance. This work suggests that specificity of defensive symbionts against different natural enemies may be driven by the evolution of toxin repertoires, and that toxin diversity may play a role in shaping host-symbiont-enemy interactions.
| Nearly all insects harbor bacterial partners. These microbes can be defensive symbionts, protecting their hosts against parasites and pathogens, and in some cases, may defend against more than one enemy, presenting opportunity to study the evolution of specificity underlying defensive interactions. What factors determine specificity and generality of defense? Can symbiont-encoded effector molecules act generally against enemies without causing harm to the host? We show here that symbiont-encoded ribosome-inactivating toxins, previously implicated in protection of a Drosophila fruit fly against its nematode parasite, are also implicated in defending flies against parasitic wasps. We quantify activity of the toxin as the proportion of ribosomes depurinated and, importantly, find that susceptible wasps are attacked early in development and are strongly affected, while hosts and a resistant wasp are not. We also show that not one, but a family of toxins is maintained and expressed by symbionts. Together, our findings implicate toxin diversity as a factor contributing to the evolution of specificity in a symbiont-mediated defense.
| Multicellular organisms commonly harbor microbial symbionts. Some of the best-studied recent examples lie in maternally transmitted bacterial endosymbionts of insects, which are ubiquitous [1–3], and rely on the successful reproduction of their hosts to ensure their own survival. Unsurprisingly, this puts them in direct conflict with their hosts' natural enemies, and recent work has documented extraordinary diversity in insect symbiont-mediated protection. For example, native inherited bacterial endosymbionts confer resistance to parasitic wasps in aphids and Drosophila [4,5], predatory spiders in rove beetles [6], parasitic nematodes and RNA viruses in Drosophila flies [7–9] and pathogenic fungi in beewolves and aphids [10–12]. While an increasing number of examples are being uncovered, very little is known about the mechanism of symbiont-mediated protection. How specific is protection? Can symbionts use the same strategies to protect against different classes of natural enemies? How do symbionts recognize and target enemies without harming their host in turn?
In this study, we address the mechanism and generality of protection conferred by Spiroplasma defensive symbionts of Drosophila flies. Spiroplasma (Mollicutes) is a diverse and widespread lineage of arthropod-associated bacteria [13] thought to infect at least ~7% of insects [3]. Spiroplasma biology is incredibly diverse. While many strains are commensal in arthropod guts, a number of pathogenic strains have been identified, including pathogens of bees (S. melliferum and S. apis), crayfish (S. eriocheiris), and plants (S. citri and S. kunkelii) [14–17]. Plant-pathogenic strains are vectored by plant-feeding Hemipteran insects. In addition, vertical transmission has evolved independently in Spiroplasma numerous times, including in Drosophila flies, pea aphids, and butterflies [18–20]. Vertically transmitted Spiroplasma have evolved a number of interesting strategies to persist in their hosts, including male-killing and protection against natural enemies. Perhaps the best-studied defensive Spiroplasma infects the hemolymph and ovarian tissues of the woodland fly Drosophila neotestacea, which it protects against a common and virulent parasitic nematode, Howardula aoronymphium [9]. Indeed, the benefits of Spiroplasma protection are so great, and nematode parasitism is so prevalent in nature, that Spiroplasma-infected flies have replaced uninfected ones, with the symbiosis currently spreading across North America [21]. The Spiroplasma symbiont of D. neotestacea (hereafter sNeo) and the closely-related strains infecting Drosophila melanogaster (hereafter sMel) and Drosophila hydei have also been shown to defend against parasitic wasps in the lab [5,22–25]. sMel is also known as the melanogaster sex ratio organism, or MSRO, because it also acts as a male-killer [26], eliminating all sons of infected females during embryogenesis by targeting male-specific components of the dosage compensation complex [27–29]. Spiroplasma is highly effective in protecting against wasps in the lab, though not always in rescuing flies. Despite high rates of fly mortality in some cases, wasp killing results in a net benefit to Spiroplasma-infected hosts [22]. Parasitic wasps are among the most important sources of mortality against Drosophila in the wild [30,31], although whether Spiroplasma protection against wasps occurs in nature has not yet been demonstrated. Interestingly, unlike protection against wasps, which appears to be quite general to Spiroplasma in Drosophila, symbiont transfection experiments showed that only sNeo is able to protect against nematodes [23].
We recently made progress toward resolving how Spiroplasma protects D. neotestacea against parasitic nematodes. We found that Spiroplasma encodes a ribosome-inactivating protein (RIP) that is implicated in nematode defense [32]. RIPs are N-glycosidases that irreversibly inactivate eukaryotic cytosolic ribosomes by cleaving a specific adenine residue from the 28S ribosomal RNA (rRNA). Well-known RIPs include potent toxins such as ricin and Shiga toxin, both of which are categorized as type 2 RIPs. These RIPs are composed of an A-chain, a polypeptide which possesses N-glycosidase activity, bonded by disulphide linkages to a B-chain, a lectin or carbohydrate binding domain that is involved in attachment to the cell surface. RIPs possessing only an A-chain are categorized as type 1 RIPs; this type was previously found in sNeo. These may be equally effective ribosome inactivators in vitro, but are often less toxic in vivo, likely because cell entry is restricted.
The best studied RIPs are those found in angiosperms, where they are extremely diverse and common, and appear to act in defensive roles against both predators and viruses [33]. In addition to inhibition of protein synthesis, both type 1 and type 2 RIPs lead to apoptosis and necrosis in animals [34]. Bacterially-encoded RIPs have been much less studied, other than those found in Shigella dysenteriae and enterohemorrhagic Escherichia coli serotypes, which are notorious for their contribution to bacterial virulence in humans [35,36]. RIP homologs have been found in many bacterial genomes but few have been characterized. A gene encoding a Shiga-like toxin B-chain is among several toxin genes associated with protective strains of APSE, the bacteriophage harboured by Hamiltonella defensa, a defensive symbiont of aphids that confers protection against parasitic wasps [37–39], but this gene has not yet been functionally characterized.
We previously found that a type 1 RIP in sNeo is transcriptionally upregulated in nematode-infected flies [32], and attacks nematode ribosomes in vitro [40]. Furthermore, infection experiments revealed that nematodes that infect flies harbouring Spiroplasma exhibit the hallmarks of attack by RIP toxins, including massive increases in levels of depurinated ribosomes [40]. Intriguingly, the genome of the Spiroplasma symbiont of D. melanogaster contains five putative RIPs [41], hinting that RIP toxins also play a role in protection against parasitic wasps, and that successful protection against a specific enemy might depend on the specific arsenal of toxins encoded by each symbiont. We therefore explored the role of RIPs in parasitic wasp defense, exposing Spiroplasma-positive and negative flies to three wasp species, quantifying RIP toxin expression, as well as intact and depurinated ribosomes in wasps and hosts, and surveying the genome of Spiroplasma in D. neotestacea for additional RIPs.
We find strong evidence implicating Spiroplasma-encoded RIPs in protection against parasitic wasps. Wasp ribosomes are depurinated as soon as eggs hatch inside the fly larva, with depurination peaking upon host pupation. We find little evidence that fly ribosomes suffer significant collateral damage from RIPs. We also report the discovery of an ectoparasitic wasp that is not killed by Spiroplasma and escapes the brunt of RIP attack. Furthermore, we identify three additional putative RIP genes in the genome of sNeo, which protects against both nematodes and wasps, and show that this symbiont encodes two RIPs with no apparent close relatives in the sMel genome. Our results contribute to a growing appreciation for the potential of symbiont-encoded toxins as important determinants of specificity in insect defensive symbioses.
To assess RIP activity on parasitic wasp 28S rRNA, we modified an established, highly-sensitive RT-qPCR-based assay [40,42,43] to quantify the abundance of ribosomes that have been depurinated at the α-sarcin/ricin loop (SRL), relative to the total ribosome pool. In separate reactions, primers specific to either intact or depurinated SRLs were used to quantify ribosomal targets in each state. The forward primers differ in sequence only at their 3’ termini, with one set of primers designed to hybridize with “intact” ribosomal cDNA synthesized from rRNA possessing an adenine at the affected site (adenine SRLs, thymine variant cDNA), and the other primer set targeting cDNA synthesized from rRNA containing the abasic position (depurinated SRLs, adenine variant cDNA). Strong specificity to wasp and fly ribosomes is conferred by the sequence of the reverse primer, which hybridizes with a less-conserved region nearby and prevents mis-amplification across primer sets. We found a remarkably strong signal of RIP attack, with elevated abundances of depurinated wasp ribosomes associated with Spiroplasma-positive flies (Fig 1A). We tested parasitized hosts on the first day of pupation and found high levels of depurination (t-test: L. heterotoma in D. melanogaster t14 = 34.17, p < .001, a >13,000-fold increase). On the second day of pupation, the extent of RIP attack was even more striking (t-tests: L. heterotoma in D. melanogaster t10 = 30.07, p < .001, a >800,000-fold increase; L. boulardi in D. melanogaster t14 = 41.21, p < .001, a >1,000,000-fold increase; L. heterotoma in D. neotestacea t9 = 21.09, p < .001, a >350,000-fold increase). These fold changes are all at the upper limit of our assay’s quantification sensitivity. The unambiguous signal of depurinated wasp ribosomes in Spiroplasma-positive hosts was accompanied by a strong reduction in the pool of intact wasp ribosomes in all symbiont-wasp encounters (Fig 1A, t-tests: L. heterotoma in D. melanogaster t14 = 5.39, p < .001, a 4.1-fold decrease; L. boulardi in D. melanogaster t14 = 2.45, p = .028, a 1.7-fold decrease; Welch’s t-test: L. heterotoma in D. neotestacea t5.47 = 3.88, p = .010, a 3.2-fold decrease). This pattern parallels our protection assays, in which we found almost complete wasp mortality in Spiroplasma-positive flies (Fig 1B), as has been previously shown [22,23]. Despite wasp death, flies parasitized by L. heterotoma (but not L. boulardi) are not rescued by Spiroplasma, as has also been previously shown [25]. These experiments clearly demonstrate that an attack involving ribosome depurination is mounted against parasitic wasps by Spiroplasma during host defense. The attack is followed by complete wasp mortality but differential host survival, possibly contingent upon the wasp species.
In order to determine the onset of depurination of wasp ribosomes, we carried out a detailed timecourse experiment, assaying wasp ribosomes for RIP activity immediately after exposing D. melanogaster to L. heterotoma and at subsequent 24-hour intervals until the second day of pupation (120 hours post-exposure). Our results reveal evidence of RIP attack remarkably early, just 48 hours after wasp exposure, with two of the six parasitized fly larvae tested showing a >3,000-fold increase in depurinated wasp ribosomes relative to Spiroplasma-free, L. heterotoma-infested controls (Fig 2; t6 = 13.2, p < .001). In the remaining four larvae, there was no significant change relative to Spiroplasma-free controls, suggesting RIP attack had not yet begun (t8 = 0.08, p = .933). All six fly larvae tested at the following time point, 72 hours after wasp exposure, showed a strong signal of wasp ribosome depurination (t9 = 9.56, p < .001).
The result at 48 hours led us to suspect that the onset of RIP activity might coincide with hatching of the wasp egg. Therefore, we carried out an additional exposure, identical in design to the previous, and dissected parasitized larvae at 48 hours (N = 20) and 72 hours (N = 11), scoring the frequency of unhatched eggs and hatched larvae at each time point. In our assays, wasps begin hatching around 48 hours after infestation. At this time, 16 fly larvae contained at least one wasp egg, just two contained a wasp larva (one of these contained both an egg and a larva), and three were unparasitized. In our 72 hour dissections, nine flies contained wasp larvae, two were unparasitized, and none contained wasp eggs.
Together, the results of these experiments suggest that RIP attack begins very early in wasp development, apparently as soon as newly hatched wasp larvae are exposed to host hemolymph. The early signal of depurination fits nicely with previous work showing wasp developmental delays in Spiroplasma-positive flies as early as 72 hours after infestation [22].
A major question in defensive symbiosis is whether and how symbionts distinguish hosts from natural enemies, and whether hosts exhibit any delayed deleterious effects of symbiont-mediated protection. Specifically, we hypothesized that host mortality in L. heterotoma-exposed, Spiroplasma-positive flies might be due to off-target RIP attack. To address this, we quantified intact and depurinated host ribosomes, using the same parasitized flies that we used in our wasp ribosome assays. Although we were able to detect significant levels of depurinated host ribosomes relative to Spiroplasma-negative controls (Fig 3B, 3D and 3F, ANOVAs: L. heterotoma-infested D. melanogaster, F2,21 = 176, p < .001, a ~950-fold increase; L. boulardi-infested D. melanogaster, F2,21 = 826, p < .001, a > 6,000-fold increase; L. heterotoma-infested D. neotestacea, F2,19 = 60.7, p < .001, a ~750-fold increase), their proportion relative to total ribosomes was much lower than in wasps and was not elevated in wasp-infested relative to uninfested flies, suggesting that RIP activity is not responsible for host mortality (Fig 3B, 3D and 3F, pink versus blue boxes; plots are labeled with the results of Tukey post hoc comparisons). We found no evidence of reduced levels of intact ribosomes in Spiroplasma-positive relative to negative flies in L. boulardi-infested D. melanogaster (ANOVA: F2,21 = 4.08, p = .032, a modest 1.4-fold increase in intact templates detected in S+ flies) or in L. heterotoma-infested D. neotestacea (ANOVA: F2,19 = 0.36, p = .705). L. heterotoma-infested D. melanogaster appeared to show a reduction in intact ribosomes (ANOVA: F2,21 = 13.9, p < .001) associated with Spiroplasma infection, as well as an apparent increase in detectable intact templates in the Spiroplasma-positive, wasp-uninfested treatment (Fig 3A, 3C and 3E). Because this effect is more pronounced in flies succumbing to wasp infestation, i.e. during L. heterotoma but not L. boulardi parasitism, it is possible that there is some loss of ribosomal integrity associated with fly mortality in D. melanogaster, a result we do not observe in D. neotestacea. Finally, to test for delayed RIP activity on adult hosts, we measured levels of intact and depurinated ribosomes in week-old adult Spiroplasma-positive D. melanogaster that had survived L. boulardi parasitism, and found no significant change in abundance of depurinated host ribosomes compared to unexposed controls (S1 Fig; t-tests: intact, t10 = 0.20, p = .849; depurinated, t10 = 1.30, p = .221).
We tested whether RIPs are encountering host ribosomes intracellularly, or if free host ribosomes are being depurinated in the hemolymph, where Spiroplasma resides, which could account for the RIP activity we found in host flies. We bled third instar D. neotestacea larvae into Ringer’s solution and separated hemolymph from hemocytes (i.e. blood cells) by centrifugation, then assayed cell pellets and supernatant for RIP activity. We found 11-fold greater levels of depurination in the hemolymph than in the cell pellet, and 125-fold greater levels of depurination in hemolymph than in the bled larval carcasses. Levels of depurination were significantly different across RNA isolation sources (Fig 3G, ANOVA: F5,29 = 32.66, p < .001) with levels of depurination in hemolymph, i.e. sourced from extracellular ribosomes, significantly greater than in hemocytes, bled larvae, and control larvae (Tukey test: p < .001). The proportion of depurinated ribosomes in the cell pellets, enriched for ribosomes within hemocytes, was also significantly greater than in bled larvae (p = .002) and not greater than controls (p = .112). Compared with the levels of depurinated host and wasp ribosomes we report from the wasp exposure experiments in D. neotestacea, the proportion of host ribosome depurination in hemolymph is not significantly different from the highest proportion of wasp depurination we observed but is greater than in wasp-infested host larvae from these experiments (p < .001). These results suggest that the low levels of host depurination we see are due to depurination of free host ribosomes that have been released into the hemolymph where there is abundant Spiroplasma.
Previous work focused on a single RIP, sNeo-RIP1, that is upregulated in response to nematode exposure in D. neotestacea, and that depurinates at the α-sarcin/ricin loop of eukaryotic 28S rRNA [32,40]. Interestingly, the Spiroplasma from D. melanogaster encodes five putative RIPs. We predicted therefore that sNeo may harbor additional RIP diversity. We generated sNeo Illumina and PacBio sequence and screened our draft genome assembly, identifying a total of four putative RIPs. Two are phylogenetically distinct from all of the RIPs of sMel (sNeo-RIP1 & 2), while the two others (sNeo-RIP3 & 4) have close relatives in sMel (Fig 4). All four of the putative sNeo RIPs return significant matches to the ricin A-chain (HMMER e-values < E-05 to pfam PF00161). We searched for putative lectin domains (B-chains) within these RIP-coding sequences but failed to identify any candidate domains. We confirmed each new sNeo RIP by PCR amplification and Sanger sequencing.
RIPs in Spiroplasma from D. melanogaster and D. neotestacea represent three distinct groups based on predicted protein domains and phylogenetic analysis (Fig 4). The first clade is composed of sNeo-RIPs 1 and 2, which contain a short predicted disordered region between the predicted signal peptide and N-glycosidase domain, as reported for sNeo-RIP1 [40]. The second clade contains sNeo-RIP3 and sMel-RIPs 3, 4, and 5, and encodes a longer N-terminal domain of ~150 aa, which is not predicted to be disordered by HMMER [44]. The third clade contains sNeo-RIP4 and sMel-RIPs 1 and 2, which appear to lack an N-terminal domain, and encode a unique ~150 aa long C-terminal tail. Neither the N- or C-terminal domains produce significant hits via HMMER to indicate putative function of these domains.
We quantified baseline transcript abundance of each sNeo and sMel RIP throughout development of unparasitized D. neotestacea and D. melanogaster, starting 24 hours after the flies entered the second larval instar stage, and at three more 24-hour intervals. Each RIP transcript was quantified for six biological replicates at each time point. We found that although all sMel and sNeo RIPs were constitutively expressed, they differed in abundance (ANOVA: sNeo RIPs, F3,92 = 109, p < .001; sMel RIPs, F3,69 = 20.1, p < .001, S2 Fig).
We were interested in examining RIP attack in a wasp that we predicted might overcome protection by Spiroplasma. Pachycrepoideus vindemmiae (Hymenoptera: Pteromalidae) is a pupal ectoparasitic wasp [45] that pierces the fly pupal case with its ovipositor and deposits an egg on the cuticle of the developing fly inside. Thus the developing wasp larva feeds on but does not develop surrounded by host hemolymph like Leptopilina. As we expected, P. vindemmiae successfully parasitizes Spiroplasma-positive flies (Fig 5A). We detected depurinated ribosomes, but much less than for L. heterotoma and L. boulardi (Fig 5B, ANOVA: F3,28 = 27.8, p < .001). Depurinated ribosomes were 67 times more abundant in Spiroplasma-positive samples than Spiroplasma-free controls (t-test: t19 = 3.27, p = .004), and there was no reduction in intact ribosomes, unlike Leptopilina (Fig 5C, Welch’s test: t16.6 = 0.29, p = .778).
In this paper, we implicate ribosome-inactivating proteins as key players in Spiroplasma-mediated protection against parasitic wasps, suggesting that similar mechanisms are used to defend against nematodes and wasps. We show very high levels of depurination at the α-sarcin/ricin loop for wasp ribosomes developing in Spiroplasma-positive flies. Depurination occurs very early, soon after the wasp larva hatches. Likewise, we show a marked decrease in levels of intact wasp ribosomes.
Interestingly, we found that the pupal ectoparasitic wasp, Pachycrepoideus vindemmiae, is not affected by Spiroplasma—nor does it suffer from high levels of ribosome attack. We do detect some depurinated P. vindemmiae ribosomes, but at significantly lower levels than L. boulardi and L. heterotoma. This suggests that RIP effectiveness may depend on symbiont location and titer. Endoparasitic wasp larvae (as well as nematode motherworms) are bathed in Spiroplasma-infested hemolymph, and likely experience a much higher degree of exposure to toxins than externally-feeding parasites or gut parasites. Like Leptopilina, Pachycrepoideus feeds on fly hemolymph, but living outside of the host, it is not immersed in hemolymph upon hatching. Thus it may not be surprising that Pachycrepoideus overcomes Spiroplasma defense; although, ingesting plant RIPs has been shown to be highly toxic to some insects, such as Callosobruchus and Anthonomus beetles [46], and Pachycrepoideus may be ingesting significant levels of toxin, since Spiroplasma titer increases dramatically during fly pupation [25]. Resistance to ricin has been demonstrated in houseflies [47] and grasshoppers [48] and to ricin and saporin (a type 1 RIP) in Spodoptera and Heliothis moths [46], where reduced RIP toxicity has been attributed to serine protease-mediated hydrolysis in the digestive tract, and it is possible that RIPs are also broken down in Pachycrepoideus' gut.
What determines a symbiont's ability to defend against one type of natural enemy and not another, and to avoid harming hosts? A potential clue may be found in the diversity of RIPs encoded by sNeo and sMel, which encode 4 and 5 divergent RIPs, respectively, despite having highly reduced genomes typical of vertically transmitted endosymbionts [2,41]. Furthermore, some of these divergent RIPs are not shared between the two symbionts, even though they are both strains of Spiroplasma poulsonii. We speculate that the different RIPs are specialized for targeting different cell types. This may explain why sNeo but not sMel defends against nematodes [23], as it encodes two divergent RIPs that sMel lacks. It will be interesting to determine whether Spiroplasma RIPs respond differentially to different host natural enemies or stresses.
We do not yet know how Spiroplasma RIPs target and enter specific host or parasite cells. Although Spiroplasma RIPs are type 1 and do not contain lectin domains, some, but not all, have N- or C-terminal domains with no known homologs; perhaps these are important in target specificity. The discovery of signatures of RIP activity in Spiroplasma-infected D. melanogaster opens up the possibility of engineering transgenic flies that express Spiroplasma RIPs to directly test these questions, and to confirm a causal role of Spiroplasma RIPs in protection. Some type 1 RIPs, such as saporin and trichosanthin in plants, have been shown to require interaction with surface receptors for cell entry [49,50]. For example, in mammalian cells, endocytosis of saporin is mediated by a member of the low density lipoprotein (LDL) receptor family, called LRP1. It was recently proposed that lipids may play an important role in Spiroplasma protection against wasps [25], as both Spiroplasma and wasps take up lipids from their host [51,52]. In insects, uptake of lipophorin, the major lipid carrier in Drosophila hemolymph [53], is mediated by an insect relative of LRP1, VLDL [54]. Thus, it is tempting to speculate that parasitic wasp adaptations to increase lipid uptake, such as elevated production of VLDL receptors, could also result in enhanced vulnerability to RIPs. Recently, Mateos et al. (2016) [55] found that some endoparasitic wasps, such as Leptopilina guineaensis, are resistant to Spiroplasma, and it would be interesting to see whether resistant wasps differ in VLDL receptor sequence.
Toxins that target highly conserved cellular components may be favored for their utility against a range of natural enemies, but the benefit of this generality may also come at a cost to the host. However, we found little evidence of negative effects of Spiroplasma RIPs on hosts. This suggests that mortality of Spiroplasma-positive flies attacked by L. heterotoma (but not L. boulardi) is not due to the symbiont. Differential fly mortality is probably due instead to differences between Leptopilina species, particularly with respect to their venoms, which are very different [56,57]. A recent study found that L. heterotoma venom is extremely toxic to Drosophila, killing flies even when no wasp eggs are deposited [58]. Interestingly, even though there is little evidence of negative effects of wasp or nematode [40] protection on flies, it is worth noting that Spiroplasma does kill flies: sMel kills male embryos [26,29,59,60] and also causes pathology in old adult females [51].
We have shown that Spiroplasma symbionts encode a diverse assemblage of RIP toxins that appear to play an important role in defense against parasitic wasps, as well as parasitic nematodes. Toxin diversity and expansion may be a common feature in defensive symbionts [11,61,62], and a major outstanding question is how this toxin diversity mediates specificity [63] and coevolutionary arms races between symbionts, hosts, and enemies.
To avoid confusion, we refer to the strains of Spiroplasma poulsonii that infect D. melanogaster and D. neotestacea as sMel and sNeo, respectively. This terminology is similar to how Wolbachia strains that infect Drosophila are called (i.e. wMel [64]). sMel is also known as Spiroplasma poulsonii MSRO (melanogaster sex ratio organism) because of its male-killing phenotype. This strain was originally collected in D. melanogaster in Uganda, Africa [65] and was introduced into the Oregon-R genetic background by microinjection and provided to us by Bruno Lemaitre’s lab at EPFL. Male-killing in S. poulsonii MSRO is highly penetrative, such that every generation is 99–100% female. We therefore mated Spiroplasma-infected females to uninfected Oregon-R males every generation to maintain the culture. It is important to note that because of this penetrance, all larvae of the Spiroplasma-positive treatment of D. melanogaster are female, different to the Spiroplasma-free treatment and D. neotestacea which have unbiased sex ratios (sNeo does not kill males). D.neotestacea and its Spiroplasma symbiont (sNeo) were collected in West Hartford, CT, USA in 2006. Fly lines are maintained on Instant Drosophila Medium (Carolina Biological Supply) on a 12:12 light dark cycle, D. melanogaster at 24°C and D. neotestacea lines at 21°C with the addition of a cotton dental roll and a small slice of Agaricus bisporus mushroom for D. neotestacea. L. heterotoma (strain Lh14) is a generalist endoparasitic wasp that successfully parasitizes many species of Drosophila and L. boulardi (Lb17) is a specialist endoparasitic wasp of Drosophila melanogaster and other members of the melanogaster group [66]. These wasp strains were initially collected by Todd Schlenke in Winters, California in 2002 and were provided by the Lemaitre lab and maintained on second instar Oregon-R larvae at 24°C in a vial with a moistened dental cotton roll and white sugar. Pachycrepoideus vindemmiae is a generalist pupal ectoparasitic wasp of many cyclorrhaphous flies [67] and was provided by Dr. Joan Cossentine (Agriculture and Agri-Food Canada) and maintained on Oregon-R pupae at 24°C in vials with a dental cotton roll that had been dipped in a 10% white sugar solution.
To ensure highly efficient vertical transmission of Spiroplasma, parental flies were aged at least five days prior to egg laying. Adults were placed on fresh instant food or mushroom vials for 24 hours then removed. From these vials, three-day-old second-instar larvae were picked and placed on instant food in 35 mm petri dishes. We found that D. neotestacea pupal survival was negatively affected by picking and transplanting them to dishes, often because they preferred to crawl into the narrow lid opening to pupate. For this species we carried out wasp exposures in vials with a small piece of mushroom such that larvae could not burrow deep enough to avoid wasp ovipositors. Leptopilina exposures were done in duplicate or greater with at least 40 fly larvae per dish. Five mated female wasps with experience parasitizing Oregon-R were added to each dish and allowed to oviposit for 48 hours. Pachycrepoideus exposures were also done in duplicate petri dishes with 30 two-day-old fly pupae presented to five experienced female wasps for 48 hours. Wasp success in oviposition was inferred from detection of wasp rRNA by RT-qPCR and was 100% for L. heterotoma and L. boulardi on D. melanogaster, 69% for L. heterotoma on D. neotestacea, and in 62% for P. vindemmiae on D. melanogaster.
For timecourse experiments, the Leptopilina protocol above was altered to shorten Drosophila egg-laying time to eight hours as well as wasp exposure time to three hours in an effort to better synchronize both fly and wasp development across samples. For the L. heterotoma timecourse in D. melanogaster, 30 second-instar fly larvae were moved into a 35 mm petri dish, as described above, for each of two replicate dishes per treatment. Despite briefer exposures, parasitism rates were high; wasp rRNA was detected in 69 of 72 wasp-exposed flies.
We performed an additional exposure of D. melanogaster to L. heterotoma in order to determine the timeframe of wasp hatching within Spiroplasma-positive fly larvae. As done previously, 40 second-instar larvae were exposed to five experienced L. heterotoma females for three hours in a dish, afterward the wasps were removed. 48 h after the start of wasp exposure, 20 larvae were removed and dissected under a Leica MZ6 dissecting light microscope and each was visually scored for the presence of a hatched or unhatched wasp. At 72 h post-exposure, dissections were again carried out, on all of the remaining live larvae (n = 11), and larvae were scored for hatched or unhatched wasps.
Following wasp exposures, Drosophila pupae were collected for RNA extraction. For Leptopilina exposures, we collected first- and second-day-old pupae, four pupae from each of two replicate dishes or vials per treatment. For the L. heterotoma in D. melanogaster timecourse, three larvae or pupae were collected from each replicate dish (six per treatment) at the each of the following time points post-exposure: 3 h, 24 h, 48 h, 72 h, 96 h, and 120 h. Flies were larvae at the first four collection points, and were pupae at 96 h and 120 h.
RNA was extracted from all samples in the same way regardless of fly species or age: Flies were homogenized for 10–15 seconds with a bead-beater in a 1.5 mL microfuge tube with 300 uL TRIzol reagent (Invitrogen) and approximately ten 1 mm zirconia beads (BioSpec Products). TRIzol extractions were carried out according to the manufacturer’s recommended protocol. RNA was quantified on a NanoDrop spectrophotometer. 1 μg or, for some of the smallest larvae extracted, 0.5 μg, was used for cDNA synthesis immediately following quantification. cDNA synthesis reactions used SuperScript II reverse transcriptase (Invitrogen) and were each primed with 50 ng of random hexamer primers (Integrated DNA Technologies). RT-qPCR reactions were done in duplicate with SYBR Select Master Mix (Applied Biosystems) in a Biorad C1000 Touch Thermal Cycler with a CFX96 Real-Time System interfaced with CFX Manager 3.0 software. Ct values across replicates did not vary by more than 0.5 Ct. For timecourse reactions, all reactions for one primer set could not be amplified in a single plate so an interplate calibrator of pooled samples was used to normalize samples. Primers used throughout this study and the details of their validation are provided in S1 Table.
RIPs are N-glycosidases that target an adenine residue in the α-sarcin/ricin loop of eukaryotic 28S rRNA. We used an established RT-qPCR based assay to quantify RIP activity [40,42,43]. In brief, the N-glycosidase activity of RIPs leaves an abasic site at this adenine, and when reverse transcriptase encounters this position during cDNA synthesis, it incorporates a deoxyadenosine monophosphate (dAMP) into the nascent strand, while complementary DNA (cDNA) constructed from intact templates will receive a thymidine monophosphate (TMP). For detection of these targets by RT-qPCR, forward primers were designed such that the 3’-most primer position complements intact or depurinated variants of the RIP-targeted position, and a secondary mismatch was introduced at an adjacent position for both primer sets to improve specificity. The reverse primer targets a nearby region of 28S rRNA exhibiting high sequence divergence between Leptopilina/Pachycrepoideus and Drosophila. Primer sequences and further details of their validation are presented in S1 Table. We tested the specificity of this assay on IDT gBlocks synthetic DNA with either A or T residues at the RIP-targeted site. We found it to be highly-sensitive to detect this single nucleotide base change, i.e. cross-template amplification by our primers does not occur until 17–20 Ct (cycle threshold) values after the intended target, meaning fold-changes of 130,000–1,000,000 are the upper limit of detection for this assay. A product for each of the RT-qPCR targets was Sanger sequenced to confirm product identity and following each reaction, melt curves were examined to confirm that melting temperature matched the expectation for each product.
Fold changes were calculated by the Pfaffl ratio method [68]. Here ΔCt is calculated by subtracting the Ct of each treated sample from an untreated sample. We used the global mean Ct for each primer set in place of the untreated sample which facilitates the plotting of untreated (S-) data. Primer efficiency (E) was incorporated as 1+EΔCt for each primer and ratio of gene of interest to reference was calculated as usual. The reference target used to normalize intact and depurinated targets was a nearby region of the 28S rRNA. The reference gene used to normalize RIP transcript abundance was the Spiroplasma DNA-directed RNA polymerase subunit B (rpoB). Data were graphed in R version 3.3.3 using the ggplot2 package.
To quantify relative changes in RIP transcript abundance, we performed RT-qPCR on cDNA samples generated from unparasitized host larvae during time course experiments. We normalized RIP expression in each of six biological replicates per time point to the corresponding rpoB reference transcript data and calculated –ΔCt values to compare RIP transcript abundances to one another. We analyzed the results with ANOVAs to test for significant effects of time and RIP copy. We note that in sMel, RIPs 3,4, and 5 are nearly identical at the nucleotide level, probably the result of recent gene duplication. As a result, our primers do not differentiate between their transcripts and we refer to them collectively as sMel RIPs3-5.
To test whether fly hemolymph is enriched for depurinated host ribosomes, third instar D. neotestacea larvae were rinsed in Ringer’s solution and anesthetized on ice. Eight larvae were bled into 500 μL of ice cold Ringer’s solution for ten minutes after piercing the cuticle with a needle. This was done in separate 35 mm petri dishes to generate biological replicate pools. Larval carcasses were collected and stored in Ringer’s solution on ice while hemolymph solutions were centrifuged for 10 minutes at 3,000 rpm at 4°C to pellet hemocytes. Pellets were observed in each tube and the supernatant (hemolymph) fraction was transferred to a separate tube. The pellet was rinsed with cold Ringer’s solution and centrifuged briefly to remove residual supernatant in the tube. RNA was extracted from all fractions as well as pools of unbled control larvae using TRIzol. cDNA synthesis was done using 100 ng of input RNA per sample and RIP assays were carried out as described previously. This experiment was done once initially with two biological replicate pools per tissue type and repeated with four more biological replicate pools per tissue type.
Because sNeo does not grow in culture, we prepared sNeo genomic DNA extractions from host tissues enriched for Spiroplasma. We dissected ovaries from approximately 40 three-week-old D. neotestacea flies and extracted DNA using the phenol-chloroform method. We sequenced the sNeo genome using Pacific Biosciences RSII and Illumina HiSeq 2500 sequencing technologies (Genome Quebec). We sequenced long reads on three SMRT cells and short reads in 0.4 of a sequencing lane. A preliminary sNeo genome was assembled by Genome Quebec using long reads only via the SMRT Pipeline, with an estimated genome size of 1.8 Mb. The assembly contains Drosophila-derived sequences, decontaminating and error-correcting these data is an ongoing project. We searched for RIP genes within this preliminary assembly by tblastn using the sMel RIPs and sNeo-RIP1 as query sequences and confirmed the presence and sequence of each with PCR and Sanger sequencing.
Phylogenetic analysis was performed on Spiroplasma RIP sequences from sNeo, sMel, and other sequenced Spiroplasma genomes, which were identified by blastp and tblastn against GenBank’s non-redundant protein and genomic sequence databases. Spiroplasma matches with expected value less than 10−5 were retained. Amino acid sequences were aligned by MAFFT under the E-INS-i alignment algorithm. The putative domain divisions shown in Fig 4 were determined from this alignment. The best protein substitution model was selected using ProtTest 2.4 [69]. The selected model under Bayesian information criterion was WAG+I+G+F. A phylogram was constructed and SH-like approximate likelihood-ratio scores calculated with PhyML 3.0 [70] implemented by SeaView 4.5.4 [71].
|
10.1371/journal.pgen.1000623 | Genome-Wide Association Study Implicates Chromosome 9q21.31 as a Susceptibility Locus for Asthma in Mexican Children | Many candidate genes have been studied for asthma, but replication has varied. Novel candidate genes have been identified for various complex diseases using genome-wide association studies (GWASs). We conducted a GWAS in 492 Mexican children with asthma, predominantly atopic by skin prick test, and their parents using the Illumina HumanHap 550 K BeadChip to identify novel genetic variation for childhood asthma. The 520,767 autosomal single nucleotide polymorphisms (SNPs) passing quality control were tested for association with childhood asthma using log-linear regression with a log-additive risk model. Eleven of the most significantly associated GWAS SNPs were tested for replication in an independent study of 177 Mexican case–parent trios with childhood-onset asthma and atopy using log-linear analysis. The chromosome 9q21.31 SNP rs2378383 (p = 7.10×10−6 in the GWAS), located upstream of transducin-like enhancer of split 4 (TLE4), gave a p-value of 0.03 and the same direction and magnitude of association in the replication study (combined p = 6.79×10−7). Ancestry analysis on chromosome 9q supported an inverse association between the rs2378383 minor allele (G) and childhood asthma. This work identifies chromosome 9q21.31 as a novel susceptibility locus for childhood asthma in Mexicans. Further, analysis of genome-wide expression data in 51 human tissues from the Novartis Research Foundation showed that median GWAS significance levels for SNPs in genes expressed in the lung differed most significantly from genes not expressed in the lung when compared to 50 other tissues, supporting the biological plausibility of our overall GWAS findings and the multigenic etiology of childhood asthma.
| Asthma is a leading chronic childhood disease with a presumed strong genetic component, but no genes have been definitely shown to influence asthma development. Few genetic studies of asthma have included Hispanic populations. Here, we conducted a genome-wide association study of asthma in 492 Mexican children with asthma, predominantly atopic by skin prick test, and their parents to identify novel genetic variation for childhood asthma. We implicated several polymorphisms in or near TLE4 on chromosome 9q21.31 (a novel candidate region for childhood asthma) and replicated one polymorphism in an independent study of childhood-onset asthmatics with atopy and their parents of Mexican ethnicity. Hispanics have differing proportions of Native American, European, and African ancestries, and we found less Native American ancestry than expected at chromosome 9q21.31. This suggests that chromosome 9q21.31 may underlie ethnic differences in childhood asthma and that future replication would be most effective in populations with Native American ancestry. Analysis of publicly available genome-wide expression data revealed that association signals in genes expressed in the lung differed most significantly from genes not expressed in the lung when compared to 50 other tissues, supporting the biological plausibility of the overall GWAS findings and the multigenic etiology of asthma.
| Asthma (OMIM 600807) is a leading chronic childhood disease with prevalence rates reaching a historically high level (8.9%) in the United States and continuing to increase in many countries worldwide [1],[2]. Asthma is characterized by airway inflammation and bronchoconstriction leading to airflow obstruction, but the mechanisms leading to asthma development remain unknown. Genetic risk factors likely play a central role in asthma development. Twin studies support a strong genetic component to asthma (especially childhood asthma) with heritability estimates suggesting that 48–79% of asthma risk is attributable to genetic risk factors [3]. In an effort to localize disease susceptibility genes, genome-wide linkage studies have identified at least 20 linkage regions potentially harboring disease genes [4], and over 100 positional and biological candidate genes have been tested for association with asthma [3]. However, no genes have been definitely shown to influence this complex disease.
Genome-wide association studies (GWASs) have emerged as a powerful approach for identifying novel candidate genes for common, complex diseases. In the first asthma GWAS, using 307,328 single nucleotide polymorphisms (SNPs), Moffatt et al. found highly statistically significant associations of SNPs in adjacent genes ORM1-like (S. cerevisiae) (ORMDL3; OMIM 610075) and gasdermin B (GSDMB or GSDML; OMIM 611221) with risk of childhood asthma in German and British populations [5]. Meta-analysis of the Moffatt et al. study and five subsequent replication studies, including our own study, supports the association of ORMDL3 and GSDML SNPs with risk for childhood asthma across various populations [6]. More recently, using 518,230 SNPs, Himes et al. implicated SNPs in phosphodiesterase 4D, cAMP-specific (phosphodiesterase E3 dunce homolog, Drosophila) (PDE4D; OMIM 600129) with risk of asthma in whites from the United States and replicated this finding in two other white populations [7]. Using only 97,112 SNPs, Choudhry et al. implicated chromosome 5q23 SNPs for association with asthma in Puerto Ricans [8], but no other Puerto Rican cohorts are available for replication.
Few genetic studies of asthma have included Hispanic populations, and replication of positive genetic findings is scarce across Hispanic groups. Hispanics have differing proportions of Native American, European, and African ancestries. This admixture introduces special considerations (such as controlling for population stratification in association studies), but admixture in Hispanic populations also provides a unique opportunity to use ancestry analysis to evaluate our genetic association findings [9],[10].
Mexico City is one of the most polluted cities in the world, and its inhabitants experience chronic ozone exposure, which has been linked to asthma development in children and adults and asthma exacerbations [11]–[13]. We conducted a GWAS to identify novel candidate susceptibility genes associated with childhood asthma in case-parent trios from Mexico City and tested the most significantly associated SNPs in an independent study of trios of Mexican ethnicity. GWAS findings were then examined in the context of ancestry analysis and genome-wide expression data to provide supportive evidence for associations with childhood asthma.
In the GWAS, there were more male (58.7%) than female (41.3%) children with asthma, and the mean age at enrollment was 9.0±2.4 years (Table 1). The replication study had a similar distribution of males (59.7%) to females (40.3%), but there was an older mean age at enrollment distribution of 13.4±5.4 years. All GWAS and replication study subjects with asthma were clinically diagnosed before age 18.
Additional demographic and clinical characteristics of the 492 children with asthma evaluated in the GWAS are presented in Table 1. More children had mild asthma (72.3%) than moderate to severe asthma (27.7%). Among the 445 children with skin prick test data, nearly all (91.7%) were atopic based on having a positive skin test to at least one aeroallergen. Few mothers reported smoking during pregnancy (4.8%), but at least one parent currently smoked in about half of the families. The median annual average of the daily maximum 8 hour averages, a measure of residential ambient ozone exposure, was 67 parts-per-billion (ppb). The residence of 48.2% of the children had a low ambient ozone exposure (≤67 ppb), and the residence of 51.8% of the children had a high exposure (>67 ppb).
The 520,767 autosomal SNPs passing quality control were tested for association with childhood asthma using additive modeling with the log-linear method in 492 children with asthma and their biological parents from Mexico City. Not surprisingly given the study size, no SNP met genome-wide significance using a conservative Bonferroni adjustment. Nonetheless, the comparison of observed and expected p-values in the quantile-quantile plot (Figure 1) shows several top SNPs with some deviation from expectation. These deviations may occur by chance or may represent a true excess of small p-values.
Figure 2 shows the observed p-values plotted against chromosomal location. An intergenic SNP on chromosome 16 had the most significant association with childhood asthma [rs1867612 (p = 1.55×10−6)], followed by an intronic SNP in potassium voltage-gated channel, Shab-related subfamily, member 2 (KCNB2; OMIM 607738) on chromosome 8 [rs2247572 (p = 1.94×10−6)] and two intergenic SNPs on chromosome 20 [rs6063725 (p = 3.52×10−6)] and rs720810 (p = 5.13×10−6)] with only moderate linkage disequilibrium (LD) (r2 = 0.59). The next most significant SNP (rs2378383) highlights a cluster of SNPs on chromosome 9q21.31 ranking among the top GWAS SNPs. This cluster of SNPs spans transducin-like enhancer of split 4 (E(sp1) homolog, Drosophila) (TLE4; OMIM 605132) and its upstream region. LD analysis of SNPs with p≤0.001 in this cluster shows two large LD blocks in this region with one block encompassing TLE4 and the other block encompassing the upstream region (Figure S1).
Eleven of the 18 most significantly associated SNPs met our criteria to be selected for replication in 177 case-parent trios of Mexican ethnicity from the Genetics of Asthma in Latino Americans (GALA) study [14]. The GWAS p-values for the 11 SNPs selected for replication testing ranged from 3.30×10−5 to 1.55×10−6 (Figure 2). There were no significant deviations in Hardy-Weinberg equilibrium (HWE) for the replication SNPs in either the GWAS or replication study (p>0.12), and minor allele frequencies (MAFs) were similar between the two studies (Table 2). The replication study had at least 70% power to detect an association with four SNPs (rs2378377, rs4674039, rs1830206, and rs3814593) and at least 80% power to detect an association with the remaining seven SNPs (rs1867612, rs2247572, rs6063725, rs720810, rs2378383, rs6951506, and rs3734083) for similar MAFs and relative risk (RR) estimates observed in the GWAS.
Association results in the GWAS and replication studies are compared in Table 2. No SNPs were significant with conservative Bonferroni correction for multiple testing, but two SNPs were associated with childhood asthma in the replication study with a p-value close to 0.05. The chromosome 9q21.31 SNP rs2378383, which is located 147 kb upstream of TLE4 in an intergenic region between coiled-coil-helix-coiled-coil-helix domain containing 9 (CHCHD9) and TLE4, was associated with childhood asthma in the replication study with p = 0.03. Meta-analysis of rs2378383 in the two studies gave a combined p-value of 6.79×10−7, and the RR estimate for carrying one copy of the rs2378383 minor allele (G) compared to carrying no copies in the GWAS [RR, 0.61; 95% confidence interval (CI), 0.49–0.76] was quite similar to the RR estimate in the replication study (RR, 0.63; 95% CI, 0.41–0.96). The SNP rs2378377, a neighboring intergenic SNP in moderate LD with rs2378383 (r2 = 0.73), had a marginal association with childhood asthma in the replication study with p = 0.06 (combined p = 2.68×10−6). RR estimates for the rs2378377 minor allele (G) were also similar between the GWAS (RR, 0.64; 95% CI, 0.53–0.79) and the replication study (RR, 0.71; 95%CI, 0.50–1.02). None of the other nine SNPs were associated in the replication study (Table 2).
Association results from additive modeling for SNPs in the region spanning TLE4 and its upstream region (chromosome 9 nucleotide positions from 81,114,500 to 81,531,500, NCBI build 36.3) were obtained from the previous GWASs of asthma [5],[7],[8]. Our top two SNPs from this region were genotyped only in the GWAS in whites from the United States [7], where they were not associated with asthma (p = 0.59 for rs2378383 and p = 0.65 for rs2378377). Eighty-nine other SNPs were available in this region, and the smallest p-value was observed for rs1328406 (p = 0.056). There were 54 SNPs available in this region from the GWAS in German and British populations [5], with the smallest p-values observed for rs2807312 (p = 0.0041), rs7849719 (p = 0.018), rs7862187 (p = 0.033), rs10491790 (p = 0.043), and rs946808 (p = 0.049). From the 26 SNPs available from the GWAS in Puerto Ricans [8], the smallest p-value was 0.19. In our GWAS in Mexicans, there were several SNPs in TLE4 and its upstream region with small p-values, and the SNPs listed above are located in close proximity to many of our associated SNPs. Similar to our GWAS and replication study, only cases with childhood-onset asthma were included in both GWASs in white populations [5],[7], and the cases from Himes et al. were predominantly atopic (91.2%) as defined by at least one positive skin prick test [7]. In contrast, the GWAS in Puerto Ricans included both childhood-onset and adulthood-onset asthma cases, and 83% of the cases were considered atopic as defined by total IgE>100 IU/mL [8].
Associations of rs2378383 and rs2378377 were examined in data from the GWAS population stratified by residential ambient ozone exposure (199 trios with ≤67 ppb and 214 trios with>67 ppb annual average of the maximum 8 hour averages) and by current parental smoking (253 trios with and 233 trios without current parental smoking). The minor alleles of both SNPs were inversely associated with asthma at p<0.05 in all strata (results not shown), thus giving no evidence for effect modification in the presence of these environmental exposures.
Among the 445 cases with skin test data available, 408 can be classified as atopic by virtue of having at least one positive skin test. We repeated the GWA scan in this subset of 408 trios. Chromosome 9q21.31 SNPs predominated among the top ranked SNPs, with rs2378383 (p = 7.18×10−7) and rs2378377 (p = 1.08×10−6) being the two top ranking SNPs. In addition to smaller p-values, the magnitudes of association with asthma were slightly stronger for rs2378383 (RR, 0.54; 95% CI, 0.42–0.69) and rs2378377 (RR, 0.54; 95% CI, 0.42–0.72) in the subset of trios where the asthmatic child is also known to be atopic.
The chromosome 9q21.31 SNPs rs2378383 and rs2378377 were tested for association with the number of positive skin tests as a quantitative measure of the degree of atopy in the trios with skin test data. Both SNPs were associated with degree of atopy (p = 0.0018 for rs2378383 and p = 0.0010 for rs2378377). Their RR estimates indicate an inverse association in which carrying one copy of the minor allele is associated with a decreasing number of positive skin tests (RR, 0.92; 95% CI, 0.87–0.97 for rs2378383 and RR, 0.92; 95% CI, 0.88–0.97 for rs2378377), consistent with the direction of association for asthma.
The Mexican subjects in the GWAS had mean ancestral proportions of 69.5±15.6% for Native American, 27.3±14.3% for European, and 3.2±3.0% for African ancestries. Given the predominance of Native American ancestry, we evaluated Native American transmission in the GWAS along the chromosomal arm (9q) containing the replicated SNP (rs2378383) by ancestry analysis. As shown in Figure 3, there is a significant under-transmission of Native American ancestry at rs2378383 (z-score = −2.21 and two-sided p = 0.028) and surrounding SNPs.
The deficiency in Native American ancestry at this locus suggests that a protective allele occurred at a higher frequency in the Native American ancestral population than in the European and African ancestral populations. An examination of this SNP in the HapMap and Human Genome Diversity Panel (HGDP) data reveals that the frequency of the G allele is generally low in European, African, and East Asian populations (0.125 in HapMap European, 0.033 in HapMap African, 0.122 in HapMap Chinese, and 0.273 in HapMap Japanese), while its frequency is much higher in Native American populations (0.57 in HGDP Pima and 0.36 in HGDP Mayan). This pattern suggests that the G allele may be tagging a protective allele in the Native American ancestral population. This conclusion is consistent with the finding that the G allele is associated with a decreased risk for childhood asthma in the GWAS and replication analyses (Table 2).
We examined gene expression patterns in 51 diverse human tissues in the context of GWAS findings to determine whether genes expressed in asthma relevant tissues ranked higher than genes not expressed in such tissues and thus to assess the biological plausibility of our overall GWAS findings. These results are presented in Figure 4. In the 14,330 genes with GWAS SNPs in the gene or nearby, median false discovery rate q-values (derived from the GWAS p-values) were compared between genes expressed versus genes not expressed in each of 51 human tissues.
Among the 51 tissues, the most significant deviation between the median q-values was found between 3,618 genes expressed in the lung versus 10,712 genes not expressed in the lung (Figure 4, p = 0.00025). This finding remains significant even after a conservative Bonferroni correction for multiple testing. The q-values for the 3,618 lung-expressed genes are presented in Table S1. TLE4 did not contribute to the observation of significantly lower GWAS q-values in lung-expressed genes, as TLE4 was classified as not expressed. TLE4 displays a nearly ubiquitous expression pattern with similar low intensity levels across many tissues, so even though it is present in the lung, its expression level in the lung did not exceed our 75th percentile threshold to be classified as expressed.
Other tissues in the respiratory and immune system also showed deviations in the median GWAS q-values for expressed versus not expressed genes, including thymus and lymph node at the p<0.01 level (uncorrected) and fetal lung, trachea, tonsil, smooth muscle, and bronchial epithelium at the p<0.05 level (uncorrected). Thus, SNPs in genes more highly expressed in tissues related to the pathogenesis of asthma and allergies tend to give more significant GWAS results than genes more highly expressed in other tissues. These results give biological credibility to our overall GWAS findings and are consistent with the multigenic etiology of asthma.
A two-dimensional cluster analysis was conducted to identify the implicated tissues with correlated gene expression patterns. As shown in Figure S2, lung tissue is grouped in a cluster with fetal lung and placenta tissues, thus suggesting that gene expression patterns in lung are most similar to gene expression patterns in fetal lung and placenta and that their signals are correlated. There are 2,385 genes classified as expressed in all three tissues – lung, fetal lung, and placenta. The gene expression patterns of other implicated tissues are also highly correlated, including the immune tissues tonsil, lymph node, and thymus (Figure S2).
Genetic studies of asthma are few in Hispanic populations, and to our knowledge, this work presents the first asthma GWAS in Mexicans and the most extensive coverage of genetic variation for an asthma GWAS in any Hispanic population. The GWAS included 492 Mexican case-parent trios. Given the moderate GWAS sample size, no SNP met genome-wide significance. However, the ranking of GWAS SNPs highlighted a potentially important candidate region for childhood asthma susceptibility, chromosome 9q21.31. Several chromosome 9q21.31 SNPs with small GWAS p-values were located in TLE4 and its upstream region, and two of these SNPs (rs2378383 and rs2378377) were tested for replication in an independent study of 177 case-parent trios of Mexican ethnicity. Despite the small sample size for replication, both SNPs gave p-values close to 0.05 and the same direction and magnitude of association as the GWAS. Neither rs2378383 nor rs2378377 have a known impact on TLE4 expression, but given their location upstream of TLE4, it is possible that these SNPs reside in a TLE4 regulatory region.
Ancestry analysis in this chromosomal region provided supportive evidence that rs2378383 (G) is associated with a decreased risk of childhood asthma in Mexicans. Ancestry and transmission-based association analyses provide complementary but not completely independent lines of evidence. At each SNP, the log-linear method only used parents who were heterozygous in genotype, while ancestry analysis used all parents who are heterozygous in ancestry, including parents who are homozygous in genotype. We did not a priori expect that ancestry analysis results would corroborate log-linear association results.
Our ancestry analysis uses the same principles that underlie admixture mapping and relies on the key assumption of different risk allele frequencies between the ancestral populations, primarily Native American and European in this study. Under this assumption, individuals with disease in the admixed population would be expected to share an excess of ancestry from the population with the highest risk allele frequency at the disease locus [9]. In contrast, individuals with disease in the admixed population would be expected to share a shortage of ancestry from the population with the highest protective allele frequency at the disease locus. At chromosome 9q21.31, there was less Native American ancestry than expected, suggesting that the Native American ancestral population had a higher frequency of the protective rs2378383 allele (G).
Ancestry analysis implicated chromosome 9q21.31 as a chromosomal region that may underlie ethnic differences in childhood asthma. Complex diseases with differing disease prevalence rates in the ancestral populations are most suitable for ancestry analysis [15]. Prevalence rates of childhood asthma in the true ancestral Native American, European, and African populations are unknowable, but it is interesting to note that Mexicans have the highest Native American ancestry and the lowest asthma prevalence rate among Hispanic populations [16]. Differing frequencies of genetic risk factors in the ancestral populations presumably contribute to the differing prevalence rates of childhood asthma in modern populations. Our study found an association between Native American ancestry and a lower disease risk. Similarly, Native American ancestry was associated with milder asthma in a previous study of subjects of Mexican ethnicity from the GALA study [17]. These findings collectively suggest that the Native American ancestral population had higher frequencies of alleles that decrease prevalence and severity of asthma in the modern Mexican population. A comparison of asthma prevalence and severity among modern Native Americans, Europeans, and Africans would further support this interpretation, but such data are scarce [17].
The evidence for locus-specific ancestry around rs2378383 has implications for replication. Because rs2378383 (G) occurs at relatively low frequency in European, African and East Asian populations, genetic association studies in these populations are likely to suffer from lack of power at this locus. In contrast, the G allele occurs at moderate frequency in the Native American populations surveyed in HGDP [18]. Such disparate allele frequencies facilitate ancestry analysis in the region and improve the statistical power of transmission-based tests, as there are many more heterozygous parents in the Mexican population than a European, African or East Asian population. In fact, we obtained association results for SNPs in the chromosome 9q21.31 region from previous GWASs and found that SNPs in this region had only nominal evidence for association with asthma in the GWASs in white populations [5],[7]. It is not surprising that substantial evidence for replication was not found given the ethnicity differences (whites for Moffatt et al. and Himes et al. [5],[7] and Puerto Ricans for Choudhry et al. [8]). Future replication and fine-mapping of the region would be most effective if performed in Native American populations, or admixed populations with high Native American ancestral contributions.
The chromosome 9q21.31 SNPs associated with childhood asthma in the GWAS map to TLE4 and its upstream region. The TLE family of proteins in humans is homologous to the Drosophila Groucho protein, which participates in cell fate determination for neurogenesis and segmentation. The highly conserved structure among the Drosophila Groucho and human TLE gene products suggest similar functions as transcriptional regulators in cell fate determination and differentiation [19]. Six genes encode the TLE family of proteins in humans (TLE1, TLE2, TLE3, TLE4, TLE5, TLE6), as deposited in the NCBI database. The distinct expression patterns among the TLE genes suggest a complex mechanism in humans involving non-redundant roles for the TLE genes [20]. The TLE4 gene, in particular, shows ubiquitous expression across many tissues [19],[21], and TLE4 functions as a transcriptional co-repressor in several key developmental pathways [22]. More specifically, TLE4 has been implicated in early B-cell differentiation. TLE4 interacts with the transcription factor Paired box 5 (PAX5; OMIM 167414), which activates B-cell specific genes and represses alternative lineage fates [23]. A spliced version of TLE4 acts as a negative regulator for the PAX5/TLE4 function [23]. An alteration of B-cell differentiation involving TLE4 could be relevant to immune development and thus asthma.
TLE interacts with Runt-related transcription factor 3 (RUNX3; OMIM 600210) in a manner that may be directly relevant to asthma. In mice, loss of RUNX3 function results in an allergic asthma phenotype due to accelerated dendritic cell maturation and resulting increased efficacy to stimulate T cells [24]. Interaction with TLE is required for RUNX3 to inhibit dendritic cell maturation [25]. A recent paper provides support for the interaction of RUNX3 specifically with TLE4 [26]. Interestingly, the chromosome 9q21.31 SNPs rs2378383 and rs2378377 near TLE4 are associated with asthma as well as degree of atopy in our data, and their associations with asthma became more pronounced when considering only the asthmatic children with atopy and their parents. These findings suggest that the influence of TLE4 on asthma may be related to its influence on immune system development.
Childhood asthma is a complex disease, and there are likely many susceptibility genes influencing immune system development and asthma in the Mexican population. The examination of GWAS in the context of genome-wide expression illustrated the biological plausibility of our GWAS findings and showed consistency with the involvement of multiple genes. Genes expressed in the lung show association signals that differ most significantly from the association signals from genes not expressed in the lung when compared to 50 other human tissues. The lung represents a major pathogenic site for asthma, and this finding implies that multiple genes expressed in the lung are collectively associated with an increased risk of childhood asthma. Some of the other implicated tissues may represent false positives, but several of the highlighted tissues are biologically plausible for childhood asthma, including trachea, bronchial epithelium, smooth muscle, and immune tissues such as thymus, tonsil, and lymph node.
Other GWASs have implicated different susceptibility loci. Several SNPs implicated in the first asthma GWAS by Moffatt et al. in the ORMDL3 region [5] were associated with childhood asthma in our GWAS [including rs9303277 (p = 0.036), rs11557467 (p = 0.014), rs8067378 (p = 0.020), rs2290400 (p = 0.037), and rs7216389 (p = 0.042)] but were not ranked among our top 5,000 SNPs. More recent GWASs have implicated loci other than ORMDL3. The PDE4D SNPs implicated by Himes et al. [7] were not associated with childhood asthma in our GWAS at p<0.05. Two nearby SNPs, not in LD with the implicated SNPs, were associated [rs13158277 (p = 0.030) and rs7717864 (p = 0.015)] but were also not ranked among our top 5,000 SNPs. Chromosome 5q23 SNPs implicated by Choudhry et al. [8] were not associated with childhood asthma in our GWAS at p<0.05. Initial GWAS findings regarded as replicated may not be ranked among the front runners in a genome-wide scan in the replication populations for statistical [27] as well as other biological reasons (such as ethnic differences, phenotypic heterogeneity, genetic heterogeneity, differing patterns of interacting environmental exposures, or multigenic etiology). This trend in discordant GWAS findings is quite common for various complex diseases [28], and follow-up studies are crucial in separating true genetic associations from false positives.
The major limitation of this study is the sample size for the GWAS and replication study. The Mexican population is largely under-studied given its size, and only moderate sample sizes are currently available for the study of asthma genetics. In our study, no SNPs met genome-wide significance, and no replication SNPs met the significance threshold when using a conservative Bonferroni correction for multiple testing. Despite this limitation, top GWAS findings, replication in an independent population, and ancestry analysis taken together implicate a novel region for association with asthma in Mexican children.
This study has several strengths. The case-parent trio design and the log-linear analysis protects against bias due to population stratification [29], so our GWAS results are not confounded by population stratification in this admixed population. Also, disease misclassification is minimal. Although bronchial hyper-reactivity was not tested, children with asthma were given reliable diagnoses based on clinical grounds by pediatric allergists at a pediatric allergy specialty clinic. The allergy clinic is a tertiary referral clinic, so the children with asthma were previously seen by a generalist and a pediatrician over time for recurrent asthma symptoms. Physician diagnosis of asthma has been shown to have a high level of validity in children after the first few years of life [30]. Further, the asthmatic children were predominantly atopic to aeroallergens based on skin prick testing limiting heterogeneity of the disease phenotype.
The GWAS and replication association results and the supporting ancestry analysis implicate chromosome 9q21.31 as a novel susceptibility locus for childhood asthma in the Mexican population. This region contains a biologically plausible novel susceptibility gene for childhood asthma, TLE4, but further work is needed to decipher whether TLE4 or a nearby gene explain the signals from the chromosome 9q21.31 region. Further, childhood asthma is a complex disease with a proposed multigenic etiology, but most single studies will not have sufficient power to examine such complex relationships. Identification of important interacting risk factors in childhood asthma and other complex diseases will require very large sample sizes. This work identifies chromosome 9q21.31 (including TLE4) as a novel candidate susceptibility locus for childhood asthma, suggests that this region may underlie ethnic differences in childhood asthma, and emphasizes the presence of multiple genetic risk factors in the complex mechanism leading to childhood asthma.
The study protocol was approved by the Institutional Review Boards of the Mexican National Institute of Public Health, Hospital Infantil de Mexico Frederico Gomez, and the US National Institute of Environmental Health Sciences (NIEHS). Parents gave written informed consent for the children's participation, and children gave their assent.
Children with asthma (aged 5–17) and their biological parents were recruited between June 1998 and November 2003 from a pediatric allergy specialty clinic at a large public hospital in central Mexico City, Hospital Infantil de Mexico Frederico Gomez. The case-parent trio design protects against bias due to population stratification in this admixed population [29],[31]. Blood samples were collected from enrolled children and their parents for DNA extraction.
Children were diagnosed with asthma by a pediatric allergist at the referral clinic based on clinical symptoms and response to treatment [32]. Asthma severity was rated as mild (intermittent or persistent), moderate, or severe by the pediatric allergist according to symptoms in the Global Initiative on Asthma schema [33]. Questionnaires on the children's asthma symptoms and risk factors, including environmental tobacco smoking, were completed by parents, nearly always the mother.
The clinical evaluation also included skin prick testing to measure atopy and pulmonary function testing, as previously described [6]. A battery of 24 aeroallergens common in Mexico City was used for skin prick testing. Histamine was used as a positive control, and the test was considered valid if the histamine reaction was 6 mm or greater [34]. Glycerin was used as a negative control. Children were considered atopic if the diameter of skin reaction to at least one allergen exceeded 4 mm. Pulmonary function testing was performed at a later date using the EasyOne spirometer (ndd Medical Technologies, Andover, Massachusetts) according to American Thoracic Society guidelines [35]. Children were asked to withhold asthma medications on the morning of the test. The best test of three technically acceptable tests was selected. Percent predicted forced expiratory volume in 1 second (FEV1) was calculated using spirometric prediction equations from a childhood population in Mexico City [36].
Measurements of ambient ozone were obtained from the Mexican government's air monitoring station closest to each child's residence (within 5 km). The annual average of the daily maximum 8 hour averages of the ozone level was collected for the year prior to study entry and dichotomized at the median for stratified analyses. Further details on the ozone measurement protocol have been described elsewhere [6].
Peripheral blood lymphocytes were isolated from whole blood, and DNA was extracted using Gentra Puregene kits (Gentra Systems, Minneapolis, Minnesota). A total of 498 complete case-parent trios with previously confirmed parentage and sufficient amounts of DNA were genotyped for 561,466 SNPs using the Illumina HumanHap 550 K BeadChip, version 3 (Illumina, San Diego, California) at the University of Washington, Department of Genome Sciences. Genotypes were determined using the Illumina BeadStudio Genotyping Module, following the recommended conditions. Results for three unrelated study subjects fell below the genotyping call rate threshold of 95% resulting in exclusion of three trios. The remaining study subjects were genotyped with an average call rate of 99.7%.
Quality control analyses were conducted using PLINK (http://pngu.mgh.harvard.edu/~purcell/plink/) [37], unless otherwise stated. In preliminary SNP-level quality control, SNPs were excluded due to poor chromosomal mapping (N = 173), missingness>10% (N = 988), MAF<0.001% (N = 253), HWE p-value (in parents only)<1×10−10 (N = 557), Mendelian errors in more than two families (N = 4,945), and heterozygous genotype calls for chromosome X SNPs in more than one male (N = 380). All SNP exclusions were made sequentially in the above order.
Subject-level quality control verified that no subjects had unusual autosomal homozygosity or an inconsistent sex between genotype and collected phenotype data. Subject-level quality control next assessed subject relatedness to identify unknown intra- and inter-family relationships. This identified two duplicated trios and one trio with first-degree relative parents requiring exclusion. Trio exclusions were not necessary for other identified relationships, including parents in different trios being first-degree or second-degree relatives and nuclear families with two children with asthma being split into two case-parent trios. There were 492 complete case-parent trios (1,468 study subjects) in the final analysis data set.
Final SNP-level quality control made exclusions due to one or more discordant genotypes across 14 HapMap replicate samples identified using the Genotyping Library and Utilities application (http://code.google.com/p/glu-genetics/) (N = 921) [38]. Final SNP exclusions were also made due to more stringent missingness and MAF thresholds, missingness>5% (N = 3,137) and MAF<1% (N = 16,696). Of the 533,416 SNPs passing all quality control criteria (95.0%), the 520,767 autosomal SNPs were analyzed for purposes of this study.
Subjects of Mexican ethnicity from the GALA study were used for replication testing. The GALA study protocol has been described elsewhere [14]. Subjects of Mexican ethnicity with physician-diagnosed asthma and presence of two or more asthma symptoms in the past two years (wheezing, coughing, and shortness of breath) were enrolled along with both biological parents in San Francisco, California, US, and Mexico City, Mexico. Total plasma IgE was measured for all subjects with asthma.
The children with asthma in the GWAS (less than 18 years old) were enrolled at a pediatric allergy clinic, so nearly all had allergic asthma. To maximize comparability with the GWAS, the replication study included only the 177 complete case-parent trios comprising subjects of Mexican ethnicity having childhood-onset (age of onset<18 years) asthma and atopy (total IgE>100 IU/mL) and their parents.
SNPs were ranked by GWAS p-value. The top SNP and 10 other top ranking GWAS SNPs were tested for replication in the GALA study. The highest ranking SNP along with 10 other top ranking SNPs with no strong LD (r2<0.9) with other higher-ranked SNPs and MAF>10% were selected for replication. Statistical power to detect associations in the replication study of 177 case-parent trios was calculated for each selected SNP using QUANTO (http://hydra.usc.edu/gxe) [39]. The MAFs and RR estimates observed from the GWAS with a log-additive model were specified in the power calculation for each selected SNP.
Genotyping for replication SNPs was performed on the Applied Biosystems (ABI, Foster City, California) PRISM 7500 Real-Time PCR System using primers and probes from ABI's Assay-by-Demand. The assay was performed under universal conditions, with each reaction containing 3.75 ng DNA, 0.125 µL 40X Assay Mix and 2.5 µL TaqMan Universal PCR master mix brought to a final volume of 5 µL with sterile water. Thermal cycling conditions began at 95°C for 10 minutes and then proceeded with 60 cycles of 92°C for 15 seconds and 60°C for 2 minutes. After the PCR reaction, plates were scanned by the ABI PRISM 7500 PCR system to determine genotypes by allelic discrimination.
The log-linear likelihood approach was used to examine associations of individual SNPs with childhood asthma in the GWAS and replication study [29],[40]. The log-linear method is a generalization of the classic family-based test for association between genetic variants and disease, the transmission disequilibrium test [41], which compares the distribution of alleles transmitted from parents to affected offspring with the distribution of alleles not transmitted. An asymmetry of allele distributions implies that the variant under study is associated with disease within families. This inference requires the assumption of Mendelian inheritance, such that the allele under study is not related to the parents' fertility or to the offspring's survival [40]. The log-linear method and other transmission-based methods test the same null hypothesis of no linkage or no association (i.e. no LD) between the allele and disease [40]. Unlike other transmission-based methods, the log-linear method provides risk estimates to assess the direction and magnitude of association.
Robustness to population stratification is a well-known property of the case-parent trio design and transmission-based methods. The log-linear method achieves robustness to population stratification by stratifying on the six possible parental mating types, which are defined by the number of copies of the allele carried by each of the two parents [40]. The assumption of HWE is not required for the log-linear method, but we tested for HWE in parents as a check for genotyping error.
The log-linear method was implemented using the LEM computer program [42] with a one degree-of-freedom log-additive risk model specified. When missing genotypes occur, the log-linear method uses the expectation-maximization algorithm to maximize the likelihood, allowing incomplete trios to contribute information and minimizing loss of statistical power [31]. P-values were generated to assess statistical significance, and the RR of carrying one copy of the risk allele was calculated to assess the direction and magnitude of association. For the most significantly associated GWAS SNPs, pair-wise LD was assessed in the parents using PLINK (http://pngu.mgh.harvard.edu/~purcell/plink/) [37] or HAPLOVIEW (http://www.broad.mit.edu/mpg/haploview/) [43]. A combined p-value from meta-analysis of the GWAS and replication association results was computed using MANTEL [44].
Interesting SNPs from the GWAS and replication study were tested for effect modification with environmental exposures relating to air pollution and environmental tobacco smoking. Data from the GWAS population were stratified by current parental smoking (yes/no) and residential ambient ozone exposure (stratified at the median of 67 ppb annual average of the daily maximum 8 hour averages), and the log-linear method was used to test for genetic associations in each stratum.
Additional analyses were conducted in the GWAS population using the skin prick testing data. The genome-wide association scan was repeated in the subset of trios with children classified as atopic (one or more positive skin tests) in an effort to reduce phenotypic heterogeneity and thus reduce genetic heterogeneity. Then, an extension of the log-linear method for quantitative traits [45] was used to test for associations of interesting SNPs with the number of positive skin tests to a battery of 24 aeroallergens as a measure of atopy. This analysis was performed only on the case-parent trios with skin test data and complete genotype data.
The program FRAPPE was used to estimate individual ancestry proportions, assuming three ancestral populations: Native American, European, and African [46]. HapMap (phase 3) genotype data from 109 individuals from the United States with northern and western European descent and 108 individuals from Nigeria were included to represent ancestral European and African individuals, respectively. Genotype data from 35 Mayan and Pima Indians were taken from the HGDP, which have been genotyped using Illumina 650 K arrays [18], as the best available representation of ancestral Native American individuals. Individual ancestry proportions were averaged across all study subjects to determine the mean Native American, European, and African ancestral contributions.
The program SABER was used to infer locus-specific ancestry in each individual [47]. Our goal was to elucidate the pattern of ancestry segregation near the replicated chromosome 9q SNP (rs2378383). SNPs along chromosome 9q were analyzed to more accurately infer ancestry, but since we were only interested in examining the pattern of ancestry segregation at rs2378383 a priori, a multiple testing correction was not applied. A trio-based ancestry analysis test was implemented similar to that described in Clarke and Whittemore [48], which parallels the transmission disequilibrium test [41]. Because the African ancestry in this population is quite low (<5%), we tested Native American versus non-Native American ancestry. At each SNP, parents were considered to have no Native American ancestral alleles if their posterior Native American ancestry was less than 0.1, one ancestral allele if between 0.4 and 0.6, and two ancestral alleles if greater than 0.9. Conservative thresholds were set to minimize misclassification. Parents with intermediate ancestry estimates falling outside the specified ranges were excluded, resulting in a 10% missing rate. For parents who were heterozygous in ancestry, the null hypothesis that the Native American allele is transmitted to the children with probability equal to ½ was tested at each SNP [48].
The Genomics Institute of the Novartis Research Foundation maintains a freely accessible database (http://symatlas.gnf.org) of genome-wide expression profiles of the protein-encoding transcriptome in many diverse human and mouse tissues and cell lines [21]. As reported, tissue samples were predominantly obtained from the normal physiological state [21]. The custom expression array for humans targeted 44,775 transcripts corresponding to known, predicted, and poorly characterized protein-encoding genes [21]. We obtained the expression data for the 44,775 transcripts in 51 diverse human tissues and mapped these transcripts to 15,047 unique genes after accounting for multiple transcripts per gene and mapping to current nomenclature.
Gene expression patterns in the Novartis data were examined in the context of GWAS results. Of the 15,047 genes with expression data, 12,199 genes had at least one genotyped GWAS SNP mapping within the gene, and an additional 2,131 genes had at least one genotyped SNP mapping near the gene for a total of 14,330 genes. SNPs mapping within the 5′- most extent to the 3′- most extent over all isoforms for a gene or within a larger region expanded by 50 kb in both directions were considered. This broader region was considered in order to capture potential regulatory regions. For each gene, one false discovery rate q-value was calculated using the log-linear p-values of SNPs mapping within or near the gene based on a method for combining p-values by Peng et al. [49]. Genes were then categorized as expressed or not expressed in each of the 51 tissues examined. The expression threshold was the 75th percentile of normalized intensity values for each tissue. The global median q-values across genes expressed versus genes not expressed were calculated for each tissue, and a two-tailed Wilcoxon rank-sum test was conducted to generate a test of significance for this difference in median q-values.
Gene expression patterns are correlated across different tissues. We performed a two-dimensional hierarchical clustering to describe the correlation of expression patterns using Spearman's correlation coefficient across genes and across tissues. Genes and tissues with similar gene expression patterns were grouped into clusters using Ward's distance as the linkage function to be optimized. Partek Genomics Suite 6.08.1010 (Partek Inc., St. Louis, Missouri) software was used to perform this analysis using the continuous expression values in the 15,047 genes with expression data.
|
10.1371/journal.ppat.1005297 | The N-terminal Helical Region of the Hepatitis C Virus p7 Ion Channel Protein Is Critical for Infectious Virus Production | The hepatitis C virus (HCV) p7 protein is required for infectious virus production via its role in assembly and ion channel activity. Although NMR structures of p7 have been reported, the location of secondary structural elements and orientation of the p7 transmembrane domains differ among models. Furthermore, the p7 structure-function relationship remains unclear. Here, extensive mutagenesis, coupled with infectious virus production phenotyping and molecular modeling, demonstrates that the N-terminal helical region plays a previously underappreciated yet critical functional role, especially with respect to E2/p7 cleavage efficiency. Interrogation of specific N-terminal helix residues identified as having p7-specific defects and predicted to point toward the channel pore, in a context of independent E2/p7 cleavage, further supports p7 as a structurally plastic, minimalist ion channel. Together, our findings indicate that the p7 N-terminal helical region is critical for E2/p7 processing, protein-protein interactions, ion channel activity, and infectious HCV production.
| Hepatitis C virus (HCV) infection can lead to significant liver disease and, without a vaccine, continues to pose a significant public health threat. The viral p7 protein is a multifunctional protein that is required for infectious virus production via its role in orchestrating virion assembly and its activity as an ion channel. However, while there is accumulating structural information on p7, there is no consensus on which conformation(s) exist during a natural infection or how structural elements relate to p7 functions. By comparing two prominent, yet highly divergent models of p7, we identified one region of structural similarity–the N-terminal helical region. While mutagenesis screening of other regions of the protein are in keeping with p7 conformational flexibility, mutations within the N-terminal helical region had a significant impact on infectious virus production, due in part to a loss of efficient E2/p7 cleavage. We further postulated the precise functional impact of mutations throughout p7 by homology modeling and demonstrated tolerance for diverse amino acid substitutions for specific N-terminal helix residues with putative ion channel defects. Together, these data not only support p7 as a structurally plastic, minimalistic ion channel, but also provide extensive insight into the p7 structure-function relationship and highlight the importance of the N-terminal helical region in E2/p7 processing, protein-protein interactions, ion channel activity, and infectious HCV production.
| Over 130 million people worldwide are at risk for liver fibrosis, cirrhosis, hepatocellular carcinoma, and end stage liver disease as a result of hepatitis C virus (HCV) infection [1]. These complications of infection have made hepatitis C the most common indication for liver transplantation [2]. Further, while novel direct-acting antivirals targeting HCV have dramatically improved clinical outcomes, no vaccine exists to date, and the disease burden is expected to increase over the next decade [3].
HCV is a hepatotropic, plus-strand RNA virus of the Hepacivirus genus and Flaviviridae family [4,5]. IRES-mediated translation of the 9.6 kb HCV genome yields a single polyprotein that is proteolytically cleaved to produce 10 mature viral proteins that participate in viral replication and assembly of nascent virions [6]. The p7 protein, located at the junction between the structural and non-structural proteins [7], is a small, 63 amino acid integral membrane protein [8], predominantly localized to the endoplasmic reticulum (ER) [9].
In the context of the HCV life cycle, p7 is dispensable for viral RNA replication [10] but required for infectious virus production [11,12], although it does not appear to be a structural component of the virion nor is it required for HCV glycoprotein-mediated entry [9,13,14]. Accumulating evidence suggests that p7 orchestrates intracellular viral protein distribution [15–17], at least in part, via an (direct or indirect) interaction with the viral NS2 protein [16,18–22]. Additional interactions have been suggested with core at the genetic level [23] and with E2 by immunofluorescence-based colocalization and FACS-FRET methods, although coimmunoprecipitation of p7 with HCV glycoproteins in HCV-replicating cells has yielded disparate results [9,24]. Further, yeast two-hybrid and bioinformatically-predicted cellular binding partners have not been further validated [25–28].
Based on the ability of p7 to alter membrane permeability, it has been classified as a viroporin along with HIV-1 vpu and influenza virus M2, among others (reviewed in [29]). p7 ion channels are sensitive to hexamethylene amiloride [30], long-alkyl-chain iminosugar derivatives [31], and–depending on genotype [32,33]–amantidine [34], all of which inhibit cation channel activity in artificial membranes [34,35]. The importance of p7 ion channel function for HCV has been demonstrated by correlation of intravesicular pH modulation and infectious virus production in cell culture [36]. This activity has been hypothesized to enable proper glycoprotein folding, protect against premature degradation [37], or guard against acid-induced conformational changes [14,36,38,39].
Structurally, initial computational modeling predictions [18,40], refined by NMR experiments [22,41,42], indicate that p7 monomers adopt a “hairpin-like” topology consisting of an N-terminal helix and “turn” sequence upstream of two transmembrane segments that are connected by a hydrophilic, positively-charged cytosolic loop containing two highly conserved basic resides. The N- and C-termini are oriented towards the ER lumen and may provide a platform for interactions with viral or host proteins [18,43].
The intricacy of p7 structure is further complicated by p7 homo-oligomerization. Based on the typical oligomeric structures of viroporins, p7 subunits reside side-by-side in classical hexameric and heptameric models [40,42,44,45]. Molecular dynamic simulation of p7 oligomers, based on the monomeric model put forth by Montserret et al. [41], suggest that multiple oligomeric states are feasible and that p7 is structurally plastic and may adopt multiple conformations during oligomerization and/or as a function of its lipid environment [44,46]. In contrast, the recent NMR structure of hexameric p7 [47] exhibits an unusual architecture where part of each p7 subunit crosses over to interact with all the five other p7 subunits. The resulting rigid structure is reminiscent–albeit comparatively inverted–of single-particle electron micrographs of p7 that depicted a “flower-shaped” architecture [43].
Despite the increasing amount of structural data on p7, there is no consensus on which conformation(s) exist during a natural infection or how structural elements relate to p7 protein-protein interactions, cation selectivity, and ion channel gating. Influenza virus M2 and HCV p7 can partially functionally complement each other [36,48], yet analogy to HIV-1 vpu or influenza virus M2 provides limited mechanistic insight given the divergent structural features and diverse functions described [29]. Modeling of homo-oligomeric assembly [35,40,44] and electrophysiology experiments [49] indicate that the first transmembrane helix of p7 lines the pore, and the C-terminus (including TMD2 and unstructured termini) has been proposed to interact with other proteins [41]. Further, residues potentially involved in cation selectively and gating or intra/intermolecular stability have been postulated [12,41,47]. However, while mutation of two basic residues, K33 and R35, within the cytosolic loop, supported their role in ion channel function [36], none of the putative pore-lining residues studied to date by mutagenesis are essential for p7 ion channeling in vitro [12,41,49–51].
A comprehensive analysis of residues in key structural regions has not been performed. While the amino acid sequence of p7 is not highly conserved, extensive physico-chemical conservation [41] suggests that the overall p7 structure is similar across genotypes despite variability among individual amino acids. Here, we aimed to probe p7 plasticity and functionality using a combination of mutagenesis and molecular modeling approaches. Our data indicate a critical role for the N-terminal helix region of p7 in modulating E2/p7 cleavage and further support p7 as a structurally plastic, minimalist ion channel through interrogation of specific N-terminal helix residues predicted to point toward the channel pore.
Previous reports indicate that p7 is not required for viral RNA replication but is required for infectious virus production. Modeling data indicate that in addition to the hexameric and heptameric forms of p7 demonstrated experimentally, tetrameric and pentameric oligomers may also exist, at least transiently [44]. To provide biological evidence of p7 structural features and define regions important for functionality, we generated two p7 mutant panels in the context of the J6/JFH infectious clone–one in which an alanine was inserted after every third amino acid throughout the entire length of the protein to perturb p7 structure and a second in which tryptophan substitutions were made throughout the transmembrane domain regions at residues 19–29, 31–32, and 36–43 to probe intra- and intermolecular interactions as well as amino acid hydrophilic pore- vs. hydrophobic bilayer-facing orientation (Fig 1A). Mutation of conserved basic residues K33 and R35 in the cytosolic loop was previously shown to impede ion channel activity and block infectious virus production both in vitro and in vivo [11,12,17,36,48,52]; thus, we excluded these from our analysis. Quantification of cell-associated HCV RNA at 8 and 72 hours post-electroporation indicated that over this time frame all mutants replicated with wild-type (WT) efficiency (Fig 1B), exhibiting a mean 495-fold increase in RNA copies per 50 ng of total RNA. Several mutants exhibited marked reductions (>1 log decrease vs. WT) in extracellular infectious titers (A6, A27, A39, A51, A60 and F26W, A28W, V36W) (Fig 1C), and these data correlated with slightly lower levels of HCV RNA at 72 hours post-electroporation (Fig 1B), likely due to reduced virus spread within the culture. Of these, A6 and A60, located near the N- and C-termini, failed to produce any infectious virus. Surprisingly, the majority of p7 mutants were competent for infectious virus production, including mutants with alanine insertions within the first transmembrane domain, a region considered important for ion channeling. Further, several mutations (e.g. A21, F25W and Y31W) even yielded titers above those obtained for WT virus indicating that p7 can accommodate these genetic changes (S1 Fig). These results support a model of p7 structural plasticity in human hepatoma cells replicating full-length HCV genomes.
The first structure of monomeric p7 was obtained by combining NMR experiments performed in a 2,2,2-trifluoroethanol (TFE) / water mixture with molecular dynamics (MD) simulations [41]. A full-length, FLAG-tagged monomeric p7 structure was later determined in methanol [42], and recently, structures of p7 from two different genotypes were determined in 1,2-Dihexanoyl-sn-glycero-3-phosphocholine (DHPC) micelles [22] and dodecylphosphocholine (DPC) [47], illustrating the monomeric and hexameric p7 forms, respectively. Importantly, these p7 NMR structure models differ on the location of secondary structural elements and orientation of p7 transmembrane domain regions (Fig 2A), most notably for segment 33–47. These discrepancies may be due in part to differences in the HCV genotype tested (1b vs. 5a) and/or the lipid-mimicking environment used (TFE, DHPC, DPC, or methanol), the latter of which has been shown to impact p7 activity [46]. To better visualize reported p7 structural elements, we used hexameric p7 models in DPC as described by OuYang and colleagues [47] (model 1; Fig 2B) and in 1-palmitoyl-2-oleoyl-sn-glycero-3-phosphocholine (POPC) as described by Chandler et al. [44] that employs the monomeric p7 structure put forth by Montserret et al. [41] (model 2; Fig 2B). These two models were selected based on the availability of hexamer structure coordinates and our effort to compare divergent models. Interestingly, in model 1, the p7 subunits are crossed such that part of each monomer interacts with all other subunits, while in model 2, they reside side-by-side, illustrating the topology typical of two transmembrane helical proteins (Fig 2B). However, despite the huge differences in the organization of the central part of p7 subunits between these two models, the N-terminal helix (1–18) is close to the C-terminus segment of p7 subunits in both models and forms a hexameric helix bundle with a similar organization of residues; notably, the side chains of amino acids 9 and 12 in both models point to the pore lumen (Fig 2C). Importantly, this organization is also observed in other theoretical and NMR-based models [40,42].
Given these similarities in N-terminal helical packing and the conserved hydropathic pattern in this region, we extended our tryptophan mutagenesis of p7 to screen positions 1–18 (Fig 3A). Similar to the mutants tested in Fig 1, all of the N-terminal mutants replicated to similar wild-type levels (Fig 3B), yet strikingly, mutagenesis in this region had a more profound impact on infectious virus production illustrated by a greater than 1 log decrease in titer for half of the mutants tested including A1W, L2W, E3W, K4W, V6W, H9W, A10W, A11W, and S12W. While varying levels of infectious HCV were detected for genomes harboring mutations at positions 5, 7, 8, 11, and 13, six of the viruses tested (A1W, E3W, K4W, H9W, A10W, and S12W) failed to produce any detectible virus by 72 hours post-electroporation (Fig 3C).
Previous studies have shown that mutations in the N-terminal region of p7 can modulate the partial cleavage at the E2/p7 and p7/NS2 junctions [55]. To assess a potential impact of Trp substitution on host signal peptidase cleavage efficiency, we probed for E2 antibody-reactive proteins in Huh-7.5 cells replicating WT or p7 mutant genomes by western blot. To demonstrate E2 antibody specificity and distinguish between incompletely processed E2-containing protein species (E2p7NS2, E2p7, and E2), we utilized monocistronic, wild-type J6/JFH1 and ΔE1E2 genomes along with bicistronic genomes that contain an IRES between E2 and p7 or p7 and NS2 to remove the requirement for protein processing at these junctions. Parallel analysis of N-terminal helix mutants illustrated an E2/p7 processing defect for A1W, E3W, and K4W. Surprisingly, these data also suggested a similar defect may contribute to the deleterious phenotypes of other downstream N-terminal helix mutants, most notably for H9W through S12W (Fig 3D).
To identify second site amino acid changes that could compensate for E2/p7 cleavage or p7-specific defects introduced by tryptophan substitutions, we serially passaged Huh-7.5 cells harboring deleterious N-terminal mutant genomes to allow for the emergence of variants that are competent for infectious virus production (Fig 4). After three to seven passages, virus was detected in the supernatant for all genomes except H9W. Despite several attempts at electroporating hepatoma cells with this mutant genome, we were unable to select for a virus capable of spread. This was not due to a high genetic barrier (i.e. the requirement of multiple nucleotide changes to obtain a viable virus), as both serine and glycine (amino acids that are one nucleotide change away from tryptophan) function in this position (Fig 7; see also S1 Fig and S6 Fig for models). Further analysis by titrating WT RNA into a constant amount of either control (Δp7) or H9W RNA at the time of electroporation resulted in 3.7-fold less infectious virus production at a WT:H9W RNA ratio of 1:2 (compared to WT:Δp7 at the same ratio). While these data hinted that the mutant p7 might act via a dominant negative mechanism to suppress WT p7 and the production of infectious virus (S2 Fig), the effect was less dramatic when the ratio was further increased in favor of the mutant.
We next sequenced the p7 region of HCV RNA extracted from naïve Huh-7.5 cells inoculated with supernatant from passaged cells replicating mutant HCV genomes and identified conservative same-site changes in five of the ten viruses analyzed (Fig 4A). These viruses–with mutations at positions 1, 3, 4, 10, and 12 –failed to produce any detectable infectious virus in our original characterization, and together, these data suggest that certain physico-chemical characteristics of the amino acid side chain at these positions are critical. Replacement of tryptophan at positions 1 and 10 with cysteine represents a reversion to a small residue, while glycine at polar positions 3, 4, and 12 likely represents a release of hydrophobic steric constrains (S3 Fig). Re-engineering of these amino acid changes into the original mutant genome confirmed their ability to rescue WT-levels of virus production (Fig 4C and 4D). L2W, V6W, and A11W revealed putative second-site mutations in E2; however, as we chose to focus our analysis on p7, whether these mutations are responsible for rescuing infectious virus production remains to be determined.
To investigate whether rescue of infectious virus production correlated with enhanced E2/p7 cleavage, we probed for E2 by western blot, comparing original mutant genomes with those re-engineered to contain p7 mutations identified after passage. Strikingly, all viruses harboring p7 mutations identified after passage yielded a marked increase in the amount of ‘free’ E2 relative to E2p7 compared to their original mutation counterparts (Fig 4B). These data suggest that diminished cleavage at this junction contributed to our original deleterious phenotypes for these N-terminal helix mutant viruses and consequently impede our ability to evaluate their impact with respect to p7-specific functions.
To evaluate the impact of tryptophan substitutions in the N-terminal helix independent of E2/p7 cleavage, we engineered these mutations at positions 1–13 into a bicistronic genome containing the EMCV IRES between E2 and p7 (J6/JFH E2-IRES-p7 [11]; Fig 5A), thus eliminating the need for polyprotein processing at this junction. We then phenotyped these bicistronic N-terminal tryptophan substitution mutants with respect to replication and infectious virus production after electroporation into Huh-7.5 cells. As expected, all mutant viruses replicated to similar wild-type levels (Fig 5B). Furthermore, as predicted by our western blot data indicating defects in E2/p7 cleavage were at least partially responsible for abrogation of infectious virus production, the majority of bicistronic N-terminal helix mutants now yielded infectious virus titers that were comparable to wild-type (Fig 5C). Nonetheless, several mutants, including A1W, V6W, H9W, A10W and S12W, remained impaired, suggesting a cleavage-independent, p7-specific defect also impacts infectious particle production for these viruses.
These phenotypes indicated a deleterious impact of tryptophan on p7 function at positions 1, 6, 9, 10 and 12 but do not provide evidence for a rational hypothesis regarding the mechanism of the defect. Thus, to gain insight into the impact of these mutations on p7 structure, we modeled these tryptophan substitutions via homology molecular modeling. Because the structural impact, and hence, proposed functional consequence of our mutations, may differ depending on the 3D model, we aimed to develop hypotheses based on both model 1 and model 2 (Fig 2B). Comparing the sequence of the J6 (genotype 2a) p7 used in our study to the genotype 1b and 5a p7 used to study the p7 structure by NMR indicated sufficient similarity at the amino acid level (Fig 2A) to enable the generation of J6 p7 models by homology using Swiss-Model facilities [56]. (Coordinates of homology models 1 and 2 for p7 HCV J6 strain are available as supplementary.pdb files; S1 File and S2 File). Introduction of any of our tryptophan mutations in these homology models yielded energetically stable hexamer structures without significant structural changes indicating that p7 structure models 1 and 2 readily accommodate these mutations (S1 Fig). We then closely examined each model to evaluate/predict the structural/functional consequence of the tryptophan substitution (Fig 6A). Not surprisingly, given the similarity of these two models at the N-terminus (Fig 2), our predictions were generally consistent between model 1 and model 2 (S1 Fig). Indeed, the orientation of the tryptophan side chain toward the lumen of the pore in both models suggests a likely ion channel defect for H9W and S12W. The A10W mutation could disturb p7 intramolecular interactions or interrupt p7 interactions with binding partners, as also predicted for the A1W mutant because of its N-terminal position (S1 Fig). Interestingly, model 1 and model 2 did give rise to incongruent hypotheses for some mutants (e.g. V6W), suggesting these residues may impact multiple aspects of p7 function. Alternatively, such mutants could be used as tools to test the accuracy of one model over the other by directly assessing the functional defect in cell culture.
We next sought to corroborate the hypothesized p7 functional consequences of tryptophan substitution based on our homology models by further assessing selected mutants in cell culture. Specifically, we aimed to rescue infectious virus production by putative ion channel defective mutants in Huh-7.5 cells using bafilomycin A1. Bafilomycin A1 prevents vesicular acidification and thus may compensate for a loss of p7 channel activity. Further, this inhibitor was previously shown to compensate for a defective p7 mutant harboring mutations K33A and R35A [36]. In our initial experiments, 8 nM bafilomycin was found to be both relatively non-toxic to cells and extremely effective in alkalinizing cellular compartments and over a 24 hr time period, retaining 80% cellular viability with complete loss of acidic organelle labeling with LysoTracker Red DND-99 (S4 Fig). Since bafilomycin A1 can also prevent endosomal acidification and thus impede HCV entry into cells used for subsequent infectivity analysis, we further optimized methods to concentrate virus- and bafilomycin A1-containing supernatants 5-fold while simultaneously removing a sufficient amount of the inhibitor to enable infectious virus quantification by limiting dilution assay (S4 Fig).
We selected V6W, H9W, and S12W bicistronic p7 mutant viruses specifically for analysis based on our homology models that suggested an ion channel defect for both H9W and S12W. For V6W, the interpretation differed between model 1 and model 2, providing a potential opportunity to decipher between them. Following electroporation and incubation of Huh-7.5 cells with selected viral genomes, cells were provided with media containing either bafilomycin A1 or DMSO. Cell culture supernatants were collected 24 hours later and infectivity was assessed (Fig 6B). Notably, bafilomycin A1 treatment resulted in a boost of viral titers for all genomes tested that were capable of making detectible levels of infectious particles under DMSO conditions. However, only in the case of the KRAA mutant and our S12W mutant was this increase significant (Fig 6C). These data support our structure model-based hypothesis that tryptophan substitution at position 12 abrogates ion channeling, and also indicate that altering intracellular pH via bafilomycin A1 treatment is insufficient to ‘rescue’ the impact of tryptophan substitution at position 6. Interestingly, despite both models pointing to an ion channel defect for H9W, we were unable to recover any infectious virus for this mutant in our assay. One explanation is that our methods are not conducive to detection of low levels of infectious virus production; indeed, the limit of quantification for our limiting dilution assay in this context is 10-fold higher than previous experiments due to some residue bafilomycin A1 carryover in the supernatant. Thus, a small, but significant increase in infectious virus production, as was previously shown in a similar experiment with the KRAA mutant [36], may not be uncovered.
In both models interrogated in this study, residues 9 and 12 point to the pore formed by p7 oligomerization (Fig 2C). These structural data, supported by our ability to significantly increase S12W mutant infectious viral titers by altering vesicular pH, indicate the amino acids at these positions could contribute to cation selectivity and flux. To further analyze the requirements at these residues, as well the residue at position 6, which is oriented towards the pore in model 1, we expanded the amino acid repertoire at these positions and analyzed the impact of polarity, charge, and hydrophobicity on viral replication and infectious virus production (Fig 7). Amino acids were analyzed in both monocistronic and bicistronic viral genetic backgrounds in order to segregate between amino acids that impact E2/p7 cleavage versus those that influence p7-specific functions. Our data indicate that although position 6 tolerates both hydrophilic and hydrophobic residues, bulky residues (Leu, Phe and Trp) are more detrimental in the monocistronic context, suggesting these amino acids have a negative impact on E2/p7 cleavage, whereas residues with smaller side chains (Ala, Ser, Thr) have almost no effect on infectious virus production (Fig 7A and 7D, and S5 Fig). Specifically, we observed a correlation between the increasing size of hydrophobic residue side chains (S6 Fig) and the inhibition of virus production.
Extending our analysis to position 9, we observed that polar amino acids Gln, Asn, and Ser all function at this position to support infectious virus production while hydrophobic residues (Ala, Cys, Leu and Tyr) do not (Fig 7B). In agreement with these data, H9A mutation in JFH-1 p7 (genotype 2a) was previously shown to reduce channel conductance by ~70% [47]. However, the impact of these hydrophobic residues was less significant in Huh-7.5 cells replicating bicistronic genomes, indicating the primary impact of these substitutions is on E2/p7 cleavage (Fig 7E). Surprisingly, both acidic (Asp) and basic (Arg) residues support infectious virus production, albeit to low levels.
Similar to position 9, we also observed that amino acids at position 12 with bulky hydrophobic side chains inhibited virus production in the monocistronic context, but this was again less apparent for bicistronic genomes harboring the same amino acid changes (Fig 7C and 7F), indicating again that the primary impact of these substitutions is on E2/p7 cleavage. In both cases, amino acids at positions 9 and 12 with polar character supported infectious virus production (Fig 7B, 7C, 7E and 7F). Interestingly, substitution with negatively charged Asp at position 12 yielded an increase in viral titers above those obtained for WT, potentially via enhanced cation recruitment at the pore entry. However, positively charged Arg also functions at this position while Ala and Ser are the only natural amino acids found at this position. Together these data indicate that E2/p7 cleavage efficiency is sensitive to downstream mutations within the N-terminal helix region of p7 while the tolerance of positions 6, 9, and 12 to amino acids of different nature in a bicistronic context further supports p7 as a structurally plastic, minimalist ion channel.
In this report, we have extensively interrogated the HCV p7 protein via mutagenesis and determined the effects of these mutations on virus replication and infection in cell culture. In addition, we have modeled these mutations using p7 structure information based on previous NMR experiments. Our data confirm previous reports that p7 is not required for viral replication, as all p7 mutants tested replicated with wild-type efficiency. Importantly, our large-scale, structure-function analyses illustrate a global tolerance for amino acid sequence alterations, either by insertion or individual amino acid substitution in the J6/JFH background. These results underscore the structural flexibility of p7 [44] that has been similarly described for other viroporins such as HIV-1 Vpu [57]. Our data are further in line with the conservation of p7 amino acid physico-chemical properties and hydropathic character but not precise sequence across genotypes [41]. Notably, two of the nine conserved amino acids (G18 and Y42) were directly assessed in this study by Trp substitution and resulted in an increase and decrease in infectious virus titer, respectively, although these phenotypes were not the most dramatic in our panel. Interestingly, structure models give rise to incongruent hypotheses regarding the impact of G18W mutation (S1 Fig), suggesting this residue may provide another opportunity to further probe these structure models and test p7 function in cell culture and p7 ion channel activity after reconstitution in artificial membranes. Together our data indicate that escape mutants with significant fitness could be readily generated in the context of p7-targeting antiviral compounds, potentially limiting the efficacy of this class of inhibitors in the clinic. Still, there were several positions tested that did show a marked impact on infectious virus production; this was most pronounced when the residues within the first eighteen amino acids, comprising the N-terminal helical region, were interrogated.
Our functional predictions are based on available hexameric p7 models; however, previous studies indicate that the oligomerization of seven p7 subunits is also feasible [35,44], possibly even resulting in a mix of oligomeric states within the infected cell. Nonetheless, computational analyses for models where p7 subunits reside side-by-side [35,44] indicate that amino acid positions are similar and pore lining residues are retained with the addition of the 7th monomer; thus, our data interpretation would likely be consistent in this context. Interestingly, molecular dynamics simulations in a hydrated POPC bilayer showed that hexameric p7 model 2 formed a pore that was transiently permissive to solvent, potentially linked to a hydrophobic barrier formed by F25 [44]. In our initial mutant panels we observed a decrease in titer following Trp substitution at position F26 and an increase in titer after F25W mutation. Both hexameric models assayed here suggest residue 26 has multiple contacts within p7, thus likely playing a role in stabilizing the protein, whereas the amino acid at position 25 lines the pore. Naturally occurring residues at both of these positions across genotypes are invariably hydrophobic, and while the higher polarity of the Trp side chain may facilitate the passage of ions at position 25, resulting in a higher production of virus particles, the same substitution at position 26 may destabilize p7 assembly. The ability of Trp substitution to boost titers at several positions (including residue 25, as well as 29 and 31) is interesting given that this amino acid does not naturally occur at any of these positions. This suggests that the increased titers we observed in Huh-7.5 cells are not advantageous in a more physiologically relevant system (e.g. primary human hepatocytes) and may negatively impact viral fitness in vivo, although this was not directly tested in this report.
A side-by-side structural comparison of the models proposed by OuYang et al. [47] (model 1) and Chandler et al. [44] (model 2) revealed similar helical packing at the N-terminus in these otherwise incongruent models; hence, we focused our studies on this region of p7. Surprisingly, several lines of investigation, including western blot analyses of N-terminal helix mutants and related pseudorevertants, as well as infectious virus production phenotyping of bicistronic mutant genomes, all indicated that mutation within this region has a significant, detrimental impact on E2/p7 processing. This supports previous studies that have implicated this region in modulating the partial cleavage at E2/p7 and p7/NS2 junctions [58]. Interestingly, A13W, a mutant that originally demonstrated a 1-log attenuation compared to WT, acquired an additional mutation at position 17 (N17D) after passage that correlated with enhanced E2/p7 processing and increase in infectious virus production. In model 1, residue 13 points towards the pore, while 17 lies at the p7 protein surface within the hydrophobic region of the membrane–making the identification of negatively charged aspartic acid at this position energetically counter intuitive (S3 Fig). In model 2 (as well as the model presented by Foster et al. [42]), however, residues 13 and 17 both point to the p7 pore, one directly above the other, indicating these residues could be related in function.
Overall, our work suggests that tryptophan substitution (or potentially bulky or hydrophobic residues in general, as demonstrated for residues 6, 9, and 12) negatively modulates important interactions between the C-terminus of E2 and the N-terminus of p7 that play a regulatory role in cleavage efficiency of the E2/p7 junction, possibly through mediating correct presentation of the cleavage site to the signal peptidase. In uncleaved E2p7 species, the topology of p7 may be inverted [59], and while a specific role for E2p7 in the HCV life-cycle remains questionable, it has been hypothesized that the proper timing of E2/p7 cleavage may be critical to avoid spontaneous ion channel formation in the ER membrane and to promote the assembly process [44]. Nonetheless, the separation of E2 and p7 is absolutely required for infectious virus production, specifically for proper NS2 localization near assembly complexes [58] and presumably p7 oligomerization.
Beyond deficiencies in protein processing, our mutagenesis data using bicistronic constructs, further informed by homology modeling, identified several N-terminal helix mutants with p7-specific defects, including positions 6, 9, and 12. Interestingly, modeling of V6W (Fig 6A), which is located within a conserved hydrophobic cluster spanning from position 5–8 and is either a Val or Ile in all HCV genotypes, indicated this mutation would face the pore in model 1 and has been proposed to play a role in closing the pore via the formation of a hydrophobic ring [47]. In contrast, this V6W mutation would likely affect helix-helix interactions (i.e. oligomeric structural stability) in model 2. In fact, our inability to rescue this bicistronic mutant using bafilomycin A1, in addition to the fact that position 6 tolerates both hydrophilic and hydrophobic residues suggests this position does not play a critical direct role in ion channeling.
Position 9 is an Asn or, in genotype 2 viruses, a His–both of which have an affinity for monovalent and divalent cations. The hydrophilic nature of this position, as well as its location at the pore entry in both models, has implicated this residue in cation selectivity [41]. OuYang et al. [47] further propose it serves as a “filter” to dehydrate cations, allowing them to pass through the hydrophobic ring formed by position 6, the more narrow part of the channel in model 1. In accordance with these hypotheses, polar residues supported virus production in our study, while substitution with hydrophobic or charged residues resulted in significantly decreased infectious virus titers compared to wild-type. Our results for this residue were reminiscent of those obtained for position 17, which is located within the turn sequence and found to line the pore. Position 17 naturally occurs as a histidine or asparagine (like position 9) as well as glutamine–all of which share polar characteristics–and has been implicated in ion channel function [49], as our present data suggest for position 9. Similar to our results at position 9, mutation of H17 to A or G by Meshkat et al. resulted in a decrease in titer while H17E (polar) boosted infectious virus in the supernatant [51]. Interestingly, we did not see a major phenotype following Trp substitution at position 17 (N17W) in the context of the J6 p7 sequence, perhaps due to conservation of some polarity by this substitution.
Beyond positions 6 and 9, we also investigated the previously overlooked residue at position 12, which is only Ser or Ala in natural variants and points to the pore in both models. Our data show a striking complete loss of infectious virus production upon Trp substitution–a phenotype that was only partially rescued when this mutation was subsequently engineered into a bicistronic background. Modeling data indicated aromatic ring-mediated pore obstruction while subsequent rescue with bafilomycin A1 further suggest a novel role for this position in ion channel function. Still, the global tolerance of both position 9 and position 12 for amino acid substitution of different characteristics supports p7 as a structurally plastic, minimalist ion channel.
Our present study identified N-terminal mutants that are defective for infectious virus production but did not distinguish between a defect in particle assembly versus infectivity. The identification of p7 mutants that are competent for particle assembly, but exhibit a profound defect in specific infectivity would provide a unique tool to probe the impact of p7 function on the viral particle at later stages of the viral life cycle. Further, deleterious mutants for which our models generated incongruent functional defect hypotheses offer additional opportunities to delineate between these models by probing p7 functional defects in cell culture or reconstituted in artificial membranes and correlating these with structure-based predictions to support or refute the available structural data. Importantly, our study highlights a potential regulatory role of the p7 N-terminal helix residues in the cleavage efficiency of the E2/p7 junction, although the precise underlying mechanisms remain elusive. In sum, our work illustrates the convergence of current p7 models at the N-terminal helix and demonstrates the biological impact of amino acid perturbation in this region, offering extensive insight into the relationship between p7 structure and function in the context of HCVcc.
J6/JFH [60], the K33A/R35A p7 mutant and bicistronic genomes J6/JFH E2-IRES-p7 and p7-IRES-NS2 [11] have all been previously described. To facilitate the generation of p7 mutant viruses, two silent restriction sites (NotI and BglII) were engineered into the wild type monocistronic J6/JFH sequence in E2 and NS2, respectively. The resulting genome is termed J6/JFH 1.1 and referred to here simply as J6/JFH or wild-type. Mutations in p7 were introduced by overlap PCR using standard procedures and engineered into monocistronic or bicistronic constructs using NotI and BgIII or MluI and NotI, respectively. Plasmid and primer sequences are available upon request. All constructs were confirmed by sequencing.
Huh-7.5 cells [61] were propagated in Dulbecco’s modified minimal essential medium (DMEM) supplemented with 10% heat-inactivated fetal bovine serum (FBS) and 0.1 mM nonessential amino acids (NEAA). Cells were grown at 37°C in a humidified 5% CO2 atmosphere.
Viral cDNAs were linearized with XBa1 and purified using a MinElute PCR purification kit (Qiagen). In vitro RNA transcription was performed using a T7 RiboMAX Express large-scale RNA production system (Promega) and newly synthesized RNAs were isolated with an RNeasy RNA isolation kit with a second DNase I digestion (Qiagen) according to the manufacturer’s protocol. RNAs were eluted in nuclease free water and integrity and concentration were determined by agarose gel electrophoresis and absorbance at 260 nm, respectively.
In vitro-transcribed RNAs were electroporated into cells using a 4 mm gap 96-well plate format (BTX ElectroSquare Porator ECM830 with Plate Handler; Harvard Apparatus). Briefly, Huh-7.5 cells were trypsinized, washed in cold, RNase-free Dulbecco’s phosphate-buffered saline (D-PBS) without Ca2+ / Mg2+ (Gibco–Invitrogen) and resuspended at a concentration of 1.5 x 107 cells / ml in D-PBS. Two hundred microliters (3 x 106 cells) was then mixed with 5 ug RNA and loaded into the cuvette. Electroporation was performed using the following settings: 0.80 kV, 99 ms, 5 pulses. Pulsed cells were transferred into 1.5 ml DMEM with 10% FBS and 0.1 mM NEAA before plating. Cells were plated in 24-well plates at a density of 5.3 x 104 cells / well in a final volume of 0.5 ml.
Eight hours post-electroporation, cells were washed twice with D-PBS. One well from each electroporation was then harvested in 0.35 ml RLT buffer containing 0.01 ml beta-mercaptoethanol (βME) per ml, applied to a Qiashredder and spun at 16,300 x g for 2 min before storage at -80°C. A second well from each electroporation was provided with 0.5 ml fresh complete medium and returned to the incubator until 72 hours post-electroporation when the cell culture supernatant was collected and stored at -80°C until analysis. The cells were then washed and collected in RLT buffer as described above. HCV infectious titers in the supernatants were determined by a limiting dilution assay on naïve Huh-7.5 cells as previously described [60]. Total cellular RNA was isolated using an RNeasy kit (Qiagen) and 50 ng of total RNA was then assayed for HCV genomes using a one-step quantitative RT-PCR assay (Multicode-RTx HCV RNA kit, Luminex Corp.) targeting the 3’ UTR of the viral genome and a Roche LC480 light cycler, according to manufacture’s instructions.
Selected p7 genomes shown to be defective for infectious virus production were electroporated into Huh-7.5 cells as described above. Cells were plated (1.3 x 106 cells) in 100 mm dishes and maintained in 10 ml DMEM 10% FBS with 0.1 mM NEAA. Supernatants were collected before each passage and stored at -80°C until analysis. Supernatants found to contain infectious virus were then applied to naïve Huh-7.5 cells (300,000 cells / 100 mm dish plated 24 hrs prior to inoculation), split once, and harvested in 0.6 ml RLT containing βME for RNA extraction and HCV RNA sequencing or fixed in 4% paraformaldehyde (PFA) and stained with anti-NS5A antibody (9E10 [60]-alexafluor 647) to determine the frequency of HCV antigen-positive cells by flow cytometry.
The relatively high amino acid sequence similarities between p7 of HCV strain J6 and that of strains EUH1480 (40% identity, 80% overall similarity) and HC-J4 (62% identity, 92% overall similarity) allowed us to construct three-dimensional homology models 1 and 2 for p7 hexamers, respectively, using the NMR structure of HCV p7 of OuYang et al. [47] as template (PDB accession number 2M6X) for model 1, and the NMR/MD model of Chandler et al. [44] as template for model 2. Models of p7 were constructed with the Swiss-Model automated protein structure homology modeling server (http://www.expasy.org/spdbv/ [56]) using the p7 HCV strain J6 sequence as input. p7 model 1 was directly obtained as a hexamer by the automated procedure. For model 2, raw amino acid sequence of p7 from strain J6 was first loaded in Swiss-PdbViewer software [56] and fitted to the NMR/MD p7 hexamer model of Chandler et al. [44] before submission for model building to Swiss-Model using the SwissModel Project Mode. All p7 mutants were constructed using the latter protocol, i.e., fitting of the raw amino acid sequence of p7 mutants to wild type hexamer models 1 and 2 from the J6 strain. Coordinates of homology models 1 and 2 for p7 HCV J6 strain are available as supplementary.pdb files. These coordinates are derived directly from the automated model building with no further minimization or manual manipulation.
Electroporated Huh-7.5 cells were plated in 6-well plates and lysed 72 hpe using modified radioimmunoprecipitation assay (RIPA) buffer (50 mM Tris-HCl (pH 8.0), 1% (v/v) nonyl phenoxypolyethoxylethanol, 0.5% (w/v) Na-deoxycholate, 150 mM NaCl, and 0.1% sodium dodecyl sulfate). Protein (10 μg) was then denatured and subsequently deglycosylated using PNGase F according to the manufacturer’s protocol (New England BioLabs, Inc.) before being separated on 4–12% Bis-Tris NuPAGE polyacrylamide gels (ThermoFisher Scientific) and transferred to 0.2 micron nitrocellulose membranes. Membranes were blocked with 5% milk in Tris-buffered saline with 0.1% Tween-20 and E2-containing protein species were detected using rat anti-E2 antibody (clone 3/11 [62]; 2 μg/ml final concentration). Following secondary antibody staining with Peroxidase AffiniPure donkey anti-rat IgG (H+L; 1:10,000), blots were visualized using SuperSignal West Dura reagent (Thermo Scientific).
To establish the bafilomycin concentration to be used in subsequent virus rescue experiments, bafilomycin A1 (Sigma Aldrich; or DMSO vehicle control) was titrated onto mock-electroporated cells and both viability and intracellular pH assessed 24 hrs post-treatment. Cellular viability was determined using CellTiter-Glo luminescent cell viability assay (Promega) according to the manufacturer’s protocol. Parallel wells were washed in HEPES buffer [36] and loaded with 50 nM LysoTracker Red DND-99 (ThermoFisher Scientific) diluted in HEPES buffer for 30 min at 37°C to label acid organelles. Cells were then washed with PBS, trypsinized, and LysoTracker Red content analyzed by flow cytometry after gating on live cell singlets.
For rescue experiments, selected p7 mutant genomes were electroporated into Huh-7.5 cells as described above. Forty-eight hours post-electroporation, the cell culture media was removed and replaced with media containing bafilomycin A1 (8 nM final concentration) or DMSO (vehicle control). Supernatants were harvested 24 hours post-treatment, pooled across identical wells and applied to Millipore centrifugal filters (100 MW cutoff). Samples were centrifuged at 930 x g for 15 min at 4°C and then dialyzed with 4 ml serum free medium by centrifuging again at 930 x g for 12 min to remove bafilomycin A1. The remaining sample volume (250–500 μl) was brought up to 600 μl with DMEM containing 10% FBS and 0.1 mM nonessential amino acids and infectious virus was quantified by standard limiting dilution assay performed on naïve Huh-7.5 cells.
Statistical analysis of virological data was performed with GraphPad Prism 5. Specific tests are noted in figure legends.
|
10.1371/journal.pgen.1004784 | Phylogenetically Driven Sequencing of Extremely Halophilic Archaea Reveals Strategies for Static and Dynamic Osmo-response | Organisms across the tree of life use a variety of mechanisms to respond to stress-inducing fluctuations in osmotic conditions. Cellular response mechanisms and phenotypes associated with osmoadaptation also play important roles in bacterial virulence, human health, agricultural production and many other biological systems. To improve understanding of osmoadaptive strategies, we have generated 59 high-quality draft genomes for the haloarchaea (a euryarchaeal clade whose members thrive in hypersaline environments and routinely experience drastic changes in environmental salinity) and analyzed these new genomes in combination with those from 21 previously sequenced haloarchaeal isolates. We propose a generalized model for haloarchaeal management of cytoplasmic osmolarity in response to osmotic shifts, where potassium accumulation and sodium expulsion during osmotic upshock are accomplished via secondary transport using the proton gradient as an energy source, and potassium loss during downshock is via a combination of secondary transport and non-specific ion loss through mechanosensitive channels. We also propose new mechanisms for magnesium and chloride accumulation. We describe the expansion and differentiation of haloarchaeal general transcription factor families, including two novel expansions of the TATA-binding protein family, and discuss their potential for enabling rapid adaptation to environmental fluxes. We challenge a recent high-profile proposal regarding the evolutionary origins of the haloarchaea by showing that inclusion of additional genomes significantly reduces support for a proposed large-scale horizontal gene transfer into the ancestral haloarchaeon from the bacterial domain. The combination of broad (17 genera) and deep (≥5 species in four genera) sampling of a phenotypically unified clade has enabled us to uncover both highly conserved and specialized features of osmoadaptation. Finally, we demonstrate the broad utility of such datasets, for metagenomics, improvements to automated gene annotation and investigations of evolutionary processes.
| The ability to adjust to changing osmotic conditions (osmoadaptation) is crucial to the survival of organisms across the tree of life. However, significant gaps still exist in our understanding of this important phenomenon. To help fill some of these gaps, we have produced high-quality draft genomes for 59 osmoadaptation “experts” (extreme halophiles of the euryarchaeal family Halobacteriaceae). We describe the dispersal of osmoadaptive protein families across the haloarchaeal evolutionary tree. We use this data to suggest a generalized model for haloarchaeal ion transport in response to changing osmotic conditions, including proposed new mechanisms for magnesium and chloride accumulation. We describe the evolutionary expansion and differentiation of haloarchaeal general transcription factor families and discuss their potential for enabling rapid adaptation to environmental fluxes. Lastly, we challenge a recent high-profile proposal regarding the evolutionary origins of the haloarchaea by showing that inclusion of additional genomes significantly reduces support for a proposed large-scale horizontal gene transfer into the ancestral haloarchaeon from the bacterial domain. This result highlights the power of our dataset for making evolutionary inferences, a feature which will make it useful to the broader evolutionary community. We distribute our genomic dataset through a user-friendly graphical interface.
| Organisms across the tree of life routinely experience changes in osmotic conditions. The ability to adjust physiological responses to these osmotic fluxes plays a role in processes ranging from desiccation tolerance and virulence of pathogenic bacteria [1], [2], to drought resistance in food crops [3], to mammalian reproduction [4]. In humans, osmotic response is essential for proper functioning of the heart [5], kidneys [6] and nervous system [7], and defects in osmo-response are implicated in a variety of chronic disorders [8]. Although there exists a large body of work on osmoadaptation, there remain a number of gaps in our knowledge. For example, how are different osmoadaptation strategies dispersed across phylogenetic space? Does there exist a strict delimitation between obligate halophiles and halotolerant organisms, or do these designations obscure a more nuanced biological reality? How do organisms with a wide range of salinity tolerances regulate the large physiological changes required to rapidly adapt to fluctuations in environmental salinity? Is there a fitness trade-off between static adaptation to constant level of high-salinity and the ability to adapt to changing salinity levels? How did the halophilic phenotype arise in evolutionary history? Here we use comparative genomics of a large number of extreme halophiles to begin to fill in some of the gaps in our current understanding of osmoadaptation.
The haloarchaea (a family of microorganisms belonging to the domain Archaea) have mastered the art of osmoadaptation. Members of this family thrive in extremely saline environments (up to NaCl saturation), and must constantly adapt to large shifts in salinity due to rainfall and evaporation. Although united by their ability to live in hypersaline environments (salinity greater than that of ocean water), the haloarchaea exhibit a diverse set of metabolic capabilities and span a broad range of environmental phenotypes [9], [10], including psychrotolerance (growth below 10°C), thermotolerance (growth above 45°C), and alkaliphilicity (maximum growth in basic environments). This diversity, along with the presence of well-developed genetic and biochemical toolkits [11], [12], makes this clade an excellent target of study for expanding our understanding of osmo-response.
To investigate the genetic potential for osmo-response in this clade, we have sequenced high-quality draft genomes for 59 species of haloarchaea isolated from 20 countries across six continents and environments ranging from fermented fish sauce to Permian age salt deposits (see Table S1). Combined with 21 previously sequenced haloarchaeal genomes, this dataset provides a rich opportunity for insight into osmoadaptation. Here we present an analysis highlighting adaptations to high salt at the gene and protein levels as well as analysis of ion transport capabilities and transcriptional machinery likely to play a role in mediating responses to changing osmotic conditions.
This genomic dataset will also be of use to the broader genomic and archaeal research communities. Although the archaea play major roles in global element cycling and ecosystem stability, this domain has been understudied. Our sequencing project nearly quintuples the number of available genomes for the haloarchaea and increases by ∼30% the number of sequenced archaea. We demonstrate several utilities of this dataset for defining community structure in metagenomic studies, improving automated genome annotation, and inferring the timing of evolutionary events along the haloarchaeal tree.
To facilitate large-scale use of this dataset, we have made sequence and annotation information available through an SQL database as well as the NCBI genome repository (for accession numbers, see Table S2). We provide gene calls and annotations derived using two independent automated annotation pipelines - the Rapid Annotation using Subsystems Technology (RAST) server [13] and NCBI's Prokaryotic Genome Annotation Pipeline (PGAAP) [14]. Using a combination of BLAST [15] and TRIBE-MCL [16], we have generated clusters of homologous proteins representing distinct protein families. We have made genome data and homology clusters for all 80 sequenced haloarchaea available through the genome context visualization tool JContextExplorer [17] (see Text S1 for access instructions). We believe that this genomic data will provide a rich source of information for the archaeal, genomics, evolutionary biology, and systems biology research communities for many years to come.
Fifty-nine haloarchaeal isolates from 17 genera were sequenced on Illumina GAII and HiSeq platforms using a combination of paired-end (85 nt reads), mate-pair (6 Kbp fragments), PCR-free paired-end, and unbarcoded “SOUP” libraries. SOUP libraries were prepared by pooling unbarcoded libraries for several species, which were phylogenetically distant enough to enable unambiguous read mapping. For library preparation details, see Materials and Methods; for information on methods used for each genome, see Table S2. Mean per base sequencing depth ranged from 13x to 188x, with a mean coverage of 90x (Figure 1). Sequence reads were assembled into contigs using the a5 pipeline [18], with a cross-species mean of 75 and median of 66 contigs. Genome assemblies ranged from 3.06 to 4.94 Mbp in size and all demonstrated high G+C content (mean of 62%) and high coding density (mean of 82%), as expected. Contig assemblies have been deposited in the NCBI genomic database along with annotations derived from the Prokaryotic Genome Annotation Pipeline (PGAAP) [14]. This annotation pipeline called between 2,945 and 4,645 putative protein coding regions, depending on the species.
A large fraction of proteins (∼41%) were annotated as hypothetical or of unknown function, likely a consequence of low experimental coverage of the archaeal domain. As sequencing projects study organisms of increasingly distant relationship to experimentally characterized model organisms, our ability to accurately analogize functions based on homology to previously characterized proteins declines. As such, these 124,149 unannotated haloarchaeal proteins represent a rich set of potential experimental targets for uncovering mechanisms of salt adaptation and other aspects of archaeal biology. Progress in understanding these mechanisms will benefit from an experimental focus on highly-conserved haloarchaeal proteins, as these are most likely to be involved in physiological processes integral to haloarchaeal biology. A significant fraction of these proteins are widely distributed, with 44% present in at least 10 of the 23 haloarchaeal genera with sequenced members, and 34% present in at least half of the included 80 genomes. To facilitate informed selection of targets for experimental work, we provide the distribution of these proteins in Dataset S1. Incorporation of our dataset into existing curated databases and automated workflows will facilitate downstream extrapolation of functional information learned from experimental approaches.
Since the beginning of our study, genome data from independently conducted sequencing projects have been released for several species included in the present study. For a comparison of sequencing statistics for these independently sequenced genomes see Table S3.
Previous phylogenetic studies have described two major haloarchaeal clades and several smaller groups with poorly defined relationships to these clades [19]. Here we update this previous work based on a phylogeny constructed using a concatenated set of 40 highly conserved genes (Figure 2) [20]. We expand the previously defined clades as follows: we consider a species to belong to a clade if a member of that genus was previously assigned to that clade and the genus is not paraphyletic or polyphyletic, and we also include any species which group with that clade with at least 75% support. Using this process for determining clade relationship, the increased resolution in the multi-marker phylogeny allows us to assign Halovivax to Clade 1. On the basis of the same phylogeny, we also propose designation of a third haloarchaeal clade, including the Halobacterium, Natronomonas, Halorhabdus, Halosimplex, Halomicrobium, and Haloarcula genera.
Previous studies have commented on the poorly resolved relationship between the Haloterrigena and Natrinema genera, which were originally designated based on lipid composition and DNA-DNA hybridization patterns [21]. Although Tindall [21] suggests that difficulties in genera-level assignment of some Haloterrigena and Natrinema species are simply the result of experimental error (including faulty DNA-DNA hybridization data), our results suggests that these genera, as currently defined, are actually polyphyletic. Species within these genera should therefore be reassigned using modern phylogenetic metrics. The multi-marker phylogeny was also instrumental in resolving other apparent genera-level paraphylies and polyphylies. These include the Natronorubrum and Halobiforma genera, which appear to be non-monophyletic when only rpoB' DNA or protein sequence similarity is considered [22].
To estimate the fraction of haloarchaeal phylogenetic diversity represented by this set of 80 haloarchaea, we performed rarefaction analysis, plotting the number of unique protein families against the number of randomly drawn genomes (Figure 3). Accurately grouping proteins into families is a non-trivial problem that has sparked the development of a large number of protein clustering algorithms [23]. As there is no experimental data for the vast majority of haloarchaeal proteins, clustering must rely on computational sequence similarity metrics. We therefore selected three methods to define protein families and generated rarefaction curves for each. The methods were as follows: (1) COG orthology groups [24], (2) in-house homology clusters defined using the clustering algorithm TRIBE-MCL [16] (see Materials and Methods, Dataset S2, Figures S1 & S2)), and (3) TRIBE-MCL defined homology clusters excluding those with only a single member (singletons).
As the COG database is limited to proteins with putative orthologs in at least three major phylogenetic groups, and only two archaeal lineages are included in the genomic dataset used to build the COGs (Euryarchaeota and Crenarchaeota), protein families unique to the archaea are not present in this database. Thus, only protein families that existed prior to the eukaryotic-archaeal split or were subsequently exchanged by horizontal gene transfer between eukaryotes or bacteria and archaea are included. As such, the rarefaction curve of COG proteins in the haloarchaea saturates very quickly, with only 2,172 COGs being present in the haloarchaea, and only 17 genomes being required to discover 90% of these families.
In contrast, using the TRIBE-MCL derived homology clusters, which do not exclude proteins specific to the archaea, the number of unique protein families in the sequenced set of haloarchaeal genomes was 17,223. Forty-two genomes were required to discover the first 90% of these families, when singleton genes were excluded. Including singletons, however, it is apparent that much more haloarchaeal diversity remains to be discovered, as more than 300 new protein families are added with each new genome. Taking into account the likelihood that many singleton gene families are the result of spurious gene calls, the true sampled phylogenetic diversity of the haloarchaea lies between the singleton-excluded and the singleton-included cases. By revealing the diversity of the haloarchaeal pangenome, this analysis highlights the importance of deep sequencing for phylogenetically informed selection of experimental targets.
A recent paper analyzing 10 haloarchaeal genomes posited that the evolution of the haloarchaeal phenotype was the result of a single mass horizontal transfer of ∼1,000 bacterial genes into an ancestral archaeal methanogen [25]. This conclusion was based on 1) the finding that a large number of gene trees (1,089), built from a set of protein families with both archaeal and bacterial members (1,479), placed haloarchaeal genes in a monophyletic group with bacterial rather than archaeal homologs and 2) diverse phylogenetic evidence supporting the Methanomicrobia as sister group to the Haloarchaea [26]–[28] (see also Dataset S3). Further supporting this hypothesis, several of these transferred genes were associated with functions required for the proposed physiological transformation of an obligately anaerobic, autotrophic methanogen to a heterotrophic, facultatively aerobic haloarchaeon [25].
Accuracy of inferred evolutionary events is strongly influenced by the selection of representative species on which the inference is based. We tested whether the conclusions made by Nelson-Sathi, et al. [25] from an analysis of 10 haloarchaeal genomes were robust to a reanalysis against our more phylogenetically diverse dataset. In our analysis we also observed a large amount of horizontal gene transfer from the bacteria, however, we found that this exchange of genetic material was not limited to a single acquisition event at the haloarchaeal root, but rather occurred in many different transfer events throughout haloarchaeal evolution.
For each of the protein families in the original set reported by Nelson-Sathi, et al. [25], we added homologs from 65 additional haloarchaeal genomes and rebuilt gene trees using the same software tools and parameters (obtained via correspondence with the authors). More than two-thirds (67.2%) of the protein families originally designated as basal acquisitions no longer retained this characteristic after incorporating the additional haloarchaeal homologs (Table 1). Analysis of the re-computed gene trees revealed that, not only did most transfers not happen near the base of the haloarchaeal clade, but, for many protein families, multiple independent transfer events from bacteria to the haloarchaea have occurred. Depending on the gene, these additional transfers either predate (Figure S3) or follow (Figure S4) the acquisition discovered by Nelson-Sathi, et al. Both cases are inconsistent with a single, basal transfer scenario: rather, our results are consistent with previous findings that horizontal gene transfer is rampant among bacteria and archaea [29]. This interpretation is further supported by the fact that the putatively transferred genes do not appear to have been transferred from a common bacterial phylum, as indicated by the phylogenetic affiliation of the most closely related bacterial homolog for each protein family (see Table S2 in [25]).
The simplest explanation for the difference in findings between the original study and our reanalysis is rooted in the nature of the two genomic datasets investigated. Due to the limited number of genomes available at the time, Nelson-Sathi, et al. worked under the assumption that the 10 haloarchaeal genomes they sampled reasonably represented haloarchaeal diversity. This lead to the identification of genes as bacterial transfers to the last common ancestor of the haloarchaea on the basis of as few as two haloarchaeal homologs. Based on the distribution of genomes used (two from Clade 1, two from Clade 2, and six from Clade 3, see Figure 2), a gene present in only two haloarchaea will often represent a transfer to a single clade, rather than to the haloarchaeal root. By contrast, our genomic dataset, representing a more even phylogenetic sampling of the haloarchaea, revealed multiple, clade specific transfer events indicative of a complicated history of gene transfer between the haloarchaea and bacteria. This analysis highlights the value of large, phylogenetically informed genomic data sets for increasing the accuracy with which we can make evolutionary inferences, and reveals the dangers in making wide-reaching evolutionary claims based on limited genomic data.
Despite their metabolic and physiological diversity [9], [10], all members of the haloarchaeal clade share an obligately halophilic lifestyle. To understand the common mechanisms underlying this lifestyle, we investigated the haloarchaeal core genome. A total of 304 of the in-house defined protein families (see Materials and Methods) were present in all of the 80 investigated genomes (Table S4). Of these 304 core proteins, 55 (18.1%) were predicted to be involved in translation, transcription, or regulation thereof. In addition to the expected ribosomal proteins, RNA polymerase subunits, and known general transcription factors, the core genome included a number of predicted transcription factors whose functional importance in regulating haloarchaeal gene expression is underexplored. These include a ArsR-family transcription factor involved in alleviation of heavy metal toxicity, an AsnC-family member involved in feast/famine response, a CBS domain-containing protein of unknown function, and a PadR-family protein, possibly involved in regulation of phenolic acid metabolism [30]. As the level of functional specificity provided by domain-level matches is limited (for example, PadR-family proteins have also been shown to be involved in regulation of multidrug pumps [31]) the contributions of these conserved transcriptional regulators to haloarchaeal biology will need to be experimentally determined. However, their wide distribution across 23 haloarchaeal genera suggests that these proteins likely play important roles in regulating physiological responses to shifting environmental parameters routinely experienced in hypersaline environments, including changes in oxygen availability, salinity, and concentrations of heavy metals.
The core genome includes 10 protein families predicted to be involved in stress response, including both cold shock and heat shock members as well as stress response proteins with no predicted specific function. Previous studies of haloarchaea have shown upregulation of heat and cold shock genes in response to salinity changes, indicating that these chaperones may play a wider role in mediating stress responses than previously believed [32], [33]. As such, these 10 core haloarchaeal stress response proteins are candidates for experimental work to study the complex interplay among different stress response mechanisms.
DNA mismatch (MutSL), homologous recombination (RadAB) and base excision repair mechanisms are also universally conserved in this clade. Conspicuously lacking from the haloarchaeal core genome is a photolyase, responsible for correcting UV induced thymidine-dimers. Seven species were missing an annotated photolyase, all of which were Clade 1 haloarchaea (see Text S1 and Figure 2). If these genes are indeed absent – not hiding in unassembled regions of the genome - their absence would be surprising, given that many haloarchaeal species are routinely exposed to high levels of UV radiation due to evaporation of shallow hypersaline lagoons. Only two of these species (Natrialba taiwanensis and Natrialba aegyptia) encode genes annotated as belonging to the UVR system of UV damage repair, with each encoding only one of the five proteins in this repair system. Although haloarchaeal high G+C content has been proposed as an adaptation for avoiding UV-induced dimerization [34], potentially reducing the need for a photolyase repair system, the link between G+C content and UV damage in this clade is far from certain and alternate explanations for high G+C content have been proposed [35].
A number of transport-related protein families were also universally conserved, including several ABC transporters with peptides, amino acids and/or metals as predicted substrates. ABC transporters were extremely abundant, making up six of only eleven protein families with greater than 400 members. Although precautions were taken to filter out spurious domain-level matches (see Materials and Methods), each of these ABC transporter families may be composed of many members with divergent substrate specificities. Four protein families related to phosphate transport were also conserved, including two low-affinity phosphate transporters and two regulators of phosphate transport.
Not surprisingly, a number of proteins involved in biosynthesis of isoprenoid lipids were conserved across the haloarchaea. Isoprenoids are characteristic of haloarchaeal cell membranes [36], and are known to reduce membrane permeability to Na+ and Cl− ions [37], a necessary prerequisite for regulating ionic composition at high salinities. The committed step in isoprenoid synthesis may be upregulated under high salt conditions [38], further suggesting their importance to cell maintenance in hypersaline environments. However, as isoprenoids serve as precursors for a number of other compounds, alternative hypotheses must also be considered.
Fifty-five of the 304 conserved haloarchaeal proteins (∼20%) had either no assigned functional annotation or only a domain-level match to a previously characterized protein. Based on their high conservation, these proteins likely play important roles in the haloarchaeal biology, including adaptation to hypersaline environments. These 55 protein families, therefore, represent a manageable set of targets for exploring the genetic mechanisms of halophilicity.
Haloarchaea are generally considered “salt-in” strategists – actively accumulating potassium and chloride ions to prevent water efflux in hypersaline environments. In contrast, the “salt-out” strategy entails accumulation or synthesis of organic compatible solutes to increase internal osmolarity without increasing cytoplasmic salinity. Although recent work has shown that some haloarchaea may utilize compatible solutes in some situations [39]–[41], and many halotolerant organisms transiently accumulate moderate levels of intracellular K+ ions in the initial stage of osmoadaptation [42], [43], distinguishing between salt-in and salt-out strategists remains useful for differentiating between obligate and facultative halophiles.
Due to the dynamic nature of hypersaline environments, the haloarchaea possess a range of ion transporters for accommodating fluctuating salinity levels. We investigated the phylogenetic distribution of a number of ion transporter genes potentially involved in osmoadaptation to hyper-osmotic or hypo-osmotic shock, as well as compatible solute import and biosynthesis genes (Figure S5). This analysis enabled us to propose a generalized haloarchaeal strategy for dynamic osmoadaptation (Figure 4A).
During osmotic upshock, potassium import and sodium extrusion are mediated by secondary transport, using the proton gradient generated either by direct light-activated proton translocation through bacteriorhodopsin (BR) or through respiration. All 80 haloarchaea investigated possess a H+/K+ symporter of the Trk family for potassium uptake, while few (10 species) also possess the closely related Na+/K+ symporter of the Ktr family (Figure 4B). This observation is consistent with Corratgé-Faillie et al.'s. prediction that use of the Trk system may be evolutionarily advantageous in hypersaline environments by avoiding Na+ uptake [44]. Many haloarchaea (66 species) may power sodium extrusion through YrbG Na+/Ca2+ antiporters, although this strategy would be limited to environments with high calcium concentration. During osmotic downshock, it is vital for organisms using a salt-in strategy to rapidly rid the cytoplasm of excess salts to avoid hypertonic cell lysis. The haloarchaea have the genetic potential to export excess potassium ions through a combination of secondary transport with Kef-like H+/K+ antiporters and non-specific ion loss through the mechanosensitive channel MscS, which has been shown to play an important role in potassium efflux during osmotic downshock in E. coli [45].
In contrast to the prevalence of these predicted secondary transporters, only a small number of genomes encode ATP-dependent transporters for osmoadaptation (Figure 4B). Prominently lacking in most haloarchaeal genomes are the outwardly rectifying Na+ pump NatABC and the inwardly rectifying K+ pump KdpABC, both of which depend on ATP hydrolysis to power ion transport. Previous researchers have calculated a large difference in energetic cost between salt-in and salt-out strategies, with production of compatible solutes being more costly than establishment of ionic gradients [46]. We propose that, in situations where large amounts of material must be transported, even a small difference in energetic efficiency between secondary and primary transport would result in a bias towards secondary transport systems, which require fewer steps. If true, this would explain the observed bias of haloarchaeal genomes for secondary transport systems for K+ accumulation and Na+ extrusion. However, experimental work will be required to calculate the ion exchange stoichiometry of these transporters and to determine the relative efficiency of secondary versus primary transport in this system.
Haloarchaeal strategies for uptake of chloride are difficult to decipher from the genomic data, as metabolism of this important counterion is not well understood [47]. The most recent review of the topic states that chloride is imported through “cotransport with sodium ions and/or using the light-driven primary chloride pump halorhodopsin” [47]. However, although experimental work has suggested the presence of a light-independent chloride uptake system in Halobacterium sp. NRC-1 [48], neither the energy source for this system, nor a genetic mechanism for its implementation have been identified. As only 41 of the organisms analyzed here possess a halorhodopsin homolog, some alternate strategy for chloride import must exist. We screened for homologs to archaeal and bacterial chloride transport proteins, including a predicted (Na+/K+)/Cl− symporter from Methanosarcina acetivorans [49], a predicted bacterial cation chloride transporter from Aminomonas paucivorans, and EriC, a H+/Cl− antiporter [50] involved in acidic shock tolerance in Escherichia coli [51]. Of these, we discovered only two EriC homologs, belonging to the alkalitolerant Natrialba aegyptia [52] and the alkaliphilic Natrialba magadii [53]. Based on these observations we propose a possible role for EriC homologs in Cl− uptake in alkaliphilic environments where export of protons down their concentration gradient would enable accumulation of a high intracellular level of chloride ions.
Due to recent interest in haloarchaeal use of compatible solutes (as part of a broader osmoadaptation strategy which also includes potassium accumulation) [39]–[41], we analyzed the phylogenetic distribution of many genes involved in compatible solute transport and biosynthesis (Dataset S4). This analysis reveals prevalent uptake mechanisms for the compatible solutes glycine betaine, ectoine, and proline (Figure 4C), commonly used as osmoprotectants by facultative halophiles. Chemotaxis towards, and accumulation of the trimethylammonium compounds glycine betaine, carnitine, and choline, has been demonstrated in a model haloarchaeaon [39]. We find that, although putative homologs to the binding and transducer proteins for this system (CosB and CosT) are widely distributed, only 10 species possess both members (Figure S5). Due to high levels of sequence similarity between CosB and the trimethylammonium compound transporters OpuCC/OpuBC [39], we cannot confidently assert presence of a trimethylammonium compatible solute chemotaxis system in these species, and caution that experimental work must be done to validate these results. The abundance of compatible solute transporters within haloarchaeal genomes does not, however, necessarily indicate utilization of a salt-out strategy. During periods of decreased environmental salinity haloarchaea often coexist with halotolerant microorganisms, which are generally salt-out strategists. As evaporation increases salinity, these organisms lyse and the released compatible solutes may then be used as carbon and nitrogen sources by extreme halophiles [54]. Compatible solutes have also been shown to have thermoprotective effects in both mesophilic bacteria [55] and hyperthermophilic archaea [56]. Experimental evidence of compatible solute accumulation suggests that some haloarchaea may utilize a salt-out strategy under certain conditions [39]–[41].
Recently, biosynthesis of the compatible solute trehalose was found to be widespread in the haloarchaea [41]. We interrogated our genomes for genes associated with compatible solute biosynthesis, including the ProJH pathway for proline synthesis during osmotic upshock [57], the Ne-acetyl-β-lysine and cyclic 2,3-bisphosphoglycerate (cBPG) synthesis pathways [43], [58], and the BetAB/GbsBA pathways for oxidation of choline to glycine betaine [59], [60] (Figure S5). We found that, in contrast to the widespread mechanisms for compatible solute uptake, pathways for biosynthesis of these compatible solutes were rare. Complete pathways for osmotically regulated proline synthesis and Ne-acetyl-β-lysine production were absent in all 80 genomes. Although putative homologs to ProH were present in six species of Halorubrum, their function is unclear in the absence of ProJ, which catalyzes initial transformation of glutamate for proline biosynthesis [57]. One species (Natronorubrum tibetense) was found to encode an intact pathway for biosynthesis of cBPG, previously thought to be restricted to methanogens [58]. Only nine species were found to possess homologs for the three components required for glycine betaine synthesis from extracellular choline: 1) a choline transporter, 2) a choline dehydrogenase, and 3) a glycine betaine aldehyde dehydrogenase. In eight of these species, the preferred strategy appears to be choline uptake via OpuB, oxidation to glycine betaine aldehyde using GbsB, and final oxidation to glycine betaine via GbsA/BetB. The ninth species (Halococcus saccharolyticus) appears to utilize the flavin adenine dinucleotide-bound choline dehydrogenase BetA rather than the type III alcohol dehydrogenase GbsB [58] for initial choline oxidation.
We also investigated uptake mechanisms for the biologically important magnesium and phosphate ions (Figure 4D & E). For phosphate accumulation, all 80 sequenced haloarchaea encode two members of the PHO4 superfamily (PF01384), annotated as an anion permease (TRIBE-MCL cluster 105; Tribe105), and a sodium dependent phosphate transporter (Tribe63). This second annotation potentially enables active phosphate uptake at the expense of the sodium gradient in hypersaline environments. As discussed above in the context of potassium and sodium transport, ATP-dependent phosphate uptake appears not to be favored, with only 22 homologs of ABC-type phosphate transporters encoded in this genome set. The opposite was found to be the case with magnesium transport, with the ATP-dependent transporters MgtA and MgtB being far more common than the inwardly rectifying Mg2+ channels MgtE and CorA. However, with only 52 of 80 sequenced haloarchaea encoding at least one of these Mg2+ transport mechanisms, it is clear that alternate strategies for magnesium uptake remain to be discovered. As magnesium concentrations have been shown to play important roles in stabilizing halophilic enzymes [61], discovery of these alternative magnesium uptake strategies is vital to understanding the nature of halophilic proteins.
Our investigations into haloarchaeal ion transport reveal both a core set of highly conserved strategies (eg. Trk-based K+ uptake, Na+-mediated phosphate accumulation) as well as more sparsely distributed abilities (eg. alkaliphilic chloride import via EriC). We have observed that secondary transport seems to be preferred to ATP-dependent primary transport for maintenance of large ion gradients in hypersaline environments, and have identified important gaps in our understanding of chloride and magnesium accumulation. The generalized nature of our model of dynamic osmoadaptation in haloarchaea, based on a broad and deep sampling scheme, contrasts with the specificity of previous models, which were largely limited to single model systems widely spaced across the phylogenetic tree. By highlighting strategies conserved across the haloarchaea, we hope to help build a general understanding of osmoadaptation across the tree of life.
In addition to the ability to rapidly adapt to changing environmental salinities, haloarchaeal adaptations to high salt also include intrinsic physiological features present under all environmental conditions. These adaptations include an acidified proteome and high genomic G+C content [35], [62], [63], discussed here and in the subsequent section, respectively. Proteome acidification may be beneficial for salt-in strategists by prevention of protein aggregation via the ability of acidic amino acid residues to reorganize protein-solvent interactions [61], [64], [65], although alternative explanations have been proposed [66]. As expected, all 59 organisms whose draft genome we report here have both high genomic G+C content (ranging from 59–69%) and a highly acidified proteome. Histograms of predicted isoelectric points of haloarchaeal proteomes revealed an asymmetric bimodal distribution, with a dramatically larger major mode around pH 4.5, a minor mode around pH 10.0, and a consistent overall shape across haloarchaeal species (Figure 5 and Data Dryad package [67]). Corroborating previous work, this major mode was shifted towards lower pI values compared to non-haloarchaea [68].
Certain haloarchaeal proteins were not acidified, including many ribosomal subunits, membrane proteins, and DNA-binding proteins (Table S5). Potential reasons for non-acidification include 1) shielding from the hypersaline cytoplasm (eg. internal ribosome subunits, membrane proteins), and 2) presence of a selective force against acidification (eg. DNA-binding proteins). Structural mapping of the ribosome demonstrates that subunits exposed to the hypersaline cytosol tend to be acidified, while shielded internal subunits have high pI (Figure 6). Large regions of transporters and other membrane–associated proteins are likewise shielded from the saline cytoplasm by the cell membrane, mitigating selective pressure to acidify. DNA-binding proteins must retain positively charged residues to interact efficiently with negatively charged DNA, and so also tend not to acidify (shown for the general transcription factor TATA-binding protein (TBP), Figure S6, and ribosome elongation factor α-1, Figure S7). A large number (11,087) of non-acidified proteins were unannotated. Based on functional consistency we identified among other high pI proteins, we propose these proteins as candidates for exploratory research seeking novel DNA-binding or membrane associated proteins such as transcription factors, transporters, and chemotaxis/sensory receptors.
In addition to their acidified proteomes, the highly G+C biased genomes of the haloarchaea are also predicted to be an adaptation for life in high salt. Although the mechanism for this adaptive benefit is unknown, several possibilities have been proposed, including decreased risk of thymine dimers resulting from high UV exposure in shallow brine pools [34], or selective pressure driven by A+T bias of insertion sequence elements [35]. Notably, the only known haloarchaeon lacking a G+C biased genome – Haloquadratum walsbyi (48%) – possesses a large number of photolyase genes, postulated to enable it to mitigate the effects of UV induced pyrimidine dimerization [69]. The Nanohaloarchaea, an uncultured clade of halophilic archaea proposed as a sister group to the Haloarchaea, also have low G+C content (43 and 56% for the two members of this clade with draft genomes), although they inhabit the same hypersaline environments as the high G+C Haloarchaea [70]. The evolutionary rationale behind this difference is unknown.
Regardless of the mechanism for its maintenance, genome-wide G+C bias offers a method for identification of candidates for horizontal gene transfer from organisms with G+C content differing from the recipient species, as horizontally transferred genes are often A+T shifted relative to the host genome [71]. We examined the G+C content of the 80 haloarchaeal genomes, using a sliding 100 bp window, and conducted change-point analysis to extract regions with local G+C content differing from the genome average. It is important to note here that haloarchaeal plasmids, including minichromosomes and megaplasmids, are known to have decreased G+C content compared with primary replicons (“chromosomes”) [63]. As the mechanisms for maintaining decreased G+C content in smaller replicons are unknown, and in order to accommodate draft genomes where the identity of the primary replicons are unknown, we have chosen to be replicon size neutral. Some regions of the genome are also expected to have low local G+C content due to selective pressure for maintaining a higher A+T percentage (eg. origin of replication sites). In addition, we have been neutral as to the direction of divergence from genome-wide G+C average, in order to allow detection of regions of unusually high as well as unusually low G+C (Figure 7, Figure S8, Datasets S6 & S7).
We found these regions to be highly enriched in protein families involved in DNA metabolism and transcriptional regulation, transmembrane transport (and other membrane proteins), and horizontal transfer of genetic information. Specifically, of the seventy-nine functionally annotated protein families with at least five members which were enriched at least eight-fold in the divergent G+C regions, thirty-six (45%) were annotated with DNA/RNA-binding capabilities, eleven (14%) were associated with horizontal gene transfer mechanisms, and seven (9%) were associated with the cell membrane or cell surface (Table S6). These results are consistent with our investigations into non-acidified haloarchaeal proteins, in that both analyses identified nucleic acid binding and transmembrane proteins as potentially shielded from selective pressure to acidify and accumulate high G+C content. However, the specific proteins identified by these analyses were not identical and previous work has indicated that G+C bias and acidification are not correlated for individual proteins [35]. We speculate that many of the 138 unannotated protein families enriched in these regions of abnormal G+C content may be involved in DNA or RNA binding. We provide these regions in our Data Dryad package [67], as a rich source of data for identification of novel nucleic acid binding proteins and investigation of functionally important horizontal gene transfer events into the haloarchaea.
In addition to local variation in G+C content, we also investigated variation at the genus level. We found that, although some genera display little variability in genomic G+C content (eg. Halorubrum, Haloarcula), others exhibit a wide range (eg. Haloferax, Halococcus) (Figure 8C). This wide deviation in G+C content cannot be attributed to tolerance of a wide range of salinities, as the known NaCl tolerance range of Haloarcula and Halococcus species are very similar (3.2 M and 3.5 M respectively), as are those for Haloferax and Halorubrum species (4.1 M and 4.2 M respectively) [72]. Thus, the link between high G+C content and salinity tolerance in the haloarchaea appears to be more complex than previously appreciated.
Recent work has uncovered surprising roles for eukaryotic and archaeal general transcription factors in mediating differential gene regulation during cellular differentiation and environmental response [73]–[75]. In the haloarchaea, both the TATA-binding protein (TBP) and transcription factor B (TFB, known as transcription factor IIB in eukaryotes) families have undergone extensive expansion [75], [76]. TFB paralogs of Halobacterium sp. NRC-1 have been shown to differentially contribute to fitness under stresses commonly encountered in hypersaline environments, including variations in salinity and heavy metal concentration [75]. Multiple TBP and TFB paralogs may enable haloarchaeal species to quickly and efficiently modify transcriptional response to these environmental fluxes.
We examined the evolutionary history of haloarchaeal TBP and TFB homologs in order to understand their potential impact on environmental response. Phylogenetic distribution of paralog classes suggest that expansions of the TFB family are ancient, with several duplications occurring prior to haloarchaeal diversification (Figure 9, Dataset S5, Figure S9, Dataset S6). Homologs of five of the seven TFB paralogs from the model haloarchaeon Halobacterium sp. NRC-1 were present in at least 79 of 80 sequenced isolates, while another (tfbA) was present in 74 isolates. The remaining paralog, tfbE, was found in only 39 of the 80 genomes sequenced, suggesting either that this paralog emerged from a relatively late gene duplication event, or has been lost from a large number of genomes.
Evolutionary expansion of the TBP family appears to be a more recent phenomenon, with three haloarchaeal lineages showing distinct patterns of duplication and divergence (Figure 10, Dataset S7, Figure S10, Dataset S8). Phylogenetic distribution of TBP paralogs suggests that the ancestral haloarchaeal TBP was most similar to tbpE of Hbt. sp. NRC-1, with only one species (Natrinema pallidum) apparently lacking this homolog. We speculate that this gene may be present at a contig boundary in the assembly for this organism (which consists of 116 contigs), and may later be uncovered by additional sequencing. The ancestral nature of the tbpE homolog is also supported by it being the only TBP in the natural TBP knockout strain Halobacterium salinarum PHH4 [77].
The previously recognized tbpD/B/F expansion appears to be limited to Halobacterium species and three haloalkalitolerant Clade 1 haloarchaea (Natrialba aegyptia, Natrialba taiwanensis, and Natrinema pellirubrum). In addition to this well-known expansion in the Halobacterium clade, we have uncovered two additional clade-specific diversifications. The smaller of these two expansions appears to be the result of a single gene duplication event (giving rise to tbpW) at the base of the Halococcus genus. An additional expansion has occurred near the base of the Clade 2 haloarchaea, with each species possessing at least one (tbpX) and in the Haloferax genus, up to three (tbpX, tbpY, and tbpZ) TBP homologs derived from this duplication event.
Finally, our analysis of haloarchaeal TBPs revealed a large number of tbpC-like homologs. For eukaryotes and archaea, TATA-binding protein normally consists of two domains derived from a duplication event. In each domain, DNA-binding is dependent upon a pair of intercalating phenylalanines [78]. In Halobacterium spp., the tbpC gene has lost the N-terminal phenylalanine pair, while retaining the C-terminal pair. The tbpC gene is easily knocked-out in Hbt. sp. NRC-1 [76] and was not detected at the transcriptional level under any growth condition tested in Hbt. salinarum PHH1 [77]. These data suggest that TbpC may be either nonfunctional or may play a very specialized role in transcriptional regulation under non-laboratory conditions. It is unclear whether defunctionalization may be a result of loss of DNA-binding ability by the N-terminal TBP domain or if loss of the intercalating phenylalanines may be part of an overall loss of function resulting from relaxed selective pressure. Our phylogenetic analysis grouped 21 TBP homologs from 18 species with Halobacterium spp. tbpC, of which 18 sequences were missing the N-terminal phenylalanine pair, one was missing a single phenylalanine from the N-terminal pair, and two possessed all four phenylalanine residues. Interestingly, sequences missing only the C-terminal phenylalanines, and several sequences missing the N-terminal pair, were not grouped with the tbpC homologs, suggesting formation of this clade is not merely an artifact of long-branch attraction. Collectively, this evidence suggests multiple losses of DNA-binding ability in either the N-terminal or C-terminal TBP domain (presumably, sequences having lost both pairs of DNA-intercalating phenylalanines have lost transcription factor function).
Specialization of haloarchaeal general transcription factor paralogs has been implicated in regulating response to a number of environmental perturbations, including variations in temperature, salinity, pH and concentration of heavy metals [75]. Understanding the complicated history of haloarchaeal TBP and TFB diversification will facilitate design of evolutionarily informed experiments for investigating the contribution of general transcription factor paralogs to fitness in the dynamic environments in which these species live.
In addition to the applications we have already discussed (primarily focused on learning about osmoadaptation), our dataset has the power to address diverse problems in genomics, metagenomics, and other areas of bioinformatics. Here we illustrate some examples highlighting the diverse applicability of our dataset.
Strains were acquired as desiccated cells from the American Type Culture Collection (ATCC) in Manassas, Virginia, USA; the Leibniz Institute DSMZ German Collection of Microorganisms and Cell Cultures (DSM) in Braunschweig, Germany; and the Japan Collection of Microorganisms (JCM) in Ibaraki, Japan, as indicated in Table S2. Cells were rehydrated in recommended media according to culture collection center protocols and grown to stationary phase at 37°C in liquid culture. Genomic DNA was harvested with Wizard Genomic DNA purification kit (Promega).
Sequence libraries were constructed using a combination of standard fragmentation (200–500 bp), mate-pair fragmentation (6 Kbp), and PCR-free transposon-mediated insertion of sequencing primers (Epicentre Nextera) [81]. Transposase was purified from E. coli BL21 (DE3) containing the cloning vector pWH1891. For mate-pair libraries, 6 Kb pair-end libraries were constructed and the terminal 50 bases of each end were sequenced, according to standard protocols. Additional libraries were constructed by combining DNA fragments from haloarchaeal species distantly enough related to enable unambiguous assignment of non-barcoded reads (“SOUP”). All sequencing was performed on Illumina HiSeq and GAII platforms. The paired-end information and trimming information were specified using annotation strings on the description line of the reads. Reads were assembled using the a5 pipeline [18]. Following assembly, genomic DNA contamination arising from transpose-mediated library preparation was removed by searching assembled reads against a local BLAST database consisting of E. coli BL21 (DE3) genomic DNA and the cloning vector pWH1891. BLAST hits with an E-value ≤ 10−20 were considered to be significant matches and candidates for contamination. Contigs with matches covering ≥80% of the contig length and contigs ≤1 Kbp with matches covering any portion of the contig were treated as contamination and discarded from further analysis. Long contigs for which only a small portion of the contig matched to the local BLAST database were also discarded if there were either no annotated features, or if the annotated features were E. coli genes. These criteria resulted in a total of 497 contigs equaling 265.57 Kbp being removed from the final assemblies.
A dual annotation pipeline was implemented in order to take advantage of the strengths of different existing automated annotation tools. Assembled genomes were first submitted to the Rapid Annotation using Subsystem Technology (RAST) server at the National Microbial Pathogen Data Resource. RAST-based gene calls and annotations were used for building of protein families, core genome and pan genome analyses, phylogenetic reconstruction, analysis of general transcription factor expansions, GC-bias analysis, building of molecular marker sets and phylogenetically informed re-annotation. These annotations can be accessed as a loadable “popular genome set” through the genome context viewer JContextExplorer [17] as well as a custom MySQL database (see Text S1 for instructions). The RAST annotation system was particularly useful in enabling comparison of our genomes with previously sequenced haloarchaea, by allowing standardization via rapid reannotation of existing genomes.
In addition, the newly sequenced genomes were annotated using NCBI's Prokaryotic Genome Annotation Pipeline (PGAAP) [14]. PGAAP gene calls and annotations were used for COG analysis, proteome acidification calculations, and genera-based genomic feature comparisons. These annotations can be accessed through the NCBI website using the accession numbers listed in Table S2 as well as through our custom MySQL database.
A phylogeny was constructed for all archaeal genomes available through the Integrated Microbial Genomes database along with our sequenced haloarchaea using a concatenated set of 40 conserved marker genes [20]. Peptide sequences were downloaded for all archaeal genomes from the IMG 4.0 database on January 4, 2013. HMM profiles of 40 bacterial and archaeal PhyEco markers were searched against these peptide sequences. We excluded genomes with less than 35 of these markers, as well as duplicate genomes which were included in both our analysis and the IMG database. A total of 151 IMG archaeal genomes and all 80 haloarchaeal genomes were included in the phylogenetic tree building. For each PhyEco marker family, only single-copy members from the genomes were included. Independent alignments were built using MUSCLE [82] for each gene families and then concatenated. A phylogenetic tree was built from the concatenated alignment using PHYML 3.0 [83] with the LG substitution model. Tree topology and branch lengths were optimized by the program and aLRT SH-like statistics was used for branch support estimation. The R package ape was used to remove duplicate genomes [84] and the final tree was visualized using FigTree [85]. The tree files are available as Dataset S3 (all archaea) and Dataset S11 (haloarchaea only).
Homologous protein families were constructed using a Markov clustering-based approach. First, an all-vs-all BLASTp search of all protein coding genes called by RAST in the 80 haloarchaeal genomes was conducted using an E-value cutoff of 10−5, soft masking, and the Smith-Waterman alignment algorithm. These options have been shown to improve homolog detection over other BLAST methods [86]. To enable detection of large homology groups, the number of returned matches was set at 100,000. To remove spurious, domain-level matches, BLAST results were post-filtered and matches retained only if bi-directional query-match coverage was ≥80%, and each sequence was ≥75% of the length of the other. Matches with E-values>10−10 were also excluded at this stage (Text S2).
BLAST results were then clustered into homology groups using the clustering algorithm TRIBE-MCL [16], which utilizes the protein-protein similarity network implicit in the BLAST scores. This method has the benefit of building inclusive homology families including orthologs, xenologs and both in- and out-paralogs, unlike reciprocal best BLAST methods which often overlook complex gene relationships in the search for one-to-one ortholog mapping. To determine an appropriate inflation parameter for TRIBE-MCL clustering, homology clusters derived from inflation parameters ranging from 1.4 to 8.0 were compared and benchmarked using manually curated protein families for haloarchaeal TATA-binding proteins and opsins. The inflation parameter I = 2.5 was selected, as it most closely recapitulated known ortholog/paralog relationships for the selected protein families. A total of 17,591 homology clusters (protein families) were obtained with this method, comprising 276,364 of 303,129 total haloarchaeal proteins. The remaining proteins were singletons, lacking significant sequence similarity to other sequences in the dataset. These sequences may represent species-specific innovation, recent gene influx from horizontal gene transfer, or simply a lack of sequencing depth along some sampled haloarchaeal lineages.
Random subsets of haloarchaeal genomes (ranging from zero to eight genomes) were selected from the set of RAST-annotated haloarchaea, using the Java Samplers package [87]. For each selection the following quantities were tabulated: (1) the number of unique COGs, (2) the number of unique TRIBE-MCL determined homology clusters, and (3) the number of unique non-singleton TRIBE-MCL determined homology clusters. Ten thousand random selections were performed for each sample size and the average number of unique protein families obtained for each sample were plotted in Microsoft Excel (Figure 3).
The 1,479 protein families used in the analysis described in [25] were obtained from the authors. All genes were renamed with an “E”, “A”, or “H” tag prepended to their original NCBI gene ID number, according to their taxonomic classification as, respectively, (eu)bacteria, archaea, or haloarchaea. Proteins from 65 haloarchaeal species not included in the original study were assembled into a set of “additional haloarchaeal proteins”, and named with an “H” tag and unique ID number. Five haloarchaeal genomes in our set were either different strains of a species included in the original study, or a re-sequencing of the same strain, and so were excluded from re-analysis. Gene trees were re-constructed for all 1,479 of the original protein families, using the alignment program MAFFT [88], with command line options –legacygappenalty, –anysymbol, and –quiet; the protein model determination tool ProtTest [89], using command line options –all-matrices, all-distributions, -S 2 –F –t1; and the gene tree creation tool PhyML [83]. All command line options were obtained from the authors of [25]). Gene trees were determined to be single transfers from the bacteria only if (1) all haloarchaeal homologs formed a monophyletic group and (2) this monophyletic group rooted with bacterial rather than archaeal homologs. For gene trees matching these criteria, corresponding protein families were extended to include homologs predicted from the set of “additional haloarchaeal proteins”, using BLAST with an E-value cutoff of 10−10 and a minimum percent identity of 30%. Gene trees were generated for these extended protein families using the parameters described above. Extended gene trees were re-evaluated according to the criteria described above to determine whether they could still be classified as single bacterial transfers. Some (17.2%) of the extended protein families contained a sufficiently large number of members such that determination of gene trees was infeasible, and so were excluded from re-analysis. Gene tree classification results are provided in Table 1. Phylogenetic trees visualized in Figures S4 and S3 were made using Phyfi [90].
A list of 74 genes involved in ion transport and compatible solute transport or synthesis was compiled. Representative protein sequences for each of the selected genes were obtained by searching the NCBI protein database by annotation (Dataset S12). When available, sequences from the model organisms Escherichia coli and Bacillus subtilis were selected. Genes with “MM” prefix derive from Methanosarcina mazei Go1. BLASTp searches for these query sequences were conducted against a local BLAST database containing 80 haloarchaeal genomes. In order to obtain all potential homologs, the maximum target sequences parameter was set to 100,000 and an E-value cutoff of 1 was used. Resulting matches were then filtered using custom perl scripts to remove spurious domain-level matches by requiring a bidirectional query-target coverage of ≥80%, and a bidirectional query-target length of ≥75%. To reduce the number of non-specific family-level matches (for example, to differentiate between the closely related Trk H+/K+ and Ktr Na+/K+ symporters), sequences were also filtered to remove matches with an E-value of>1e−20. A presence/absence matrix was constructed from these filtered BLAST results using custom bash and R scripts, and visualized in iTOL [91], [92]. Genes were grouped into functional categories based on substrate specificity and directionality for ion transporters, substrate imported for compatible solute transporters, or product for compatible solute biosynthesis proteins. As 20 of the query genes were not detected in the set of investigated species, only 54 columns appear in Figure S5. Gene presence/absence data was scaffold on the multi-marker concatenated haloarchaeal phylogeny (Dataset S11).
Whole-genome feature data for the 59 newly sequenced, PGAAP-annotated haloarchaea were retrieved from an in-house MySQL database using custom scripts [67]. For each organism the genome size, number of contigs, percent G+C content, percent coding, number of coding regions, and number of insertion sequence elements were extracted. Genomes were organized into genera and box plots for each of the above traits were drawn for each genus using MATLAB's Statistics toolbox [93]. Images were exported using the supplementary export_fig function [94]. See Figure S23.
All isoelectric point analyses were carried out exclusively on the PGAAP gene call set. The pI of every protein coding sequence was computationally predicted using the protein sequence isoelectric point prediction method included in the BioJava framework [95]. pI histograms for each genome were computed using MATLAB's statistics toolbox [93], sorting proteins into 100 equally spaced bins ranging from a pI of 2.0 to 13.0. All proteins with a pI of 7.5 or greater were collected and a list of all represented gene annotations was procured. For each gene annotation in this list, all instances were tabulated, including the number of instances with a pI greater than or equal to 7.5. A table showing these results, sorted by frequency of instances with high pI can be found in Table S5. All proteins annotated as ribosomal subunits were collected, and all subunits which could confidently be mapped to experimentally characterized structures for the large (1QVG) [96] and small (1FKA) [97] ribosomal subunits were mapped. Subunits with a pI>7.5 in 60% or more of cases were colored red, and subunits with a pI<7.5 in 60% or more of cases were colored blue. Any subunits which could not be confidently mapped or which had variable pI across species were left the default color of green. All mappings were done using the PyMOL protein visualization program [98].
Comparative acidification visualizations of TBP and ribosome elongation factor α-1 proteins for haloarchaea and non-halophilic organisms were generated using the SWISS-MODEL web interface [99] and 1D3U [100] and 3VMF [101] as the structural templates, respectively.
To investigate localized variations in G+C content across each genome, custom scripts were created (Text S3 & S4). First, %G+C was calculated for 100 bp windows across the genome, with a 20 bp step size. The terminal <100 bps of each contig were not included in calculation, so as not to artificially inflate the final G+C calculation for each contig. Windows with>10% ambiguous nucleotides were assumed to have %G+C equal to the contig mean. Plots were generated showing a) the overall mean G+C percent for the entire genome as a horizontal green line, b) the contig boundaries as vertical red lines, c) the mean G+C percent for each contig as horizontal blue lines, and d) all G+C percentages for individual 100 bp segments as a black line [67]. Steps where the mean %G+C inflects away from the local mean were calculated using the R changepoint package [102], and these changepoints were plotted over the stepwise G+C percentages as vertical green lines [67]. After manual curation of changepoints for each species, annotated features between changepoints were retrieved from each species' general feature format (GFF) file. Ten species were excluded from this analysis, either due to having a very large number of contigs, or having no observable inflections in local mean G+C content. For a list of excluded organisms and reasons, see Text S1. The set of features extracted from regions of abnormal G+C content were analyzed for enrichment of homologous protein families (TRIBE-MCL clusters). The frequency of each protein family in the abnormal G+C feature set was compared to its frequency in the entire genome set, and families with at least eight-fold enrichment were investigated further. To avoid including families which were enriched artificially by having a small number of members which are all present in abnormal G+C regions, families with fewer than five members were excluded from enrichment analysis.
A local BLAST database was constructed using RAST-derived gene calls for the 80 haloarchaea. This database, and the NCBI non-redundant protein database (as of October 30th, 2012), were queried for homologs to the archaeal and eukaryotic general transcription factors TATA-binding protein (TBP) and transcription factor B/IIB (TFB) using a query set of 19 curated TBP homologs and 27 curated TFB homologs from six archaeal and one eukaryotic species (Datasets S18 & S19). A BLASTp search was conducted (BLAST+2.2.27) using a maximum expect value of 10−5 and maximum target sequence of 100,000. For each set of homologs, a multi-sequence alignment was constructed using MUSCLE v3.8.31 [82] and manually curated to remove poorly aligning regions. Highly divergent and fragmentary sequences were manually removed. One thousand bootstrapped alignments were created by resampling the curated alignment using Seqboot in the Phylip toolkit (v3.69) [103]. Phylogenetic trees were created for each bootstrapped alignment using FastTree v2.1.5 SSE3 [104] and a consensus tree was produced by comparing these trees to a guide tree, also constructed using FastTree, using the tree comparison tool CompareToBootstrap.pl [105]. For all clades in the guide tree, the fraction of bootstrapped trees in which that clade appeared were tabulated and recorded as the bootstrap support values for that clade. Because this approach is limited in that only clades appearing in the initial guide tree are considered, we also constructed a consensus tree with the Consense program in the Phylip package [105] using the extended majority rule option. Each clade in the resulting consensus tree represents the most frequent grouping of those species in the 1000 bootstrap replicate trees, independent of any guide tree. For Consense tree files, see Datasets S20–S23. Branch lengths and node labels in Consense tree represent the number of bootstrapped tree replicates for which each clade was observed, rather than sequence divergence. Trees were visualized in FigTree [85].
Marker genes were identified by the following procedure: Homology clusters were identified that were (1) universal to the genus of interest, (2) not found in any other haloarchaeal species, and (3) single copy. Homologs were aligned using MUSCLE [82] with default parameters, and a hidden Markov model was generated from each alignment using HMMer [106] with default parameters. Additionally, a list of marker homology clusters was generated, as were the genomic coordinates for each marker gene [67]. This list may be loaded into JContextExplorer [17] as a custom context set.
Putative missed gene calls were identified using JContextExplorer [17], with the Haloarchaea genome set loaded. To retrieve the context surrounding Tribe4688 (with gene calls in 12 Haloferax species) a “Between” context set was created with a 5 Kbp limit, and the query “966; 458” was carried out under “Cluster Search”. Regions were visualized by selecting all instances and pushing the “View Contexts” button. This produced a set of either two or three gene groupings, in the order Tribe458 (always) – Tribe4688 (sometimes) – Tribe966 (always). The sequence between instances of Tribe458 and Tribe966 in the three Haloferax species that did not contain an instance of Tribe4688 was extracted, and every contiguous region of 117 nucleotides was translated into a protein using MATLAB's bioinformatics toolbox [107]. Each match was manually compared to the gene members of protein sequences of Tribe4688, and three matches were identified, one in each of the three organisms lacking an instance of Tribe4688 (Haloferax alexandrinus, Hfx. elongans, and Hfx. sulfurifontis). The sequence alignments were visualized, with amino acids colored according to the properties of their side chains. To retrieve the region of poor gene call consistency in Haloarcula, the same between context set was used with a 5 Kbp limit, with the query “254; 369” under “Cluster Search”. Genomic segments were visualized by selecting all instances and pushing the “View Contexts” button.
A presence/absence matrix was constructed for each TRIBE-MCL protein family across the 80 haloarchaea and transformed into a distance matrix using Euclidean distances. Hierarchical clustering of this distance matrix was performed using Mev [108], and the resulting matrix was visualized. Groups of protein families which clustered together have a similar phylogenetic dispersal pattern, and may therefore perform functionally related tasks. This distribution was manually interrogated to find groups of clustered protein families with functionally related annotations. Unannotated protein families in that cluster were then hypothesized to also be involved in that task.
|
10.1371/journal.pcbi.1001001 | A Computational Approach to Analyze the Mechanism of Action of the Kinase Inhibitor Bafetinib | Prediction of drug action in human cells is a major challenge in biomedical research. Additionally, there is strong interest in finding new applications for approved drugs and identifying potential side effects. We present a computational strategy to predict mechanisms, risks and potential new domains of drug treatment on the basis of target profiles acquired through chemical proteomics. Functional protein-protein interaction networks that share one biological function are constructed and their crosstalk with the drug is scored regarding function disruption. We apply this procedure to the target profile of the second-generation BCR-ABL inhibitor bafetinib which is in development for the treatment of imatinib-resistant chronic myeloid leukemia. Beside the well known effect on apoptosis, we propose potential treatment of lung cancer and IGF1R expressing blast crisis.
| Protein interaction data are accumulating rapidly and, although imperfect and incomplete, they provide a valuable global description of the complex interplay of proteins in a human cell. In parallel, modern proteomics technologies make it possible to measure in an unbiased manner the protein targets of a drug. Such data reveal multiple targets in a view that contrasts with a previously prevalent paradigm that drugs had single – or a very limited number of – targets. In this context of newly available systems level data and more precise and complete information about drug interactions, it is natural to try to determine the global perturbation exerted by a drug on a human cell to identify potential side effects and additional indications. We present a computational method that aims at making such predictions and apply it to bafetinib, a recently developed leukemia drug. We show that meaningful predictions of additional applications to other cancers or resistant cases and likely side effects are obtained that are not straightforward to determine with existing algorithms. Our method has a strong potential to be applicable to other drugs.
| Biomedical research is changing towards a systems pharmacology view of drug action [1]. In parallel, chemical proteomics (Figure 1), a postgenomic version of classical drug affinity purifications which use is growing rapidly, has been developed to measure drug target profiles in an unbiased manner [2]–[5]. It usually reveals larger than expected spectra of targets which are causing both therapeutic and adverse effects. Such unbiased target profiles are very valuable entry points to understand which regions of the cell machinery are perturbed by a drug. It is hence desirable to develop new specific algorithms exploiting chemical proteomics profiles. Generally, it is natural that protein interaction networks are involved to characterize drug targets, action on diseases, and potential side effects [6]–[11]. Existing methods are mainly based on the network topology and on an integration of gene expression data and phenotype similarities [12]–[14].
Alternatively, precise modeling of perturbations which change the protein interaction network has the potential to predict new drug targets and to provide a detailed mechanism of action simultaneously [15]–[17]. Beside network approaches, classical gene ontology (GO) enrichment analyses of drug targets are commonly used which result in no detailed mechanism but identify different processes and functions of direct involvement [11], [18]. However, one pivotal aspect is that drug targets can perturb protein interaction networks and biological processes without being directly part of the latter. Therefore, we present a new algorithm which combines direct and peripheral perturbations of functional sub-networks and exploits chemical proteomics drug target profiles. The idea of functional sub-networks is based on the finding that genes associated with the same disease often share protein-protein interactions and gene ontology terms [19]. Our algorithm estimates the drug impact on biological processes and the detailed perturbation effects can be visualized as a network, which facilitates interpretation. Furthermore, we introduce an affinity score to weigh the drug target profile on the basis of interaction strengths.
We applied our algorithm to the bafetinib (NS-187, INNO-406) target profile. Bafetinib is a small molecule tyrosine kinase inhibitor in development for chronic myeloid leukemia (CML) [20]. It has been designed to potently and specifically inhibit BCR-ABL and the SRC family kinase (SFK) LYN, but no other SFKs, with the purpose of displaying an improved safety profile over multi-kinase and pan-SFK inhibitors, such as dasatinib, while retaining the advantageous dual mechanism of action. We have recently characterized the detailed target profile of bafetinib by chemical proteomics and to interpret the complex dataset obtained is challenging. One of the most popular methods for distinguishing (and potentially quantifying) specific drug targets from non-specific background proteins is the competition of soluble drug molecules with the affinity matrix for drug binding proteins (Figure 1) [21]–[23]. Comparison of the protein eluates from a competed and a non-competed drug pulldowns will highlight specific binders, while non-specific binding proteins will not be affected. However, even after correct identification and potentially determination of quantitative interaction parameters for distinct drug-protein pairs, a global or mechanistic understanding of drug effects is but a distant goal requiring some sophisticated experimental and/or theoretical follow-up. Our theoretical effort advances significantly our mechanistic understanding of the effects of bafetinib and provides others with a computational strategy applicable to different drug profiles.
Our computational approach to predict the impact of bafetinib on a functional network is based on the human protein-protein interaction network, on the annotation of its nodes and on a drug target profile associated with an affinity measure.
The network is constructed from protein-protein interactions found in the public interaction databases HPRD, MINT, Intact, DIP and BioGRID [24]–[28]. Furthermore, it is supplemented with published interactions of the BCR-ABL core complex which is the primary target of bafetinib in chronic myeloid leukemia (CML) [29]. The resulting undirected network contains 11505 proteins and 80363 interactions.
The human network of all known protein-protein interactions is associated with its biological processes of gene ontology (GO) derived from UniProtKB and Entrez Gene [30]–[32]. All ancestors of the GO tree are assigned in addition to achieve a complete and consistent annotation. In total, the human interaction network consists of 6390 different BP terms. 8939 (78%) nodes of the human interactome are at least associated with one biological process. A uniform functional sub-network is a connected fraction of the interactome, in which all the proteins share the same function, i.e., one unique GO term. The interactome can contain multiple disjoint functional sub-networks for the same annotation.
The recently published drug target profile of the kinase inhibitor bafetinib measured in the cell line K562 is used [33]. Rix et al took three quality criteria into account: (1) The drug target profile is devoid of proteins in the K562 core proteome. (2) No frequent hitters are included. (3) The proteins must be seen in replicates. In addition, splice variants and protein fragments are excluded. The 33 proteins are listed in Table S1.
Bafetinib can impact the uniform functional sub-networks in two ways via its targets (Figure 2):
It is difficult to predict which mode of perturbation has a higher impact. Directly inhibiting a pivotal sub-network member can completely disrupt a function. Nonetheless, biological signaling networks often have multiple alternative routes and protein isoforms to rescue the cell. Drug targets acting at the periphery can modify significantly the function through interaction or modulation of a modification, e.g., phosphorylation. By this mechanism, the inactivation of different branches and isoforms is possible. Furthermore, functional boundaries are often loosely defined and incompletely annotated. We thus treat both perturbation modes equally and therefore the perturbed functional sub-network (Figure 2) combines the uniform functional sub-networks with all interactions to the peripheral drug targets. This combination could result in joining otherwise disjoint uniform functional sub-networks via a drug target as linker (see MAPK14 in Figure 3). The direct targets are already members of the uniform functional sub-networks.
We define a score snet which predicts how strong functional sub-networks are perturbed by bafetinib. For this purpose different features of the perturbed functional sub-network are combined.
The first feature describes how frequent the annotation is present in the sub-network. Peripheral drug targets don't share the functional annotation (Figure 2), hence they dilute the functional annotation of the sub-networks. To ensure that the function is not underrepresented in the network, a first factor of the score is the ratio of the number nannot,net of nodes which have a specific annotation to the total size nnet of the perturbed functional sub-network.
The second feature puts the drug impact in relation to the sub-network size. Generic biological functions result in very big sub-networks, in which the drug targets play overall no important role anymore. Furthermore, the drug should preferentially perturb a function at several different points. Hence, the proportion of the number ndrug,net of drug target nodes to the number nnet of all nodes in the perturbed functional sub-network resembles a good measure.
Lastly, the binding affinity at of bafetinib to its targets or the potency of inhibition is important for effective perturbation. In theory, the affinities can be measured in biochemical assays which are not always available. However, we propose hereafter an ad hoc affinity measure derived from chemical proteomics data directly. The impact is summarized in the sum of drug affinities to its targets in the perturbed functional sub-network Tdrug,net divided by the overall affinity of all possible drug targets Tdurg. Combining these factors results in a score for each disrupted functional network:(1)The last two factors of equation (1) have an additional role and benefit. Mass spectrometry detection as used in chemical proteomics does not detect direct drug interactors only; it can also detect secondary interactors, i.e. proteins that bind to direct drug interactors. Without prior knowledge, it is difficult to distinguish between direct and indirect interactors but we believe that it is advantageous to use the complete target profile of bafetinib as it embeds the true drug targets into a specific context and increases the crosstalk with annotated nodes of the sub-networks (the second factor in equation (1) increases). The affinity factor ensures that also true and strong drug targets are part of the sub-network.
The affinity of bafetinib to its targets is used to score the impact on sub-networks in equation (1). The higher the protein amount in mass spectrometry analysis, the higher the number of different detected peptides covering the protein sequence [34]. Hence, the peptide count pt of each protein is a rough estimate of the amount of pulled-down protein. If soluble bafetinib is supplemented, the soluble drug blocks the binding pocket of its target yielding a reduced amount of pulled-down proteins that are specific drug binders (Figure 1) and thus, their peptide counts pt,comp decrease. This observation is expressed in the first factor of the affinity score. Since in chemical proteomics the drug is always present at a large excess of constant concentration, it is only possible to distinguish the affinities of completely competed proteins by taking the protein amount into account. To down weigh this parameter influence, the logarithm is applied to pt. Thus, the affinity score can be expressed by the following equation:(2)Due to the reduced complexity of the competed pull-down, it can sporadically happen that pt,comp>pt. This case is seen as no competition and thus the affinity is set to 0.
An empirical p-value is calculated via randomization of the interactome. First, the interaction partners of each node are randomly selected. It is ensured that the degree of each node remains constant. Second, the annotation is randomly assigned to the nodes, while the total number of each term is preserved. The presented algorithm is applied to 500 random instances of the interactome. The empirical p-value is calculated from the fraction of randomized interactomes containing a sub-network with a score equal or better to the tested score divided by the total number of random instances. The highest score of all the random instances smax, rand is 0.124.
The presented approach is programmed in the statistical environment R/Bioconductor and available at http://bioinformatics.cemm.oeaw.ac.at/drugDisruptNet [35], [36]. The provided R package depends on graph, RBGL, snow and GO.db [37]–[39]. Parallelization is done with snow to generate and score random instances [40]. Additionally, the above described data and the results are stored as R data objects.
Networks are visualized with Cytoscape [41]. For comparison, classical GO/KEGG/Biocarta enrichment analysis of sets are performed with DAVID [18].
We present a novel strategy to analyze the mechanisms of action of bafetinib. The target profile is weighted with respect to its drug affinity and its impact on protein interaction networks is scored. Ten perturbed functional sub-networks are scored higher than any sub-network of the 500 randomized interactomes (smax, rand = 0.124), see Table 1 and Figure 3, 4 and supplementary Figure S1, S2. The sub-networks do not necessarily contain all the components of a specific function since several disjoint functional sub-networks can be constructed. Bafetinib is designed to treat BCR-ABL dependent chronic myeloid leukemia (CML). Constitutively active BCR-ABL interferes strongly with apoptosis in malignant cells. We catch this process in our significantly perturbed sub-networks at rank 6 (Table 1). Furthermore, MAP kinase signaling can also be brought together with pathogenesis and treatment of CML. The top ranked perturbation of “Epidermal growth factor receptor (EGFR) signaling pathways” and “Insulin receptor signaling pathway” suggest potential novel domains of treatment for bafetinib and “heart development” indicates a putative side effect. The hit signaling pathways further play important roles in the general perturbed processes of aging, extracelluar structure organization and cell cycle. Finally, phosphorylation is an obvious process to be perturbed by a kinase inhibitor. We discuss pathogenesis, potential new domains of drug treatment and putative side effects of bafetinib in more details.
Inactivated apoptosis signaling plays a pivotal role in BCR-ABL dependent CML pathogenesis [42] and is well represented in the significant sub-networks. The perturbed functional sub-network of apoptosis (Figure 3) is disrupted by inhibition of ABL2, MLTK, LYN and MAPK14 (p38α). These kinases are not annotated themselves as “induction of apoptosis by intracellular signals” but act at the periphery of the uniform functional sub-network. K562 cells express ABL1, a central node of the network, and its fusion protein BCR-ABL. High amounts of BCR-ABL hide specific ABL1 detection with mass spectrometry. However, western blots proved ABL1 as a competed target of bafetinib in K562 [33]. Hence, the score of perturbation underestimates the impact of bafetinib on apoptosis in CML.
The impact of bafetinib on apoptosis in CML is manifested with 5 targeted kinases at the periphery (Figure 3). The method strongly prefers networks which are attacked by several high affinity drug targets. In theory, a single perturbation might be enough to significantly interfere with a biological function. However, biological signaling networks are often highly redundant thus requiring perturbation at several points in order to observe an effect [43]. Hence, promiscuous drugs like dasatinib are very successful in CML and other cancers and the multi-targeted networks are likely to be of high relevance in drug treatment.
Even if we know that the drug has an inhibitory effect on the target kinases, we cannot predict without additional knowledge whether missing phosphorylation has an enhancing or decreasing effect on the biological process. The constitutively active kinase BCR-ABL results in a strong anti-apoptotic phenotype. Inhibition counteracts this behavior [44]. Inhibition of LYN has a similar effect in this context [45]. Contrary to this, MAPK14 inhibition rescues cells from apoptosis [46]. Only through the complex interplay of different signals, the malignant cells die upon treatment as desired. Hence, visualization of the network together with its disturbers strongly aids in interpreting their influence. This is a great advantage compared to simple GO enrichment analysis which does not display the relationship of the proteins to each other.
The top ranked perturbed functional sub-network is based on the epidermal growth factor receptor (EGFR) signaling pathway (Figure 4). It is peripherally interacting with six kinases of the drug profile. Three additional kinases are directly interacting with EGFR but also interfering with 7 further proteins of the signaling cascade. Additionally, the crosstalk between the pulled down non-kinase members and the functional network is very high. In total 13 out of 33 EGFR signaling components (39%) are interacting with the drug profile.
EGFR is not expressed in hematopoietic cells (such as K562) but this sub-network strongly suggests that bafetinib has the potential to interfere with EGFR signaling for instance in lung cancer cells. Recently, it was shown through the combination of chemical proteomics, phosphoproteomics and functional genomics that dasatinib, a broad-spectrum kinase inhibitor, leads to apoptosis in lung cancer cells via inhibition of SRC, EGFR, FYN and, notably, LYN [47]. Therefore, it is possible that also bafetinib might have a pro-apoptotic effect on these cells as it is also a potent inhibitor of LYN. While expression of dasatinib-insensitive gatekeeper mutants of DDR1 (or ABL1) did not rescue the H292 lung cancer cell line from dasatinib action, the role of DDR1 might be quite different in primary lung cancer cells as several recent reports described this receptor tyrosine kinase to be one of the most highly expressed and phosphorylated kinases in primary lung tumor specimens [48], [49]. Thus, it is conceivable that bafetinib might exert pro-apoptotic effects on lung cancer cells, and it might do so through simultaneous inhibition of LYN and DDR1.
Second highest is the perturbation of insulin receptor signaling pathway. It was suggested that bafetinib, CGP76030 and nilotinib might overcome imatinib resistance in blast crisis patients which feature BCR-ABL gene amplification [50]. Phase 1 studies could not verify this yet [51]. However, we propose to treat only the subgroup of CML blast crisis patients which expresses IGF1R with bafetinib. The drug targets are strongly interacting with the insulin receptor signaling pathway which maintains survival of hematopoietic cells through IGF1R (supplementary Figure S1). The IGF1R expression frequency is strongly increased in blast crisis patients (73%). Inhibition of IGF1R was shown in imatinib-resistant CML to induce apoptosis [52]. IGF1R is not a known direct target of bafetinib but attacking several downstream components simultaneously might show a similar effect as a direct IGF1R inhibition.
A potential side effect of several tyrosine kinase inhibitors, like sunitinib and dasatinib, is an increased risk for cardiotoxicity [53]. Observed toxicity in rats can be a result of higher concentration than used in patients [54]. Nevertheless, perturbation of the “heart development” network (Figure S2) indicates some possible risks which should be closely monitored during clinical trials.
We validated the robustness of the algorithm by following the rank of the biological process upon leaving-one-out (supplementary Figure S3). The ranks of the first five sub-networks (Table 1) are generally stable upon loss of a node. High affinity targets are essential to the phenotype which results in increased sensitivity of highly ranked terms to high affinity targets. On the contrary, weaker binders, which are not competed away with free drug, have only a modest effect on the rank.
Furthermore, we investigated the effect of hubs on the sub-network ranks, which might exert an influence on the phenotype upon inhibition. It is not clear whether hubs are, in the context of our analysis, highly important or “general signal diluters”. Therefore, we weighted up and then down the affinity of the targets by log10 of their node degree. Multiply by this factor, i.e. increasing hubs importance, the top 6 sub-networks remain unchanged and “cell cycle arrest” even improved its rank by one. The others sub-networks were substituted by “response to insulin stimulus”, “response to peptide hormone stimulus” and “cellular response to hormone stimulus”, which are in line with insulin receptor signaling. Upon down-weighting by division, the top 4 sub-networks still remained unchanged. We conclude for robustness and reasonable independence of local topology. In other words, the function of the sub-network at hand seems to play a strong role in scoring, which is appropriate.
To show the general interest of our method we applied the algorithm to data we published recently analyzing lung cancer (HCC297) treatment with dasatinib [47], another kinase inhibitor. Interestingly, dasatinib is highly promiscuous and pulls down 176 proteins (33 kinases) compared to the 33 proteins of bafetinib. In addition, free compound competition data were not available in this case; we thus exploited IC50s of autophosphorylation, which were available instead (Table S2A of Ref. [47]). The log10 of the IC50s (in the nM range) were used to weight the effect of dasatinib on the kinases. Analysis results in 681 significantly hit sub-networks due to the huge kinase profile (P-value<0.002, Table S2). Even though more than 5% of the human kinases are targeted by dasatinib and, subsequently, many sub-networks are significantly impacted, the top disrupted sub-networks are insightful. For instance, the top 10 ranked biological processes are centered on cell cycle arrest, cell growth and apoptosis. In the highest ranked sub-networks, SRC, LYN and EGFR play a pivotal role, which is absolutely consistent with our experimental data where these three proteins were shown with dasatinib gate-keeper mutants to strongly contribute to cell viability of HCC297 [47]. These results show that the algorithm can provide informative data even in very challenging situations.
In comparison to our approach, classical GO enrichment analysis (p-value<0.01) of the 33 bafetinib drug targets result in 33 significant biological processes with high redundancy in the GO tree (supplementary Table S3). Basically, they represent 3 GO terms: cytoskeleton organization (especially actin filament), phosphorylation and regulation of stress-activated protein kinase signaling pathway. Except phosphorylation which is obvious in a target profile of a kinase inhibitor there is no overlap with the perturbed functional sub-networks. The GO term of cytoskeleton organization contains competed and non-competed members of the target profile. The combined attack power of few competed kinases is too low to see perturbation of the large uniform functional sub-network (387 members) which is based on cytoskeleton organization. Enrichment analysis with KEGG and Biocarta pathways (p-value<0.01) yielded no hit.
In contrast to GO enrichment analysis, the presented method does not as much rely on accurate annotations. Possible missing annotations of drug targets interacting at the periphery with a functional sub-network have only a minor effect on the score. However, we would like to point out that boundaries of pathways and biological processes are very diffuse. Crosstalk between different signaling cascades and metabolic pathways is essential for a living cell. Integrating protein interactions to peripheral drug targets provides a way out of this dilemma and can catch therefore more relevant processes than GO enrichment. Alternatively, augmenting the drug target profile with their direct interactors, results in a set of 831 proteins. GO enrichment analysis (p-value<0.01) of this set results in 676 biological processes (Table S4). Again the first hits are related to general phosphorylation which is obvious for a kinase inhibitor profile. At the ninth rank “regulation of programmed cell death” which is related to perturbed apoptosis is presented with 127 proteins of the augmented set. The highest scored perturbed sub-network of EGFR signaling is only found at position 368. Even though the disrupted processes are detected with the augmented GO analysis, their ranks are so bad that they would not be considered as relevant. Our approach thus picks the most relevant perturbed functional networks and allows for insights beyond traditional GO enrichment analysis.
Competition experiments in chemical proteomics provide an additional layer of security to the drug target profile. Secondary and unspecific binders are difficult to distinguish from true drug targets. They are often similar in the range of peptide counts and other properties. The competition with a soluble drug and our affinity score helps in identifying biological target proteins. Interestingly, unspecific binders influence the perturbation algorithm only marginally since the proteins are dispersed all-over the interactome and have no affinity to a specific uniform functional sub-network. Furthermore, their binding affinity score is 0. On the contrary, secondary binders of true drug targets increase the crosstalk to the functional sub-network which is attacked by the true target. Hence they can be used advantageously embedding the true targets in a specific context.
In conclusion, we identified successfully known mechanisms in CML as well as potential new applications and possible side-effects. We believe that the proposed computational approach can shed light in mechanisms of other drugs including highly promiscuous compounds and when soluble compound competition data are lacking. Hence, we provide an R package at http://bioinformatics.cemm.oeaw.ac.at/drugDisruptNet.
|
10.1371/journal.pgen.1004500 | VIB1, a Link between Glucose Signaling and Carbon Catabolite Repression, Is Essential for Plant Cell Wall Degradation by Neurospora crassa | Filamentous fungi that thrive on plant biomass are the major producers of hydrolytic enzymes used to decompose lignocellulose for biofuel production. Although induction of cellulases is regulated at the transcriptional level, how filamentous fungi sense and signal carbon-limited conditions to coordinate cell metabolism and regulate cellulolytic enzyme production is not well characterized. By screening a transcription factor deletion set in the filamentous fungus Neurospora crassa for mutants unable to grow on cellulosic materials, we identified a role for the transcription factor, VIB1, as essential for cellulose utilization. VIB1 does not directly regulate hydrolytic enzyme gene expression or function in cellulosic inducer signaling/processing, but affects the expression level of an essential regulator of hydrolytic enzyme genes, CLR2. Transcriptional profiling of a Δvib-1 mutant suggests that it has an improper expression of genes functioning in metabolism and energy and a deregulation of carbon catabolite repression (CCR). By characterizing new genes, we demonstrate that the transcription factor, COL26, is critical for intracellular glucose sensing/metabolism and plays a role in CCR by negatively regulating cre-1 expression. Deletion of the major player in CCR, cre-1, or a deletion of col-26, did not rescue the growth of Δvib-1 on cellulose. However, the synergistic effect of the Δcre-1; Δcol-26 mutations circumvented the requirement of VIB1 for cellulase gene expression, enzyme secretion and cellulose deconstruction. Our findings support a function of VIB1 in repressing both glucose signaling and CCR under carbon-limited conditions, thus enabling a proper cellular response for plant biomass deconstruction and utilization.
| Many filamentous fungi that grow on plant biomass are capable of producing lignocellulase enzymes to break down plant cell walls into utilizable sugars, thus holding great potential in reducing the cost of the next-generation biofuels. Cellulase production is subject to induction by the presence of plant biomass components and to repression by the availability of easily metabolized sugars, such as glucose. Genes required for repression of cellulase gene expression when preferred carbon sources are present (carbon catabolite repression) and those that play a role in mediating glucose sensing/metabolism have been identified in filamentous fungi, but the mechanisms involved in crosstalk between repression versus induction of cellulase gene expression is poorly understood. Here, we report the identification and functional characterization of VIB1, a transcription factor essential for plant cell wall deconstruction in Neurospora crassa and COL26, a transcription factor that functions in glucose sensing/metabolism and regulation of CCR. We show that disabling CRE1 repression and modulating the glucose response by deletion of col-26 restored growth of the Δvib-1 mutant on cellulose. Our findings are particularly important in understanding the molecular basis of enzyme production that could allow a further strain improvement for plant biomass deconstruction.
| Bioconversion of lignocellulosic biomass to simple sugars holds great promise in next-generation biofuel production and relies on a complex repertoire of proteins for enzymatic deconstruction of plant cell walls [1]. Many filamentous fungi have evolved to utilize cellulosic materials and are capable of producing a wide spectrum of enzymes, but only a few species have been harnessed for industrial usage [2]. Further improvement in fungal cellulolytic enzyme production is desired to make biofuel production cost-competitive, but this relies on a better understanding of the molecular basis of networks involved in carbon sensing and regulatory aspects associated with induction of gene expression of hydrolytic enzymes [3].
Cellulolytic enzyme production and secretion is a unique attribute of filamentous fungi, and efforts to identify important factors in enzyme production led to the discovery of a number of transcriptional activators and repressors. For example, the transcription factor XlnR/XYR1 positively regulates expression of cellulase and hemicellulase genes in Aspergillus niger and Trichoderma reesei, respectively [4]–[7]. In Neurospora crassa, the transcription factors CLR1 and CLR2 are essential for growth on cellulose and are required for expression of a ∼212 gene regulon that is induced in response to cellodextrins, such as cellobiose [8], [9] (Figure 1). In Aspergillus nidulans and A. oryzae, a clr-2 homolog, called clrB/manR, respectively, is also essential for cellulase gene expression and activity [8], [10], [11]. Additional transcriptional regulators that promote expression of some genes encoding hydrolytic enzymes have also been identified, including mcmA in A. nidulans [12], clbR in A. aculeatus [13], and aceII and bglR in T. reesei [14], [15].
In addition to induction, cellulase gene expression is also subject to carbon catabolite repression (CCR), which functions when a favorable carbon source, such as glucose, is present [3], [16], [17]. The most well-characterized transcription factor involved in CCR in filamentous fungi is CreA/CRE1. Deletion of creA/cre-1 alleviates some aspects of CCR for cellulolytic enzyme expression in Aspergilli [18]–[22], T. reesei [23]–[25], Penicillium decumbens [2] and N. crassa [26], [27]. In A. nidulans, repression by CreA occurs both by binding to promoters of hemicellulase genes as well as repressing expression of transcriptional activators [28]. Other factors including creB/cre2, creC, creD, lim1, and aceI were also reported to promote CCR in different fungal species via unknown mechanisms [29]–[37]. The strength of CCR is tuned by glucose sensing and signaling, although crosstalk between these two regulatory systems is not well understood. In N. crassa, RCO3, a predicted sugar transporter was proposed to function as a glucose sensor [38], [39]. In A. nidulans, phosphorylation of glucose triggers CreA repression [29], [40]. In Magnaporthe oryzae, trehalose-6-phosphate synthase (Tps1) promotes glucose metabolism and CCR through inhibition of Nmr (nitrogen metabolite repression) proteins (Nmr1, Nmr2, Nmr3) [41]. Downstream, a multidrug and toxin extrusion pump, Mdt1, promotes citrate efflux to relieve CCR. To what extent these mechanisms are shared among cellulolytic fungi and whether they all converge to regulate CreA/CRE1-mediated CCR is currently unclear.
N. crassa is an early colonizer of burnt vegetation [42], [43], grows robustly on plant biomass and secretes a broad spectrum of enzymes to degrade plant cell walls [44], [45]. By screening the N. crassa near-full genome deletion strain set [46] for growth on Avicel (crystalline cellulose), we identified a transcription factor, vib-1, that is essential for cellulose utilization. VIB1 (vegetative incompatibility blocked) is a p53-like transcription factor that is conserved among filamentous ascomycete fungi. Characterized as a mediator of nonself recognition and cell death in N. crassa [47], [48], VIB1 is also required for extracellular protease secretion in response to both carbon and nitrogen starvation [48]. Here, we demonstrated that vib-1 functions upstream of cellulolytic gene induction and its absence leads to a weak induction of clr-2 and cellulase genes but increased expression of genes predicted to function in CCR. Functional analysis of one such predicted transcription factor gene, col-26, an N. crassa bglR homolog, showed that COL26 regulates glucose sensing/metabolism and which is separate from CRE1-mediated CCR. Deletion of both col-26 and cre-1 leads to a synergistic effect in rescuing Δvib-1 utilization of cellulose and cellulolytic activity. Our data support a function for VIB1 in repression of glucose signaling and CCR and which is critical for fungal utilization of plant biomass.
Screening of a transcription factor deletion set of N. crassa strains [46] for ability to deconstruct crystalline cellulose showed that a strain carrying a deletion of the vib-1 gene (FGSC11309) failed to grow on Avicel (Figure 2A). Since functional vib-1 is required for extracellular protease secretion in response to carbon and nitrogen starvation in N. crassa [48], [49], we hypothesized that the Δvib-1 mutant might be unable to respond to complex extracellular carbon sources. In support of this hypothesis, the Δvib-1 mutant also exhibited slow growth on xylan. Growth defects were accompanied by barely detectable extracellular enzyme activity towards crystalline cellulose and low extracellular xylanase activity (Figure 2B and S1A). In contrast, the Δvib-1 mutant accumulated a similar amount of mycelial biomass as the WT strain when inoculated into minimal media containing simple sugars (sucrose, cellobiose or xylose) (Figure S1B). The introduction of an ectopic copy of vib-1 (Pvib-1) completely restored the growth defects on Avicel of the Δvib-1 mutant, as well as the secretome and cellulolytic enzyme activity of culture supernatants (Figure 2B).
To test the hypothesis that the role of VIB1 in cellulose utilization is conserved in other filamentous fungi, especially in fungi used in industrial production of cellulolytic enzymes, we carried out complementation tests using the vib-1 ortholog from T. reesei (EGR52133; Trvib1); TrVIB1 and N. crassa VIB1 share 49% amino acid identity. Constitutive expression of Trvib1 in a N. crassa Δvib-1 mutant fully restored the growth and cellulolytic enzyme activity (Figure 2B). The Trvib1 strain also recapitulated most of the secretome of N. crassa WT and Pvib-1 strains on Avicel (Figure 2C). These results suggest that vib-1 is functionally conserved for the utilization of cellulose in filamentous ascomycete fungi.
The Δvib-1 mutant shows an inappropriate temporal and spatial conidiation pattern. These phenotypes are correlated with differential localization of VIB1-GFP in vegetative hyphae versus conidiophores [48]. As conidiation is regulated by glucose limitation [50], we assessed whether differential localization of VIB1 was also associated with cellulose utilization. We examined a strain in which we replaced the resident vib-1 gene with a functional vib-1-gfp construct. Nuclear localization of VIB1-GFP was observed in hyphae and localization was independent of carbon source, either following a shift to Avicel for 2 hrs (Figure S1C) or after prolonged growth.
Previous comparative RNA-seq analysis of WT revealed that 212 genes are significantly differentially expressed under cellulose conditions, a gene set referred to as the “Avicel regulon” [8]. To determine whether the defect in cellulase secretion and activity in the Δvib-1 mutant was due to failure to induce cellulase gene expression versus a defect in cellulase secretion, we assessed genome wide expression differences via RNA-seq between the WT and Δvib-1 strains following a shift for 4 hrs from sucrose medium to either carbon-free or Avicel medium. Of the 212 genes in the Avicel regulon, 91 genes were expressed at a significantly lower level in the Δvib-1 mutant versus WT under Avicel conditions (cutoff: Padj <0.05 and fold change >2; Table S1). This gene set includes the essential cellulase transcription factor gene, clr-2 and 43 carbohydrate-active enzymes (CAZy) from 27 different families (Carbohydrate Active Enzymes database: http://www.cazy.org/) [51].
CLR1 and CLR2 are strictly required for full expression of 140 genes within the Avicel regulon [8], [10]; 62 of these genes were identified in the 91-gene set that showed low expression in the Δvib-1 mutant. Although expression of clr-1 was not significantly different from WT in the Δvib-1 mutant, the expression of clr-2 was significantly reduced (FPKM: 107±43 in WT; 66±27 in Δvib-1 for clr-1 versus 171±10 in WT; 39±14 in Δvib-1 for clr-2) (Table S1). Importantly, constitutive expression of clr-2 (Pc clr-2) in minimal medium without cellulosic inducers recapitulates the response of N. crassa to crystalline cellulose, including the secretion of active cellulolytic enzymes [10]. The reduced transcription of clr-2 in the Δvib-1 mutant (Table S1) suggested that constitutive expression of clr-2 might suppress the cellulose utilization defect in the Δvib-1 mutant. To test this hypothesis, we constructed a Pc clr-2; Δvib-1 strain and evaluated its ability to secrete cellulases and utilize Avicel in comparison to the Pc clr-2, the Δvib-1 and WT strains. In support of our hypothesis, the Pc clr-2; Δvib-1 strain showed restoration of protein secretion and cellulolytic activity to near WT levels (Figure 3A). Although the clr-2 expression levels in the Pc clr-2 strain were at a similar level to a WT strain after a 4 hr shift to Avicel (Figure S4), the Pc clr-2; Δvib-1 mutant showed a ∼3.5 fold increase in clr-2 expression level under the same conditions.
To evaluate the functions of VIB1 versus CLR2 in regulating the 91 genes in the Avicel-regulon, we generated RNA-seq data from the Pc clr-2; Δvib-1 mutant that was shifted for 4 hrs from sucrose medium to carbon-free medium and compared it to previously obtained data for Pc clr-2 [10]. Analysis of genes encoding different CAZy family proteins revealed a similar pattern of expression between the Pc clr-2 and the Pc clr-2; Δvib-1 strains (Figure 3B), consistent with our hypothesis that VIB1 functions upstream of clr-2 in response to cellulose. To differentiate whether any Avicel-regulon genes that showed decreased expression levels in Δvib-1 mutant were due to the vib-1 deletion rather than a low level of clr-2 expression, we performed hierarchical clustering of expression patterns of the Avicel-regulon genes across 6 RNA-seq experiments (WT and Δvib-1 shifted to no carbon or Avicel and the Pc clr-2 and the Pc clr-2; Δvib-1 strains shifted to no carbon); 4 major expression groups were identified (Figure 3C and Table S1). Two groups (group 1 and 3) were CLR2-regulon genes that were vib-1 independent. Group 1 consisted of 54 genes whose expression was fully induced by constitutive expression of clr-2 regardless of the presence or absence of vib-1, including 33 of the 43 CAZy proteins (Table S1). Group 3 consisted of 8 genes whose expression was partially induced by clr-2, but still in a vib-1-independent manner. The fourth group of genes included 14 vib-1 modulated genes. These genes were partially induced in Δvib-1 on Avicel, but remained repressed in both the Pc clr-2 and the Pc clr-2; Δvib-1 strains under no carbon conditions (Table S1). Expression of these genes is likely induced by the cellulolytic cascade pathways upstream of CLR2 or other components present in commercial Avicel preparations, such as a low concentration of hemicellulose [9]. The second group consisted of 15 genes that were induced by constitutive clr-2 expression under no carbon conditions but in vib-1-dependent manner. This gene set included a pectate lyase (NCU06326), a BNR/Asp-box repeat protein predicted to have exo-α-L-1,5-arabinanase activity (NCU09924), a β−xylosidase (NCU09923/gh3-7), an extracellular β−1,4-D-glucosidase (NCU04952/gh3-4), a β−1,3-glucosidase (NCU09904), a starch binding domain-containing protein (NCU08746), a LysM domain-containing protein (NCU05319), a putative methyltransferase (NCU05501), and 6 hypothetical proteins. Six genes in this set encode proteins predicted to enter the secretory pathway (Table S1).
Our epistasis experiments indicated that vib-1 functions upstream of clr-2, suggesting that VIB1 could be involved in signal molecule processing that leads to CLR1 activation and thus clr-2 expression (Figure 1). In N. crassa, a strain carrying deletions of genes encoding two extracellular β-glucosidases and an intracellular β-glucosidase (Δ3βG), recapitulates the cellulolytic response when the Δ3βG strain is exposed to cellobiose [9]. These data indicate that cellobiose (or a derivative) functions as a cellulose signal that results in the induction of cellulolytic genes and subsequent secretion of cellulase enzymes. This cellobiose-induced cellulase gene expression and secretion is dependent upon functional clr-2 gene, as the Δ3βG; Δclr-2 mutant is unable to produce cellulolytic enzymes in response to Avicel or cellobiose (unpublished data). We therefore asked if VIB1 plays a role in induction via signal processing. To test this hypothesis, we created a Δ3βG; Δvib-1 quadruple mutant and asked whether the Δ3βG; Δvib-1 mutant could induce cellulase gene expression in response to cellobiose. Following a switch from sucrose to either no carbon, or 0.2% cellobiose, or 2% Avicel for 4 hrs, the induction of two major cellulase genes, cbh-1/NCU07340 and gh5-1/NCU00762 were significantly induced in the Δ3βG; Δvib-1 and the Δ3βG strains, but not in the Δvib-1 strain (p<0.05)(Figure 4A).
The restoration of cellulase gene expression in the Δ3βG; Δvib-1 strain when exposed to cellobiose was accompanied by enzyme production and activity. Similar to the Δ3βG mutant, the Δ3βG; Δvib-1 strain accumulated biomass more slowly on cellobiose than WT or the Δvib-1 mutant due to the slow conversion of cellobiose to glucose (0.51±0.11 g/L and 0.63±0.0 g/L for the Δ3βG and the Δ3βG; Δvib-1 strains, respectively, versus 3.83±0.19 g/L and 3.62±0.11 g/L for WT and Δvib-1, respectively). However, despite less biomass accumulation, both the Δ3βG and the Δ3βG; Δvib-1 strains showed significantly more enzyme activity than WT and the Δvib-1 strains on 2% cellobiose (Figure 4B). When grown on medium containing 2% Avicel as a sole carbon source, the Δ3βG; Δvib-1 strain showed significantly higher enzyme activity than Δvib-1 (Figure S2). These expression and activity data indicate VIB1 does not play a role in signal processing or signal transduction mechanisms that lead to activation of CLR1 and transcription of the cellulase activator, CLR2.
In addition to induction, cellulolytic enzyme production requires proper nutrient sensing and relief from carbon catabolite repression (CCR) (reviewed in [17], [52]). We therefore hypothesized that the Δvib-1 mutant might be defective in either nutrient sensing and/or relieving CCR in response to Avicel. To test this hypothesis, we first compared RNA-seq data of the Δvib-1 mutant when shifted from sucrose to carbon-free media versus a shift from sucrose to Avicel media. This comparison revealed 770 differentially expressed genes (cutoff: Padj<0.01 and fold change >2) (Table S2). We then compared how these genes were expressed in WT under no carbon versus Avicel conditions using a previously published RNA-seq dataset [8]. Hierarchical clustering analysis of expression patterns of these 770 genes revealed three gene clusters (Figure 5) (Table S2).
The first cluster contained 237 genes whose expression pattern was similar between the Δvib-1 and WT strains. This gene set was expressed at low levels under no carbon conditions but induced to higher levels upon exposure to Avicel. This group contained 51 CAZy proteins, clr-1 and clr-2, all three cellodextrin transporters (cdt-1, cdt-2, and cbt-1) [44], [53], [54] and 102 hypothetical proteins. This gene set overlapped the WT Avicel-regulon for 143 genes, suggesting that cellulosic induction still occurred in the Δvib-1 mutant albeit at a low level.
The second cluster consisted of 173 genes whose expression pattern was also similar between WT and the Δvib-1 strain. However, in contrast to the first gene set, the expression level of these 173 genes was higher under carbon-free conditions. This set included 7 CAZy proteins, three conidiation-specific proteins (NCU08769/con-6, NCU07325/con-10, NCU09235/con-8), a high affinity glucose transporter/NCU08152, and 103 hypothetical proteins. Genes in this cluster may encode proteins that function in a general response to carbon starvation.
The third cluster consisted of 360 genes whose expression pattern between no carbon and Avicel conditions was different in the Δvib-1 mutant as compared to the WT strain. This gene set showed consistently higher expression in the Δvib-1 mutant on Avicel medium as compared to carbon-free medium (Figure 5). Only 7 genes encoding CAZy proteins were in this set and 169 genes were annotated as hypothetical. An enrichment in the categories of metabolism and energy, particularly, degradation of glycine (p = 2.37e-03), nitrogen, sulfur and selenium metabolism (p = 8.00e-03), purine nucleotide/nucleoside/nucleobase catabolism (p = 2.49e-05), isoprenoid metabolism (p = 8.63e-04), respiration (p = 3.34e-04), metal binding (p = 6.18e-04), and mitochondrial transport (p = 2.94e-03) was observed. These data suggested that the Δvib-1 mutant was improperly responding to carbon-limited conditions as compared to a WT strain.
Within the gene set that showed increased expression level in the Δvib-1 mutant on Avicel were genes involved in CCR. This gene set included cre-1/NCU08807, creD/NCU03887, creB/NCU08378 and bglR/NCU07788 (Table S2). Although the role of cre-1 in CCR and cellulose utilization is established in N. crassa [26], [27], the function of the creB and creD homologs in cellulolytic enzyme production were uncharacterized. In N. crassa, NCU07788/BglR was previously characterized in a transcription factor deletion screen and was named col-26 for its colonial phenotype on minimal sucrose medium [46].
To determine whether homologs of the CCR genes that showed increased expression in the Δvib-1 mutant play a role in cellulose deconstruction, we first measured protein concentration and cellulase enzyme activity in supernatants from the Δcol-26, ΔNCU08378/creB, and ΔNCU03887/creD mutants grown on Avicel for 7 days: none of the mutants showed significantly different cellulase activity than WT (Figure S3). To test if these genes are involved in CCR, we evaluated resistance of WT and the mutants to 2-deoxy-glucose (2-DG). The compound 2-DG is an analogue of glucose that cannot be metabolized and is often used to select for, or evaluate, impairment of CCR and glucose repression in filamentous fungi [39], [55]–[57]. In strains with functional CCR, 2-DG is phosphorylated, thus activating CCR, resulting in the inability of the strain to grow on alternative carbon sources; strains with impaired CCR are insensitive to 2-DG exposure. When 2% cellobiose and 0.2% 2-DG were used as carbon sources, only the Δcre-1 and the Δcol-26 mutants showed 2-DG resistance, which was more obvious when Avicel instead of cellobiose was used as a carbon source (Figure 6A). These data implicated COL26 in CCR in N. crassa.
To confirm the role of COL26 in CCR, we tested CCR functionality using allyl alcohol (AA). As reported for M. oryzae [41], when CCR is impaired, alcohol dehydrogenase is expressed and will convert AA into toxic acrylaldehyde. Thus, strains with impaired CCR exhibit AA sensitivity, while strains with functional CCR are insensitive. As predicted, the Δcre-1 mutant was sensitive to AA, but the Δcol-26 mutant, similar to WT, was insensitive (Figure 6B). These data indicated that CCR was still functional in the Δcol-26 mutant. To reconcile the different results for the Δcol-26 mutant with respect to CCR, we analyzed growth of the Δcre-1 and the Δcol-26 mutants on different simple carbon sources. When grown on MM media with 2% glucose, fructose, sucrose, or cellobiose as the sole carbon source, the Δcre-1 mutant accumulated a similar amount of biomass to the WT strain (Figure 7A). However, the Δcol-26 mutant exhibited a severe growth defect on glucose, fructose and sucrose, consistent with its colonial designation [46], but only a moderate growth defect on cellobiose (Figure 7A).
The fact that the Δcol-26 mutant grew much better on cellobiose as compared to glucose, fructose, and sucrose and was insensitive to 2-DG suggested that the Δcol-26 mutant might have defects in sugar transport and/or metabolism. To test this hypothesis, we measured glucose uptake rates in WT, the Δcre-1, and the Δcol-26 mutants. Within the first 5 minutes, extracellular glucose was reduced to a similar level in all strains (Figure 7B), suggesting similar glucose transporting capacity. However, over the remaining 55 minutes, glucose uptake rates decreased dramatically in the Δcol-26 mutant (Figure 7B). These data indicate that the Δcol-26 mutant has defects in glucose sensing/metabolism, rather than in glucose transport.
Our data supported a role for CRE1 in CCR and a role for COL26 in the regulation of glucose utilization. We therefore tested sensitivity of the Δcre-1; Δvib-1 and the Δcol-26; Δvib-1 mutants to AA. The Δcre-1; Δvib-1 and the Δcre-1 mutants were both sensitive to AA (Figure 6B), indicating the Δcre-1 mutation is epistatic for CCR to Δvib-1, while the Δcol-26; Δvib-1 mutant was insensitive to AA, consistent with the active CCR phenotype of the col-26 and the Δvib-1 mutants. However, although CCR was impaired in the Δcre-1; Δvib-1 mutant, the double mutant was still unable to produce cellulolytic enzymes and grow on Avicel (Figure 8A). Similar to the Δcol-26 mutant, the Δcol-26; Δvib-1 mutant also showed defects in glucose consumption (Figure 7B). Although the Δcol-26; Δvib-1 mutant was unable to utilize Avicel, it showed slightly higher enzyme levels than that of the Δvib-1 mutant (Figure 8A). We therefore hypothesized that simultaneously preventing CRE1-mediated CCR and reducing glucose sensing/metabolism via inactivation of col-26 would restore cellulase gene expression and enzyme activity in a Δvib-1 mutant. As predicted, a Δcre-1; Δcol-26; Δvib-1 triple mutant utilized Avicel, produced significant cellulase activity and displayed a secretome similar to WT after 5 days of growth on Avicel (Figure 8A and S5). RT-PCR experiments from the Δcre-1; Δcol-26; Δvib-1 Avicel cultures showed that expression levels of clr-2 and cbh-1 were restored in the triple mutant (Figure 8B).
Although simultaneous deletion of cre-1 and col-26 restored utilization of cellulose in the Δvib-1 mutant, a significant lag in growth and enzyme activity in the triple mutant was observed as compared to the WT, Δcre-1, or Δcol-26 mutants (Figure 8C). To assess whether the Δcre-1; Δcol-26; Δvib-1 mutant was also delayed in transcriptional response upon exposure to cellulose, we measured expression levels of clr-2, cbh-1, cre-1, vib-1 and col-26 in the Δvib-1, Δcol-26, Δcre-1, Δcre-1; Δvib-1, Δcol-26; Δvib-1, and Δcre-1; Δcol-26; Δvib-1 mutants as compared to the WT strain at 4 hrs and 24 hrs after cultures were shifted to Avicel conditions. Consistent with the enzyme activity assay and growth phenotype (Figure 8C), induction of clr-2 and cbh-1 was delayed in the Δcre-1; Δcol-26; Δvib-1 mutant (Figure 9). However, in the Δcol-26 mutant at the 4 hr time point, expression levels of cre-1 were significantly higher than in the Δvib-1 mutant, with the Δcol-26; Δvib-1 mutant showing an additive phenotype of significantly increased cre-1 expression levels. At the 24 hr time point, expression levels of cre-1 were only maintained in the Δvib-1 and Δcol-26; Δvib-1 mutants, but not in the Δcol-26 mutant. These data suggest that COL26 may function to repress cre-1 transcription to promote relief of CCR during the initial response to cellulolytic induction. Surprisingly, although the Δcre-1; Δvib-1 mutant was unable to utilize cellulose, induction of both clr-2 and cbh-1 were near WT levels at the 4 hr time point, unlike the Δvib-1 mutant (Figure 9A). However, at the 24 hr time point, expression levels of clr-2 were low and cbh-1 was undetectable in Δcre-1; Δvib-1 mutant (Figure 9B). These data suggest that although the Δcre-1; Δvib-1 can respond to cellulolytic induction by increasing clr-2 and thus cbh-1 expression levels, induction signaling cannot be maintained, perhaps due to repression by COL26 or by other factors present/absent in a Δvib-1 mutant background. The fact that the Δcre-1; Δcol-26; Δvib-1 mutant does not show WT restoration of initial cellulolytic induction (Figure 8C; Figure 9A) supports the hypothesis that additional unknown factors remain to be identified that play a role in nutrient sensing/signaling and the regulation of cellulose utilization in N. crassa.
In this study, we showed that a Δvib-1 mutant displayed severe growth defects on cellulose, which was correlated with a lack of cellulolytic enzyme activity. By using RNA-seq data, we showed that expression of the Avicel regulon was significantly decreased in the Δvib-1 mutant, a phenotype that was rescued by constitutive expression of clr-2. Induction of clr-2 is dependent upon a signal cascade from cellobiose or derivative and functional CLR1 (Figure 1) [8]. Here we showed that VIB1 is not involved in inducer signal processing or perception because the Δ3βG; Δvib-1 mutant produced cellulolytic enzymes in response to cellobiose. These data indicated that VIB1 functions upstream of regulators that mediate inducer-dependent signal transduction and cellulase gene expression and activity.
Our transcriptional profiling revealed that, under Avicel conditions, a deletion of vib-1 led to an increase in transcription of genes in metabolism and energy as well as genes reported to mediate CCR. These results suggested that cellulolytic induction was mis-regulated in the Δvib-1 mutant. In the presence of glucose, N. crassa adjusts its metabolism for a high rate of glycolysis and directs carbon flux to respiration and fermentation for biosynthesis and energy production [58], while genes involved in utilization of alternative carbon sources are repressed in a CRE1-dependent manner [26], [27]. When lignocellulose is the only carbon source, CCR is relieved to allow the synthesis of “scouting” enzymes that liberate inducer molecules, such as cellobiose [9], [44], [59]. In S. cerevisiae, glucose is sensed through a multifaceted mechanism including direct detection of glucose by glucose receptors/transporters on the plasma membrane and by the sensing of glucose-6-P and other metabolites by metabolic enzymes. The glucose signals are transmitted to CCR mainly through the Snf1 complex and the Mig1 (CreA/Cre1 homolog) transcriptional repressor complex [60], [61]. In A. nidulans, mutations in two hexose kinase genes (hxkA/glkA4) results in inappropriate de-repression of genes under glucose growth conditions, although to a lesser extent than a creA mutant strain [29]. Here we show that simply eliminating CRE1-mediated CCR did not rescue the growth defect of Δvib-1 mutant on Avicel, but that a deletion of col-26 was also required.
The Δcol-26 mutant exhibited a growth defect on glucose, fructose and sucrose, which was not associated with a deficiency in glucose transport (Figure 7B). In T. reesei, a strain carrying a mutation in bglR shows reduced expression of β-glucosidase genes, suggesting the BglR plays a positive role in CCR by increasing glucose release from cellobiose [15]. However, our analyses of cellulolytic activity of secreted enzymes in the Δcol-26 mutant showed no difference in glucose versus cellobiose release (Figure 8A), a result that is in contrast to the strongly reduced glucose release from culture supernatants in the Δ3βG mutant (which lacks extracellular β-glucosidase activity) (Figure S2). Although we have not determined how glucose metabolism is changed in the Δcol-26 mutant, the resistance of Δcol-26 to 2-DG inhibition suggests a defect in glucose sensing/metabolism; CRE1-mediated CCR was still functional (as shown by insensitivity to AA). The fact that a deletion of col-26 and cre-1 restored growth of Δvib-1 on Avicel suggests a synergistic effect between glucose sensing/metabolism mediated by COL26 and CRE1-regulated CCR in repressing cellulolytic induction (Figure 10). However, other unknown factors in addition to CRE1 and COL26 play a role in the Δvib-1 mutant, because the Δcre-1; Δcol-26; Δvib-1 mutant showed a significant lag in gene induction and enzyme secretion under cellulolytic conditions (Figure 8C). Future experiments to identify additional mutations that fully suppress the Δvib-1 cellulolytic phenotype and the identification of direct targets of VIB1 will be most informative for further dissection of glucose sensing and CCR in filamentous fungi.
Our data supports the model that the regulatory function of VIB1 on CRE1-mediated CCR and COL26-mediated glucose sensing/metabolism functions during different stages of the cellulolytic response (Figure 10). At induction stage, both VIB1 and COL26 negatively regulate CRE1-mediated CCR (Figure 9), thus allowing a relief of CCR and efficient induction of cellulolytic genes in response to cellulose. During the utilization phase, glucose is released from cellulose, and glucose sensing/signaling via COL26 may repress cellulolytic responses, with VIB1 functioning to dampen this inhibition. As many cellulolytic genes are subject to carbon catabolite repression and a requirement for CLR2 for induction, the cellular response to plant biomass may depend on the relative strength of these two antagonizing forces (Figure 10). Mechanistically, how VIB1 exerts its function on glucose sensing/metabolism via COL26 and CCR via CRE1 remain to be elucidated.
In the hyper-secreting T. reesei strain, RUT-C30, disruption of phosphoglucose isomerase gene (pgi1) blocks formation of fructose-6-P from glucose-6-P and increased cellulase production on glucose. This increase relies on a genetic interaction between the Δpgi1 mutation and the cre1-1-1 mutation in the RUT-C30 background [62]. Interestingly, both the hyper-secreting T. reesei RUT-C30 and PC-3-7 strains have mutations in cre1 and bglR/col-26 [15], [63], [64], but whether a synergy exists between Δcre1 and ΔbglR in T. reesei, as in N. crassa, and its relationship to T. reesei vib1 is unclear. Many cellulolytic enzyme hyper-producers such as T. reesei RUT-C30 and PC-3-7, and P. decumbens JU-A10-T show relief from CCR, but contain a large number of mutations in additional genes that contribute to the hyper-production phenotype [2], [15], [63]–[65]. Identifying and characterizing possible synergistic effects of the different mutations on hyper-production of lignocellulose enzymes, as shown in this study, will be a challenge.
The function of VIB1 in regulating glucose sensing/metabolism and CCR plays a role in the utilization of other complex substrates. VIB1 is required for extracellular protease production in response to carbon and nitrogen starvation, a function shared by its homolog in A. nidulans, xprG [66]–[70]. The Δvib-1 mutant also exhibits inappropriate temporal and spatial conidiation and has defects in protoperithecia formation [48], [70], two developmental events that are regulated by nitrogen and glucose limitation and signaling [71]. A shotgun proteomic analysis of culture supernatant of the Pvib-1 strains under carbon source depletion showed a higher amount of intracellular proteins relative to WT (Table S3). These data are in consistent with a role of VIB1 in promoting cell death [47], [48], [72], and in autolysis in A. nidulans [73], perhaps via perturbed nutritional signaling. Autolysis is frequently observed in submerged batch cultures in industrial bioprocessing, and promotes cryptic growth for survival and protein production under nutrient-depleted conditions [74]. Further manipulations of vib-1 and its homologs in filamentous fungi may yield economic benefits via the regulation of autolysis under industrial settings.
In summary, our data show that VIB1 is an essential regulator for cellulase production under inductive conditions and identifies COL26 as an important player in glucose sensing/metabolism. As VIB1 mediates metabolic changes as well as programmed death, two properties shared by mammalian tumor suppressor p53 [75], [76], the molecular mechanism in linking the two could be conserved, and further investigation of vib-1 function and its homologs in filamentous fungi may also shed light on cancer research.
FGSC 2489 was used as the WT reference strain and background for mutant strains [46]. FGSC 11308 (Δvib-1; mat a), FGSC 11309 (Δvib-1; mat A), FGSC 11030 (Δcol-26; mat a), FGSC 11031 (Δcol-26; mat A) were obtained from the Fungal Genetics Stock Center (http://www.fgsc.net/) [50]. The vib-1 mis-expression strain Pvib-1 (Pvib-1; Δvib-1) was constructed by transforming FGSC 11308 with a DNA fragment containing the promoter of the clock controlled gene 1 (ccg-1) and the open reading frame and 3′ untranslated region (UTR) of vib-1 and homologous and flanking regions from the coding sequence of the his-3 gene. Transformants were selected for histidine prototrophy [77] and backcrossed to FGSC 2489 to obtain a his-3::pccg-1-vib-1; Δvib-1 homokaryotic strain. The Tr vib1 mis-expression strain PTr-vib1 (PTr vib1; Δvib-1) was created in the same way except that the open reading frame and 3′UTR of Tr vib1 was used. The Pc clr-2, the Δcre-1, the Δ3βG and the Δ3βG Δcre-1 strains were from previous studies [9], [10], [26]. The Pc clr-2; Δvib-1 strain, the Δ3βG; Δvib-1 strain, the Δcol-26; Δvib-1 strain, the Δcre-1; Δcol-26 strain, and the Δcre-1; Δcol-26; Δvib-1 strain were created through crosses.
N. crassa cultures were grown on Vogel's minimal medium (VMM) [78]. Unless noted, 2% (w/v) sucrose was used as a carbon source. Strains were pre-grown on 3 mL VMM slants at 30°C in dark for 24 hrs, then at 25°C in constant light for 4–10 days to stimulate conidia production. For flask cultures, conidia were inoculated into 100 mL of liquid media at 106 conidia/mL and grown at 25°C in constant light and shaking (200 rpm). To test 2-DG and allyl alcohol sensitivity, 3 mL of liquid media containing either 0.2% (w/v) 2-DG (Sigma Aldrich, MO) or 100 mM allyl alcohol were inoculated with 106 conidia/mL and grown in 24-well plates at 25°C in constant light and shaking (200 rpm).
For crosses, one parental strain was grown on synthetic crossing medium [79] as the female for 2 weeks at room temperature for protoperithecial development. The other parental strain was used as the male to fertilize the protoperithecia. Crosses were kept for 3 weeks at room temperature. Ascospores were collected and activated as described [80], plated on 1% VMM, and incubated at room temperature for 18 hrs. Germinated ascospores were selected and transferred to selective slants for further screen and confirmation.
Cultures were grown on sucrose for 16 hrs, centrifuged at 2000 g for 10 min and washed in VMM or MM without a carbon source, followed by 4 hrs growth in 100 mL VMM or MM with 2% carbon source (sucrose, cellobiose, Avicel PH-101 (Sigma Aldrich, MO)) or with no carbon source added.
Mycelia were harvested by filtration and flash frozen in liquid nitrogen. RNA was extracted using the Trizol method (Invitrogen) and further purified using RNeasy kits (QIAGEN). Four ng of RNA was used as template in each quantitative RT-PCR (qRT-PCR) reaction. qRT-PCR was carried out using EXPRESS One-Step SYBR GreenER kit (Invitrogen) and Applied Biosystems Step One Plus Real Time PCR system. qRT-PCR were done in biological duplicates or triplicates with actin as the endogenous control. Relative expression levels were normalized to actin, and fold changes in RNA level were the ratios of the relative expression level on inducing conditions to no carbon conditions.
Libraries were prepared according to standard protocols from Illumina Inc (San Diego, CA) and sequenced on the HiSeq 2000 platforms at QB3 Vincent J. Coates Genomics Sequencing Laboratory (CA). Sequenced reads were mapped against predicted transcripts from the N. crassa OR74A genome [81](Neurospora crassa Sequencing Project, Broad Institute of Harvard and MIT http://www.broadinstitute.org/) with Tophat v2.0.4 [82]. Transcript abundance (FPKM) was estimated with Cufflinks v2.0.2 mapping against reference isoforms and differential gene expression were analyzed with Cuffdiff v2.0.2 [83]. Biological replicates used for RNA-seq showed high reproducibility. The Pearson correlation of FPKM on log basis (p-value<2.2e-16): rp≥0.96 between WT (Nc) replicates, rp≥0.91 between WT (Av) replicates, rp≥0.99 between Δvib-1 (Nc) replicates, and rp≥0.96 between Δvib-1 (Av) replicates.
For hierarchical clustering analysis, FPKM were log transformed, normalized and centered on a per gene basis with Cluster 3.0 [84] so that values from each gene ranged from −1 (minimum) to 1 (maximum). Average linkage clustering was performed with Euclidean distance as the similarity metric. Functional category analysis was done as described in [8]. Lists of genes were matched against the MIPS Functional Category Database [85], and significance of enrichment was calculated.
For CMCase and xylanase activity assays, Azo-CM-Cellulose and Azo-xylan (Beechwood) from Megazyme (Wicklow, Ireland) were used as substrates. Protein concentration was measured with the Bradford assay (BioRad). Cellulase assays were conducted by mixing 500 µL of culture supernatant with 500 µL 0.5% (w/v) Avicel in 100 mM sodium acetate, pH 5.0, and incubated with shaking at 37°C for 5 hrs. Reactions were stopped by centrifugation at 2000 g for 5 min and by addition of 9 volumes of 0.1 M NaOH to the reaction supernatants. Released glucose and cellobiose were separated on a PA-200 HPAEC column and analyzed on Dionex ICS-3000 as described in [45].
Strains were grown in 3 mL VMM with 2% cellobiose as the carbon source in the well of 24-well plates at 25°C in constant light with shaking (200 rpm) for 40 hrs to reach the same mycelial biomass, then glucose was added into each culture such that the culture was grown in MM with 1% (w/v) glucose for 1 hr. The cultures were thoroughly washed with MES buffer (10 mM 2-(N-morpholino)ethanesulfonic acid, 100 mM NaCl), and each washed culture was transferred into 4 ml of MES buffer supplemented with 10 mM glucose and grown at room temperature for 1 hr with shaking at 550 rpm. Culture supernatants were sampled at 5, 20, and 60 min, and diluted in 50 volumes of 0.1 M NaOH. Glucose levels were measured using Dionex ICS-3000 HPAEC-PA 200 and MES buffer instead of VMM was used to avoid precipitation that interferes with downstream analysis.
Culture supernatants were mixed with 4× SDS loading buffer and boiled for 10 min before loading onto Criterion 4–15% Tris-HCl Precast Gel (Bio-Rad). GelCode Blue Stain Reagent (Thermo Scientific) was used for gel staining.
Strains were inoculated in 2% sucrose VMM and grown at 25°C for 12 hrs in eight-chamber Lab-Tek chambered cover glass (Nalge Nunc International, Naperville, IL). Localization of VIB1-GFP was observed using a 100×1.4 NA oil immersion objective on a Leica SD6000 spinning disk confocal with 488 nm laser and controlled by Metamorph software. Z-series stacks were collected and maximum intensity projections were created using ImageJ. For medium shift experiment, the cultures in the chamber were washed with VMM without carbon sources and VMM with 0.5% Avicel was added, followed by immediate time-lapse recordings with an interval of 15 min.
Equal volume of culture supernatants of WT and Pvib-1 strains was subjected to SDS-PAGE and secretome proteins identified as described in [86]. In-gel trypsin-digestion was performed according to manufacture protocol (Promega, Trypsin Gold). Digested peptides were separated using ProtID-Chip-43 (II) and analyzed using the Agilent 6510 Q-TOF LC/MS as in [9].
|
10.1371/journal.pgen.1004936 | Protein Poly(ADP-ribosyl)ation Regulates Arabidopsis Immune Gene Expression and Defense Responses | Perception of microbe-associated molecular patterns (MAMPs) elicits transcriptional reprogramming in hosts and activates defense to pathogen attacks. The molecular mechanisms underlying plant pattern-triggered immunity remain elusive. A genetic screen identified Arabidopsis poly(ADP-ribose) glycohydrolase 1 (atparg1) mutant with elevated immune gene expression upon multiple MAMP and pathogen treatments. Poly(ADP-ribose) glycohydrolase (PARG) is predicted to remove poly(ADP-ribose) polymers on acceptor proteins modified by poly(ADP-ribose) polymerases (PARPs) with three PARPs and two PARGs in Arabidopsis genome. AtPARP1 and AtPARP2 possess poly(ADP-ribose) polymerase activity, and the activity of AtPARP2 was enhanced by MAMP treatment. AtPARG1, but not AtPARG2, carries glycohydrolase activity in vivo and in vitro. Importantly, mutation (G450R) in atparg1 blocks its activity and the corresponding residue is highly conserved and essential for human HsPARG activity. Consistently, mutant atparp1atparp2 plants exhibited compromised immune gene activation and enhanced susceptibility to pathogen infections. Our study indicates that protein poly(ADP-ribosyl)ation plays critical roles in plant immune gene expression and defense to pathogen attacks.
| Fine-tuning of gene expression is a key feature of successful immune responses. However, the underlying mechanisms are not fully understood. Through a genetic screen in model plant Arabidopsis, we reveal that protein poly(ADP-ribosyl)ation (PARylation) post-translational modification plays a pivotal role in controlling plant immune gene expression and defense to pathogen attacks. PARylation is primarily mediated by poly(ADP-ribose) polymerase (PARP), which transfers ADP-ribose moieties from NAD+ to acceptor proteins. The covalently attached poly(ADP-ribose) polymers on the accept proteins could be hydrolyzed by poly(ADP-ribose) glycohydrolase (PARG). We further show that members of Arabidopsis PARPs and PARGs possess differential in vivo and in vitro enzymatic activities. Importantly, the Arabidopsis parp mutant displayed reduced, whereas parg mutant displayed enhanced, immune gene activation and immunity to pathogen infection. Moreover, Arabidopsis PARP2 activity is elevated upon pathogen signal perception. Compared to the lethality of their mammalian counterparts, the viability and normal growth of Arabidopsis parp and parg null mutants provide a unique genetic system to understand protein PARylation in diverse biological processes at the whole organism level.
| Plants sense the presence of pathogens by the cell surface-localized pattern recognition receptors (PRRs), which perceive evolutionarily conserved pathogen- or microbe-associated molecular patterns (PAMPs or MAMPs), including bacterial flagellin, lipopolysaccharide (LPS), peptidoglycan (PGN), elongation factor Tu (EF-Tu), and fungal chitin [1]–[3]. A 22-amino-acid peptide corresponding to a region near the amino-terminus of flagellin (flg22) is recognized by the Arabidopsis PRR Flagellin-Sensing 2 (FLS2), a leucine-rich repeat receptor-like kinase (LRR-RLK) [4], [5]. Perception of flg22 by FLS2 induces instantaneous association with another LRR-RLK, Brassinosteroid Insensitive 1 (BRI1)-Associated Kinase 1 (BAK1), mainly through ectodomain heterodimerization of flg22-activated FLS2/BAK1 complex [6]–[9]. The receptor-like cytoplasmic kinases (RLCKs), BIK1 and its homolog PBL1, constitutively associate with FLS2 and BAK1, and are released from the receptor complex upon flg22 perception [10]–[12]. BAK1 directly interacts and phosphorylates BIK1 at both serine, threonine and tyrosine residues, thereby activating downstream signaling [12], [13]. In addition, both BAK1 and BIK1 complex with PRR EFR (receptor for EF-Tu) [11], [14], AtPEPR1 (receptor for endogenous danger signal Pep1) [12], [15], [16], and plant brassinosteroid hormone receptor BRI1 [17]–[19]. Activation of PRR complex by the corresponding MAMP triggers a series of defense responses, including rapid activation of MAP kinases (MAPKs) and calcium-dependent protein kinases, transient reactive oxygen species (ROS) production and calcium influx, stomatal closure, callose deposition and massive transcriptional reprogramming [1]–[3]. It has been shown recently that BIK1 is able to phosphorylate plasma membrane-resident NADPH oxidase family member respiratory burst oxidase homolog D (RBOHD), thereby contributing to ROS production [20], [21]. However, it remains largely unknown how PRR complex activation leads to profound immune gene transcriptional reprograming.
Protein poly(ADP-ribosyl)ation (PARylation), an important post-translational modification process, plays a crucial role in a broad array of cellular responses including DNA damage detection and repair, cell division and death, chromatin modification and gene transcriptional regulation [22]–[24] (S1 Fig.). PARylation is primarily mediated by members of poly(ADP-ribose) polymerases (PARPs), which transfer ADP-ribose moieties from nicotinamide adenine dinucleotide (NAD+) to different acceptor proteins at glutamate (Glu), aspartate (Asp) or lysine (Lys) residues resulting in the formation of linear or branched poly(ADP-ribose) (PAR) polymers on acceptor proteins (S1 Fig.). PAR activities and PARPs have been found in a wide variety of organisms from archaebacteria to mammals and plants, but they are apparently absent in yeast [25]. Human PARP-1 (HsPARP-1) is the most abundant and ubiquitous PARP among a family of 17 members, and it catalyzes the covalent attachment of PAR polymers on itself (auto-PARylation) and other target proteins, including histones, DNA repair proteins, transcription factors, and chromatin modulators [22]. HsPARP-1 possesses three functional domains with a DNA binding domain at N-terminus, auto-modification domain in the middle and a catalytic domain at C-terminus (S2A Fig.). PARylation is a reversible reaction and the covalently attached PAR on the target proteins can be hydrolyzed to free PAR or mono-(ADP-ribose) by poly (ADP-ribose) glycohydrolase (PARG) [22], [23] (S1 Fig.). PARG contains both endo- and exo-glycohydrolase activities that promote rapid catabolic destruction of PAR of target proteins [26]. There is only one PARG gene in humans with three different isoforms: PARG99 and PARG102 in the cytoplasm and PARG110 in the nucleus [26]. Mammalian PARG possesses a regulatory and targeting domain (A-domain) at the N-terminus, a mitochondrial targeting sequence (MTS) in the middle and a conserved catalytic domain at the C-terminus [27] (S2B Fig.). The catalytic core containing “GGG-X6-8-QEE” PARG signature motif interacts with PAR and executes hydrolysis activity [28]. Despite of their apparently opposing activities, members of PARPs and PARGs coordinately regulate protein PARylation and play essential roles in a wide range of cellular processes and contribute to the pathogenicity of various diseases, including cancer, cardiovascular diseases, stroke, metabolic disorders, diabetes and autoimmunity [25].
The Arabidopsis genome encodes three members of PARPs, AtPARP1 (At2g31320), AtPARP2 (At4g02390) and AtPARP3 (At5g22470) and two members of PARGs, AtPARG1 (At2g31870) and AtPARG2 (At2g31865) [23], [29] (S2 Fig.). AtPARP1 (it was originally named as AtPARP2) shares the conserved domain structure with HsPARP-1, whereas AtPARP2 (it was originally named as AtPARP1) and AtPARP3 more closely resemble HsPARP-2 and HsPARP-3 [29] (S2A Fig.). As their mammalian counterparts, plant PARPs are implicated in DNA repair, cell cycle and genotoxic stress [29]-[32]. Importantly, plant PARPs play an essential role in response to abiotic stresses. Transgenic Arabidopsis or oilseed rape (Brassica napus) plants with reduced PARP gene expression were more resistant to various abiotic stresses, including drought, high light and heat, partially attributed to a maintained energy homeostasis of reduced NAD+ and ATP consumption and alternation in plant hormone abscisic acid (ABA) levels in the transgenic plants [33], [34]. The two Arabidopsis PARG genes, AtPARG1 and AtPARG2, which were likely derived from a tandem duplication event, locates next to each other on the same chromosome [23]. AtPARG1 (TEJ) was originally identified as a regulator of circadian rhythm and flowering in Arabidopsis [35]. Interestingly, the AtPARG2 gene was robustly induced by the treatments of MAMPs and various pathogens [36]. The plants carrying mutation in AtPARG1, but not AtPARG2, showed the elevated elf18 (a 18-amino-acid peptide of EF-Tu)-mediated seedling growth inhibition and phenylpropanoid pigment accumulation, suggesting a negative role of Arabidopsis PARG in certain plant immune responses [37]. Similar to AtPARP1, AtPARG1 also plays a role in plant drought, osmotic and oxidative stress tolerance [38]. In contrast to the extensive research efforts on PARPs/PARGs in animal systems, the biochemical activities and molecular actions of plant PARPs/PARGs remain poorly characterized.
To elucidate the signaling networks regulating immune gene activation, we developed a sensitive genetic screen with an ethyl methanesulfonate (EMS)-mutagenized population of Arabidopsis transgenic plants carrying a luciferase reporter gene under the control of the FRK1 promoter (pFRK1::LUC). The FRK1 (flg22-induced receptor-like kinase 1) gene is a specific and early immune responsive gene activated by multiple MAMPs [39], [40]. A series of mutants with altered pFRK1::LUC activity upon flg22 treatment were identified and named as Arabidopsis genes governing immune gene expression (aggie). In this study, we isolated and characterized the aggie2 mutant, which exhibited elevated immune gene expression upon multiple MAMP treatments. Map-based cloning coupled with next generation sequencing revealed that Aggie2 encodes AtPARG1. Extensive biochemical analysis demonstrates that both AtPARP1 and AtPARP2 carry poly(ADP-ribose) polymerase activity, whereas AtPARG1, but not AtPARG2, possesses poly(ADP-ribose) glycohydrolase activity in vivo and in vitro. Significantly, the enzymatic activity of AtPARP2 is enhanced upon flg22 perception, suggesting the potential involvement of protein PARylation in MAMP-triggered immunity. The aggie2 mutation (G450R) occurs at a highly conserved PARG residue which is essential for both Arabidopsis AtPARG1 and human HsPARG enzymatic activity. Consistent with the negative role of AtPARG1 in plant innate immunity, AtPARP1 and AtPARP2 positively regulate immune gene activation and plant resistance to virulent bacterial pathogen infection. Our results indicate that the reversible posttranslational PARylation process mediated by AtPARPs and AtPARGs plays a crucial role in mounting successful innate immune responses upon MAMP perception in Arabidopsis.
The aggie2 mutant isolated from a genetic screen of the EMS-mutagenized pFRK1::LUC transgenic plants exhibits elevated FRK1 promoter activity upon flg22 treatment compared to its parental line, pFRK1::LUC (WT) (Fig. 1A). The elevated luciferase activity in the aggie2 mutant was observed over a 48-hr time course period upon flg22 treatment (Fig. 1B). Notably, the aggie2 mutant did not display detectable enhanced FRK1 promoter activity in the absence of flg22 treatment, suggesting its specific regulation in plant defense. In addition to flg22, other MAMPs, including elf18, LPS, PGN and fungal chitin, also elicited the enhanced FRK1 promoter activity in the aggie2 mutant (Fig. 1C), indicating that Aggie2 functions as a convergent component downstream of multiple MAMP receptors. Consistently, the aggie2 mutant displayed the enhanced FRK1 promoter activity in response to the non-pathogenic bacterium Pseudomonas syringae pv. tomato (Pst) DC3000 hrcC defective in type III secretion of effectors, and a non-adaptive bacterium P. syringae pv. phaseolicola NPS3121 (Fig. 1D). The pathogenic bacterium Pst DC3000 failed to activate pFRK1::LUC, likely due to the suppression function of multiple effectors secreted from virulent bacterium [40]. Pathogen infection or purified MAMPs could induce callose deposits in leaves or cotyledons of Arabidopsis, which has emerged as an indicator of plant immune responses [41]. We compared callose deposits by aniline blue staining in WT and aggie2 mutant plants upon flg22 treatment. The aggie2 mutant deposited more callose than WT plants 12 hr after flg22 treatment, and the size of each callose deposit appeared bigger in the aggie2 mutant than that in WT plants (Fig. 1E).
We also detected MAPK activation and ROS production, two early events triggered by multiple MAMPs, in WT and aggie2 mutant. The flg22-induced MAPK activation detected by an α-pERK antibody did not show significant and reproducible difference in WT and aggie2 seedlings (Fig. 1F), suggesting that Aggie2 acts either independently or downstream of MAPK cascade. The flg22-induced ROS burst appeared to be similar in the aggie2 mutant compared to that in WT plants (Fig. 1G). We did not observe reproducible disease alternation in the aggie2 mutant compared to WT plants in response to Pst DC3000 infection either by hand-infiltration or spray-inoculation with various inoculums and conditions (S3A Fig.). Among 7 times of disease assays with Pst DC3000 hand-infiltration, we observed that aggie2 was slightly more resistant than WT plants for 4 times, whereas we did not see the significant difference between aggie2 and WT for other 3 times (S3A Fig). By contrast, the aggie2 mutant showed enhanced susceptibility to a necrotrophic fungus Botrytis cinerea compared to WT plants as evidenced by symptom development and lesion progression after infection (S3B Fig).
We further detected endogenous FRK1 expression in flg22-treated seedlings of WT and aggie2 mutant with quantitative reverse transcription-polymerase chain reaction (qRT-PCR) analysis. The FRK1 expression was significantly elevated in the aggie2 mutant compared to that of WT pFRK1::LUC transgenic plants at both 30 min and 90 min after flg22 treatment (Fig. 1H). Similarly, the expression of several other early MAMP marker genes, including MYB15 and At2g17740 was also enhanced in the aggie2 mutant (Fig. 1H). Taken together, the results indicate that Aggie2 negatively regulates the expression of certain flg22-induced genes.
To isolate the causative mutation in aggie2, we crossed aggie2 (in the Col-0 accession background) with the Ler accession and mapped aggie2 to an 88 kilobase pair (kb) region between markers F20M17 and F22D22 on Chromosome 2 (Fig. 2A). We then performed Illumina whole genome sequencing of aggie2 and WT pFRK1::LUC transgenic plants. The comparative sequence analysis identified a G to A mutation at the position 1348 bp of At2g31870 within this 88 kb region. The mutation was further confirmed by Sanger sequencing of the genomic DNA of At2g31870. At2g31870 encodes AtPARG1 and the mutation in the aggie2 mutant causes an amino acid change of Glycine (G) at 450 to Arginine (R) (G450R) (Fig. 2B). The G450 in AtPARG1 resides in a highly conserved region at the C-terminus with unknown function. Notably, this residue is invariable in different species of plants and animals, including Arabidopsis, poplar, tomato, maize, sorghum, rice, moss, rat, mouse, human and fruit fly, suggesting the essential role of this residue in PARG functions (Fig. 2B).
To confirm that the G450R lesion in AtPARG1 is the causative mutation in aggie2, we complemented the aggie2 mutant with a construct carrying AtPARG1 cDNA fused with a FLAG epitope tag under the control of its native promoter (pAtPARG1::AtPARG1-FLAG). Two homozygous T3 transgenic lines, one line with relatively low (C2-3) and another line with moderate (C4-1) expression of AtPARG1-FLAG, were chosen for complementation assays. Both lines restored WT level of pFRK1::LUC activity upon flg22 treatment either imaged with an EMCCD camera (Fig. 2C) or quantified by a luminometer (Fig. 2D), confirming that the enhanced FRK1 promoter activity in aggie2 is caused by the mutation in AtPARG1. We also isolated T-DNA insertion line of AtPARG1, parg1-1 (SALK_147805) and parg1-2 (SALK_116088), and examined flg22-induced immune gene activation. Similar to the aggie2 mutant, parg1-1 and parg1-2 displayed the elevated activation of FRK1, MYB15 and At2g17740 after flg22 treatment compared to WT Col-0 plants (Fig. 2E). PARP inhibitor disrupted MAMP-induced cell wall lignification [37]. We found that both parg1-1 and aggie2 mutants showed the enhanced accumulation of lignin biosynthesis precursors, O-4-linked-coniferyl and sinapyl aldehydes, upon flg22 treatment by Wiesner staining (Fig. 2F). The complementation line C2-3 restored accumulation of these lignin biosynthesis precursors to the WT level (Fig. 2F). Consistent with a previous report [36], the transcript of AtPARG2, but not AtPARG1, was induced by flg22 treatment (S4A Fig.).
AtPARG1 encodes a putative poly(ADP-ribose) glycohydrolase with a predicated activity to remove poly(ADP-ribose) polymers on the acceptor proteins catalyzed by poly(ADP-ribose) polymerases (PARPs). To elucidate the biochemical activity and function of AtPARGs, we first characterized the function of AtPARPs and established in vivo and in vitro protein PARylation assays. The Arabidopsis genome encodes three PARPs, AtPARP1, AtPARP2 and AtPARP3, with each consisting of a conserved PARP catalytic domain and a variable DNA binding domain (S2 Fig.). AtPARP1 and AtPARP3 carry zinc-finger domains for DNA binding, which is similar with human HsPARP-1, whereas AtPARP2 contains two SAP domains with putative DNA binding activity. The SAP domain was named after scaffold attachment factor A/B (SAF-A/B), apoptotic chromatin condensation inducer in the nucleus (Acinus) and protein inhibitors of activated STAT (PIAS), which all have DNA and chromatin binding ability and regulate chromatin structure and/or transcription [42]. Analysis of their tissue expression pattern suggests that AtPARP1 and AtPARP2 are expressed in leaves, whereas AtPARP3 is primarily expressed in developing seeds (S4B-S4C Fig.). Thus, we focused on AtPARP1 and AtPARP2 for the functional studies.
We first tested whether AtPARP1 and AtPARP2 carry poly(ADP-ribose) polymerase activity with recombinant proteins of AtPARP1 and AtPARP2 fused with Maltose Binding Protein (MBP). In the presence of activated DNA, both AtPARP1 and AtPARP2 could catalyze PARylation reaction by repeatedly transferring ADP-ribose groups from NAD+ to itself (auto-PARylation) as appeared a ladder-like smear with high-molecular-weight proteins in a Western blot using an α-PAR antibody which detects the PAR polymers of PARylated proteins (Fig. 3A). Apparently, AtPARP2 exhibited stronger in vitro enzymatic activity than AtPARP1 when detected by α-PAR antibody. The enzymatic activity of AtPARP1 and AtPARP2 was blocked by 3-AB, a competitive inhibitor of PARP (Fig. 3A). The activity of AtPARP2 is comparable with that of human HsPARP-1 (S5A Fig.). In addition, both AtPARP1 and AtPARP2 were able to transfer ADP-ribose from Biotin-NAD+ to itself and a relatively discrete band could be detected by horseradish peroxidase (HRP) conjugated streptavidin (Fig. 3B). The specificity of PARP activity was confirmed with 3-AB treatment, which dramatically reduced auto-PARylation. Similar with the observation using α-PAR antibody, AtPARP2 exhibited stronger in vitro enzymatic activity than AtPARP1 when detected by streptavidin-HRP for biotinylated NAD+. We further developed a PARylation assay with radiolabeled 32P-NAD+ as the ADP-ribose donor (Fig. 3C). Clearly, both AtPARP1 and AtPARP2 were able to transfer ADP-ribose from 32P-NAD+ to itself as shown with SDS-PAGE autoradiograph (Fig. 3C). The formation of relatively discrete band was likely caused by these assay conditions, which favor synthesis of short polymers due to limited amount of NAD+ [43]. Together, the data support that both AtPARP1 and AtPARP2 are active poly(ADP-ribose) polymerases in vitro. It has been shown that human HsPARP-1 could modify linker histone H1 proteins and thereby create a chromatin structure more accessible to RNA polymerase II (RNAPII) to regulate transcription [44]. We further examined whether AtPARP2 was also able to PARylate Arabidopsis histone proteins. As detected with radiolabeled 32P-NAD+, AtPARP2 could PARylate two Arabidopsis histone proteins H1.1 and H1.3 (Fig. 3D). It is possible that Arabidopsis PARPs may use a similar mechanism for transcriptional regulation.
We further developed an in vivo PARylation assay with transiently expressed AtPARP2 tagged with an HA epitope at the C-terminus in Arabidopsis protoplasts. After feeding the cells with 32P-NAD+, the AtPARP2 proteins were immunoprecipitated with an α-HA antibody and separated in SDS-PAGE. A band corresponding to the predicated molecular weight of AtPARP2 was observed with autoradiograph, indicating in vivo AtPARP2 activity (Fig. 3E). This band is specific to AtPARP2 since it was absent in the vector control transfected cells. Strikingly, the flg22 treatment enhanced AtPARP2 in vivo PARylation activity as detected by increased band intensity with autoradiograph. Apparently, the flg22-mediated enhancement of AtPARP2 activity was not due to the increase of protein expression after treatment (Fig. 3E). The data demonstrate that AtPARP2 possesses poly(ADP-ribose) polymerase activity in vivo and AtPARP2-mediated protein PARylation is regulated by flg22 signaling. We further examined AtPARP2-GFP localization with Agrobacterium-mediated Nicotiana benthamiana transient assay. A strong fluorescence signal from AtPARP2-GFP was exclusively detected in the nucleus (Fig. 3F), which is consistent with its potential role in DNA repair, chromatin modulation and transcriptional regulation.
We next tested whether AtPARG1 and AtPARG2 possess poly(ADP-ribose) glycohydrolase activity (Fig. 4A). We isolated and purified AtPARG1 and AtPARG2 proteins fused with glutathione S-transferase (GST) expressed from E. coli, and established an in vitro PARG assays to examine whether AtPARGs could remove PAR from auto-PARylated AtPARP2 in vitro. As shown in Fig. 4B, AtPARG1 diminished the formation of the ladder-like smear of auto-PARylated AtPARP2 detected in a Western blot with an α-PAR antibody, suggesting the PARG activity of AtPARG1 towards AtPARP2. However, AtPARG2 appeared to be inactive towards auto-PARylated AtPARP2 in this assay (Fig. 4B). Similarly, AtPARG1, but not AtPARG2, could remove PAR polymers from auto-ADP-ribosylated AtPARP2 as detected with 32P-NAD+ autoradiograph (Fig. 4C). We further examined whether AtPARG2 may possess PARG activity specifically towards AtPARP1 but not AtPARP2. As shown in Fig. 4C, AtPARG2 did not remove PAR polymers from auto-ADP-ribosylated AtPARP1. The 6xHistidine (His6)-tagged AtPARG2 also did not display in vitro enzymatic activity (S5B Fig.). Similar to the above assays using in vitro expressed AtPARG1 proteins (Fig. 4B & 4C), the immunoprecipitated AtPARG1 expressed in Arabidopsis protoplasts almost completely removed PAR polymers from in vitro PARylated AtPARP2 (Fig. 4D). Furthermore, AtPARG1 was able to remove PAR polymers from auto-PARylated human HsPARP-1 (S5C Fig.). Similarly, human HsPARG was also able to remove PAR polymers from AtPARP2 (S5D Fig.), suggesting the functional conservation of human and Arabidopsis PARPs/PARGs.
To test whether AtPARGs carry enzymatic activity in vivo, HA-tagged AtPARG1 or AtPARG2 was co-expressed with FLAG-tagged AtPARP2 transiently expressed in Arabidopsis protoplasts. After feeding the protoplasts with 32P-NAD+, AtPARP2 activity was detected with autoradiograph after immunoprecipitation with an α-FLAG antibody. Significantly, co-expression of AtPARG1, but not AtPARG2, substantially removed PAR polymers from in vivo PARylated AtPARP2 (Fig. 4E). The expression level of AtPARG1 and AtPARG2 was similar in protoplasts as detected by an α-HA Western blot (Fig. 4E). Taken together, our data indicate that AtPARG1 has in vivo and in vitro poly(ADP-ribose) glycohydrolase activity and AtPARG2 activity was not detected. Consistently, the parg1 mutant, but not parg2 mutant, accumulated higher PAR polymers than Col-0 with dot blotting of nuclear proteins by α-PAR antibody (S5E Fig.). Subcellular localization study indicates that AtPARG1-GFP and AtPARG2-GFP reside mainly in nucleus, but also in plasma membrane and cytoplasm when transiently expressed in Arabidopsis protoplasts (Fig. 4F).
Notably, AtPARG1 protein possesses a PARG signature motif with the conserved sequence of “GGG-X7-QEE”. Mutation of E273 (the last E in the signature motif) in AtPARG1 to glycine (E273G) blocked its enzymatic activity, implicating the importance of this signature motif in PARG enzymatic activity (Fig. 5A). Examination of AtPARG2 sequence revealed that AtPARG2 has a polymorphism in the PARG signature motif. Instead of the conserved sequence “GGG-X7-QEE”, AtPARG2 possesses “GGL-X7-QEE” (Fig. 4A and 5A). The importance of this residue was shown by that the mutation of G264 (third G in the signature motif) in AtPARG1 to leucine (G264L) blocked its PARG enzymatic activity (Fig. 5A). We further determined whether lack of enzymatic activity of AtPARG2 (Fig. 4B, 4C & 4E) is due to this polymorphism in the PARG signature motif. We mutated leucine (L275) in AtPARG2 to glycine and generated the conserved “GGG-X7-QEE” motif. However, AtPARG2L275G mutant with a perfectly conserved PARG signature motif still did not show any detectable poly(ADP-ribose) glycohydrolase activity (Fig. 5A). The data suggest that the polymorphism of the PARG signature motif in AtPARG2 is not the sole determinant of its lack of detectable enzymatic activity and additional polymorphisms/deletions also account for its loss of PARG functions. There are only about 52% amino acid identity and 66% similarity between AtPARG1 and AtPARG2 (S6 Fig.).
We further addressed whether the aggie2 (G450R) mutation affected its PARG activity. Significantly, the aggie2 (G450R) mutant of AtPARG1 completely abolished its enzymatic activity detected by either α-PAR antibody (Fig. 4B) or 32P-NAD+ autoradiograph-based assay (Fig. 4C). Notably, the G450 in AtPARG1 is highly conserved among PARGs of different species (Fig. 2B). Interestingly, the corresponding mutation in human HsPARG (G867R) also abolished its activity towards HsPARP-1 and AtPARP2, suggesting the essential role of this highly conserved residue in different PARGs (Fig. 5B and S5D Fig.). The phenylalanine (F) at position 227 in bacterium Thermomonospora curvata PARG is implicated in positioning the terminal ribose and the mutation of which rendered the enzyme inactive [28]. Surprisingly, mutation of the corresponding residue F457 to glycine (F457G) in AtPARG1 did not affect its enzymatic activity (Fig. 5A), suggesting a possible distinct function mediated by this residue in different PARGs and potentially divergent evolution.
We tested the involvement of AtPARPs in plant innate immunity and immune gene activation. Because of the potential functional redundancy of AtPARP1 and AtPARP2 [30], [31], we performed disease assay and analyzed defense gene expression in atparp1atparp2 (atparp1/2) double mutant. The atparp1/2 mutant plants were more susceptible to virulent P. syringae pv. maculicola ES4326 (Psm) infection compared to WT plants as indicated by more than 10 fold increase of bacterial growth in the atparp1/2 mutant (Fig. 6A). The disease symptom development was more pronounced in the atparp1/2 mutant than WT plants (Fig. 6A). Similarly, the atparp1/2 mutant plants showed the enhanced susceptibility with bacterial growth and symptom development to the infections by Pst DC3000 and a less virulent bacterium Pst DC3000ΔavrPtoavrPtoB (Fig. 6B & S7 Fig.). In addition, the atparp1/2 mutant plants showed the reduced induction of MAMP marker genes, including FRK1 and At2g17740, compared to WT plants at 90 min after flg22 treatment (Fig. 6C). Together, these data indicate that AtPARP1 and AtPARP2 are positive regulators in plant immunity and defense gene activation to bacterial infections.
Protein PARylation mediated by PARPs and PARGs is an important, but less understood posttranslational modification process implicated in the regulation of diverse cellular processes and physiological responses [26]. In this study, an unbiased genetic screen revealed that Arabidopsis AtPARG1 plays an important role in regulating immune gene expression upon pathogen infection. We established and performed extensive in vitro and in vivo biochemical assays of PARP and PARG enzymatic activities. We have shown for the first time that Arabidopsis AtPARP1 and AtPARP2 are able to transfer ADP-ribose moieties from NAD+ to itself and acceptor proteins in vitro and in vivo. Thus, they are bona fide poly(ADP-ribose) polymerases. Interestingly, in contrast to their mammalian counterparts, AtPARP2 is more enzymatically active than AtPARP1. Significantly, MAMP perception promotes substantial enhancement of AtPARP2 enzymatic activity in vivo, reconciling the biological importance of PARPs/PARGs in regulating immune gene expression. AtPARG1, but not AtPARG2, is able to remove PAR polymers from PARylated proteins in vivo and in vitro and it is a bona fide poly(ADP-ribose) glycohydrolase. The Arabidopsis parg1 (aggie2) mutant plants exhibited elevated expression of several MAMP-induced genes and callose deposition. Conversely, the Arabidopsis atparp1/2 mutant showed reduced expression of MAMP-induced genes and enhanced susceptibility to virulent Pseudomonas infections. Thus, the data suggest that protein PARylation positively regulates certain aspects of plant immune responses. Notably, the viability and normal growth of Arabidopsis parp and parg null mutants represent a unique opportunity to study protein PARylation regulatory mechanisms in diverse biological processes at the whole organismal level.
Our results lend support to a previous study that treatment of pharmacological inhibitor of PARPs, 3-AB, disrupted elf18- and/or flg22-induced callose and lignin deposition, pigment accumulation and phenylalanine ammonia lyase activity [37]. However, the flg22-induced defense genes (FRK1 and WRKY29) were not affected by 3-AB treatment [37]. Our study with Arabidopsis parg and parp genetic mutants revealed a previously unrecognized function of protein PARylation in regulating immune gene expression upon pathogen infection. This is consistent with the general role of human PARPs and PARG in transcriptional regulation and chromatin modification [43], [45] and further substantiates the hypothesis that plant PARPs could ameliorate the cellular stresses caused by antimicrobial defenses (e.g. the effects of elevated ROS levels) [23]. Interestingly, ADP-ribosylation has also been exploited by pathogens as a means to quell plant immunity. Two Pseudomonas syringae effectors, HopU1 and HopF2, mono-ADP-ribosylate RNA-binding protein GRP7 and MAPK kinase MKK5 respectively, and interfere with their activities in plant defense transcription regulation and signaling [46], [47].
Unlike mammals and most other animals that encode a single PARG gene, the Arabidopsis genome encodes two adjacent PARG genes, AtPARG1 and AtPARG2, as well as a pseudogene At2g31860. Surprisingly, only AtPARG1, but not AtPARG2, possesses detectable poly(ADP-ribose) glycohydrolase activity in vitro and in vivo with our extensive biochemical assays. Sequence analysis identified a polymorphism in the conserved PARG signature motif “GGG-X7-QEE”, where the third G is replaced with an L in AtPARG2. The PARG signature motif is absolutely required for its enzymatic activity as mutations at this motif in AtPARG1 completely abolished its activity. However, creation of the conserved signature motif in AtPARG2 was unable to gain its PARG activity suggesting that other polymorphisms in AtPARG2 are also responsible for its lack of enzymatic activity. Consistent with our biochemical assays, the PAR polymer concentration was much higher in atparg1 mutant than that in WT plants. A similar conclusion was reached on tej mutant, which carries a G262E mutation in the invariable signature motif of AtPARG1 [35]. The atparg1, but not atparg2 mutant, affected elf18-induced seedling growth inhibition and pigment formation, and sensitivity to DNA-damaging agent [37]. Interestingly, AtPARG2 is substantially induced in multiple plant-pathogen interactions [36] and it is required for plant resistance to B. cinerea infections [37]. Thus, despite of lacking detectable enzymatic activity, AtPARG2 may still play certain role in plant immunity. It is possible that AtPARG2 may regulate AtPARG1 activity. It is also possible that AtPARG2 has evolved novel functions in plant immune responses. Several other plant species, including rice, poplar, tomato and maize, are also predicted to encode multiple PARGs [23] (S8 Fig.). Unlike Arabidopsis PARGs, different PARG members in other species have invariant signature motif. For example, all three PARGs in poplar contain GGG-X7-QEE signature motif (S8 Fig.). However, a few other species such as Eutrema salsugineum, Capsella rubella, Phaseolus vulgaris, Oikopleura dioica, and Xenopus laevis, contain PARGs with an AtPARG2-like signature GGL-X7-QEE. It remains unknown how many PARGs are enzymatic active in the species with multiple PARGs.
Although there are 17 PARPs in mammals, the parp-1parp-2 double mutant mice are not viable and die at the onset of gastrulation, suggesting the essential role of protein PARylation during early embryogenesis [48]. The lethality of parp-1parp-2 double mutant mice might be due to genomic instability. However, Arabidopsis atparp1/2 double mutant is largely morphologically similar with WT plants and does not display any obvious growth defects. Although Arabidopsis atparp1/2 double mutant was hypersensitive to genotoxic stress, they did not have significant changes in telomere length nor end-to-end chromosome fusions [30]. Albeit mainly expressed in developing seeds, AtPARP3 may have redundant functions with AtPARP1 and AtPARP2 in maintaining genome stability. It remains interesting whether atparp1/2/3 triple mutant will exert abnormal plant growth and development. Consistent with the essential function of PARylation during embryogenesis, PARG-deficient mice and Drosophila are embryonic lethal which is probably due to the accumulation of PAR polymers and uncontrolled PAR-dependent signaling [49], [50]. The normal plant growth phenotype of atparg1 mutant might be due to the redundant function of AtPARG2. However, our extensive biochemical analysis indicates that AtPARG1, but not AtPARG2, accounts for most of PARG enzymatic activity. As AtPARG1 and AtPARG2 reside next to each other on the same chromosome, it is challenging to generate the double mutant. It remains possible that other PAR-degrading enzymes with distinct sequences exist in Arabidopsis. In vertebrate, ADP-ribosyl hydrolase 3 (ARH3), a structurally distinct enzyme from PARG, could also degrade PAR polymers associated with the mitochondrial matrix [26].
We observed that AtPARP2 activity was rapidly and substantially stimulated by flg22 treatment. In line with this observation, it has been shown that bacterial infections induced the increase of PAR polymers in Arabidopsis [37]. It is well established that damaged DNA stimulates PARP activity. Recent studies have shown that pathogen treatments induce DNA damage [51], [52], which could potentially serve as a trigger to activate PARP. Treatments with virulent or avirulent Pst strains for hours could induce DNA damage in Arabidopsis as detected by abundance of histone γ-H2AX, a sensitive indicator of DNA double-strand breaks or by DNA comet assays [51]. Prolonged pathogen treatment is often accompanied with the elevated accumulation of plant defense hormone salicylic acid (SA). It has also been shown that SA can also trigger DNA damage in the absence of a genotoxic agent [53]. However, treatments of flg22 or elf18 did not induce detectable DNA damage [51]. In addition, flg22-mediated stimulation of AtPARP2 activity occurs rather rapidly and within 30 min after treatment. Apparently, flg22 signaling could directly activate AtPARP2. It is well known that human HsPARP-1 is regulated by different posttranslational modification processes, such as phosphorylation, ubiquitination, SUMOylation and cleavage [22]. HsPARP-1 could be activated by phosphorylated MAPK ERK2 in a broken DNA-independent manner, thereby enhancing ERK-induced Elk1 phosphorylation, core histone acetylation, and transcription of the Elk1-target genes [54]. MAPK cascade plays a central role functioning downstream of multiple MAMP receptors. It will be interesting to test whether flg22-activated MAPKs directly modulate PARP and/or PARG activities.
Our genetic and biochemical analyses revealed that PARP/PARG-mediated PAR dynamics regulates immune gene expression in Arabidopsis. Mammalian PARPs/PARG regulate gene expression through a variety of mechanisms including modulating chromatin, functioning as transcriptional co-regulators and mediating DNA methylation [55]. PARylation of histone lysine demethylase KDM5B maintains histone H3 lysine 4 trimethyl (H3K4me3), a histone mark associated with active promoters, by inhibiting KDM5B demethylase activity and interactions with chromatin. In addition, HsPARP-1 is able to promote exclusion of H1 and opening of promoter chromatin, which collectively lead to a permissive chromatin environment that allows loading of the RNAPII machinery [45]. HsPARG is also able to promote the formation of a chromatin environment suitable for retinoic acid receptor (RAR)-mediated transcription by removing PAR polymer from PARylated H3K9 demethylase KDM4D/JMJD2D thereby activating KDM4D/JMJD2D to inhibit H3K9me2, a histone mark associated with transcriptional repression [43]. Arabidopsis PARPs and PARGs are localized in the nucleus, and AtPARP2 could PARylate Histone H1. It is plausible to speculate that similar modes of action of protein PARylation-mediated transcriptional regulation exist in plants. Future identification of PARP/PARG targets (promoters and proteins) and PAR-associated proteins, especially during plant immune responses, will elucidate how protein PARylation modulates plant immune gene expression.
Arabidopsis accession Col-0, pFRK1::LUC transgenic plants, aggie2 mutant, atparg1-1 (SALK_147805), atparg1-2 (SALK_16088), atparg2 (GABI-Kat 072B04), atparp1/atparp2 (GABI-Kat 692A05/SALK_640400), pPARG1::PARG1-FLAG transgenic plants were grown in soil (Metro Mix 366) at 23°C, 60% humidity and 75 µE m−2s−1 light with a 12-hr light/12-hr dark photoperiod. Four-week-old plants were used for protoplast isolation and transient expression assays according to the standard procedure [56]. Seedlings were germinated on ½ Murashige and Skoog (MS) plate containing 1% sucrose, 0.8% Agar and grown at 23°C and 75 µE m-2s−1 light with a 12-hr light/12-hr dark photoperiod for 12 days, transferred to a 6-well tissue culture plate with 2 ml H2O for overnight, and then treated with 100 nM flg22 or H2O for indicated time.
Pseudomonas syringae pv. tomato (Pst) DC3000, hrcC, ΔavrPtoavrPtoB, P. syringae pv. maculicola ES4326 (Psm), or P. syringae pv. phaseolicola NPS3121 strains were cultured overnight at 28°C in the KB medium with 50 µg/ml rifampicin or streptomycin. Bacteria were harvested by centrifugation, washed, and adjusted to the desired density with 10 mM MgCl2. Leaves of 4-week-old plants were hand-infiltrated with bacterial suspension using a 1-ml needleless syringe and collected at the indicated time for luciferase activity or bacterial growth assays. To measure bacterial growth, two leaf discs were ground in 100 µl H2O and serial dilutions were plated on TSA medium (1% Bacto tryptone, 1% sucrose, 0.1% glutamic acid, 1.5% agar) with appropriate antibiotics. Bacterial colony forming units (cfu) were counted 2 days after incubation at 28°C. Each data point is shown as triplicates. Botrytis cinerea strain BO5 was cultured on Potato Dextrose Agar (Difco) and incubated at room temperature. Conidia were re-suspended in distilled water and spore concentration was adjusted to 2.5 × 105 spores/ml. Gelatin (0.5%) was added to conidial suspension before inoculation. Leaves of six-week-old plants were drop-inoculated with B. cinerea at the concentration of 2.5 × 105 spores/ml. Lesion size was measured 2 days post-inoculation.
The pFRK1::LUC construct in a binary vector was transformed into Arabidopsis Col-0 plants. The homozygous transgenic plants with flg22-inducible pFRK1::LUC were selected for mutagenesis. The seeds were mutagenized with 0.4% ethane methyl sulfonate (EMS). Approximately 6,000 M2 seedlings were screened for their responsiveness to flg22 treatment. The seedlings were germinated in liquid ½ MS medium for 14 days, and then transferred to water for overnight and treated with 10 nM flg22. After 12 hr flg22 treatment, the individual seedlings were transferred to a 96-well plate, sprayed with 0.2 mM luciferin and kept in dark for 20 min. The bioluminescence from induced pFRK1::LUC expression was recorded by a luminometer (Perkin Elmer, 2030 Multilabel Reader, Victor X3). The candidate mutants with altered flg22 responsiveness were recovered on ½ MS plate for 10 days, and then transferred to soil for seeds.
The aggie2 mutant was crossed with Arabidopsis Ler accession, and an F2 population was used for map-based cloning. Mapping with 270 F2 plants with aggie2 mutant phenotype placed the causal mutation in an 88 kb region between marker F20F17 and F22D22 on chromosome 2. The aggie2 genomic DNA was sequenced with the 100 nt paired-end sequencing on an Illumina HiSeq 2000 platform at Texas AgriLife Genomics and Bioinformatics Service (TAGS) (College Station, TX, USA). Ten-fold genome coverage was obtained with 11M reads. The Illumina reads were analyzed using CLC Genomics Workbench 6.0.1 software. By mapping to Col-0 genomic sequence (TAIR10 release), SNPs were identified as candidates of aggie2 mutation. In the aforementioned 88 kb region, a G to A mutation at the position of 1348 nt of At2g31870 was identified with 100% frequency. The mutation was confirmed by Sanger sequencing of aggie2 genomic DNA.
The AtPARP1, AtPARP2, AtPARG1, AtPARG2 and Histone H1.1 (At1g06760) genes were amplified from Arabidopsis Col-0 cDNA and cloned into a plant transient expression vector (pHBT vector) with an HA, FLAG or GFP epitope tag at the C-terminus via restriction sites NcoI or BamHI and StuI respectively. The oligos used to amplify aforementioned cDNAs are listed in S1 Table. The target genes were confirmed by Sanger sequencing. The cloned genes in plant expression vector were then sub-cloned into protein fusion vectors, pGEX-4T (Pharmacia, USA), pMAL-c (NEB, USA) or pET28a (EMD Millipore, USA), for protein expression in bacteria. For Histone H1.3 (At2g18050), we ordered cDNA from ABRC (G13366) and cloned it into a modified pMAL-c via SfiI site. Point mutations were introduced by site-directed mutagenesis PCR. The AtPARG1 promoter (1163 bp upstream of start codon ATG) was amplified from the genomic DNA of Col-0 and digested with KpnI and NcoI. The AtPARG1-FLAG-NOS terminator fragment was released from pHBT-AtPARG1-FLAG via NcoI and EcoRI digestion. The two fragments were ligated and sub-cloned into a binary vector, pCAMBIA2300 via KpnI and EcoRI sites to yield expression construct (pAtPARG1::AtPARG1-FLAG). The resulting binary vector was transformed into aggie2 via Agrobacterium-mediated transformation.
The primers for cloning and point mutations were listed in the S1 Table.
Expression and purification of GST, His6 and MBP fusion proteins were performed according to the manufacturer's manuals. For in vitro auto-PARylation reaction, 1.2 µg of MBP-AtPARP2 or MBP-AtPARP1 proteins were incubated in a 20 µl reaction with 1 × PAR reaction buffer (50 mM Tris-HCl, pH8.0, 50 mM NaCl) with 0.2 mM NAD+, and 1 × activated DNA (Trevigen, USA). To inhibit PAR reaction, 2.5 mM PARP inhibitor, 3-Aminobenzamide (3-AB, Sigma, USA), was added to the reaction. The reactions were kept at room temperature for 30 min and stopped by adding SDS loading buffer. To detect PARG activity, about 1.0 µg of purified GST, GST-AtPARG1 or GST-AtPARG2 proteins together with 2.5 mM 3-AB were added to auto-PARylated AtPARP2 proteins derived from the above PAR reactions and incubated at room temperature for another 30 min. PARylated proteins were separated in 7.5% SDS-PAGE and detected with an α-PAR polyclonal antibody (Trevigen, USA). For Biotin NAD+ PAR assay, 25 µM Biotin-NAD+ (Trevigen, USA) was added to replace NAD+ in the reaction described above. The PAR polymer formation was detected by Streptavidin-HRP (Pierce, USA). For in vitro 32P-NAD+-mediated PAR assays, 1.0 µg of MBP-AtPARP2 or MBP-AtPARP1 proteins were incubated in a 20 µl reaction in the buffer containing 50 mM Tris-HCl, pH8.0, 4 mM MgCl2, 300 mM NaCl, 1 mM DTT, 0.1 µg/ml BSA, 1 × activated DNA, 1 µCi 32P-NAD+ (Perkin Elmer, USA) and 100 nM cold NAD+ for 30 min at room temperature. For Histone PARylation assays, 2.0 µg of MBP-H1.1 or MBP-H1.3 proteins were added in the above reactions. The radiolabeled proteins were separated in SDS-PAGE and visualized by autoradiography.
For in vivo PAR assays, 500 µl Arabidopsis protoplasts at the concentration of 2 × 105/ml were transfected with 100 µg of plasmid DNA of pHBT-AtPARP2-HA. After 12 hr incubation, the protoplasts were treated with 100 nM flg22 for 30 min and fed with 1 µCi 32P-NAD+ for 1 hr. The protoplasts were then lysed in IP buffer (50 mM Tris-HCl, pH7.5, 150 mM NaCl, 5 mM EDTA, 1% Triton, 1 × protease inhibitor, 1 mM DTT, 2 mM NaF and 2 mM Na3VO4) and the AtPARP2-HA proteins were immunoprecipitated with α-HA antibody (Roche, USA) and protein-G-agarose (Roche, USA) in a shaker for 3 hr at 4°C. In vivo PARylated proteins enriched on the beads were then separated in 10% SDS-PAGE and visualized by autoradiography. For in vivo PARG assay, AtPARG1-HA or AtPARG2-HA plasmid DNA was co-transfected with AtPARP2-FLAG plasmid DNA into protoplasts, and expressed for 12 hr. The protoplasts were fed with 32P-NAD+ and subjected to immunoprecipitation as described above. The AtPARP2-FLAG proteins were immunoprecipitated with α-FLAG agarose gel (Sigma, USA), separated in 10% SDS-PAGE and visualized by autoradiography. The expression of AtPARPs and AtPARGs was detected with Western blot (WB) using the corresponding antibodies.
The 12-day old seedlings grown on ½ MS plates were harvested and ground into fine powder in liquid nitrogen. Isolation of nuclei with Honda buffer was performed according to published procedure [57]. Nuclear proteins were released in 1xPBS buffer with 1% SDS and spotted on nitrocellulose membrane. The protein loaded on the membrane was normalized by using α-Histone H3 antibody (Abcam, USA), and the PAR polymers were detected by α-PAR antibody. The relative PAR level was determined by calculating the ratio of PAR signal to Histone H3 signal after quantification of hybridization intensity with ImageJ software.
For RNA isolation, 12-day-old seedlings grown on ½ MS plate were transferred to 2 ml H2O in a 6-well plate to recover for 1 day, and then treated with 100 nm flg22 for 30 or 90 min. RNA was extracted using TRIzol reagent (Life Technologies, USA) and quantified with NanoDrop. The RNA was treated with RQ1 RNase-free DNase I (Promega, USA) for 30 min at 37°C, and then reverse transcribed with M-MuLV Reverse Transcriptase (NEB, USA). Real-time RT-PCR was carried out using iTaq Universal SYBR Green Supermix (Bio-Rad, USA) on 7900HT Fast Real-Time PCR System (Applied Biosystems, USA). The primers used to detect specific transcript by real-time RT-PCR are listed in S2 Table.
Leaves of six-week-old plants grown in soil were hand-inoculated with 0.5 µM flg22 or H2O for 12 hr. The leaves were then transferred into FAA solution (10% formaldehyde, 5% acetic acid and 50% ethanol) for 12 hr, de-stained in 95% ethanol for 6 hr, washed twice with ddH2O, and incubated in 0.01% aniline blue solution (150 mM KH2PO4, pH 9.5) for 1 hr. The callose deposits were visualized with a fluorescence microscope. Callose deposits were counted using ImageJ 1.43U software (http://rsb.info.nih.gov/ij/).
Leaves of six-week-old plants grown in soil were surface-sterilized by 70% ethanol, rinsed with H2O and incubated with 100 nM flg22 or H2O for 12 hr. The leaves were then de-stained in 95% ethanol with 2% chloroform for 12 hr and 95% ethanol for 6 hr, washed twice with 95% ethanol, and incubated in 2% phloroglucinol solution (20% ethanol, 20% HCl) for 5 min. The images were scanned by HP officejet Pro 8600 Premium.
Ten-day-old seedlings germinated on ½MS plate were transferred to 2ml H2O in a 6-well plate to recover for 1 day, and then treated with 100 nM flg22 for 5, 15 or 45 min. The seedlings were grinded in IP buffer. The cleared lysate was mixed with SDS sample buffer and loaded onto 12.5% SDS-PAGE. Activated MAPKs were detected with α-pErk1/2 antibody (Cell Signaling, USA).
ROS burst was determined by a luminol-based assay. At least 10 leaves of four-week-old Arabidopsis plants for each genotype were excised into leaf discs of 0.25 cm2, followed by an overnight incubation in 96-well plate with 100 µl of H2O to eliminate the wounding effect. H2O was replaced by 100 µl of reaction solution containing 50 µM luminol and 10 µg/ml horseradish peroxidase (Sigma, USA) supplemented with or without 100 nM flg22. The measurement was conducted immediately after adding the solution with a luminometer (Perkin Elmer, 2030 Multilabel Reader, Victor X3), with a 1.5 min interval reading time for a period of 30 min. The measurement values for ROS production from 20 leaf discs per treatment were indicated as means of RLU (Relative Light Units).
Arabidopsis protoplasts were transfected with various GFP-tagged pHBT constructs as indicated in the figures. Fluorescence signals in the protoplasts were visualized under a confocal microscope 12 hr after transfection. To construct 35S::AtPARP2-GFP binary plasmid for Agrobacterium-mediated transient assay, the NcoI-PstI fragment containing AtPARP2-GFP was released from pHBT-35S::AtPARP2-GFP and ligated into pCB302 binary vector. For tobacco transient expression, Agrobacterium tumefaciens strain GV3101 containing pCB302-35S::AtPARP2-GFP was cultured at 28°C for 18 hr. Bacteria were harvested by centrifugation at a speed of 3500 rpm and re-suspended with infiltration buffer (10 mM MES pH = 5.7, 10 mM MgCl2, 200 µM acetosyringone). Cell solution at OD600 = 0.75 was used to infiltrate 3-week-old Nicotiana benthamiana leaves. Fluorescence signals were detected 2 days post-infiltration. Fluorescence images were taken with Nikon-A1 confocal laser microscope systems and images were processed using NIS-Elements Microscope Imaging Software. The excitation lines for imaging GFP, RFP and chloroplast were 488, 561 and 640 nm, respectively.
|
10.1371/journal.pgen.1005930 | Regulation of Budding Yeast CENP-A levels Prevents Misincorporation at Promoter Nucleosomes and Transcriptional Defects | The exclusive localization of the histone H3 variant CENP-A to centromeres is essential for accurate chromosome segregation. Ubiquitin-mediated proteolysis helps to ensure that CENP-A does not mislocalize to euchromatin, which can lead to genomic instability. Consistent with this, overexpression of the budding yeast CENP-ACse4 is lethal in cells lacking Psh1, the E3 ubiquitin ligase that targets CENP-ACse4 for degradation. To identify additional mechanisms that prevent CENP-ACse4 misincorporation and lethality, we analyzed the genome-wide mislocalization pattern of overexpressed CENP-ACse4 in the presence and absence of Psh1 by chromatin immunoprecipitation followed by high throughput sequencing. We found that ectopic CENP-ACse4 is enriched at promoters that contain histone H2A.ZHtz1 nucleosomes, but that H2A.ZHtz1 is not required for CENP-ACse4 mislocalization. Instead, the INO80 complex, which removes H2A.ZHtz1 from nucleosomes, promotes the ectopic deposition of CENP-ACse4. Transcriptional profiling revealed gene expression changes in the psh1Δ cells overexpressing CENP-ACse4. The down-regulated genes are enriched for CENP-ACse4 mislocalization to promoters, while the up-regulated genes correlate with those that are also transcriptionally up-regulated in an htz1Δ strain. Together, these data show that regulating centromeric nucleosome localization is not only critical for maintaining centromere function, but also for ensuring accurate promoter function and transcriptional regulation.
| Chromosomes carry the genetic material in cells. When cells divide, each daughter cell must inherit a single copy of each chromosome. The centromere is the locus on each chromosome that ensures the equal distribution of chromosomes during cell division. One essential protein involved in this task is CENP-ACse4, which normally localizes exclusively to centromeres. Here, we investigated where CENP-ACse4 spreads in the genome when parts of its regulatory machinery are removed. We found that CENP-ACse4 becomes mislocalized to promoters, the region upstream of each gene that controls the activity of the gene. Consistent with this, the mislocalization of CENP-ACse4 to promoters leads to problems with gene activity. Our work shows that mislocalization of centromeric proteins can have effects beyond chromosome segregation defects, such as interfering with gene expression on chromosome arms.
| The eukaryotic genome is packaged into chromatin, which consists of 147 bp repeating units of DNA wrapped around histone proteins to form nucleosomes [1]. Chromatin is important not only for packaging and protecting DNA, but also for regulating access of genes and other DNA elements to nuclear proteins involved in processes such as transcription, replication, and chromosome segregation. Most nucleosomes are composed of the canonical histone proteins, H2A, H2B, H3, and H4 [2]. However, the behavior and functions of nucleosomes can be altered both by chemically modifying canonical histones through post-translational modifications and by exchanging canonical histones for histone variants that alter nucleosome composition [2]. For example, H2A.Z is a variant of histone H2A and is found at promoter nucleosomes genome-wide where it regulates transcription [2–4]. In contrast, the conserved CENP-A variant (also called CenH3) replaces H3 in nucleosomes exclusively at the centromere where it regulates chromosome segregation [5–7]. Because changes in nucleosome composition can have a major impact on the underlying functions of the genome, it is critical to understand the mechanisms that control the localization of histone modifications and variants.
The genomic incorporation of the budding yeast H2A.ZHtz1 (SGD ID: S000005372) histone variant is regulated by the SWR1 (SWR-C) and INO80 (INO80-C) chromatin remodeling complexes [8]. H2A.ZHtz1 localizes to intergenic regions, specifically near transcription start sites (TSS) at the +1 and -1 nucleosomes surrounding nucleosome-depleted regions (NDRs) [3, 4, 9–12]. In budding yeast, H2A.ZHtz1 nucleosomes are correlated with high nucleosome turnover [13], which is proposed to assist transcriptional initiation or rapid changes between transcriptional states [14–16]. SWR-C incorporates H2A.ZHtz1 into nucleosomes by exchanging H2A/H2B dimers for H2A.ZHtz1/H2B dimers [17–19]. In contrast, the mechanism of H2A.ZHtz1 removal from nucleosomes by INO80-C is less well understood because it has two reported activities that both lead to H2A.ZHtz1 exchange, either by swapping H2A.ZHtz1/H2B dimers for H2A/H2B dimers [20] or by promoting turnover of the entire nucleosome [8, 19].
The localization of the CENP-A variant is regulated by the histone chaperone HJURP (Scm3 in budding yeast), which is targeted specifically to centromeres [21–25]. Centromeric sequence and size are highly variable throughout eukaryotes and can be specified by either an underlying sequence or through epigenetic inheritance [26, 27]. Despite the diversity of centromeres, CENP-A is a conserved hallmark of all centromeres. The presence of CENP-A directs the formation of the kinetochore, a large protein complex that mediates attachments between the microtubules of the mitotic spindle and the chromosome during cell division [26, 28, 29]. CENP-A mislocalization to euchromatin through overexpression or tethering can lead to ectopic kinetochore formation and genomic instability [30–32]. However, CENP-A mislocalization has not been reported to disrupt other genomic processes [33, 34].
Multiple mechanisms ensure that CENP-A does not mislocalize to euchromatin. A number of chromatin remodelers and histone chaperones are reported to help maintain centromeric chromatin or prevent CENP-A mislocalization, including Fun30, RSC, INO80-C, CAF-1, HIR, FACT, and RbAp48 and SWI/SNF [35–41]. In addition, ubiquitin-mediated proteolysis prevents ectopic CENP-A localization by controlling total CENP-A protein levels [42–45]. In budding yeast, proteolysis of CENP-ACse4 (SGD ID: S000001532) is mediated by an E3 ubiquitin ligase called Psh1 (SGD ID: S000005415) [46, 47]. When CENP-ACse4 is overexpressed in the absence of Psh1-mediated proteolysis [42, 46–48], cells accumulate high levels of CENP-ACse4 in euchromatin. This also results in lethality, although the underlying cause has not been determined [42, 46, 47].
Similar to CENP-A, H2A.Z also contributes to chromosome segregation. In human cells, H2A.Z is found at pericentromeric regions, where it is incorporated at the inner centromere between the CENP-A nucleosome domains, and helps to establish centromeric heterochromatin [49, 50]. Similarly, H2A.ZHtz1 is also a component of pericentromeric chromatin in budding yeast, where it localizes to nucleosomes flanking the CENP-ACse4 nucleosome and is important for chromosome segregation through unknown mechanisms [4, 51, 52]. However, it is unclear whether there is a connection between the localization of the histone variants. In human cells, overexpressed CENP-A was found to mislocalize to regions enriched for H2A.Z, although no physical interaction was detected between these two histone variants [33]. In contrast, studies in S. pombe have shown that CENP-ACnp1 tends to mislocalize to ectopic regions that are depleted of H2A.ZHtz1 [53].
We set out to determine whether there are features of euchromatin that normally prevent budding yeast CENP-A misincorporation as well as to identify the functional consequences of CENP-A mislocalization to euchromatin. The identification of euchromatic sites that strongly misincorporate CENP-A may also shed light on the underlying cause of the lethality. To address these questions, we performed the first genome-wide analysis of CENP-A overexpression in the absence of ubiquitin-mediated degradation. We found that overexpressed CENP-ACse4 mislocalizes to promoters that are enriched for NDRs flanked by H2A.ZHtz1, and this mislocalization is dramatically enhanced in cells that cannot degrade CENP-ACse4. This localization pattern appears to be due in part to co-opting of INO80-C to incorporate excess CENP-ACse4 into promoter nucleosomes that normally contain H2A.ZHtz1. Consistent with this, there was a significant correlation between transcripts that were misregulated in cells lacking H2A.ZHtz1 and those with high levels of CENP-ACse4 mislocalization. We also found that a subset of promoters that misincorporate CENP-ACse4 have decreased transcription, which may be the underlying cause of lethality. Together, these data suggest that it is essential that cells regulate CENP-ACse4 localization not only to ensure proper chromosome segregation, but also to protect cells from promoter nucleosome disruption and transcriptional misregulation.
To identify the precise genomic sites of CENP-ACse4 mislocalization in budding yeast, we performed ChIP-seq on endogenous and overexpressed CENP-ACse4 in the presence and absence of Psh1-mediated proteolysis. All strains contained a fully functional ectopic 3Flag-CSE4 gene integrated at the URA3 locus under the endogenous promoter and were deleted for the endogenous CSE4 gene. Cells overexpressing CENP-ACse4 contained an additional copy under the control of the GAL promoter (pGAL-3Flag-CSE4). As seen previously, CENP-ACse4 overexpression inhibited the growth of WT cells and resulted in lethality in psh1Δ cells (S1A Fig) [46, 47]. The growth inhibition correlated with the total amount of chromatin-bound CENP-ACse4 protein (Fig 1A, S1B Fig).
To analyze CENP-ACse4 localization, cells were crosslinked with formaldehyde and the chromatin was isolated and subsequently digested with Micrococcal nuclease (MNase), which cuts linker DNA between nucleosomes. The CENP-ACse4 nucleosomes were purified from the MNase-treated chromatin by immunoprecipitation of 3Flag-Cse4. The amount of CENP-ACse4 recovered in the ChIP samples reflected the starting levels in the chromatin (S1C Fig). The input samples (MNase-digested chromatin) and ChIP samples (3Flag-Cse4-bound chromatin after immunoprecipitation) were made into paired-end sequencing libraries using a modified Solexa library preparation protocol that captures DNA particles down to ~25 bp (S1D Fig) [54, 55]. Paired-end sequencing resulted in greater than 1.5 million reads/sample, with an average read length ranging from 147–164 bp (S1 Table). The mononucleosome-sized sequencing reads from the input and ChIP samples for each strain were mapped to the S. cerevisiae reference genome version SacCer3 [56].
The peaks of CENP-ACse4 enrichment genome-wide correlated with the levels of chromatin-bound CENP-ACse4 (Fig 1B, Table 1). Seventeen peaks were identified for the 3Flag-CSE4 strain, representing the sixteen centromeres as well as a peak 150 bp from CEN9. A small amount of CENP-ACse4 mislocalization was seen starting in the psh1Δ strain with 66 peaks, and was further increased in cells overexpressing CENP-ACse4 with 4043 peaks. The greatest enrichment in the euchromatin was detected in the psh1Δ cells overexpressing CENP-ACse4 with 14,199 peaks. An example of the coverage data and corresponding peaks for a representative region around Centromere 4 shows a single centromere peak for the WT strain and additional peaks around the centromere in the other strains (Fig 1C). The increased CENP-ACse4 mislocalization in surrounding euchromatin is especially apparent in the pGAL-3Flag-CSE4 and psh1Δ pGAL-3Flag-CSE4 strains that have the highest levels of CENP-ACse4. We independently confirmed the CENP-ACse4 enrichment at CEN4 and at other representative peaks by ChIP-qPCR (S2A Fig). Our initial analysis also identified a CENP-ACse4 peak at the rDNA locus in all strains. This did not show significant enrichment in the 3Flag-CSE4 strain by ChIP-qPCR but did in the cells with overexpressed CENP-ACse4, similar to previously reported data [46, 57] (S2A Fig). However, due to the difficulty in analyzing this repetitive region by standard mapping algorithms, ChIP coverage of this region was excluded from further computational analyses.
To determine if mislocalized CENP-ACse4 favors certain genomic regions, we analyzed the percentage of CENP-ACse4 peaks in various functional regions of the genome, including centromeres, pericentromeres, telomeres, replication origins, genes, and intergenic regions (Fig 1D). We defined pericentromeres as 20 Kilobases (Kb) flanking each centromere, consistent with the 20–50 Kb size of cohesin enrichment around each centromere in budding yeast [58, 59]. As expected, the majority of CENP-ACse4 peaks in WT cells were at centromeres, with an increase in pericentromeric peaks in the psh1Δ mutant. However, the majority of peaks in the strains overexpressing CENP-ACse4 were in the intergenic regions, with a smaller percentage within genes. As intergenic regions make up less than 30% of the entire genome, these data indicate a strong enrichment of CENP-ACse4 in intergenic regions in cells overexpressing CENP-ACse4.
We next asked whether the intergenic enrichment correlates with features known to be associated with centromeres. ChIP-seq of mildly overexpressed CENP-ACse4 previously identified 23 centromere-like regions (CLRs) on chromosome arms that are enriched for mislocalized CENP-ACse4 and other kinetochore proteins [60]. These CLRs share characteristics with centromeric sequences such as having a high AT% and conferring stability to plasmid DNA. As expected, most of the CLRs have CENP-ACse4 peaks in the psh1Δ pGAL-3Flag-CSE4 strain (S2B Fig). However, CENP-ACse4 was overexpressed to much higher levels in our study (150-fold compared to 3-fold), so the CLRs are a small fraction of the total peaks. Consistent with this, there was also enrichment in low confidence negative control regions (LCNCRs), indicating there is no preference for CLR localization. We also analyzed the AT content of the DNA bound by mislocalized CENP-ACse4, as this is a defining characteristic of centromeric DNA in budding yeast. As expected, CENP-ACse4 peaks were highly enriched for AT nucleotides in the WT strain. However there was only a moderate increase in AT% in the psh1Δ strain compared to the input nucleosomes, and almost no AT bias in the strains with overexpressed CENP-ACse4 (S2C–S2F Fig). Together, these data indicate that the mislocalization of CENP-ACse4 is due to a more widespread effect than just centromere-like characteristics.
We next asked whether the intergenic enrichment of overexpressed CENP-ACse4 was specific to either promoters (defined as 500 bp upstream of the transcription start site (TSS)) or transcription terminators (defined as 500 bp downstream of the transcription termination site (TTS)) by calculating the number of peaks in these regions (Table 1). CENP-ACse4 was enriched in both regions when overexpressed, so we more precisely analyzed the pattern by plotting the average coverage in 10 bp windows for regions 500 bp upstream and downstream of all TSS or TTS (the TSS or TTS is plotted at position 0 based on previously reported RNA-seq transcription start positions [61]). In the psh1Δ cells overexpressing CENP-ACse4, there was enrichment -200 bp from the TSS and directly over the TSS, which correspond to the -1 and +1 nucleosomes respectively (Fig 2A). At the TTS, CENP-ACse4 was enriched in the nucleosome just after the termination site, and was shifted slightly into the NDR compared to the WT nucleosomes. Although the level of CENP-ACse4 enrichment in the other three strains was much lower overall, the trend is similar in the cells with increased CENP-ACse4. This pattern is reminiscent of the pattern of CENP-ACse4 mislocalization upon deletion of CAC1 and HIR1, which leads to ectopic CENP-ACse4 enrichment at promoters in the presence of Psh1 [38].
We next asked whether the accumulation of CENP-ACse4 in promoters and terminators is associated with the basal level of transcription in WT cells. We plotted CENP-ACse4 enrichment at the TSS and TTS of genes binned into quartiles by the published transcription levels in a WT strain, ranked from lowest transcription to highest transcription [62]. However, there was no correlation between CENP-ACse4 enrichment and the different transcription levels (Fig 2B, S3 Fig). Therefore, the CENP-ACse4 localization to promoters in the psh1Δ pGAL-3Flag-CSE4 strain was not an artifact of increased chromatin accessibility in areas of high transcription, such as was found in the previously reported CENP-ACse4 ChIP-seq for slightly overexpressed or hypomorphic CENP-ACse4 [63]. We also analyzed whether CENP-ACse4 mislocalization correlated with the direction of transcription of the surrounding genes, since this has been shown for cohesin localization, which is specifically enriched in convergent intergenic regions outside of the pericentromere [58, 64]. We classified the intergenic regions as tandem (between two genes transcribed in the same direction), convergent (between two genes transcribed towards each other), or divergent (between two genes transcribed away from each other) (Fig 2C). In promoters, CENP-ACse4 was enriched at the tandem and divergent genes (Fig 2D, S4 Fig). At the terminators, CENP-ACse4 was enriched at the tandem TTS and depleted at the convergent TTS. Because convergent regions lack promoters, these data are consistent with the enrichment of CENP-ACse4 to promoter regions.
Since CENP-ACse4 mislocalization to promoters was not correlated with transcription levels, we looked for another chromatin feature specific to promoters that might enhance CENP-ACse4 incorporation. One characteristic of promoters that is less commonly found at the 3’ ends of genes is the NDR between the -1 and +1 nucleosomes at the TSS [65]. We therefore compared CENP-ACse4 profiles centered on all NDRs and found a strong CENP-ACse4 enrichment in the nucleosomes flanking the NDRs in the psh1Δ pGAL-3Flag-CSE4 strain (Fig 3A). Because NDRs vary in length up to 557 bp, we asked whether there was a specific NDR length that correlated with CENP-ACse4 mislocalization and found the highest enrichment in NDRs longer than 65 bp (Fig 3B). We obtained similar results when the analysis was centered on the TSS instead of the NDR (S5A–S5D Fig), consistent with the enrichment of CENP-ACse4 in NDR containing promoters.
The localization of CENP-ACse4 to the nucleosomes flanking the NDRs is similar to H2A.ZHtz1, the only other histone variant in budding yeast [4]. In addition, the SWR-C chromatin-remodeling complex that incorporates H2A.ZHtz1 preferentially binds to NDRs greater than 50 bp [66], similar to the length of NDRs that have the highest CENP-ACse4 enrichment (greater than 65 bp) (Fig 3B). We therefore investigated the relationship between previously reported H2A.ZHtz1 localization [4] and the mislocalization of overexpressed CENP-ACse4 in psh1Δ cells. There was a striking similarity in their enrichment at NDRs (Fig 3C), as well as a similar trend of co-enrichment in the nucleosomes flanking replication origins (S5E and S5F Fig) and centromeres (Fig 3D, S5G Fig). The CENP-ACse4 coverage at the TSS was also similar to H2A.ZHtz1 coverage, while at the TTS CENP-ACse4 was shifted more into the 3’ NDR than H2A.ZHtz1 (S6A and S6B Fig). The histone variants exhibited a genome-wide trend to co-localize, as seen in a representative region of the arm of Chromosome 4 (Fig 3E and S6C Fig). There was a high coincidence of overlap between CENP-ACse4 peaks in the experimental strains with H2A.ZHtz1 peaks in WT cells, although they were not specifically enriched in any of the genomic features correlated with CENP-ACse4 mislocalization (Fig 3F, S6D and S6E Fig) Together, these data indicate a significant enrichment of misincorporated CENP-ACse4 at sites where H2A.ZHtz1 nucleosomes are normally located genome-wide in psh1Δ cells overexpressing CENP-ACse4.
The co-localization of the histone variants led us to further analyze their relationship. First, we tested whether H2A.ZHtz1 promotes CENP-ACse4 localization by performing ChIP on WT, psh1Δ, and psh1Δ htz1Δ cells overexpressing CENP-ACse4. htz1Δ cells are defective in induction from the GAL promoter [67], so we used a tetracycline promoter to control CSE4 levels. Overexpressed CENP-ACse4 bound to promoter regions in the psh1Δ htz1Δ double mutant, at levels similar to or even higher than the psh1Δ strain (Fig 4A). These data indicate that H2A.ZHtz1 is not required for CENP-ACse4 mislocalization, so we next asked whether the H2A.ZHtz1 incorporation machinery is involved. Swr1 (SGD ID: S000002742) is the Swi/Snf related ATPase in SWR-C that deposits H2A.ZHtz1 into nucleosomes [11, 19, 68], so we measured the levels of chromatin-bound CENP-ACse4 in swr1Δ cells. We confirmed that H2A.ZHtz1 was reduced at a previously reported promoter nucleosome locus by ChIP-PCR (Fig 4B)[17, 68, 69]. Similar to our findings with the htz1Δ mutant, bulk H2A.ZHtz1 was not depleted in the chromatin fraction in swr1Δ, but CENP-ACse4 chromatin levels were somewhat higher in the swr1Δ psh1Δ cells compared to psh1Δ (Fig 4C, S7A and S7B Fig). In addition, there was no change in CENP-ACse4 stability in swr1Δ cells (S7C Fig). We also tested whether CENP-ACse4 overexpression in the psh1Δ mutant affects H2A.ZHtz1 promoter occupancy, but did not detect an effect at the loci analyzed (S7D Fig). However, given that H2A.ZHtz1 is estimated to occupy only a small proportion of nucleosomes at any given locus in the population, it may be difficult to detect a significant difference [11, 19]. Together, our data suggest that although ectopic CENP-ACse4 and WT H2A.ZHtz1 localize to similar sites, the H2A.ZHtz1 incorporation machinery does not promote CENP-ACse4 mislocalization and may instead help to prevent CENP-ACse4 promoter incorporation.
Since the ectopic localization of CENP-ACse4 does not depend on H2A.ZHtz1 incorporation, we asked whether chromatin remodelers that remove H2A.ZHtz1 are involved. INO80-C has been reported to act preferentially on H2A.ZHtz1-containing +1 nucleosomes and to promote full nucleosome turnover [19, 20]. We therefore hypothesized that CENP-ACse4 might be incorporated into chromatin when canonical H3 is removed by INO80-C-mediated nucleosome turnover. Previous work showed that deletion of the ATPase Ino80 (SGD ID: S000003118) leads to a global alteration of H2A.ZHtz1 localization patterns genome-wide without affecting the overall levels of H2A.ZHtz1 incorporation in the genome [20, 70]. However, this deletion mutant is not viable in the strain background we used in this study [71]. We therefore used a deletion of NHP10 (SGD ID: S000002160), a non-essential INO80-C subunit that facilitates binding to nucleosomes and DNA, but that does not affect catalytic activity in vitro [72–74]. To analyze CENP-ACse4 levels, we performed chromatin fractionation in WT and nhp10Δ cells overexpressing CENP-ACse4. Similar to previously reported work, we did not detect a change in total H2A.ZHtz1 levels in the chromatin in the nhp10Δ strain (S8A and S8B Fig)[20, 70]. However, CENP-ACse4 chromatin levels were somewhat reduced when NHP10 was deleted (Fig 5A and S8B Fig), suggesting that INO80-C histone exchange activity contributes to CENP-ACse4 misincorporation. To more directly test this possibility, we asked whether Ino80 associates with CENP-ACse4 in vivo. CENP-ACse4 co-immunoprecipitated with Ino80 (Fig 5B), and this interaction increased in the absence of Psh1. To determine how this affects cell viability, we also analyzed the growth of nhp10Δ mutant cells overexpressing CENP-ACse4. Although strong CENP-ACse4 overexpression is lethal to psh1Δ cells regardless of the presence of NHP10 (S8C Fig), a deletion of NHP10 improved the growth of psh1Δ mutant cells that were moderately overexpressing CENP-ACse4 (Fig 5C). We confirmed these effects were not due to altered levels or stability of CENP-ACse4 in nhp10Δ mutant cells (S8D and S8E Fig). Together, these data suggest that at least some of the ectopic CENP-ACse4 deposition is likely coupled to the chromatin remodeling activity of INO80-C.
The mislocalization of CENP-ACse4 to promoters suggested that it could lead to transcriptional changes in the downstream genes. In addition, the relationship between CENP-ACse4 incorporation and H2A.ZHtz1 removal by INO80-C suggested that any transcriptional changes might correlate with those in htz1Δ cells. We therefore performed RNA-seq on WT, psh1Δ, pGAL-3Flag-CSE4, psh1Δ pGAL-3Flag-CSE4 and htz1Δ strains that were treated with galactose for two hours. As a control, we also included a pGAL-H3 strain to ensure any effects were specific to CENP-ACse4 overexpression and not just an effect of increased histone turnover. Cells containing just a PSH1 deletion or overexpressing CENP-ACse4 or H3 had very little change in transcription (Fig 6A and 6B). However, a large number of genes were misregulated in psh1Δ cells overexpressing CENP-ACse4, as well as in htz1Δ cells as previously described [75, 76]. We confirmed that these gene expression changes were not due to an indirect effect of CENP-ACse4 mislocalization to the rDNA by measuring the rDNA copy number and rRNA transcript levels, which were not significantly different between the strains (S9A and S9B Fig). We also confirmed that the differentially transcribed genes in the psh1Δ pGAL-3Flag-CSE4 strain are not a consequence of altered cell cycle progression [47, 77] (S9C Fig).
To determine whether CENP-ACse4 mislocalization to promoters correlates with transcriptional misregulation of downstream genes, we compared the promoters with CENP-ACse4 peaks to the genes showing altered transcription in the psh1Δ strain overexpressing CENP-ACse4. While there was a significant overlap (p = 0.0009, hypergeometric distribution) between the down-regulated genes and those with promoter CENP-ACse4 peaks (S9D Fig), the vast majority of genes with CENP-ACse4 promoter peaks do not have changes in transcription. This is similar to the relationship between H2A.ZHtz1 peaks and the genes that are differentially regulated in htz1Δ [9], confirming that changes in the histone composition of promoters does not always lead to direct transcriptional effects. However, the downregulated genes have much higher CENP-ACse4 coverage at the +1 nucleosome compared to other promoters, suggesting that both the amount and position of CENP-ACse4 misincorporation may determine which downstream genes become misregulated (Fig 6C). Analysis of transcription factor binding sites enriched at promoters of the downregulated genes with CENP-ACse4 promoter peaks identified Cse2 (SGDID: S000005293) as the most significantly enriched transcription factor (S2 File). Cse2 is a subunit of the RNA Polymerase II Mediator complex, and has also been shown to be required for chromosome segregation [79, 80], leading to the possibility that the transcriptional defects are correlated with altered Cse2 function.
Given the relationship between CENP-ACse4 and H2A.ZHtz1 localization, we also asked whether there was a correlation between the transcriptional changes in psh1Δ pGAL-3Flag-CSE4 and htz1Δ mutant cells. Interestingly, there was a significant overlap between the genes that increased transcription in both strains (Fig 6D), and these were also enriched for CENP-ACse4 in the NDR (S9E Fig). We analyzed the promoters of the affected genes for common transcription factors and found 24 that are enriched at the promoters of these genes (S2 File), so the underlying mechanism for the misregulation is not clearly associated with one factor. However, these data are consistent with the relationship between CENP-ACse4 mislocalization and the INO80-C chromatin remodeling machinery that controls H2A.ZHtz1.
In this study, we performed the first genome-wide localization of the centromeric histone variant CENP-ACse4 in the absence of Psh1-mediated proteolysis and found that it mislocalizes to intergenic regions when overexpressed. There was a significant correlation between the sites of CENP-ACse4 mislocalization and nucleosomes that normally incorporate the H2A.ZHtz1 variant. Consistent with this, we found that INO80-C, which acts on H2A.ZHtz1 nucleosomes, also contributes to the ectopic localization of CENP-ACse4, identifying another mechanism that promotes CENP-ACse4 mislocalization. We also found that the number of CENP-ACse4 ectopic peaks is significantly enhanced and leads to transcriptional defects when Psh1 is absent, underscoring the importance of proteolysis in maintaining genome stability through the exclusive localization of the centromeric histone variant.
The intergenic mislocalization of CENP-ACse4 is similar to what has been observed with mild CENP-ACse4 overexpression [54, 60] although we found that the mislocalization of overexpressed CENP-ACse4 is much stronger in the absence of proteolysis. In human cells, CENP-A overexpression misincorporates at CTCF binding sites, which are associated with the histone variants H2A.Z and H3.3 and have high levels of histone turnover [33]. In budding yeast, the connection between histone turnover and CENP-ACse4 mislocalization is less clear. High histone turnover and more open chromatin have been shown to be permissive for CENP-ACse4 mislocalization [54, 60]. This is consistent with our results, as promoters have a higher level of turnover than intragenic regions [13]. However, a caf1Δ hir1Δ double mutant that decreases histone turnover genome-wide still mislocalizes even endogenous levels of CENP-ACse4 to promoters [38]. Therefore, histone turnover is not strictly required for CENP-ACse4 mislocalization and there must be additional mechanisms that promote the ectopic deposition of CENP-ACse4.
We identified a strong similarity between H2A.ZHtz1 localization and CENP-ACse4 mislocalization in nucleosomes flanking NDRs, such as replication origins, centromeres, and +1 nucleosomes at promoters. We also found that INO80-C contributes to CENP-ACse4 mislocalization. CENP-ACse4 co-immunoprecipitates with INO80-C, and this interaction is increased in the psh1Δ mutant where there are higher levels of CENP-ACse4. Consistent with this, an nhp10Δ mutant reduced the ectopic localization and partially rescued the growth defect of the psh1Δ mutant when CENP-ACse4 was overexpressed. However, nhp10Δ does not fully rescue the lethality or ectopic deposition, so additional chromatin remodelers or histone chaperones must also contribute to ectopic CENP-ACse4 incorporation. In humans, the chaperone activity of DAXX is involved in CENP-A deposition in euchromatin [33], but there is no ortholog of this protein in budding yeast.
H2A.ZHtz1 localization to nucleosomes flanking NDRs requires SWR-C binding, and SWR-C enrichment is increased with longer NDRs in vivo [66]. Similarly, we found that CENP-ACse4 is enriched at longer NDRs. However, we determined that H2A.ZHtz1 and SWR-C are not required for CENP-ACse4 deposition. Our work is instead consistent with the possibility that the two yeast histone variants could have an antagonistic relationship, such that they are found at the same places in the genome, but never at the same time. This is reminiscent of the relationship between CENP-ACnp1 and H2A.ZHtz1 in fission yeast, where CENP-ACnp1 forms neocentromeres in regions with low H2A.ZHtz1 when the endogenous centromere is deleted [53]. However, we detect CENP-ACse4 mislocalization at nucleosomes that normally have high H2A.ZHtz1 enrichment. We speculate that this is due to different mechanisms leading to ectopic deposition. In fission yeast, the ectopic CENP-ACnp1 localization to neocentromeres depended on the centromeric chaperone [53], while our data suggests a role for INO80-C in the ectopic deposition of highly expressed CENP-ACse4. Given that INO80-C acts in opposition to SWR-C to remove H2A.ZHtz1 from nucleosomes, we propose that the full nucleosome turnover activity of INO80-C leads to the removal of H3 and the incorporation of CENP-ACse4 into promoter nucleosomes (Fig 6E). This model explains both the co-localization of the histone variants and the potentially antagonistic relationship between H2A.ZHtz1 and CENP-ACse4 in the chromatin.
Although there is a significantly higher level of euchromatic CENP-ACse4 in the absence of Psh1, the locations of the ectopic nucleosome positions are similar regardless of Psh1 activity. In both cases, overexpressed CENP-ACse4 is enriched intergenically, suggesting that Psh1 does not have preferential sites of action genome-wide. However, CENP-ACse4 was not significantly incorporated into genes even in the absence of Psh1, suggesting that additional mechanisms control its localization. We previously showed that the FACT complex, which was recently demonstrated to remove H2A.ZHtz1 from genes, interacts with Psh1 to facilitate CENP-ACse4 degradation [47, 70, 81]. However, FACT does not interact with CENP-ACse4 in the absence of Psh1 [47]. One possibility is that FACT could indirectly antagonize CENP-ACse4 mislocalization into genes by ensuring that H3 is quickly reincorporated into nucleosomes following transcription, similar to its role in fission yeast [40]. In the future, it will be important to understand how intragenic regions are protected from CENP-ACse4 deposition.
For the first time in any organism, we detected large-scale changes in transcription when CENP-A mislocalized to euchromatin. This only occurred in cells lacking Psh1, and the downregulated genes had very high levels of CENP-ACse4 in their promoters. This suggests that strong misincorporation of CENP-ACse4 at a promoter may be required to cause transcriptional defects, and may explain why this has not been previously observed. The levels of CENP-ACse4 overexpression achieved in the absence of proteolysis are much higher than previous studies that have analyzed CENP-ACse4 mislocalization. It is not clear whether mislocalization of CENP-ACse4 at a given promoter is sufficient to directly decrease transcription. We found a significant enrichment of the Cse2 transcription factor in the promoters of the downregulated genes, leading to the intriguing possibility that CENP-ACse4 incorporation alters Cse2 function at a subset of genes to inhibit transcription. It is interesting to note that Cse2 and Cse4 were identified in the same genetic screen for mutants in chromosome segregation [5, 79], and it will be important to further explore their relationship in the future.
We also identified genes that increased transcription when CENP-ACse4 was mislocalized, and these significantly overlap with those altered in htz1Δ mutant cells. This further confirms the potential antagonistic relationship between the yeast histone variants, and suggests that high levels of CENP-ACse4 may lead to similar chromatin changes at a subset of promoters as cells lacking H2A.ZHtz1. The underlying mechanism for why only a fraction of promoters that contain H2A.ZHtz1 are transcriptionally up-regulated in its absence is not known. We speculate that a change in nucleosome positioning or stability occurs at these promoters that facilitates the access of transcriptional machinery. Consistent with this, we found that the up-regulated gene promoters have CENP-ACse4 enrichment within rather than flanking the NDR and lack strong +1 enrichment.
We found that regulating the levels and localization of the centromeric histone variant is critical to prevent transcriptional misregulation in budding yeast. Although CENP-A mislocalization leads to the formation of ectopic kinetochores in other organisms, we have not been able to determine whether this occurs in budding yeast due to the difficulty of detecting ectopic kinetochores [47]. Our work suggests the possibility that transcriptional defects due to the mislocalization of CENP-ACse4 in the absence of proteolysis may be the underlying cause of lethality in these cells. These data highlight the need to accurately regulate the localization of the centromeric histone variant CENP-ACse4 to both ensure genomic stability through its centromeric functions, as well as to prevent the disruption of euchromatic functions.
Microbial techniques and media were as described [82, 83]. For all experiments involving induction of pGAL-3Flag-CSE4 or pGAL-H3, budding yeast cells of indicated strains were grown to log phase (OD 0.55–0.8, Bio-Rad SmartSpec 3000) in lactic acid media at 23°C and induced for 2 hours with 2% galactose. Yeast strains were constructed using standard genetic techniques. Epitope-tagged proteins were constructed using either a PCR integration technique [84] or by the integration of plasmids after restriction digestion. Specific plasmids and yeast strains used in this study are described in the S2 and S3 Tables.
Protein extracts to check total CENP-ACse4 levels were prepared as described [85]. Immunoblots using chemiluminescence were performed as previously described [85]. For all immunoblots, the antibody dilutions were as follows: Mouse anti-Pgk1 monoclonal antibodies (Invitrogen Catalog # 459250) at a 1:10,000 dilution were used as a loading control. Mouse anti-Flag M2 monoclonal antibodies (Sigma-Aldrich Catalog # F3165) were used at a 1:3000 dilution, Mouse anti-HA 12CA5 monoclonal antibodies (Roche Catalog # 1-583-816) were used at a 1:10,000 dilution, and rabbit anti-H2B polyclonal antibodies (Active Motif Catalog # 39237) were used at a 1:3,000 dilution. Mouse anti-Myc 9E10 monoclonal antibodies were used at a 1:10,000 dilution (Covance Catalog # MMS-150R). Co-IP experiments were performed as previously described [81] for Psh1-Myc and Ino80-Myc strains using 5ul Protein G Dynabeads conjugated with 1.5ul anti-Myc (A-14, SC-789) and run on a gradient SDS-PAGE gel. Quantitative immunoblots were carried out according to [86] with the modification of using 4% non-fat milk in PBS as the blocking agent for the anti-Flag immunoblot. Briefly, IRDye anti-mouse and anti-rabbit secondary antibodies from LI-COR were used at a 1:15,000 dilution. The immunoblots were imaged on a LI-COR imaging system, and the protein levels were quantified using Image Studio Lite.
Chromatin fractionation assays were performed as described [81], followed by quantitative immunoblots. The mean and SEM of three independent experiments is reported. anti-PGK1 was used as a marker and loading control for the soluble fraction, and anti-H2B was used as a marker and loading control for the chromatin fraction. The Cse4:H2B and H2A.Z:H2B ratios were normalized to the pGAL-3Flag-CSE4 strain. Note that the levels of H2A.ZHtz1 and H2B are somewhat variable between strains. This may be due to differential susceptibility of the cell wall to zymolyase digestion during the chromatin fractionation procedure, which seems to vary between strains. To control for this, we used H2B to determine the level of total chromatin in each condition.
3Flag-Cse4-containing nucleosomes were isolated by ChIP of 3Flag-Cse4 using monoclonal anti-Flag M2 antibodies (Sigma-Aldrich Catalog # F3165). ChIPs were performed with Micrococcal nuclease (MNase, Worthington Biochemical Corporation Catalog # LS004798)-treated chromatin as described [55] with the following addition. Before nuclei isolation, proteins were crosslinked to DNA with 1% formaldehyde for 15 minutes. Crosslinks were then reversed before DNA extraction by the addition of 1% SDS and an overnight incubation at 65°C [87]. DNA was extracted using phenol:chloroform extraction and ethanol precipitation, and was treated with RNAse and purified using a Qiagen Reaction Clean-up kit before library construction. Paired-end sequencing libraries of both input DNA from MNase-digested chromatin and 3Flag-Cse4 ChIP DNA were prepared using a modified Solexa library preparation protocol that captures DNA particles down to ~25 bp [55]. Cluster generation, followed by 25 cycles of paired-end sequencing on an Illumina HiSeq 2000, was performed by the Fred Hutchinson Cancer Research Center Genomics Shared Resource facility, resulting in 24 bp paired end reads. Base calling was performed using Illumina's Real Time Analysis software v1.13.48.0. Raw FASTQ sequence files were deposited in the NCBI GEO Series GSE69696.
Raw reads (passing Solexa quality test) were mapped to the S. cerevisiae reference genome version SacCer3 (Saccharomyces Genome Database (SGD)/UCSC) using the Burrows-Wheeler Aligner (BWA) [88]. The resulting Binary Sequence Alignment/Map (BAM) files were filtered for proper pairs with a mapping score > = 30 using samtools [89]. Mononucleosomes were identified as paired-end reads with insert sizes between 50 bp and 240 bp using R Bioconductor packages GenomicRanges, rtracklayer, Rsamtools, nucleR, and the UCSC SacCer3 reference genome [56, 90–92]. ChIP reads were compared to the input reads for each strain using the Dynamic Analysis of Nucleosome and Protein Occupancy by Sequencing, version 2 (DANPOS2) function Dpos with background subtraction [93], and the background-subtracted ChIP signal was normalized to the coverage at centromeric regions for each strain, which contains a CENP-ACse4 nucleosome throughout the cell cycle [87], and smoothed using the default DANPOS2 Dpos smoothing parameters [93]. The resulting normalized coverage data was visualized using the Integrated Genomics Viewer (IGV) [94, 95]. Wiggle track format (WIG) files of the normalized coverage for each sample in 10 bp steps are available under NCBI GEO Series GSE69696.
To identify genomic loci enriched for CENP-ACse4, we analyzed the coverage relative to the centromere. Although CENP-ACse4 is constitutively localized to the centromere [96], its coverage at the centromere is under-represented relative to other genomic regions. This effect is likely due to the decreased solubility of the centromere to MNase digestion due to kinetochore protein binding, which makes it possible for other genomic regions to appear enriched above its occupancy at the centromere [54]. We called peaks of CENP-ACse4 occupancy in each strain as any region where the CENP-ACse4 enrichment was above the threshold of the minimum average coverage at any centromere in the 3Flag-CSE4 strain using R Bioconductor packages Genomic Ranges, rtracklayer, and the UCSC SacCer3 reference genome [56, 90–92] and the DANPOS2 function Dtriple to call peaks without any further normalization or smoothing [93]. rDNA ChIP coverage was set to 0 before peak calling due to the high copy number of this region, and this locus was excluded from subsequent computational analyses. Input nucleosome peaks were also called using DANPOS2 [93]. Browser Extensible Data (BED) files of the called peaks for each sample are available at NCBI GEO Series GSE69696.
Genomic regions were annotated using the following strategy: Saccharomyces Genome Database (SGD) annotations of the SacCer3 genome were used to call regions of centromeres, pericentromeres, telomeres, origins of replication, genes, and intergenic regions in that order, such that each base was assigned to only the first overlapping region type. To analyze the percentage of peaks from each strain in each genomic region, 1 bp regions at the center of each CENP-ACse4 peak were overlapped with each region so that each peak was counted only once using R Bioconductor packages Genomic Ranges, rtracklayer, and UCSC SacCer3 [56, 90–92]. The same analysis was performed with CENP-ACse4 peaks that either did or did not overlap WT H2A.ZHtz1 peaks.
We analyzed mean CENP-ACse4 and H2A.ZHtz1 enrichment at the starts and ends of genes as well as centered on NDRs, origins of replication, or centromeres using the DANPOS2 profile function [65, 93]. H2A.ZHtz1 ChIP data is from [4]. H2A.ZHtz1 coverage was calculated from the mapped reads with greater than 90% identity using the DANPOS2 function dpos with the default parameters [93], after lifting over the coordinates to the SacCer3 genome using R Bioconductor packages Genomic Ranges, rtracklayer, and UCSC SacCer3 [56, 90–92]. For the analysis of the transcription start sites (TSS) and transcription termination sites (TTS), the mean CENP-ACse4 or H2A.ZHtz1 coverage in 10 bp windows was calculated for 500 bp upstream and downstream of 3987 transcripts using custom gene files modified to use experimentally derived TSS data instead of open reading frame (ORF) start sites from Nagalakshmi et al, 2008 (GSE11209) [61]. For the analysis of specific groups of genes, the gene file was divided into the specified bins using R Bioconductor packages before using the DANPOS2 function. For NDRs, origins, and centromeres, DANPOS2 profile was run centered on the genomic features using bed files containing either each NDR [65], origin (from SacCer3 annotation) or centromere (from SacCer3 annotation). All plots were made using GraphPad Prism version 6.0 for OSX, GraphPad Software, La Jolla California USA, www.graphpad.com.
Coverage data was visualized using IGV [94, 95]. The fraction of overlap between CENP-ACse4 peaks for each strain and reported H2A.ZHtz1 peaks (coarse grain nucleosome positions) from wild-type (WT) cells [4] was calculated using R Bioconductor packages GenomicRanges and rtracklayer [90–92].
ChIP was performed from 50 ml formaldehyde-crosslinked cultures as described [97]. Chromatin was fragmented by sonication to approximately 500 bp fragments. For HA-Htz1 ChIP, 3HA-Htz1 was immunoprecipitated using anti-HA (12CA5) antibodies (Roche). 3-fold serial dilutions of the Input (1:100, 1:300, 1:900) and ChIP (1:3, 1:9, 1:27) DNA were used for PCR reactions to detect the amount of DNA pulled down with 3HA-Htz1 in each strain [97] and were analyzed on 1.4% agarose gels. Primers for the RDS1 promoter are from [69] and are listed in S4 Table. For CENP-ACse4 ChIP, CENP-ACse4 was immunoprecipitated using anti-Cse4 (235N) antibodies [6] from strains with CENP-ACse4 expression induced from an inducible tetracycline repressed promoter [98] after a 6-hour washout of doxycycline (5ug/ml) in YC-URA media. 3-fold serial dilutions of the Input (1:100, 1:300, 1:900) and ChIP (1x, 1:3, 1:9) DNA were used for PCR reactions at CEN3 [87], the SAP4 promoter and the SLP1 promoter.
Total RNA was extracted from each sample using a hot acid phenol extraction protocol [99], followed by DNAse I treatment (Invitrogen Amplification Grade) phenol:chloroform extraction, and ethanol precipitation. Two or three independent biological replicates of each genotype were used. Total RNA integrity was checked using an Agilent 2200 TapeStation (Agilent Technologies, Inc., Santa Clara, CA) and quantified using a Trinean DropSense96 spectrophotometer (Caliper Life Sciences, Hopkinton, MA). RNA-seq libraries were prepared from total RNA using the TruSeq RNA Sample Prep v2 Kit (Illumina, Inc., San Diego, CA, USA) and a Sciclone NGSx Workstation (PerkinElmer, Waltham, MA, USA). Library size distributions were validated using an Agilent 2200 TapeStation (Agilent Technologies, Santa Clara, CA, USA). Additional library QC, blending of pooled indexed libraries, and cluster optimization were performed using Life Technologies’ Invitrogen Qubit® 2.0 Fluorometer (Life Technologies-Invitrogen, Carlsbad, CA, USA).
RNA-seq libraries were pooled (18-plex) and clustered onto a flow cell lane. Sequencing was performed using an Illumina HiSeq 2500 in “rapid run” mode employing a single-read, 50 base read length (SR50) sequencing strategy. Image analysis and base calling was performed using Illumina's Real Time Analysis v1.18 software, followed by 'demultiplexing' of indexed reads and generation of FASTQ files, using Illumina's bcl2fastq Conversion Software v1.8.4 (http://support.illumina.com/downloads/bcl2fastq_conversion_software_184.html). Reads of low quality were filtered prior to alignment to the reference genome (UCSC SacCer3 assembly) using TopHat v2.1.0[100]. Counts were generated from TopHat alignments for each gene using the Python package HTSeq v0.6.1[101]. Genes with low counts across all samples were removed, prior to identification of differentially expressed genes using the Bioconductor package edgeR v3.12.0[78]. A false discovery rate (FDR) method was employed to correct for multiple testing[102]. Differential expression was defined as |log2 (ratio) | ≥ 0.585 (± 1.5-fold) with the FDR set to 5%. Normalized differential expression data are available as excel files (S3 File), and raw data is available under NCBI GEO Series GSE69696.
The overlap between lists of genes with significantly changed transcription compared to WT yeast in htz1Δ at t = 2 hours, and psh1Δ pGAL-3Flag-CSE4 at t = 2 hours—t = 0 vs. WT t = 2 hours—t = 0 was found using the Whitehead Institute Compare Two Lists tool (http://jura.wi.mit.edu/bioc/tools/compare.php). The number of significantly up or down regulated transcripts overlapped between the genotypes was compared using the hypergeometric distribution (p-value is probability of getting more than the observed number of successes) using the total number of genes in the edgeR result files as the total population, using the GeneProf hypergeometric distribution calculator [103].
|
10.1371/journal.ppat.1000956 | Role of Abl Kinase and the Wave2 Signaling Complex in HIV-1 Entry at a Post-Hemifusion Step | Entry of human immunodeficiency virus type 1 (HIV-1) commences with binding of the envelope glycoprotein (Env) to the receptor CD4, and one of two coreceptors, CXCR4 or CCR5. Env-mediated signaling through coreceptor results in Gαq-mediated Rac activation and actin cytoskeleton rearrangements necessary for fusion. Guanine nucleotide exchange factors (GEFs) activate Rac and regulate its downstream protein effectors. In this study we show that Env-induced Rac activation is mediated by the Rac GEF Tiam-1, which associates with the adaptor protein IRSp53 to link Rac to the Wave2 complex. Rac and the tyrosine kinase Abl then activate the Wave2 complex and promote Arp2/3-dependent actin polymerization. Env-mediated cell-cell fusion, virus-cell fusion and HIV-1 infection are dependent on Tiam-1, Abl, IRSp53, Wave2, and Arp3 as shown by attenuation of fusion and infection in cells expressing siRNA targeted to these signaling components. HIV-1 Env-dependent cell-cell fusion, virus-cell fusion and infection were also inhibited by Abl kinase inhibitors, imatinib, nilotinib, and dasatinib. Treatment of cells with Abl kinase inhibitors did not affect cell viability or surface expression of CD4 and CCR5. Similar results with inhibitors and siRNAs were obtained when Env-dependent cell-cell fusion, virus-cell fusion or infection was measured, and when cell lines or primary cells were the target. Using membrane curving agents and fluorescence microscopy, we showed that inhibition of Abl kinase activity arrests fusion at the hemifusion (lipid mixing) step, suggesting a role for Abl-mediated actin remodeling in pore formation and expansion. These results suggest a potential utility of Abl kinase inhibitors to treat HIV-1 infected patients.
| Patients infected with HIV-1 are currently treated with highly active antiretroviral therapy (HAART) that efficiently suppresses the virus but does not cure the infection. HIV-1 envelope activates Rac-mediated actin cytoskeleton rearrangements in the target cell that promote membrane fusion and entry. We discovered that these rearrangements require activation of the actin polymerization machinery including the tyrosine kinase Abl. We also showed that Abl kinase inhibitors imatinib, nilotinib, and dasatinib, current drug therapies for chronic myeloid leukemia, block HIV-1 entry and infection. These results suggest that these inhibitors might be appropriate drugs for treatment of HIV-1. This strategy of using inhibitors that disable host signaling proteins rather than viral proteins, essential for pathogen survival, may have a general efficacy in developing drugs to combat HIV-1 and other pathogens that acquire drug resistance.
| HIV-1 enters cells in a pH-independent manner by fusion at the plasma membrane or from within endosomes [1]–[3]. HIV-1 entry requires multiple conformational changes in the HIV-1 glycoprotein, and rearrangement of the actin cytoskeleton. These events are triggered by binding of the viral envelope (Env) surface subunit gp120 to the primary receptor CD4 and one of two chemokine coreceptors, CCR5 or CXCR4 [1], [4]. This interaction activates signaling events in the cell, similar to those initiated by natural ligands, such as Ca2+ mobilization, activation of RhoGTPases, and phosphorylation of tyrosine kinases, pyk2, Zap70 and p56lck [4]–[6]. Rho family GTPases, which include the Cdc42, Rac, and Rho subfamilies, are responsible for regulating signaling from membrane receptors to the actin cytoskeleton. The Rho sub-family stimulates myosin based contractility, and drives the formation of stress fibers and focal adhesions. The Rac sub-family stimulates lamellipodia and membrane ruffles, and the Cdc42 subfamily stimulates the formation of filopodia [7]–[9]. We showed that HIV-1 Env binding to target cells induces activation of Rac, stimulates membrane ruffles and lamellipodia, and fusion is inhibited by dominant negative Rac [4], [10]. Furthermore, HIV-1 Env-induced Rac activation depends on activation of Gαq, phospholipase C (PLC), Ca2+ mobilization, protein kinase C (PKC), pyk2 and the GTPase Ras [5]. In the current study we identified the fusion-specific effectors of Rac required for actin cytoskeleton rearrangements that mediate membrane fusion and entry.
Guanine nucleotide exchange factors (GEFs) activate GTPases, facilitating the GDP to GTP switch, and regulate their downstream effects by participating in scaffolding protein complexes, thereby linking GTPase activity to specific effectors [7]–[9]. HIV-1 Env-induced Rac activation is mediated by a specific Rac GEF, either Tiam-1 or Trio [10], [11]. There are multiple effectors of Rac, including serine/threonine kinases, lipid kinases, actin-binding proteins, and adaptor/scaffold molecules [7], [12]. PAK is a downstream effector of Rac and Cdc42 that promotes stabilization of actin networks. Another downstream effector of Rac that nucleates actin polymerization is the Arp2/3 complex. The Arp2/3 complex is activated by the Wave2 complex through IRSp53, an adaptor protein that binds Rac and Wave2 [7]. The Wave2 complex includes Rac-associated protein 1, Nck-associated protein, Abl-interacting protein 2, and heat shock protein C300. Wave2 also associates with Abl, and Abl-mediated phosphorylation of Wave2 promotes its activation [13], [14]. In addition to determining which Rac effectors are critical for membrane fusion, we studied the steps in the membrane fusion process affected by these signaling molecules. These data demonstrate that the Wave2 signaling complex and Abl are required for Env-mediated membrane fusion, entry, and infection and that Abl kinase inhibitors arrest the fusion process at hemifusion.
To determine whether Abl, Trio, or Tiam-1 were required for HIV-1 Env-mediated cell-cell fusion, expression of these proteins was down regulated by RNA interference (RNAi) in U87.CD4.CCR5 cells. Cells expressing siRNA were then mixed with BSC40 cells expressing different Env subtypes and Env-dependent cell-cell fusion was measured. Transfection of target cells with siRNA to Tiam-1 and Abl decreased levels of Env-mediated cell-cell fusion by an average of 79±5% and 74±5% respectively for both HIV-1 R5 and dual-tropic Env-subtypes (Figure 1A, left). There was no significant fusion observed with CCR5 expressing target cells and X4 Env expressing cells with or without siRNA, as expected. The decrease in the levels of fusion correlated well with the decreased steady-state level of Tiam-1, and Abl as detected by immunoblot (Figure 1C). A siRNA directed against Trio had no effect on Env-induced cell-cell fusion despite a 70% reduction in expression of the Trio protein (Figure 1A and C). To determine whether Tiam-1 and Abl are acting exclusively upstream of Rac, a constitutively active Rac mutant, RacV12 was expressed in siRNA transfected cells. Expression of RacV12 in cells expressing siRNA to Tiam-1 reversed the effects of this siRNA on fusion, suggesting that Tiam-1 is functioning upstream of Rac. In contrast, levels of fusion in cells expressing RacV12 and siRNA to Abl were only 53±1% that of cells expressing RacV12 and control siRNA, suggesting a role for Abl upstream and downstream of Rac (Figure 1A, right).
Tiam-1 binds to the Rac and Cdc42 effector IRSp53, enhancing IRSp53 binding to Rac and activation of the Wave2 scaffolding complex [15]. To determine the role of these Rac effectors in Env-mediated membrane fusion, their expression was down regulated by RNAi in U87.CD4.CCR5 cells. The siRNA expressing cells were mixed with Env-expressing cells and cell-cell fusion was measured. Expression of siRNA to IRSp53, Wave2, and Arp3 decreased fusion by 74±5% 77±4% and 78±4%, respectively. The decrease in fusion with these siRNAs was not overcome by expression of RacV12, suggesting that these proteins are required downstream of Rac (Figure 1B). The decrease in levels of fusion correlated with the decrease in protein expression in cells expressing these siRNAs, as seen by immunoblot (Figure 1C), and each siRNA was specific for its target protein (Figure S1A). Treatment of cells stably expressing siRNA resistant Arp3, with Arp3 targeted siRNA had no effect on Env-mediated cell-cell fusion (Figure 1D, E). In contrast, with untransfected cells, and cells stably expressing siRNA resistant Arp3, treatment with siRNA to Rac decreased fusion by 75±5% and 76±3% respectively (Figure 1D). These results show that the effects of RNAi on fusion were specific to inhibition of their target molecules.
To demonstrate the role of Tiam-1, Abl, Rac, IRSp53, Wave2 and Arp3 in virus-cell fusion, their expression was down regulated by RNAi in TZM-BL cells, a derivative of HeLa cells that express CD4, CCR5, and CXCR4, and these cells were then used in a Vpr-Blam assay [16], [17]. In this assay siRNA expressing cells were mixed for 90 min with HIV-1 strains with cores carrying a β-lactamase (BlaM)-Vpr chimera, and pseudotyped with Env from ADA (R5), YU2 (R5) or HXB2 (X4), and fusion was quantified by measuring the cytosolic activity of viral core-associated BlaM [18]. Expression of siRNA to Tiam-1, Abl, Rac, IRSp53, Wave2, and Arp3 decreased virus-cell fusion by an average of 80±4%, 83±1%, 76±4%, 82±6%, 77±3% and 82±6%, respectively, for HIV-1 R5 and X4 Env subtypes (Figure 1F). These results show that activation of the Wave2 signaling complex is required for Env-dependent cell-cell fusion and virus-cell fusion.
Since treatment of cells with Abl targeted siRNA led to a decrease in Env-dependent cell-cell fusion and virus-cell fusion we wanted to determine whether treatment of target cells with commercially available Abl kinase inhibitors, imatinib (IMB), nilotinib (NIL), and dasatinib (DAS), block fusion. IMB is a relatively specific inhibitor of Bcr-Abl, Abl, Arg, and class III receptor tyrosine kinases. NIL is an Abl kinase inhibitor 20–50 fold more potent than IMB at inhibiting Abl. DAS, originally designed as a Src family kinase inhibitor, antagonizes Abl, ephrin and platelet-derived growth factor receptor kinases, and kit. DAS is 300 fold more potent than IMB at inhibiting Abl [19], [20]. To determine the concentrations of these Abl kinase inhibitors that inhibit Abl kinase activity and Env-mediated cell-cell fusion, without non-specific effects, Abl kinase activity, trypan blue analysis, vaccinia virus infection, and T7 polymerase activity were measured in addition to Env-dependent cell-cell fusion (Figure S1B, S2, and data not shown). Treatment of U87.CD4.CCR5 cells with 10 uM IMB, 500 nM NIL, and 300 nM DAS for 1 h prior to and during 3 h incubation with Env-expressing cells decreased Env-mediated cell-cell fusion by an average of 95±2%, 92±5%, and 92±6%, respectively, and Abl kinase activity by 85–87% (Figure 2A and S1B). The CCR5 inhibitor TAK-779, which completely blocks Env-mediated cell-cell fusion and infection of CCR5 expressing cells, was included as a control, and it decreased Env-dependent cell-cell fusion by 99±1% and Env-mediated Abl kinase activation by 98% (Figure 2A and S1B). Similar results were observed with U87.CD4.CXCR4 cells treated with CXCR4 inhibitor AMD3100 and Abl kinase inhibitors and incubated with cells expressing HIV-1 X4 or dual-tropic Env subtypes (Figure S3A). There was no decrease in T7 polymerase activity, or localization of CD4 and CCR5 on the cell surface (Figure S4 and data not shown). Expression of RacV12 in U87.CD4.CCR5 cells treated with IMB, NIL and DAS increased the level of fusion by an average of 3.5-fold (*, P<0.05) compared to treated cells without RacV12, suggesting a role of Abl kinase activity upstream of Rac (Figure 2B).
To determine the effect of these Abl kinase inhibitors on Env-induced Rac activation, U87.CD4.CCR5 cells were treated with inhibitors for 1 h prior to mixing with BSC40 cells expressing no HIV-1 Env, HIV-1 X4 Env, or HIV-1 R5 Env for 30 minutes in the presence of inhibitor. The mismatched X4 Env, that does not induce Rac activation in CCR5 expressing cells, and the CCR5 inhibitor TAK-779, which completely blocks Env-mediated Rac activation in CCR5 expressing cells, were included as controls [5]. Env-induced Rac activation was abolished in cells treated with TAK-779, and all three of the Abl kinase inhibitors (Figure 2C). To validate these effects in a relevant HIV-1 target cell, peripheral blood lymphocytes (PBLs), which express CD4, CCR5 and CXCR4, were used as the target cell population in an Env-dependent cell-cell fusion assay. Treatment of PBLs with IMB, NIL, and DAS decreased fusion by an average of 92±1%, 92±3%, and 99.5±1%, respectively, for HIV-1 R5, dual-tropic and X4 Env subtypes (Figure 2D). The CCR5 inhibitor TAK-779, as expected, completely blocked fusion mediated by R5 Env-expressing cells, inhibited fusion mediated by dual-tropic Env by 56±2%, and had no effect on fusion mediated by X4 Env (Figure 2D).
A long term infection assay was also performed where PBLs were infected with 150 ng of the X4 HIVHXB2 virus after 1 h preincubation with no inhibitor, DMSO, 10 µM IMB, 250 nM NIL, or 75 nM DAS. After 3 h, virus and inhibitors were washed off, inhibitors were added back and the plate was incubated at 37° for 21 days with addition of the inhibitors every 24 h. After 21 days the samples were assayed for cell viability and p24 antigen content. Treatment with IMB, NIL, and DAS decreased cell viability of HIVHXB2 infected cells by 17±4%, 8±5%, and 8±3% respectively and decreased infection by 52%, 51% and 94% compared to DMSO treated cells (Figure S5).
To validate the specificity of these effects, we performed an Env-dependent cell-cell fusion assay with cells stably expressing two different drug resistant Bcr-Abl mutants (Y253F and T315I), or expressing wild type (WT) Bcr-Abl [21]. Expression of the drug resistant Bcr-Abl mutants but not WT Bcr-Abl resulted in recovery of fusion (Figure 2E), demonstrating that the effects of these inhibitors on Env-dependent cell-cell fusion are specific to inhibition of Abl.
To confirm these results using virus particles with relevant levels of virus-associated glycoprotein, we used a virus-dependent cell-cell fusion assay based on the ability of virus particles to bridge two cells and allow transfer of cytoplasmic contents, and we also used the Vpr-BlaM assay described above [4], [10]. For the virus-dependent cell-cell fusion assay we used two populations of U87.CD4.CCR5 cells, one expressing the T7 polymerase and the other expressing the β-galactosidase (β-gal) gene under the T7 promoter. Both populations were incubated with inhibitors for 1 h prior to 3 h incubation with R5 virus HIVYU2. In this assay, controls included untreated and inhibitor treated cells that were not incubated with virus, the CCR5 inhibitor TAK-779, and T-20 which blocks entry by inhibiting the conformational change in HIV-1 gp41 required for fusion [17]. R5 Virus-dependent cell-cell fusion was reduced by an average of 94±3% in cells treated with IMB, DAS, and NIL compared to cells treated with DMSO alone, and treatment with TAK-779 and T-20 completely inhibited fusion (Figure 2F). Treatment of U87.CD4.CXCR4 cells incubated with the X4 virus HIVHXB2 with AMD3100 IMB, NIL, and DAS decreased virus-dependent cell-cell fusion by 88±7%, 98.6±1%, 87±5%, and 96±17%, respectively (Figure S3B).
For the Vpr-BlaM assay, TZM-BL cells were treated with 1 µM AMD3100, 1 µM TAK-779, 10 µM IMB, 500 nM NIL and 150 nM DAS for 1 hr prior to and during the 90 min incubation with HIV-1 Vpr-BlaM viruses expressing R5 and X4-tropic Env. AMD3100 treatment decreased X4-Vpr-BlaM activity by 84±1%, but had no effect on R5-Vpr-BlaM activity. TAK-779 treatment decreased R5-Vpr-BlaM activity by an average of 89±2%, but had no effect on X4-Vpr-BlaM activity, as expected. However, treatment of TZM-BL cells with IMB, NIL, and DAS decreased virus-cell fusion by an average of 81±4%, 89±5%, and 90±1%, respectively, for both HIV-1 R5 and X4 Env subtypes (Figure 2G). These results together with the results of the Env-dependent and virus-cell fusion assay demonstrate that Abl kinase is required for HIV-1 entry mediated by CXCR4 and CCR5.
To determine whether the Wave2 signaling complex and Abl are required exclusively for HIV-1 entry, or virus-induced fusion and infection in general, we examined infection with HIV-1 versus A-MLV Env (A-MLV-ENV-HIV-1) or VSV-G pseudotyped HIV-1 (VSV-G-HIV-1) using the TZM-BL assay. HIV-1 Env induces pH independent virus-cell fusion to facilitate entry, whereas viruses pseudotyped with VSV-G or A-MLV Env induce pH-dependent clathrin mediated endocytosis or caveolin-mediated endocytosis, respectively [22]–[25]. TZM-BL cells, a derivative of HeLa cells that express CD4, CCR5, CXCR4, and luciferase (luc) under the control of the HIV-1 LTR, were pretreated with the 10 µM IMB, 500 nM NIL and 150 nM DAS for 1 h prior to incubation with virus for 3 h, and a subsequent 24 h incubation with inhibitor only [16], [17]. The CCR5 inhibitor TAK-779, the CXCR4 inhibitor AMD3100, and ammonium chloride (NH4Cl) which inhibits endosomal acidification required for VSV-G mediated entry, were included as controls [22], [23], [26]. The top two panels of Figure 3A shows that treatment with IMB, NIL, and DAS decreased infection with R5 HIVYU2 virus and X4 HIVHXB2 virus by an average of 91±7%, 88±4%, and 91±5%, respectively, comparable to the reductions observed with Env-dependent cell-cell fusion, virus-dependent cell-cell fusion and virus-cell fusion (Figure 2). The Abl kinase inhibitors had no effect on infection of TZM-BL cells with A-MLV-ENV-HIV-1 or VSV-G-HIV-1, but treatment of cells with NH4Cl blocked infection with VSV-G-HIV-1 as expected (Figure 3A, bottom two panels). These data show that Abl-kinase inhibitors were able to block HIV-1 Env-mediated fusion specifically and had no effect on infection via pH-dependent clathrin-mediated or caveolin-mediated endocytosis, and post-entry steps were not affected by these inhibitors.
To test the effect of Wave2 complex targeted siRNAs on infection, TZM-BL cells were transfected with 200 nM control siRNA or siRNA directed towards Tiam-1, Trio, Abl, IRSp53, Wave2 and Arp3. These cells were incubated with virus for 3 h, and media alone for 24 h. The decreased levels of HIV-1YU2 and HIV-1HXB2 infection of TZM-BL cells expressing siRNA targeted to Tiam-1, Abl, IRSp53, Wave2, and Arp3 were comparable to levels of Env-mediated cell fusion with U87.CD4.CCR5 cells expressing these siRNAs, whereas siRNA to Trio had no effect (Figure 3B, top two panels). Steady state levels of target proteins in cells expressing targeted siRNAs were decreased to similar levels as in U87 cells (Figure 1C and data not shown). Infection of TZM-BL cells with A-MLV-ENV-HIV-1 or VSV-G-HIV-1 was not affected by expression of the targeted siRNAs, suggesting that Tiam-1, Abl, IRSp53, Wave2, and Arp3 are required for HIV-1 Env-mediated entry and are not necessary for post-fusion steps in the virus life cycle (Figure 3B, bottom 2 panels).
HIV-1 Env-induced fusion, and release of the viral capsid into the cytosol is a multistep process. First, gp120 binds to CD4 inducing conformational changes in gp120, and actin cytoskeletal rearrangements in the target membrane that bring the coreceptor CCR5 or CXCR4 into close proximity with CD4. Next, coreceptor binding to gp120 triggers conformational changes in gp41 to produce a prebundle conformation that inserts into the target cell membrane, allowing lipid mixing or hemifusion, and then pore formation. Additional conformational changes induce formation of the gp41 6-helix-bundle which prevents pore closure and facilitates pore enlargement and full fusion [2], [27], [28]. To determine which step(s) in the membrane fusion process are blocked by the Abl kinase inhibitors, we examined the effect on infection of membrane curving agents. Oleic acid (OLA), chlorpromazine (CPZ), and trifluoperazine (TFP) are lipid analogs that insert into the inner leaflet of the cell membrane. OLA induces negative curvature in the membrane that promotes formation of a hemifusion intermediate (i.e. lipid mixing), but cannot induce pore formation if there is a block at hemifusion. CPZ and TFP are membrane-permeable weak bases that partition into inner leaflets of cell membranes, induce positive curvature, and relieve a block at hemifusion [29]–[32].
To determine the effect of inhibitors and lipid analogs on HIV-1 infection, TZM-BL cells were treated with 1 µM AMD3100, 1 µM TAK-779, 10 µM IMB, 500 nM NIL, and 150 nM DAS for 1 h, prior to and during 1 h incubation with no virus, HIVΔENV, R5 HIVYU2, X4 HIVHXB2, A-MLV-ENV-HIV-1, or VSV-G-HIV-1. After 1 h, cells were treated with CPZ or TFP for 1 min or OLA for 5 min, followed by 2 h incubation with inhibitor and virus, and subsequent 24 h incubation with inhibitor only. Addition of CPZ and TFP to cells treated with Abl kinase inhibitors and infected with HIVYU2 or HIVHXB2 resulted in an 8 fold increase in infection compared to inhibitor treated cells infected in the absence of lipid analogs (Figure 4A), The exogenous cone shaped lipid OLA, which induces negative curvature of the membrane resulting in lipid mixing, had no affect on infection (Figure 4A). TAK-779 mediated inhibition of HIVYU2 infection and AMD3100 mediated inhibition of HIVHXB2 infection was not affected by these lipid analogs. No increase in luc activity was observed with lipid analog treatment of cells infected with HIVΔENV versus no virus, indicating that Env is required to observe an increase in infection (Figure S6A). Treatment of A-MLV-ENV-HIV-1 and VSV-G-HIV-1 infected cells with CPZ and TFP decreased overall infection by 2 fold and had no effect on cells treated with Abl kinase inhibitors, indicating that the increase in HIV-1 infection observed with Abl kinase inhibitor treated cells was specific (Figure 4A, lower panels). CPZ also partially reversed the inhibitory effects of nilotinib as measured by the Vpr-BlaM assay (Figure S6B).
Similar increases in virus-dependent cell-cell fusion were observed when U87.CD4.CCR5 cells were treated with inhibitors and lipid analogs and HIVYU2 mediated fusion was measured after 3 h (Figure S6C). Cells were also incubated with the lipid analogs in the absence of HIVYU2 to account for the effects of these agents on the cells and on T7 polymerase activity. Addition of OLA did not increase fusion in cells treated with any of the inhibitors (Figure S6B). To confirm the results obtained with the Abl kinase inhibitors we incubated TZM-BL cells transfected with Tiam-1, Abl, Rac, IRSp53, Wave2, and Arp3 targeted siRNA, for 1 h with no virus, HIVΔENV, R5 HIVYU2, or X4 HIVHXB2. After 1 h cells were treated with CPZ for 1 min or OLA for 5 min, followed by 2 h incubation with virus, and subsequent 24 h incubation with media alone. As with the Abl kinase inhibitors, treatment of siRNA transfected cells with CPZ increased infection by an average 8.4 fold compared to untreated cells, and OLA had no effect (Figure 4B). These results suggest that inhibition of Tiam-1, Abl, Rac, IRSp53, Wave2 or Arp3 arrests fusion at hemifusion, preventing pore formation, pore enlargement and content mixing.
To confirm that Abl kinase inhibitors cause arrest at hemifusion, we used a modification of a fusion assay described previously [33]. CHO-K1 cells that lack expression of the lipid ganglioside GM1, were engineered to express GFP and the HIV-1ADA (R5) Env protein. U87.CD4.CCR5 cells were used as the target cell, and lipid mixing was detected when GM1, detected by a TRITC-conjugated form of cholera toxin β-subunit (CTX), was transferred from the target cell to CHO-K1-GFP cells. Complete fusion is detected when cells express GM1, GFP, and are multinucleated. Quantification was performed for three independent experiments and the percentage of hemifused GFP+, GM1+ cells with single nuclei and the percentage of multinucleated fully fused cells was enumerated for 68 cells from each condition (Figure 5, S7, and Table S1) There were 83.1±10.9% hemifused cells with IMB-treated cells mixed with HIVADA-expressing CHO-K1 cells (Figure 5), compared to DMSO treated cells with 22.3±4.9% hemifused cells and 75.5±6.2% fully fused cells. With no HIV-1 Env or with the addition of TAK-779 there was little or no hemifusion or full fusion (Figure 5).
To demonstrate the effects of the lipid analog CPZ on HIV-1 Env mediated cell-cell fusion and to observe the effect of CPZ and the Abl kinase inhibitors on A-MLV Env or VSV-G induced cell-cell fusion we treated U87.CD4.CCR5 cells with DMSO, TAK-779, or IMB for 1 hr prior to incubation with CHO-K1 cells expressing no Env, HIVADA, A-MLV Env or VSV-G for 1 hr. After 1 h cells were treated with CPZ for 1 min and OLA for 5 min, then washed and incubated with inhibitor for an additional 2 h prior to fixation and GM1 staining. Incubation of IMB treated cells with HIVADA and CPZ promoted the transition from hemifusion to full fusion as expected (Figure S8). Fusion of A-MLV Env and VSV-G Env expressing cells with U87.CD4.CCR5 cells was unaffected by treatment with IMB or CPZ (Figure S9) and all Env-mediated fusion was unaffected by OLA treatment (data not shown). These results confirm that Abl kinase activity is required at a post-hemifusion step for HIV-1 Env mediated fusion and entry.
Dynamic regulation of the actin cytoskeleton is required for fusion of biological membranes. Multiple reports have demonstrated that actin remodeling is required for HIV-1 mediated fusion and entry [4], [5], [10], [11], [34]–[36]. Some studies showed that treatment of target cells expressing physiologically relevant levels of receptor and coreceptor with the actin filament capping drug cytochalasin D prevented the formation of the gp120-CD4-coreceptor complex [35], [37], [38]. Another more recent study, demonstrated a role for CD4 and coreceptor-mediated filamin-A interactions in receptor clustering that is dependent on RhoA and ROCK mediated phosphorylation of ADF/cofilin [34]. Previous work from our lab with the actin filament stabilizing drug jasplakinolide and the actin monomer sequestering drug latrunculin A (LA) suggested a role for actin remodeling at a post binding step in fusion [4]. To further substantiate the role of actin polymerization in HIV-1 entry, we treated cells with 1 µM LA and 5 µM latrunculin B (LB). Both drugs blocked HIV-1 fusion for multiple cell types, as measured by the Env-dependent cell-cell fusion assay, the virus-dependent cell-cell fusion assay, the virus-cell fusion assay, and infection (Figure S10).
Our previous data demonstrated that the GTPase Rac was activated by HIV-1 Env ligation of CCR5, resulting in membrane ruffles and lamellipodia in the target cell membrane. Inhibition of this activation by dominant negative Rac or by a Rac GEF inhibitor completely abolished Env-dependent cell-cell fusion, virus dependent cell-cell fusion and infection [4], [5], [10], [11]. Our lab went on to show that Env-induced Rac activation is mediated by Gαq and its downstream effectors, including Ras. Other studies showed that Ras promotes Rac activation via direct interaction with Tiam-1, or by phosphatidylinositol 3-kinase (PI3K)-mediated activation of Tiam-1 [39]. Env-dependent Rac activation likely occurs through the first mechanism, since treatment of target cells with PI3K inhibitors had no effect on Env-dependent cell-cell fusion [40].
The nonreceptor tyrosine kinase, Abl, modulates actin upstream and downstream of Rac [41], [42]. In the current study, we used siRNAs and specific inhibitors to show that the activity of Abl kinase is required both upstream and downstream of Rac for Env-induced membrane fusion. Upstream of Rac, Abl phosphorylation of the Ras GEF complex promotes the activity of the Rac GEF Tiam-1, which was shown in the current study to be required for HIV-1 fusion. Downstream of Rac, Abl promotes phosphorylation and activation of Wave2 and its interaction with the Arp2/3 complex, events also demonstrated here to be critical for HIV-1 infection, but not VSV-G or A-MLV Env-mediated infection. These results suggest that these signaling mediators are important for HIV-1 Env mediated entry, are not necessary for pH dependent clathrin or caveolin-mediated endocytosis, and are not required at post-entry steps in the virus life cycle.
There is some conflict in the literature as to the location and mechanism of virus cell fusion. A recent report used microscopic imaging to track HIV-1 Env-pseudotyped MLV virus particles and observed virus-membrane fusion in endosomes [3]. This study also showed that virus-cell fusion and infection were inhibited in the presence of the dynamin inhibitor dynasore (DYN) which is known to block both clathrin and caveolin-mediated endocytosis [3]. The results in our current study suggest that fusion is occurring via a mechanism that is distinct from that of VSV (clathrin-mediated endocytosis) or A-MLV (caveolin-mediated endocytosis). In order to address this conundrum, we treated cells with the dynamin inhibitor DYN, and then used these cells for the Env-dependent cell-cell fusion assay, the virus-dependent cell-cell fusion assay, the virus-cell fusion assay and the TZM-BL infection assay. DYN treatment decreased HIV-1 Env-mediated infection and virus-cell fusion by an average of 58±7% and 50±3% respectively (Figure S10 and Figure S11). However treatment with DYN decreased A-MLV-Env-HIV-1 infection and VSV-G-HIV-1 infection by 75±5% and 89±1% respectively, showing that the affect on HIV-1 Env-mediated infection was not as significant (Figure S11). DYN treatment also decreased Env-dependent cell-cell fusion and virus-dependent cell-cell fusion by 53±8% and 50±10%, respectively which was unexpected since these assays both measure cell-cell plasma membrane fusion. Dynamins are a group of large GTPases that are involved in multiple processes in addition to endocytosis, such as vesicle transport, cytokinesis, organelle division, cell movement and cell signaling [43]–[45]. Therefore, the inhibition observed with the dynamin inhibitor DYN could be due to nonspecific effects on cellular processes. In support of this conclusion, a recent study used the Rev-dependent indicator cell line Rev-CEM to study the effects of DYN on HIV-1 replication and VSV-G-HIV-1 infection [44]. Using this assay they observed a dosage dependent decrease in VSV-G-HIV-1 infection with DYN treatment but did not see any decrease in HIV-1 infection [44]. These results as well as the results in Figure 3 show a clear distinction between HIV-1 Env-mediated entry and VSV-G- and A-MLV-mediated entry.
The current study also showed that the block in fusion caused by inhibition of Tiam-1, Abl, Rac, IRSp53, Wave2 and Arp3 occurs after hemifusion and before cytoplasmic mixing. This conclusion was based on the 1) confocal microscopy demonstration that addition of IMB to the fusion reaction allowed membrane but not cytoplasmic mixing, and 2) observation that lipid analogs that overcome a block at hemifusion overcame inhibition of HIV-1 virus dependent cell fusion, virus-cell fusion and infection caused by Abl kinase inhibitors and siRNA expression. These results support a model whereby HIV-1 Env binding to CCR5 stimulates activation of Gαq resulting in activation of Rac and activated Rac interacts with IRSp53. IRSp53 promotes Rac activation of the Wave2 complex, which is also activated by Abl, and activated Wave2 induces subsequent activation of Arp2/3-mediated actin rearrangements which facilitate pore formation, pore enlargement, and entry of HIV-1.
Many microbial pathogens depend on Abl family kinases to mediate efficient infection of their targeted host, including Shigella flexneri, enteropathogenic Escherichia coli, Helicobacter pylori, Anaplasma phagocytophilum, coxsackievirus, poxvirus, and murine AIDS virus. Abl kinases are involved in pathogen entry, intracellular movement, and exit from target cells; proliferation of target cells; and phosphorylation of microbial effectors. Many of these processes involve reorganization of the target cell actin cytoskeleton and depend on the same signaling pathways as HIV-1 [4], [5], [46]. Discovery of these signaling mediators as fundamental components of microbial pathogenesis provides new targets for therapeutic intervention. The clinical application of IMB, NIL, and DAS, which block deregulated Abl kinases in leukemia patients, demonstrate that inhibition in vivo is possible with manageable side effects [19], [20]. In addition IMB has been shown to be an effective inhibitor of anti-apoptotic pathways induced by HIV-1 in macrophages [47]. Most current antiviral therapies target viral proteins and mutation of the virus leads to therapy resistance. Therefore, using inhibitors that target host signaling proteins essential for HIV-1 entry may be an efficient new strategy for treatment of infected patients.
U87.CD4.CCR5 cells are astroglioma cells expressing CD4, CCR5-GFP or HA-CCR5. U87.CD4.CXCR4 cells are astroglioma cells expressing CD4 and CXCR4-GFP. CHO-K1 cells (ATCC) were grown in F-12K media with 10% serum and other cells maintained as described [48]. pMSCVneo-WT, Y253F, and T315I Bcr-Abl were gifts from Dr. R. Van Etten [21]. The siRNA resistant mutations were generated in Arp3 based on sequences obtained from Santa Cruz Biotechnology, Inc (SCBT, Santa Cruz, CA,), by PCR-mediated mutagenesis of a sub fragment that was sequenced to confirm the presence of mutations before sub cloning into the corresponding cDNA. WT and mutant cDNAs were cloned in pcDNA3.1+zeo for expression by transduction. IMB, NIL, and DAS were from LC Laboratories and were used at 10 uM, 500 nM, and 300 nM respectively unless indicated; CPZ (0.5 mM), TFP (0.3 mM), OA (100 nM), OLA (50 uM) and NH4Cl (50 mM) were from Sigma; TAK-779 (1 uM), and T-20 (10 ug/ml) were from the AIDS Research and Reference Reagent Program. The control siRNA constructs (non-targeting 20–25 nt siRNA designed as a negative control), the siRNA constructs and antibodies used for Western blots were from SCBT [5]. The siRNA constructs were transfected using GeneEraser siRNA Transfection Reagent or Lipofectamine RNAiMAX Transfection Reagent according to the manufacturer's instructions (Stratagene, La Jolla, CA, Invitrogen, Carlsbad, CA).
Wild-type (WT) vaccinia (WR strain) and recombinant vaccinia viruses expressing β-galactosidase (vCB21R), T7 polymerase (vPT7-3), constitutively active Rac GTPase (vRacV12), or HIV-1 Env proteins were described [48]. HIV with R5 YU2 or X4 HXB2 Env in HIVNL4-3 backbone were generated from 293T cells; some were pseudotyped with amphotropic murine leukemia virus (MLV) or vesicular stomatitis virus (VSV) glycoproteins [5]. TZM-BL assays were performed as described [5]. For the BlaM assay pseudoviruses were produced by co-transfecting 293T cells with HIVNL4-3ΔVpr expressing YU2, ADA, or HXB2 Env and BlaM-Vpr expressing pMM310 vector. Transfected 293T cell supernatants were harvested 48 h postlipofection, filtered, and assayed for p24 antigen content by enzyme-linked immunosorbent assay. Viruses were resuspended in culture media, aliquoted and stored at −80°C.
TZM-bl cells were serum starved for 24–36 h then plated (4×104 cells/well) in 96-well plates in complete media overnight. Cells were treated with indicated concentrations of inhibitors for 1 hr prior to and during 90 min incubation with DEAE-dextran (20 µg/ml) alone or DEAE-dextran (20 µg/ml) and 150 ng p24 HIVYU2Vpr-BlaM, HIVADAVpr-BlaM, or HIVHXB2Vpr-BlaM. After 90 min virus and media were aspirated off cells and 100 ul 1X Lysis and Detection Solution was added to wells (LyticBlazer-BODIPY FL, Invitrogen). The plate was incubated at room temperature in the dark overnight. The BlaM activity was quantified using TECAN fluorescence plate reader (Tecan, Switzerland). The extent of virus-cell fusion was measured with excitation centered at 485 nm and emission centered at 535 nm. The green signal for samples incubated with no inhibitors or inhibitors and no virus was subtracted as background from their respective virus treated samples.
TZM-BL cells were serum starved for 12–24 h then plated overnight in complete media in 96 well plate at 2×104 cells per well. Cells were treated for 1 h with indicated concentrations of inhibitors prior to addition of media alone or 150 ng p24 of HIVYU2 HIVHXB2 or VSVG or A-MLV-pseudotyped HIV in the presence of 20 ug/ml DEAE-dextran for 3 h at 37°C. After 3 h cells were washed 3 times with PBS and inhibitors were added in fresh media. Following a 24 h incubation cells were lysed and luciferase (luc) units determined. Infected wells and uninfected wells with inhibitor were compared to wells with no inhibitor. For the TZM-Bl assay with lipid analogs serum starved TZM-BL cells were treated with inhibitors for 1 h, then 150 ng of indicated virus was added for 1 h prior to treatment with CPZ or TFP for 1 min or OLA for 5 min. Cells were washed three times with PBS and virus and inhibitors were added back. After 2 h cells were washed two times with PBS and incubated in inhibitor overnight and luc activities were measured.
PBMCs that were isolated and stimulated as previously described [5]. They were plated at 5×105 cells per well in 96 well plate and were treated with 10 µM IMB, 250 nM NIL, or 75 nM DAS for 1 h prior to addition of 150 ng p24 of HIVHXB2 in the presence of 20 ug/ml DEAE-dextran for 3 h at 37°C. After 3 h cells were washed three times with PBS and incubated in inhibitor for 24 h. Inhibitors were added back at the same concentration every 24 h for three weeks. 100 ul of supernatant was collected every fourth day and all samples were assayed for p24 antigen content by enzyme-linked immunosorbent assay. Two separate plates were set up under the exact same conditions and one plate was used for p24 measurement and the other was incubated with 20 ul cell viability substrate per 100 ul of sample (Promega, Madison, WI).
Envelope-mediated and virus-dependent fusion assays were described. Average fusion compared to untreated control reactions were detected by β-galactosidase activity ± standard deviation [5]. To account for any effect of inhibitors on vaccinia virus infection and/or on T7 polymerase function, vCB21R and vPT7-3 co-infected cells were similarly treated with inhibitors. Concentration curves were performed with all of the inhibitors to determine the concentration that resulted in the maximum decrease in fusion without altering vaccinia virus infection or T7 polymerase activity. Hemifusion assays were performed with 2×106 CHO-K1 cells nucleofected with a GFP expression plasmid, and after 24 h infected with vaccinia virus expressing HIVADA Env or no Env. After 16 h, 4×105 U87.CD4.CCR5.HA cells were added for 3 h, fixed with paraformaldehyde, stained with TRITC-conjugated CTX-555 (EMD), and analyzed on a 510 Meta LSM confocal microscope.
Fusion and infectivity results were compared using a two-tailed t-test. All p values, unless indicated, were <0.03.
|
10.1371/journal.ppat.1000854 | Spatial and Temporal Association of Outbreaks of H5N1 Influenza Virus Infection in Wild Birds with the 0°C Isotherm | Wild bird movements and aggregations following spells of cold weather may have resulted in the spread of highly pathogenic avian influenza virus (HPAIV) H5N1 in Europe during the winter of 2005–2006. Waterbirds are constrained in winter to areas where bodies of water remain unfrozen in order to feed. On the one hand, waterbirds may choose to winter as close as possible to their breeding grounds in order to conserve energy for subsequent reproduction, and may be displaced by cold fronts. On the other hand, waterbirds may choose to winter in regions where adverse weather conditions are rare, and may be slowed by cold fronts upon their journey back to the breeding grounds, which typically starts before the end of winter. Waterbirds will thus tend to aggregate along cold fronts close to the 0°C isotherm during winter, creating conditions that favour HPAIV H5N1 transmission and spread. We determined that the occurrence of outbreaks of HPAIV H5N1 infection in waterbirds in Europe during the winter of 2005–2006 was associated with temperatures close to 0°C. The analysis suggests a significant spatial and temporal association of outbreaks caused by HPAIV H5N1 in wild birds with maximum surface air temperatures of 0°C–2°C on the day of the outbreaks and the two preceding days. At locations where waterbird census data have been collected since 1990, maximum mallard counts occurred when average and maximum surface air temperatures were 0°C and 3°C, respectively. Overall, the abundance of mallards (Anas platyrhynchos) and common pochards (Aythya ferina) was highest when surface air temperatures were lower than the mean temperatures of the region investigated. The analysis implies that waterbird movements associated with cold weather, and congregation of waterbirds along the 0°C isotherm likely contributed to the spread and geographical distribution of outbreaks of HPAIV H5N1 infection in wild birds in Europe during the winter of 2005–2006.
| Highly pathogenic avian influenza virus of the H5N1 subtype emerged more than a decade ago in poultry in South-East Asia. In 2005, it spread outside Asia infecting both poultry and wild birds in the Middle East, Europe and Africa. Both trade of poultry and movements of wild birds were likely implicated in the spread of the infection; however, the ability of wild birds to carry the virus to novel geographical areas is still highly debated and remains obscure. In Europe, the virus mainly infected wild birds, and emergence coincided with a spell of cold weather, which is known to result in massive movements of wild waterbirds. In this paper, we demonstrate that movements of wild waterbirds associated with cold weather contributed to the spread and geographical distribution of outbreaks in Europe during the winter of 2005–2006. Higher density of wild waterbirds on bodies of water that remain unfrozen ahead of the freezing line likely favoured transmission of the virus and resulted in distinctive distribution of outbreaks at locations where surface air temperatures were 0°C–2°C. This has important implications for surveillance, which should target areas where temperatures are close to freezing in winter, especially in poultry-dense regions close to areas where waterfowl aggregate.
| Highly pathogenic avian influenza virus (HPAIV) H5N1 spread from Asia to Europe, the Middle East and Africa during the winter of 2005–2006, with sporadic outbreaks reported in poultry and wild bird populations. Both trade of poultry and poultry products and wild bird movements may have contributed to the unprecedented geographical spread of the virus [1],[2]. In Europe, most cases in wild bird populations occurred in areas where no outbreaks had previously been detected in poultry, indicating that wild birds were likely implicated in the spread of the infection [1]–[3]. Outbreaks of HPAIV H5N1 infection in Europe followed waterbird movements associated with a spell of cold weather [3]–[5], and the route of introduction of HPAIV H5N1 in western European countries was most likely associated with movements of wild birds [1],[2]. However, the role of wild birds in the spread of HPAIV H5N1 remains highly controversial [2],[6], and the association between movements of wild birds associated with cold weather and outbreaks of HPAIV H5N1 infection in Europe during the winter of 2005–2006 remains conjectural.
Experimental infections of naïve waterbirds with HPAIV H5N1 further supported a possible role of wild birds in the spread of the virus. Certain species are highly susceptible to developing clinical signs, may rapidly succumb to severe disease (e.g., swans, Cygnus spp., and European small diving ducks Aythya spp.), and will thus highlight local outbreaks without contributing significantly to long-distance transmission [7]–[9]. In contrast, other species show no visible signs, shed virus for several days, and potentially play a significant role as spreaders of HPAIV H5N1 (e.g., mallard, Anas platyrhynchos) [8]. Migratory and within-winter movements of such species asymptomatically infected with HPAIV H5N1 may result in significant geographical spread of the virus [8]. Outbreaks of HPAIV H5N1 infection in wild birds may then occur and only be detected when spreader species aggregate with highly susceptible species.
The migratory and wintering strategies of European waterbird species vary (as do those of sub-populations of some species) [10]. Waterbird species breeding at most northern latitudes are highly migratory, wintering at southern latitudes, including sub-Saharan Africa. These species include the Eurasian wigeon (Anas Penelope) and the northern pintail (A. acuta). In contrast, waterbird species breeding both at northern latitudes and in more temperate regions of Europe are partially migratory. While sub-populations breeding at most northern latitudes migrate to southern latitudes, sub-populations breeding at more temperate latitudes remain and winter in these regions. Such species include the common teal (A. crecca), the gadwall (A. strepera), the mallard, the common pochard (Aythya ferina) and the tufted duck (A. fuligula). The garganey (Anas querquedula) is one exception, as it is a fully migratory species, breeding in temperate regions of Europe and wintering exclusively in sub-Saharan Africa [10].
Waterbirds winter in areas where bodies of water remain unfrozen, allowing them to forage. Their wintering range is thus determined by the extent of ice cover [11]. For shallow fresh waters that dabbling ducks (Anas spp.) and lighter diving ducks (Aythya spp.) depend upon [11], the occurrence of ice cover is tightly coupled to surface air temperatures [12], and severe cold spells can drive massive within-winter movements of waterbirds [10],[13],[14]. Two wintering strategies are used by migratory waterbirds in response to cold weather. On the one hand, because cold spells can entail high energy expenditures and limit food availability, migratory waterbirds may winter at southern latitudes, where adverse weather conditions are rare [11]. On the other hand, because long migration distance from their wintering grounds to their breeding grounds impairs waterbirds' reproductive success [11], wintering waterbirds may minimize this distance by congregating on unfrozen bodies of water located as close as possible to their breeding grounds [15].
In recent years, sub-populations of migratory waterbird species that typically leave their breeding grounds in northern Europe for more southern wintering grounds have remained sedentary during relatively mild winters [10]. Furthermore, migratory waterbirds, such as the Eurasian wigeon and the common teal, winter in most southern locations, such as Spain and northwestern Africa, only during harsh winters [16]. Together with the large number of partially migratory waterbird species listed above, it suggests that the latter strategy is common in Europe. In addition, migratory waterbirds using the former strategy and wintering in more southern latitudes start their spring migration towards the breeding grounds as early as February [10],[17]. They may be slowed by cold weather, and constrained to stage on unfrozen bodies of water located at the forefront of the freezing front. All these conditions cause waterbirds to congregate during the winter period in suitable habitats along the freezing front where bodies of fresh water are not frozen. This will favour the transmission and spread of HPAIV H5N1 within and between species, resulting in outbreaks in wild bird populations [5]. Such transient aggregations of waterbirds close to populations of domestic poultry will increase the chance of pathogen spill-over into these hosts, which may eventually result in transmission into other domestic animals and humans.
We hypothesize that the initial cases of HPAIV H5N1 infection in wild birds in Europe during the winter of 2005–2006 occurred at locations where surface air temperatures were close to 0°C, because of higher waterbird densities along the freezing front. To test this hypothesis, we analysed the spatio-temporal correlation of surface air temperatures with the initial outbreaks of HPAIV H5N1 infection in wild birds in Europe during the winter of 2005–2006. The spatio-temporal correlation of surface air temperatures with waterbird abundance was also assessed, using publicly available data on maximum mid-January counts of mallards obtained between 1990 and 2003 across Europe (from the French Atlantic coast to eastern Ukraine and from Spain and Greece to Denmark), and mid-January counts of abundant duck species obtained between 1993 and 2008 in eastern France and Switzerland. Census sites in Eastern France and Switzerland were chosen because this region, often hit by cold spells and surface air temperatures close to 0°C, is highly used by wintering mallards and diving ducks [10]. Mallards and a proportion of common pochards experimentally infected with HPAIV H5N1 remained clinically healthy yet shed high viral titers, potentially spreading the virus over long distances [8]. Analyzing the relationship between the abundance of these species and surface air temperatures at these census sites may thus provide useful indications for higher waterbird densities close to the 0°C isotherm.
A total of 52 locations in 15 European countries experienced initial outbreaks of HPAIV H5N1 infection in wild bird populations (Table S1). Initial outbreaks of HPAIV H5N1 infection in wild bird populations were defined as one or more cases of HPAIV H5N1 infection in wild birds in countries where the first reported case(s) occurred in wild birds, at locations with no outbreak in poultry or wild birds within ∼120 km in the preceding month. Only countries with first cases of HPAIV H5N1 infection in wild birds were selected to exclude cases resulting from introduction by and spill-back transmission from infected poultry nearby. Likewise, cases were selected at locations with no outbreak in poultry or wild birds within ∼120 km in the preceding month to exclude cases resulting from local spread independent of within-winter movement of wild birds. A perimeter of ∼120 km was used based on the typical range of dispersive movements of wild waterfowl of at least 100 km [10]. The date of initial outbreaks of HPAIV H5N1 infection in wild birds used in the analyses is the date infected birds were found moribund or dead.
Initial outbreaks of HPAIV H5N1 infection in wild bird populations occurred most often when surface air temperatures were close to 0°C (Fig. 1). Of the 52 locations, 17 to 19 (33%–37%) experienced maximum surface air temperatures of 0°C–2°C on the day wild birds infected with HPAIV H5N1 were found, and on the two preceding days (day −2 to day 0; Fig. 1A). In contrast, only 1 (2%) to 8 (15%) locations experienced maximum surface air temperatures within any other 2 degree-range between day −2 and day 0 (ANOVA test, F = 9.5, p<0.0001; Fig. 2). Furthermore, the number of locations experiencing maximum surface air temperatures of 0°C–2°C peaked to 17 to 19 on day −2 to day 0, while only 8 (15%) to 14 (27%) locations experienced maximum surface air temperatures of 0–2°C on day −7 to day −3 and day +1 to day +7 (ANOVA test, F = 15.1, p = 0.0005; Fig. 2). A similar but less pronounced trend was observed for minimum and average surface air temperatures (Fig. 1B and 1C).
Areas that did not experience surface air temperatures close to 0°C may have not reported HPAIV H5N1 outbreaks in wild birds due to reporting bias, for example associated with regional differences in the ornithological and bird-watching community, or national differences in resources allocated to surveillance programs. Because the intensity of surveillance of mortality events in wild birds may be linked to human population density in a region, we determined whether the reporting of initial outbreaks of HPAIV H5N1 infection in wild birds was biased towards more densely populated regions in Europe. The statistical distribution of population density of the first administrative regions that reported initial HPAIV H5N1 outbreaks in wild birds was not significantly different from that of regions across Europe (t = 0.58, p = 0.8; Fig. S1A). Because the intensity of surveillance of mortality events in wild birds may also be linked to a country's resources, we also determined whether the reporting of initial outbreaks of HPAIV H5N1 infection in wild birds was biased towards richer countries in Europe. Likewise, the statistical distribution of gross domestic product (GDP) per inhabitant of countries that reported initial HPAIV H5N1 outbreaks in wild birds was not statistical different from that in Europe (t = 0.25, p = 0.7; Fig. S1B).
The highest counts of mallards obtained at each of 93 locations across Europe between 1990 and 2003 were typically recorded when average and maximum mid-January surface air temperatures were 0°C and 3°C, respectively. Average and maximum mid-January surface air temperatures (but not mid-January minimum surface air temperatures) at these locations on the years of maximum mallard counts were normally distributed, around a mean of 0.0°C and 2.9°C, with a standard deviation of 4.2°C and 3.6°C, respectively (Shapiro-Wilk normality test, W = 0.98 and W = 0.98; p = 0.2, and p = 0.3, respectively; Fig. 3A). In contrast, average and maximum mid-January surface air temperatures at these locations were not normally distributed when averaged over the entire period 1990–2003 for each location (Shapiro-Wilk normality test, W = 0.94 and W = 0.96; p = 0.0004 and p = 0.01, respectively; Fig. 3B).
Mallards, common pochards (Aythya ferina), and tufted ducks (A. fuligula) were the most abundant waterbird species counted at 25 locations in Rhône-Alpes (eastern France) between 1993 and 2008, and at 18 locations across Switzerland between 2002 and 2007. Average mid-January temperatures at these locations had a mean of 0.2°C (SD = 3.4°C). Minimum mid-January temperatures at these locations had a mean of −3.2°C (SD = 4.4°C). Maximum mid-January temperatures at these locations had a mean of 3.3°C (SD = 2.9°C).
Standardized mid-January counts of mallards and common pochards (but not tufted ducks) were negatively correlated with standardized minimum, average, and maximum mid-January surface air temperatures (Fig. 4), based on generalized least square linear and non-linear models with an autocorrelation structure of first order. These models determine the linear or non-linear equations of the form ax+b and ax2+bx+c, respectively, that best fit the data (and thus minimize the distance between points from the dataset and points generated by the models). An autocorrelation structure of first order was included because of the spatial correlation that exists between the data points (surface air temperatures at one location are not independent of those at another location within the region investigated). The goodness-of-fit of linear models were slightly higher than that of non-linear models (Table 1). This analysis provides evidence for an overall negative relationship between mallard and common pochard abundance and surface air temperatures. However, because of the lack of temporal precision (the exact mid-January count date was unknown and mid-January temperatures were averaged over a period of 10 days around January 15th), more detailed conclusion on the relationship between duck abundance and ice cover likelihood at the time of the count cannot be drawn. Interestingly, higher counts of mallards and common pochards were typically recorded when surface air temperatures were slightly below the mean temperatures of the region, and their abundance tended to decrease as surface air temperatures further decreased (Fig. 4).
In the present paper, we show an association between the timing and locations of initial outbreaks of HPAIV H5N1 infection in wild birds in Europe during the winter of 2005–2006 and the 0°C isotherm. Initial outbreaks of HPAIV H5N1 infection in wild birds occurred significantly more often at locations where maximum surface air temperatures were between 0°C and 2°C, on the day of the outbreaks and the two preceding days. When visualized on a dynamical map showing daily movements of maximum surface air temperature isotherms between January and March 2006 in Europe, most initial outbreaks of HPAIV H5N1 in wild birds occur within an area delineated by the 0°C and 2°C isotherms, and appear to closely follow the surge and ebb of the cold front (Video S1). Because surveillance of avian influenza in wild birds has been up-scaled in Member States of the European Union since 2002, and in most of Europe following the introduction of HPAIV H5N1 in 2005 [18], it is unlikely that many outbreaks went unnoticed outside the geographical range of the 0°C isotherm during the winter of 2005–2006. Furthermore, the absence of significant difference between human population density and GDP per inhabitant in regions or countries where outbreaks were reported and those across Europe implies that surveillance efforts were not significantly biased towards more densely populated regions or richer countries. Therefore, the geographical distribution of reported HPAIV H5N1 outbreaks in wild birds in Europe along the 0°C isotherm is likely representative of a real and important phenomenon that should be used to refocus surveillance for future outbreaks.
Waterbirds congregate in winter on unfrozen bodies of fresh water where they can forage [10],[11], and higher waterbird densities may occur along the freezing front. On the one hand, waterbirds may choose to keep as close as possible to their breeding grounds to maximize their reproductive success, and may be displaced by cold fronts; on the other hand, waterbirds may choose to migrate to and winter in regions with rare adverse weather conditions, and may be slowed by cold fronts upon their journey back to the breeding grounds, which can start as early as February [10],[11],[17]. Maximum counts of mallards across continental Europe typically occurred when average and maximum surface air temperatures were 0°C and 3°C at the census location, respectively. This suggests that temperatures slightly above 0°C are one factor favouring maximum numbers of mallards at wintering locations, notably in central and northern Europe, where such winter temperatures are frequently recorded. Overall, mallard and common pochard counts at 43 locations across eastern France and Switzerland were negatively correlated with surface air temperatures. Higher numbers of mallards and common pochards occurred when mid-January surface air temperatures were lower than mean mid-January temperatures of the region. These results strongly imply the existence of a relationship between duck abundance and surface air temperatures, despite the coarse quality of mid-January counts and mid-January surface air temperatures. Nevertheless, a number of confounding factors, such as the abundance of food resources, waterbird breeding success in the preceding spring and summer, or changes in wetland management and disturbance, likely impact on yearly duck counts in Europe and could not be accounted for due to the nature of the data. Detailed studies of the correlation of long-term daily counts of waterbirds in multiple sites across Europe in winter with daily surface air temperatures are needed to further assess whether the abundance of wintering ducks is higher when surface air temperatures are close to 0°C. Although limited data are available on the association between waterbird abundance and surface air temperature, it is interesting to note that, for instance, lower mean temperatures in the Dombes region in France in the winter of 2005–2006 were associated with unusual early arrival of common pochards [4]. Likewise, lasting cold weather was associated with higher waterbird densities in eastern and southern Germany [5]. On the other hand however, the abundance of tufted ducks was not correlated with surface air temperatures. Ice cover impacts on the distribution and abundance of dabbling and lighter diving ducks in winter, because they require shallow waters to forage [11]. While common pochards feed on both animal prey and plants, tufted ducks feed predominantly on molluscs and thus forage typically in deeper waters than common pochards [19], possibly explaining their lower sensitivity to surface air temperatures, as found in this study.
Maximum surface air temperature close to 0°C, yet on the positive side of the isotherm, may provide wintering conditions leading to maximum congregation of waterbirds, notably Anatidae, and so favour influenza virus transmission. Cold fronts and freezing temperatures are typically associated with anticyclonic conditions and (north-)easterly winds in Europe [20]. Maximum surface air temperatures above 0°C, despite minimum surface air temperatures below 0°C, thus may determine the absence of ice cover on inland bodies of fresh water located (south-)westward from the cold front. Such temperature conditions were prevalent at locations with initial outbreaks of HPAIV H5N1 infection in wild birds in Europe during the winter of 2005–2006. Therefore, congregation of waterbirds asymptomatically infected with HPAIV H5N1 and highly susceptible waterbird species in suitable habitats along the freezing front likely promoted the onset of outbreaks during the winter of 2005–2006 in Europe. Accordingly, densities of mute swans (Cygnus olor) and Anatidae were three to six times higher in ponds of the Dombes region where outbreaks of HPAIV H5N1 infection occurred, than in unaffected ponds [4]. Likewise, in eastern and southern Germany, where outbreaks of HPAIV H5N1 infection occurred, high waterbird densities were recorded on bodies of fresh water that remained un- or partially frozen. In contrast, waterbird densities were lower in western Germany, which experienced milder weather conditions, and no cases of HPAIV H5N1 infection in wild birds [5].
Higher densities of waterbirds along the freezing front likely favoured increased transmission of HPAIV H5N1. Excretion of HPAIV H5N1 in experimentally infected waterbirds starts as soon as one day following inoculation and peaks between 1 and 3 days following inoculation, typically lasting 5 days [7]–[9]. While the dabbling ducks tested did not develop clinical disease at all, diving ducks, as well as mute and whooper swans, can develop clinical disease as soon as 2 days, and die as soon as 4 days following inoculation [7],[8], although longer incubation period and longer time to death have been described in swans [9]. Clinical disease and death may occur early during the course of infection in free-ranging waterbirds due to contributing factors, such as poor nutritional and health status or harsh environmental conditions. Thus, there may be a short time-lag of 2 to 4 days between transmission of HPAIV H5N1 to uninfected birds, and death and report of HPAIV H5N1 outbreaks in wild waterbirds. Maximum surface air temperatures of 0°C–2°C were most frequently reported at locations of HPAIV H5N1 outbreaks up to two days before the day birds were found dead, which likely accounted for the time-lag between aggregation of wild waterbirds, transmission of HPAIV H5N1 and report of morbidity or mortality. Furthermore, because the peak of the infectious period is typically short, aggregation of waterbirds during two days along the 0°C isotherm may be sufficient to result in sustained HPAIV H5N1 transmission and detectable outbreaks.
Alternatively, and potentially concomitantly, maximum surface air temperatures close to 0°C may favour the persistence of HPAIV H5N1 in the environment and enhance environmental transmission of the virus independently of waterbird density. Avian influenza viruses (AIV) have been experimentally shown to remain infective for several months in water at low temperatures (below 17°C) and low salinity levels (fresh- or brackish water) [21],[22]. Environmental transmission of LPAIV is increasingly recognized as essential in maintaining LPAIV locally and from year to year [23]–[25]. However, little is known on the role environmental transmission had on HPAIV H5N1 dynamics in Europe, and several arguments in support of environmental transmission of HPAIV H5N1 can be raised. First, HPAIV H5N1 belonging to the lineage isolated in wild waterbirds in Europe remained infective for 158 days in fresh water at 17°C, and for 26 days at 28°C [26]. Second, although HPAIV H5N1 are mainly excreted from the respiratory tract of infected waterbirds, potentially favouring direct density-dependent transmission [8], this does not preclude contamination of lake water by respiratory excretions or infected carcasses, allowing environmental transmission of the virus. Third, assuming persistence patterns similar to those of LPAIV [21], slower loss of infectivity at lower temperatures may have contributed to the geographical distribution of HPAIV H5N1 outbreaks along the freezing front. However, persistence of more than 4 months in fresh water at 17°C strongly suggests that environmental transmission of HPAIV H5N1 would occur even in warmer water away from the 0°C isotherm. Furthermore, freezing and thawing are known to substantially decrease AIV infectivity by up to 10 fold, despite having little effect on AIV RNA [27]. Thus, LPAIV RNA has been recovered in ice from Siberian lakes, yet no infectious virus could be isolated [28]. Therefore, although environmental transmission cannot be ruled out, it would most likely result in a more uniform distribution of HPAIV H5N1 outbreaks where bodies of water remained unfrozen, and thus is unlikely to account alone for the geographical distribution of HPAIV H5N1 outbreaks at locations where maximum surface air temperatures were close to 0°C.
Certain waterbird species do not develop clinical disease upon experimental infection with HPAIV H5N1, and may play a crucial role as reservoirs or spreaders of infection [8]. Conversely, asymptomatic infection of wild waterbirds with low pathogenic avian influenza viruses is speculated to result in costs that may hinder the ability of infected birds to fly over long distances [29],[30]. Also, because of the relatively short infectious period of avian influenza virus infection in waterbirds, infected waterbirds may not disperse avian influenza viruses over long distances [30]. Therefore, the role of migratory birds in the long-distance spread of HPAIV H5N1 outside Asia is still highly debated [2],[6]. However, distance, duration and efforts associated with within-winter movements of wild waterbirds are smaller than that associated with spring and autumn migratory flights, and within-winter movements can frequently be undertaken by waterbirds [10],[13],[14]. Although the poultry trade may have introduced HPAIV H5N1 into Russia, the Middle East or eastern Europe in fall and early winter of 2005–2006 [1],[3], within-winter waterbird movements associated with movements of the 0°C isotherm, following the surge and ebb of a cold spell that originated in the Black Sea area [3],[5], were likely undertaken by waterbirds asymptomatically infected with HPAIV H5N1. In conclusion, waterbird movements associated with cold weather and congregation of waterbirds along the 0°C isotherm likely contributed to the spread and geographical distribution of outbreaks of HPAIV H5N1 infection in wild birds in Europe during the winter of 2005–2006.
Movements of cold weather fronts, and in particular the 0°C isotherm of maximum surface air temperature, can readily be anticipated by operational weather forecasts. Therefore, increased active surveillance of HPAIV H5N1 infection in waterbirds should target populations occurring in areas where maximum surface air temperatures are close to freezing temperatures, i.e., along the positive side of the 0°C isotherm. Such targeted surveillance should pay special attention to poultry-dense areas, as this may increase the chance of detecting early the presence of HPAIV H5N1 in both wild and domestic bird populations in Europe during winter.
All statistical analyses were performed in the R language [35]. Statistical differences are considered significant when p<0.5.
|
10.1371/journal.pntd.0004137 | Infectiousness of Sylvatic and Synanthropic Small Rodents Implicates a Multi-host Reservoir of Leishmania (Viannia) braziliensis | The possibility that a multi-host wildlife reservoir is responsible for maintaining transmission of Leishmania (Viannia) braziliensis causing human cutaneous and mucocutaneous leishmaniasis is tested by comparative analysis of infection progression and infectiousness to sandflies in rodent host species previously shown to have high natural infection prevalences in both sylvatic or/and peridomestic habitats in close proximity to humans in northeast Brazil.
The clinical and parasitological outcomes, and infectiousness to sandflies, were observed in 54 colonized animals of three species (18 Necromys lasiurus, 18 Nectomys squamipes and 18 Rattus rattus) experimentally infected with high (5.5×106/ml) or low (2.8×105/ml) dose L. (V.) braziliensis (MBOL/BR/2000/CPqAM95) inoculum. Clinical signs of infection were monitored daily. Whole animal xenodiagnoses were performed 6 months post inoculation using Lutzomyia longipalpis originating from flies caught in Passira, Pernambuco, after this parasite evaluation was performed at necropsy. Heterogeneities in Leishmania parasite loads were measured by quantitative PCR in ear skin, liver and spleen tissues.
All three rodent species proved to establish infection characterized by short-term self-resolving skin lesions, located on ears and tail but not on footpads (one site of inoculation), and variable parasite loads detected in all three tissues with maximum burdens of 8.1×103 (skin), 2.8×103 (spleen), and 8.9×102 (liver). All three host species, 18/18 N. lasiurus, 10/18 N. squamipes and 6/18 R. rattus, also proved infectious to sandflies in cross-sectional study. R. rattus supported significantly lower tissue parasite loads compared to those in N. lasiurus and N. squamipes, and N. lasiurus appeared to be more infectious, on average, than either N. squamipes or R. rattus.
A multi-host reservoir of cutaneous leishmaniasis is indicated in this region of Brazil, though with apparent differences in the competence between the rodent species. The results provide preliminary insights into links between sylvatic and peri-domestic transmission cycles associated with overlaps in the rodent species’ ecological niches.
| Across the Americas, Leishmania (V.) braziliensis is the predominant Leishmania species causing cutaneous and mucocutaneous leishmaniasis in humans. Transmitted by Phlebotomine sandflies, questions remain about the epidemiological contributions of the numerous zoonotic and more domestic host species. Domestication of the principal vector and human infection patterns suggest that human infection risk is predominantly peridomestic, whereas control strategies will be more complex if there is a link to a wildlife transmission cycle. Almost no studies have been conducted on the transmission potential of natural hosts of L. (V.) braziliensis. This study evaluates the infectiousness of experimentally infected natural rodent host species, that in different ecological habitats are proposed to act as a single or a multi-host reservoir. Clinical and parasitological development, and the ability to transmit Leishmania to sandflies, was observed under experimental conditions using a single strain of L. (V.) braziliensis isolated from the wild rat, Necromys lasiurus. Xenodiagnoses were performed with laboratory bred sand fly females established from a local population of Lutzomyia longipalpis. All three rodent species developed disseminated subclinical parasitological infections, but clinical signs (lesions) were transient and self-resolving. N. squamipes, N. lasiurus and R. rattus were all infectious when asymptomatic, though their competence in transmission potential appears to differ with R. rattus showing signs of lower susceptibility. These results provide further evidence that a multi-host reservoir is responsible for maintaining transmission with a bridge between infectious sylvatic and peridomestic rodent populations.
| Transmission of zoonotic pathogens may involve one, or typically more than one, reservoir host. Compared to pathogens with single reservoir hosts, those involving multi-host communities usually show reduced transmission rates through a process of zooprophylaxis or “dilution effect” due to heterogeneities in their competence to support pathogen replication and in their infectious duration, resulting in reduced pathogen-host contact, or vector-infectious host contact in the case of vector-borne pathogens [1, 2, 3]. The less common case in nature is that multi-host communities are more homogeneous as competent reservoirs, such that transmission is amplified, otherwise known as zoopotentiation; complexities in these scenarios are discussed elsewhere [2, 4]. Quantification of host heterogeneity has led to a better understanding of transmission dynamics [1, 5, 6], and improved mathematical predictions of transmission hotspots towards development of disease surveillance and control strategies [7, 8].
Zoonotic cutaneous leishmaniasis (ZCL) is a prime example where infection has been detected in multiple host species in different habitats, but where the competence of hosts and sand fly vectors in putative transmission cycles, are not well defined. Across the Americas, the predominant aetiological agent of ZCL is L. (Viannia.) braziliensis causing, in humans, small simple self-healing cutaneous lesions to disfiguring and destructive lesions known as espundia or mucosal leishmaniasis that can result in irreversible disfigurement of the upper nasal tract. In Brazil the dominant parasite causing cutaneous leishmaniasis is L. (V.) braziliensis and there are approximately 26,000 reported new human cases per year but estimates of annual incidences range from 72,800 to 119,600 [9]. L. (V.) braziliensis infections have been identified in sylvatic vectors and small mammals in the Atlantic rainforest biome [10, 11], however transmission has expanded into anthropogenic habitats where infection is observed in more synanthropic and peridomestic species including rodents, marsupials, domestic dogs and equids [11, 12, 13] that may or may not be epidemiologically significant for transmission to humans. Human transmission is predominantly peridomestic as indicated by case age distributions e.g. not limited to adults, forest or plantation workers [14], and the known vector Lu. whitmani, is captured in large numbers in animals sheds [11, 15]. Control of human ZCL currently relies on human case detection and treatment, however since humans are not thought to be particularly infectious, interrupting transmission necessarily relies on reservoir or/and vector control. There are no comparative transmission studies of L. (V.) braziliensis in small mammal that are indicated as being natural hosts.
By experimental infection, this study aims to compare the reservoir competence of wild and synanthropic rodents previously implicated as reservoirs of L. (V.) braziliensis in northeast Brazil 25. These experiments provide the initial data towards defining their individual vs collective susceptibility to infection, ability to support parasite replication, and their infectiousness to phlebotomine sand flies for onward transmission.
Three rodent species (Necromys lasiurus syn. Bolomys lasiurus (Lund, 1840), Nectomys squamipes Brants, 1827 and Rattus rattus Linnaeus, 1758) were selected for comparative study as potentially important reservoirs based on demonstrating high prevalences of natural infection or/and high population densities, in previous field studies in endemic ZCL foci in Pernambuco, northeast Brazil [10, 11]. In this region, the reported incidence is 18.5 human cases per 100,000 inhabitants [16]. N. lasiurus and N. squamipes are usually associated with Atlantic rain forest and scrub/plantation habitats, whereas R. rattus is predominantly captured inside houses and animal sheds [11].
Rodent colonies were established at Fiocruz-PE from adult animals live-captured in a well studied foci, Raiz de Dentro, in the Municipality of Amaraji, Pernambuco, northeast Brazil (8° 23’S, 35° 27’W), and identified based on morphological and morphometric characters [17]. A total of 60 F1 generation 35–45 day old animals (20 R. rattus, 20 N. lasiurus and 20 N. squamipes), were selected and divided into two groups of 10 animals per species for experimental infection with a high or low dose inoculation of L. (V.) braziliensis, as described below. A control group of two hamsters (Mesocricetus auratus) per group was included to confirm the infectivity of the inoculation cultures.
The L. (V.) braziliensis strain (MBOL/BR/2000/CPqAM95) used throughout our experiments was isolated on 08/06/2000 from a N. lasiurus captured in the Amaraji region, and identified as belonging to zymodeme IOC-Z74, variant 4 and serodeme 1 [18]. This same zymodeme has been isolated from man in the endemic area of Amaraji It was cryopreserved after isolation and then ampoules were thawed and contents passaged from 3 to 4 times before inoculation. In culture and hamsters this strain behaves similarly to other strains of this parasite isolated from other wild rodents and man. Promastigotes were grown in biphasic medium of Blood Agar Base (Difco 45) [19] and Schneider's medium enriched with 20% fetal bovine serum and maintained at 26°C. For experimental infections, the inoculum was prepared from a 7 day old log phase culture containing 2.8×105/ml (low dose) or 5.5×106/ml (high dose) final concentrations according to the doses used for infecting mice with different Leishmania species[20]. 7 day old cultures are the inter phase between the log and stationary phases and are composed of infectious metacyclic promastigotes and non-infectious procyclic promastigotes. Twenty replicate animals of each species, N. lasiurus, N. squamipes and R. rattus, were inoculated experimentally with either high (n = 10) or low dose (n = 10) L. (V.) braziliensis Each animal was inoculated in the following sites: left hind paw (0.025 ml), left ear (0.025 mL), and intraperitoneal space (0.05 mL) following a protocol used for experimentally infecting Proechimys[21]. Control groups of hamsters (Mesocricetus auratus) were inoculated under the same conditions. Animals were followed up for 180 days post inoculation when submitted to xenodiagnoses, and then sacrificed as described below.
After experimental infection, animals were monitored daily to detect any clinical changes including lesions on the inoculation site, hair loss, or splenomegaly.
Xenodiagnosis was performed on 18 of each rodent species six months after inoculation using 7-day old sand flies, from the first generation Lu. longipalpis captured in a well studied foci, in the Municipality of Passira, Pernambuco, northeast Brazil (7° 56’S, 35° 35’W). The mating song of this population has been determined as a burst type, being very similar to Camara and Bacarena populations of Pará State [22]. Burst song populations are principally coastal and all have the cembrene-1 pheromone[23]. The animals were anesthetized with ketamine hydrochloride at 10% and placed in cages into which female sand flies were released and allowed to feed for about 40 minutes in the presence of a similar number of male sand flies in order to induce feeding and copulation. Blood-fed females were then transferred to plastic pots that were stored in boxes with light filter protection and kept under controlled laboratory conditions until the seventh day when they were dissected to detect promastigote forms under optical microscopy.
After conclusion of xenodiagnosis, the animals were euthanized with a CO2 inhalation process. Fragments of approximately 50mg of ear skin, spleen and liver were collected from each euthanized animals, and Leishmania parasite DNA quantified by quantitative PCR (qPCR).
DNA was extracted from tissues using DNeasy Blood & Tissue kit (Qiagen) according to the manufacturer’s protocol. The initial molecular detection protocol consisted of a nested PCR assay using two pairs of SSU rDNA (Small Subunit Ribosomal gene) derived oligonucleotides. The first PCR used SSU rDNA primers [24] that amplify a conserved region of all trypanosomatids (S12: 5’-GGTTGATTCCGTCAACGGAC-3’ and S4: 5’-GATCCAGCTGCAGGTTCACC-3’); internal oligonucleotides PCR products were analyzed by electrophoresis in agarose gel. The second reaction was a real time PCR (qPCR) to quantify the parasite load [25] using primers that amplify a common region of the Leishmania (Viannia) subgenus (S17: 5’-CCAAGCTGCCCAGTAGAAT-3’ and S18: 5’-TCGGGCGGATAAAACACC-3’). The quantification protocol consisted of a real time SYBR-Green PCR; tissue parasite loads were standardized as number of SSU rDNA copies per host glyceraldehyde-3-phosphate dehydrogenase (GAPDH) copy number. The PCR conditions were optimized to generate a single melting curve of the product.
Established experimental infection was defined as the presence of one or more condition: development of skin lesions associated with symptomatic rodent ZCL, detection of splenomegaly at necropsy, qPCR detection of Leishmania in tissue samples (ear skin, liver, spleen), or infectiousness to sand flies. For statistical analyses, Leishmania loads were log10+1 transformed and tested using general linearised Poisson models (negative binomial over-dispersion coefficient α<0.088, χ2<0.94, P>0.281 in each case). The relationships between infectiousness (proportion of sandflies infected) or presence/absence of skin lesions against independent variables were analysed using logistic regression weighted by sample sizes. Depending on the outcome of interest, multivariate analysis adjusted for covariates including inoculum size (high dose or low dose), skin tissue log10 parasite load, inoculum size × skin log10 load interaction term, times to lesion onset and lesion recovery, and rodent species. All analyses were carried out using Stata v.13.1 software (Stata Corporation, College Station, Texas, USA).
Approvals to conduct this study and to capture wild animals to establish laboratory colonies were obtained from the Animal Research Ethics Committee of Oswaldo Cruz Foundation, Rio de Janeiro (Protocol No. L-056/05), and endorsed by the Brazilian Institute of Environment (IBAMA License No. 12.749–1). All the experimental animals were handled in accordance with the recommended guidelines and safety measures;. captured animals, experimental animals and the colonies were all kept in quarantine that involved microbiological testing, safety barriers with micro- and macro-isolators, and under strict hygiene conditions [26, 27] following security standards (International Organization for Standardization—IS0/15189).
Within 3 months of being inoculated and before sampling four high dose rodents (2 N. lasiurus, 2 N. squamipes) and one low dose R. rattus died, thus final follow-up sample sizes were therefore 18 N. lasiurus, 18 N. squamipes and 19 R. rattus (55 animals in total).
Infections were confirmed in 18/18 N. lasiurus, 18/18 N. squamipes and 9/19 R. rattus by molecular methods and xenodiagnosis (Tables 1 & 2). The two experimental inoculum doses appeared similar in successfully establishing rodent infection (26/29 high dose vs 19/26 low dose animals) (χ2 = 2.55 P = 0.614), though some specific differences were observed as described below. All control hamsters developed infection, that included lesions at the inoculation site and all tissues were positive by nested PCR that confirmed the infectiousness of the high and low dose innoculum.
One or more skin lesions associated with infection were observed in 14/55 animals, where 9 and 4 of the high dose animals respectively developed 1 and 2 lesions located on the ear (1 N. lasiurus, 3 N. squamipes, 2 R.rattus), ear and tail base (4 N. lasiurus) tail base (3 N. lasiurus) and one low dose animal developed a tail base lesion (1 R.rattus); no lesions were observed on the footpads at the site of experimental inoculation in any animal. The higher experimental dose induced a higher proportion of animals to present skin lesions (13/26 50%) compared to the low dose group (1/29 3.5%) (χ2 = 9.44 P = 0.002) (Table 1 and Fig 1). Splenomegaly at necropsy was rare for both doses (Table 1). At the time that the xenodiagnoses were performed no animals had visible lesions.
In high dose animals, average times to lesion development post inoculation were 38 (95% CI: 33.9–42.6), 48 (33.3–62.0) and 51 (51.0–51.0) days respectively for N. lasiurus, N. squamipes and R. rattus. Time to lesion onset was statistically shorter in N. lasiurus than N. squamipes or R. rattus (z>2.88, P< = 0.004), but not dissimilar between N. squamipes and R. rattus (z = 0.85, P = 0.398). All lesions spontaneously recovered within one month of onset, after an average 14 (95% CI: 10.9–17.1), 21 (17.8–23.5) and 19 (6.3–31.7) days for the three species, respectively. Lesion duration was shorter (i.e. faster recovery) in N. lasiurus compared to N. squamipes or to R. rattus (z>2.01, P< = 0.044), but not statistically different between N. squamipes and R. rattus (z = -0.58, P = 0.561).
The results of tissue parasite loads were quantified by qPCR in single skin, spleen, and liver tissue samples from all follow-up animals at necropsy are show in Table 3. Maximum tissue burdens were 8.1×103 in skin, 2.8×103 in spleen, and 8.9×102 in liver samples. Substantial variation in Leishmania loads were observed between individual tissues, animals, and inocula dose (Table 3 and Fig 2). Log10 parasite loads in the three tissues were only moderately correlated (Spearman’s r = 0.64–0.67, P<0.001). Average log10 skin tissue loads were lower in low dose vs high dose animals (z = -2.87, P = 0.004), whereas the variation in liver and spleen tissues loads did not significantly differ between dose groups when adjusting for inter-species variation (z<0.84, P>0.05). Leishmania loads in R. rattus tissues were generally lower compared to those in N. lasiurus (all tissue comparisons: z>-4.1, P<0.0001) or in N. squamipes (z>-4.9, P<0.0001), whereas those of N. lasiurus and N. squamipes were not significantly different from one another (z<2.4, NS).
Six months after being infected a total of 299 and 448 female Lu. longipalpis fed respectively on 25 high and 29 low dose animals. No lesions were present on any animals at this time. Flagellates were detected in sand flies fed on 18/18 N. lasiurus, 10/18 N. squamipes and 6/18 R. rattus. The low dose inoculum tended to induce a higher proportion of animals to be infectiousness to sandflies (25/29 86.2%) compared to the high dose group (9/25 36%) (χ2 = 3.46 P = 0.063) (Table 2). A median 15 (95% C.I.: 6.9–24.1), 12 (4.3–14.7), and 17 (6.6–21.0) engorged females sand flies fed on N. lasiurus, N. squamipes and R. rattus, survived to dissection.
Mutlivariate logistic regression of the proportion of flies infected (for N = 34 animals), adjusting for skin tissue Leishmania load × inoculum group interactions, indicated that N. lasiurus tended to be more infectious on average than either N. squamipes (z = -2.91, P = 0.004) or R. rattus (z = -1.73, P = 0.084). The infectiousness of N. squamipes and R. rattus was not statstically different from each other (z = 0.43, P = 0.670). Despite the general low tissue parasite loads in R. rattus, a significant proportion of low dose animals were infectious to sandflies (Table 2). Notwithstanding, the proportion of exposed sand flies infected was generally positively associated with the log10 parasite loads in skin tissue when accounting for differences between rodent species (z = 4.69, P<0.001), but was not associated with either time to lesion onset post inoculation (z-1.56, NS) or to lesion duration (z = 0.51, NS).
This study investigated the comparative development of experimental infection in three putative rodent reservoir species, and their relative ability to transmit L. braziliensis to blood-feeding sandflies. We show that all three rodent species established infections, supported persistent Leishmania burdens in multiple tissue, and presented transient clinical lesions which developed within an average 38–51 days post inoculation, that spontaneously resolved within an average 14–19 days. All three rodent species were also able to infect sandflies as demonstrated by xenodiagnosis performed at c. 6 months post inoculation, by which time all skin lesions had visually healed. We found an association between infectiousness to sand flies and Leishmania loads in ear skin tissue, but not to lesion presence/absence, onset or duration time. Comparing rodent species, N. lasiurus tended to have a greater likelihood of being infectious (18/18 animals), compared to the other two species, and for comparable skin log10 Leishmania loads, both N. lasiurus and N. squamipes infected a greater average proportion of sand flies than did R. rattus. R. rattus also appeared less likely to establish experimental infection at either inoculum dose, evidenced by less clinical signs and lower parasite loads. Despite these observations, the low dose group of R. rattus still proved to be infectious to a small proportion of sand flies, at least at 6 months post inoculation. These collective results suggest that the investigated rodent species represent a multi-host reservoir, though with variable reservoir competence by cross-sectional comparison. Whether any single rodent species can maintain a transmission cycle independently (R0>1) requires further study and parameter estimation [28]. For example, there is likely to be a trade-off between duration and degree of infectiousness relative to the host’s life expectancy: low-level infectiousness over sustained periods could be more significant than high-level infectiousness over a shortened life expectancy resulting from acute infection; data on their comparative longitudinal profile to indicate life-long transmission potential would be informative (e.g.[29]).
Our observations of R. rattus may indicate a greater innate resistance of this species to L. (V.) braziliensis than the other rodent species. It is also possible that this host resolved higher parasitological infections within a shorter time frame than our sampling regime. R. rattus experimentally infected with another cutaneous causing species, L. tropica, presented asymptomatic infections despite ear tissue parasite loads of 4×103−106 with no significant decline over 24m follow-up, and were infectious to a low proportion of sandflies (0–7%) even when fed on the site of experimental inoculation [30]. In contrast to current results, a threshold of infectiousness is positively associated with high parasite loads in ear skin of dogs naturally infected with L. infantum [31]. Unexpectedly, we observed infectiousness to sandflies to be higher in animals inoculated with the low dose compared to the high dose, the latter was associated with skin lesions and higher parasitaemia. In nature Leishmania inoculum sizes from a single infected sandfly have been found to be in the order of 4–40,000 metacyclic promastigotes [32, 33, 34]. No such figures are available for L. (V.) braziliensis. Ideally in such experiments the infecting organisms should come from the bite of a sandfly but at the moment this is technically impossible for L. (V.) braziliensis. In the absence of this possibility the inoculum should contain a similar number of organisms to those delivered by the sandfly. We calculate that our lower dose contained a maximum of approximately 22,000 metacyclic promastigotes which is within the higher range of parasites delivered by a sandfly infected with a leishmania of the subgenus L. (Leishmania), but our higher dose did not fall within the above mentioned range. As we have already said lesions are not at all typical of natural L. (V.) braziliensis infections and because the higher dose produced lesions in many animals we decided to use an inoculum containing fewer organisms in our second experiment. Other factors influence an infection and besides the number of metacyclic promastigotes such sandfly and parasite antigens accompanying the bite and previous exposure to sand fly saliva. The latter may mount a protective response against lesion development, after subsequent challenge [35]. None of our rodents exposed to sand fly bites before the xenodiagnosis nor was any sand fly saliva associated with the inoculation. Significantly more animals, which received the higher dose, had lesions compared to only a single animal that received the lower dose developed a lesion, which mirrored more closely what we have seen in wild infections. The fact that fewer animals that received the higher dose were infectious may reflect a forced immunity produced by more parasites.
Natural infections of L. (V.) braziliensis in free-ranging small mammals are occult [11, 36, 37, 38] in contrast to rodents naturally infected with some Leishmania belonging to the subgenus L.(Leishmania) such as L. (L.) amazonensis and L. (L.) major [39] that often present parasite rich lesions. Since 1996 we have examined over 1,000 small silvatic mammals for infections of L. (V.) braziliensis and periodically isolated this parasite from blood, spleen and liver [11]. The skin from wild animals was positive in PCR tests but we have never managed to isolate the parasite from this tissue nor have we seen any leishmanial skin lesions. It’s quite feasible that flies become infected from parasites present in the skin as well as parasites liberated into the blood from the liver and spleen. L. (V.) braziliensis has also been detected molecularly in apparently normal skin of asymptomatic wild mammals captured in other endemic areas [37, 38] but no isolations were obtained. It is possible that cross-sectional field studies fail to detect short-lived clinical signs in naturally infected captured rodents, a question that can only be resolved by longitudinal follow-up studies. Indeed little is known of the tissue tropism of Leishmania species in their natural hosts, and in laboratory animals some strains of L. (V.) braziliensis are predominantly cutaneous while others also visceralize [40].
The appearance of metastatic lesions at the base of the tail in 13 of our high dose animals and 1 of the low dose animals and the complete absence of lesions at the site of inoculation on the foot pad is extremely interesting. The base of the tail is one preferential feeding site for sand flies and leishmanial lesions have been frequently observed at the base of the tail for L.(L.) mexicana and L.(L.) amazonensis [41, 42]. So why did the lesions appear at this site? A possible reason is a tissue tropism which favors the parasite being in a place where it is readily available to the vector and may be present in the absence of visible lesions. This indicates that future studies on reservoirs need to concentrate on material from the tail base irrespective of the presence of a lesion.
We detected a poor correlation between the log10 parasite loads in rodent skin, liver and spleen tissues. In longitudinal studies of Leishmania loads in dog tissues naturally infected with L. infantum, we similarly observed a poor correlation between tissue loads, however, this was explained by the observed proportional shift in parasite loads in the skin relative to in bone marrow which increased during the time course of infection [31]. Many of the clinical forms of L. (V.) braziliensis in man, such as disseminated cutaneous and mucocutaneous presentations, are considered due to metastatic spread from an initial active or cured lesion or some internal tissue. L. tropica loads of 7.5×103–6×104/cm2 were reported in cutaneous sites (tail tissue, but not in liver, spleen, blood, or bone marrow) disseminated from the experimental inoculation site in R. rattus [30]. It appears that L. (V.) braziliensis, and other Leishmania species, have adopted a strategy in the host to become persistently available to sandflies in the skin and peripheral blood following parasites dissemination from the site of inoculation or/and multiplication in liver and spleen tissues.
One caveat of the current study is that xenodiagnosis was performed using Lu. longipalpis rather than Lu. whitmani, a confirmed vector of L. (V.) braziliensis. Lu longipalpis is highly susceptible to many Leishmania species, including a number of L. (Viannia) species [43] thus being classified as a permissive vector [44], and has been considered a potential vector of L. (L.) amazonensis and L. (V.) braziliensis in Brazil [45]. Here we necessarily treat the xenodiagnosis results as comparative values, assuming that any bias associated with relevant vectorial capacity components is uniform across rodent species.
Lu. whitmani and Lu. intermedia as well as other sand fly species are considered to be competent vectors of L.(V.) braziliensis, based on epidemiological and parasitological observations of wild caught infected female flies. The absence of suitable models to assess vector competence for L.(V.) braziliensis is reflected by the fact that there is only one published account of the successful experimental transmission and this was with a naturally infected fly [46]. Lu. longipalpis is considered to be a permissive vector [44] because it supports the development and adherence of different Leishmania including species of subgenus L.(Viannia) [47] [48] as well as other promastigote producing heteroxenous parasites, such as Endotrypanum [49]. However, its capacity as a vector involving colonization of the cardial region and the production of metacyclic promastigotes has yet to be assessed for this group of parasites. Within this frame work we consider that it is perfectly valid to use Lu. longipalpis to assess the comparative infectiousness of these rodent hosts. Whether its sensitivity in detecting infection is equal to that of the natural vectors can only be determined by comparative experiments. Lu. longipalpis is a complex of sibling species[22] so another question is are there differences in their susceptibility to infection? So far there is no evidence to suggest such differences exist. Our flies belong to the burst song group which is the same as flies that have been widely used by other workers under the name Marajó.
The rodent species evaluated in the current study were selected on the basis of consistent high infection rates or/and high abundance in multiple field studies in northeast Brazil [10, 11]. N. squamipes is the largest rodent of the three species (c. 240gm vs R. rattus 50gm and N. lasiurus 160gm) which may attract relatively more sand flies [15, 50, 51]. The comparative roles of domesticated hosts, such as dogs and equids, have yet to be quantified: it is known that infection prevalences in dogs are comparable to those in rodents [11, 52], and that dogs can infect Lu. whitmani when fed on their skin lesions [53, 54]. However, there are few xenodiagnosis studies on naturally infected hosts of Leishmania, and detection of infection does not necessarily equate to transmission potential (e.g. [55]).
This study provides some preliminary insights into the likely transition from the assumed original transmission cycle of L. (V.) braziliensis involving Atlantic forest small mammals and sand fly vectors, to a more peridomestic cycle involving, not least, the infectious rodents described here, that are associated with overlapping sylvatic and peridomestic habitats (N. lasiurus and N. squamipes) and domestic habitats (R. rattus) respectively [11, 15]. The expansion of this apparent “bridge” between sylvatic and peridomestic transmission habitats are facilitated by the widespread deforestation and conversion of remaining forest to sugarcane and banana plantations. The consequence of environmental shifts on multi-host identity and diversity e.g. proportion of opportunistic and/or competent host species in anthropogenic habitats, may prove to be positive or negative for human transmission [2]. Potential changes in zoopotentiation or zooprophylaxis may be offset by the sand fly vector’s restricted feeding behaviour: Lu. whitmani demonstrates a degree of domesticity, feeding site and host choice loyalty, potentially limiting vector-host contact to more predominant competent species [15, 51, 56]. This focus lies at the southern edge of the geographical range of N. squamipes [57], with the possibility that other species inhabit it’s ecological niche elsewhere [10]. Research is now needed to place the current results in context of longitudinal field studies of natural infection and transmission and including in domesticated animal hosts.
|
10.1371/journal.pgen.1000622 | Fission Yeast Tel1ATM and Rad3ATR Promote Telomere Protection and Telomerase Recruitment | The checkpoint kinases ATM and ATR are redundantly required for maintenance of stable telomeres in diverse organisms, including budding and fission yeasts, Arabidopsis, Drosophila, and mammals. However, the molecular basis for telomere instability in cells lacking ATM and ATR has not yet been elucidated fully in organisms that utilize both the telomere protection complex shelterin and telomerase to maintain telomeres, such as fission yeast and humans. Here, we demonstrate by quantitative chromatin immunoprecipitation (ChIP) assays that simultaneous loss of Tel1ATM and Rad3ATR kinases leads to a defect in recruitment of telomerase to telomeres, reduced binding of the shelterin complex subunits Ccq1 and Tpz1, and increased binding of RPA and homologous recombination repair factors to telomeres. Moreover, we show that interaction between Tpz1-Ccq1 and telomerase, thought to be important for telomerase recruitment to telomeres, is disrupted in tel1Δ rad3Δ cells. Thus, Tel1ATM and Rad3ATR are redundantly required for both protection of telomeres against recombination and promotion of telomerase recruitment. Based on our current findings, we propose the existence of a regulatory loop between Tel1ATM/Rad3ATR kinases and Tpz1-Ccq1 to ensure proper protection and maintenance of telomeres in fission yeast.
| Stable maintenance of telomeres is critical to preserve genomic integrity and to prevent accumulation of undesired mutations that might lead to formation of tumor cells. Fission yeast cells serve as a particularly attractive model system to study telomere maintenance mechanisms, since proteins critical for telomere maintenance are highly conserved between fission yeast and humans. Previous studies have shown that the checkpoint kinases ATM (Tel1) and ATR (Rad3) are required for stable maintenance of telomeres in a wide variety of organisms. Here, we investigated the molecular basis for telomere dysfunction in fission yeast cells lacking ATM and ATR kinases. Our results show that fission yeast ATM and ATR are redundantly required to promote efficient recruitment of telomere protection complex subunits to telomeres, which in turn promote recruitment of telomerase needed to maintain telomeres. Human ATM and ATR kinases might similarly promote telomere protection and telomerase recruitment by promoting recruitment of telomere protection complex subunits.
| Telomeres, the nucleoprotein protective structures at ends of eukaryotic chromosomes, are essential for stable maintenance of eukaryotic genomes [1]. In most eukaryotic species, telomeric DNA is made up of short repetitive G-rich sequences that can be extended by the specialized reverse transcriptase telomerase, to overcome the inability of semi-conservative DNA replication machineries to fully replicate ends of linear DNA [2]. While most of the telomeric G-rich repeats are composed of double-stranded DNA (dsDNA), telomeres end with G-rich 3′ single-stranded DNA (ssDNA), known as G-tail. Both dsDNA and ssDNA portions are important for maintaining functional telomeres as they provide binding sites for telomeric repeat sequence-specific binding proteins, as well as various DNA repair and checkpoint proteins, that are critical for proper maintenance of telomeres.
In mammalian cells, the shelterin complex, composed of TRF1, TRF2, TIN2, RAP1, TPP1 and POT1, plays critical roles in the stable maintenance of telomeres [1]. TRF1 and TRF2 bind specifically to telomeric dsDNA G-rich repeats via their C-terminal myb-like DNA binding domain, while POT1 binds to the telomeric G-tail via its N-terminal OB-fold domains [1]. On the other hand, RAP1, despite the fact that it is evolutionarily related to the budding yeast dsDNA telomeric repeat-binding protein Rap1, cannot directly bind to DNA, and it is recruited to telomeres via its interaction with TRF2 [1]. Likewise, TIN2 is recruited to telomeres by its ability to interact with both TRF1 and TRF2 [3]. TIN2 plays a central role in the formation of the shelterin complex through its ability to interact with the POT1 binding partner TPP1. Previous studies have shown that TRF2 is essential for preventing fusion of telomeres by non-homologous end-joining (NHEJ) and for attenuating ATM-dependent checkpoint signaling [4]. On the other hand, POT1 is critical for protection of telomeres against nucleolytic processing and for attenuating ATR-dependent checkpoint signaling [4]. The POT1-TPP1 sub-complex was also found to interact with the telomerase complex and to increase processivity of telomerase [5],[6].
Fission yeast Schizosaccharomyces pombe is an attractive model system for understanding how the shelterin complex contributes to telomere function since this organism utilizes proteins that show a high degree of conservation to the mammalian shelterin subunits [7]. In contrast, the more extensively studied budding yeast Saccharomyces cerevisiae, while providing unparalleled detailed molecular understanding on how telomere maintenance is regulated, cannot provide much insight into how the shelterin components might contribute to telomere function, since budding yeast lacks shelterin and relies on evolutionarily unrelated alternative protein complexes to maintain telomeres [8],[9].
The S. pombe shelterin complex is composed of Taz1, Rap1, Poz1, Ccq1, Tpz1 and Pot1 [7]. Taz1 directly binds to telomeric dsDNA G-rich repeats via its myb DNA-binding domain, and is thought to fulfill functions analogous to mammalian TRF1 and TRF2 [10]. Rap1, like mammalian Rap1, does not bind directly to telomeric DNA, but it is recruited to telomeres through its interaction with Taz1 [11],[12]. Poz1, the functional counterpart of mammalian TIN2, connects Taz1 to the G-tail binding protein Pot1 by simultaneously interacting with Rap1 and the Pot1 interaction partner Tpz1 [7]. Deletion of taz1+, rap1+ or poz1+ causes massive telomerase-dependent expansion of the G-rich repeat-tract length, and thus they are implicated in the negative regulation of telomerase activity [7],[13]. Tpz1, an ortholog of mammalian TPP1, interacts with Pot1 via its N-terminus, and with Poz1 and Ccq1 via its C-terminus [7]. Thus, Tpz1 is the central protein necessary for the formation of the Pot1 sub-complex, composed of Pot1, Tpz1, Ccq1 and Poz1. Pot1 and Tpz1 are both essential for protecting telomeres in fission yeast since deletion of pot1+ or tpz1+ results in rapid and complete loss of telomeric G-rich repeats and chromosome circularization [7].
Ccq1 is required for telomerase-dependent telomere maintenance as well as inhibition of checkpoint responses and recombination at telomeres [7],[14]. While an ortholog of Ccq1 has not been identified in mammalian cells, analogous proteins that are critical for telomerase recruitment and inhibition of checkpoint and repair responses at telomeres might await discovery in mammalian cells. The telomere protection function fulfilled by Pot1 and Tpz1 appears to be provided redundantly by Poz1 and Ccq1, since poz1Δ ccq1Δ cells, but not poz1Δ or ccq1Δ single deletion cells, rapidly lose telomeres and circularize chromosomes [7].
Similar to pot1Δ or tpz1Δ cells, S. pombe cells deleted for either Stn1 or Ten1 rapidly lose telomeres and circularize chromosomes [15]. Fission yeast Stn1 and Ten1 are evolutionarily conserved to S. cerevisiae Stn1 and Ten1, which are essential for telomere capping in budding yeast. Budding yeast Stn1 and Ten1 form a complex with the telomeric G-tail binding protein Cdc13, and the Cdc13-Stn1-Ten1 complex has been proposed to represent a telomere-specific replication protein A (RPA)-like complex [16]. Since Pot1 does not appear to be in the same complex as Stn1 and Ten1, fission yeast cells seem to utilize two independent capping complexes to protect telomeres [15],[17]. Higher eukaryotic cells may also utilize both Pot1 and Stn1 complexes to protect telomeres since the Stn1 ortholog in Arabidopsis was found to be important for telomere protection, and potential Stn1 orthologs have been identified in mammalian genomes based on sequence analysis [15],[16],[18].
Telomere proteins, such as TRF2 and POT1, inhibit DNA damage and/or DNA replication checkpoint signaling regulated by ATM and ATR kinases [4]. Paradoxically, checkpoint and DNA repair proteins are also essential for stable telomere maintenance. In fact, cells simultaneously lacking both ATM and ATR pathways suffer severe telomere dysfunction in a wide variety of organisms, including budding and fission yeasts, Arabidopsis and Drosophila [19]–[23]. In budding yeast, where the shelterin complex is absent, studies have uncovered redundant roles for Tel1ATM and Mec1ATR in promoting telomerase recruitment via phosphorylation of Cdc13 to enhance the interaction between Cdc13 and the Est1 subunit of telomerase [24]. However, no molecular details of telomere defect(s) caused by simultaneous loss of ATM and ATR pathways were available for the organisms that utilize telomerase, shelterin, and the Stn1 complex to maintain telomeres. Therefore, we utilized fission yeast to define the nature of telomere dysfunction in cells lacking both Tel1ATM and Rad3ATR. Our analyses implicate a defect in efficient accumulation of the shelterin complex subunits Tpz1 and Ccq1 to telomeres as the main cause of telomere dysfunction in tel1Δ rad3Δ cells, which exhibit defects in both telomere protection and telomerase recruitment.
In budding yeast, a telomere maintenance defect observed in tel1Δ mec1Δ double mutant cells can be suppressed by deleting Rif1 or Rif2 (Rap1 interacting factors) or by reducing Rap1 accumulation at telomeres. These observations suggested that the requirement of Tel1ATM and Mec1ATR for telomere maintenance could be bypassed simply by making telomeres more accessible to telomerase by removing inhibitory regulators of telomerase [25]. Moreover, tel1Δ mec1Δ cells lost their viability slower than telomerase RNA mutant (tlc1Δ) cells, and tel1Δ mec1Δ tlc1Δ cells lost their viability with a rate comparable to tlc1Δ cells. Thus, the telomere maintenance defect observed in tel1Δ mec1Δ cells may entirely be attributable to the failure of the double mutant cells to efficiently recruit telomerase to telomeres [25].
In contrast, our previous analyses suggested that fission yeast lacking Tel1ATM and Rad3ATR are likely to be defective in telomerase recruitment and other additional functions such as telomere protection [26]. This prediction was made based on the following observations. First, tel1Δ rad3Δ cells lost their viability faster than telomerase mutant (trt1Δ) cells. Second, tel1Δ rad3Δ trt1Δ and tel1Δ rad3Δ cells lost their viability at comparable rates, suggesting telomere defects observed in tel1Δ rad3Δ cells include a defect in telomerase function. Third, Taz1 deletion (taz1Δ), which allows trt1Δ cells to stably maintain telomeres by recombination and thus should be able to suppress chromosome circularization if telomerase recruitment is the only defect caused by tel1Δ rad3Δ, could not suppress chromosome circularization of tel1Δ rad3Δ cells [26],[27].
However, since taz1Δ cells show more severe telomere defects than rap1Δ or rif1Δ cells [13],[28], we tested if rap1Δ or rif1Δ could suppress chromosome circularization of tel1Δ rad3Δ cells. Fission yeast Rap1 and Rif1 show sequence homology to budding yeast Rap1 and Rif1, respectively, and rap1Δ and rif1Δ cells carry elongated telomeres, suggesting that they are important for negative regulation of telomerase in fission yeast [11]. However, neither rap1Δ nor rif1Δ was able to suppress the chromosome circularization phenotype of tel1Δ rad3Δ cells (Figure 1A). These results thus establish that mutations of telomerase inhibitors cannot suppress the telomere maintenance defect of tel1Δ rad3Δ, and further support the notion that Tel1ATM and Rad3ATR may contribute to telomere protection.
Next, we tested more directly if loss of Tel1ATM and Rad3ATR causes defects in telomere protection. In order to reliably examine changes in telomere structure or recruitment of various telomere-associated factors in tel1Δ rad3Δ cells prior to chromosome circularization, we first developed a new plasmid-based system that allowed us to utilize younger generation tel1Δ rad3Δ cells for our experiments, rather than performing meiotic crosses to create tel1Δ rad3Δ cells (Figure 2A). In this system, we took advantage of the fact that tel1Δ rad3Δ cells carrying a Rad3-plasmid grow significantly slower upon loss of the plasmid, and thus form smaller colonies when grown on non-selective media plates (Figure 2B). For our experiments, we chose multiple small colonies, individually confirmed to be tel1Δ rad3Δ based on their inability to grow on media lacking histidine (loss of his3+ marker) or media containing hydroxyurea (loss of rad3+) (Figure 2B). These freshly derived tel1Δ rad3Δ cells were then pooled and grown in liquid culture to obtain sufficient amount of cells at early generation to perform our biochemical analyses. Based on Southern blot analysis, we estimate that the average telomere length of tel1Δ rad3Δ cells utilized in our experiments is shorter than wt cells, but comparable or even slightly longer than rad3Δ cells (Figure 2C). Furthermore, based on amplification cycle numbers for input samples in our quantitative PCR analyses for chromatin immunoprecipitation (ChIP) assays, we can ensure that tel1Δ rad3Δ cells utilized in our experiments have not yet circularized their chromosomes, since primer annealing sites are completely lost after chromosome circularization [26].
We first examined changes in telomeric G-tail length by carrying out a series of non-denaturing native dot blot hybridization experiments using G-rich or C-rich strand specific probes for genomic DNA samples prepared from wt, tel1Δ, rad3Δ and tel1Δ rad3Δ cells. We found that the native hybridization signal for the probe that specifically anneals to the G-rich strand of telomeres (normalized against denatured sample), but not for the probe specific for the C-rich strand, increased significantly in tel1Δ rad3Δ cells (Figure 3A). Thus, we conclude that the telomeric G-tail is significantly elongated in tel1Δ rad3Δ cells, compared to wt, tel1Δ, or rad3Δ cells. The increase in G-tail length may be caused by a decrease in protection of the telomeric C-rich strand against degradation, or a delay in the arrival of lagging strand DNA polymerases at telomeres [17].
We next monitored recruitment of the largest subunit of RPA (replication protein A) Rad11 and the homologous recombination (HR) DNA repair proteins Rad51 and Rad52 (Rhp51 and Rad22 in fission yeast, respectively) by quantitative ChIP assays. Based on Western blot analyses, expression levels for all analyzed proteins did not change significantly, when tel1 and/or rad3 were deleted. We found that Rad11RPA, Rhp51Rad51, and Rad22Rad52 are all recruited to telomeres at significantly higher levels in rad3Δ and tel1Δ rad3Δ cells (Figure 3B–3D). While Rad22Rad52 recruitment to telomeres was comparable between rad3Δ and tel1Δ rad3Δ cells, Rad11RPA and Rhp51Rad51 recruitment to telomeres was significantly higher in tel1Δ rad3Δ than in rad3Δ cells. Since rad3Δ cells carry much shorter telomeres than wt cells [19],[26] (Figure 2C), increased incidences of cells experiencing critically short telomeres may be responsible for increase in telomere association of RPA and HR repair factors in rad3Δ cells. In contrast to HR repair proteins, telomere recruitment of Ku80, involved in NHEJ repair, was not greatly affected by deletion of tel1 and/or rad3 (Figure 3E). The observed increase in telomere binding for RPA and Rad22Rad52, but not Ku, would be consistent with the notion that chromosome circularization in tel1Δ rad3Δ cells might occur by single strand annealing rather than NHEJ, much like in pot1Δ cells [29].
Since we observed an increase in G-tail length and recruitment of HR repair factors in tel1Δ rad3Δ cells, we suspected that the integrity and/or recruitment of telomere capping complexes might be affected by the loss of Tel1ATM and Rad3ATR. Accordingly, we monitored changes in the association of the Pot1 sub-complex (composed of Pot1, Tpz1, Poz1 and Ccq1) and the Stn1 complex (composed of Stn1 and Ten1) by quantitative ChIP assays. Previous studies have established that these complexes are likely to be independent, but both are essential for telomere protection in fission yeast [7],[15],[17],[30]. Western blot analyses indicated that expression levels for all analyzed proteins are not greatly affected by deletion of tel1 and/or rad3 (Figure 4).
While we did not observe any major changes in Stn1 recruitment to telomeres (Figure 4E), we observed subunit specific changes in recruitment of the Pot1 sub-complex to telomeres when Tel1ATM and Rad3ATR were eliminated. While Pot1 recruitment to telomeres was increased in tel1Δ rad3Δ cells (Figure 4A), recruitment of Tpz1 and Ccq1 was significantly reduced in tel1Δ rad3Δ cells (Figure 4B, 4C), and recruitment of Poz1 was not significantly affected among different genetic backgrounds (Figure 4D). Therefore, it appears that simultaneous loss of Tel1ATM and Rad3ATR differentially affects individual subunits of the Pot1 sub-complex. It is also worth noting that the increase in telomere association for RPA (∼9 fold) is much greater than for Pot1 (∼2 fold) in tel1Δ rad3Δ cells.
Given that Ccq1 and Tpz1 association to telomeres was decreased while Pot1 association was increased, we wondered if the integrity of the Pot1 sub-complex is compromised in tel1Δ rad3Δ cells. Therefore, we performed pairwise co-immunoprecipitation (IP) experiments among different subunits of the Pot1 sub-complex in wt and tel1Δ rad3Δ cells (Figure 5). Surprisingly, we did not observe any obvious changes in interactions. One possible explanation might be that asynchronous fission yeast cell cultures contain a large excess of the Pot1 sub-complex that is not bound to telomeres and thus is not regulated by Tel1/Rad3. If only the telomere-bound Pot1 sub-complex stability is affected in tel1Δ rad3Δ cells, co-IP assays may not be able to detect changes in complex stability. It is currently unknown if fission yeast cells contain a large pool of telomere unbound Pot1 sub-complex, but we have previously shown that telomere association of Pot1 is cell cycle regulated and occurs maximally during late S-phase [17]. Alternatively, since previous studies have demonstrated that Ccq1 can interact with the heterochromatin modulator SHREC complex [7],[31], loss of Ccq1 from telomeres might be caused by loss of interaction between SHREC and Ccq1 without affecting the stability of the telomere-bound Pot1 sub-complex. However, we found that tel1Δ rad3Δ cells appear to have intact heterochromatin based on the intact telomere-specific silencing of a marker gene (Figure 6). Previous studies have indicated that recruitment of Pot1 can occur independently of its N-terminal OB fold domain, required to bind G-tails at 3′ ends of telomeres, and that Rap1-Poz1 interaction can promote recruitment of the Pot1 sub-complex to the dsDNA portion of telomeres [7],[32]. In fact, based on microscopic observation [32], a majority of Pot1 may be associated with dsDNA portion of telomeric and sub-telomeric regions, and only a small fraction of the Pot1 sub-complex is bound to the extreme 3′ ends of telomeres. Therefore, we currently favor the notion that Tel1ATM and Rad3ATR are especially important for stabilizing the Pot1 sub-complex bound close to the 3′ ends of telomeres, but bulk of the Pot1 sub-complex, bound to the dsDNA portion of telomeres (or unbound to telomeres), are not significantly affected by simultaneous deletion of Tel1ATM and Rad3ATR kinases.
Ccq1 was recently found to be important for telomerase-dependent telomere maintenance in fission yeast [7],[14]. Moreover, Ccq1 and Trt1TERT can be co-immunoprecipitated, and Tpz1 pull down experiments can bring down active telomerase in a Ccq1-dependent manner. Since we found reduced association of Tpz1 and Ccq1 to telomeres in our quantitative ChIP analyses (Figure 4), we next examined if recruitment of telomerase to telomeres is affected by loss of Tel1ATM and Rad3ATR. We found that telomere association of both the telomerase catalytic subunit Trt1TERT and its regulatory subunit Est1 are significantly reduced in tel1Δ rad3Δ cells (Figure 7), much like in ccq1Δ cells [14] (Figure S1). The loss of ChIP signals were not due to loss of the telomerase complex subunits since comparable expression levels of Trt1TERT and Est1 were detected by Western blots in all genetic backgrounds tested.
We next examined if interactions among telomerase, Tpz1 and Ccq1 are disrupted in tel1Δ rad3Δ. Indeed, the Ccq1-dependent interaction between the telomerase RNA subunit TER1 and Tpz1, as well as interaction between Ccq1 and TER1 were abolished in tel1Δ rad3Δ cells (Figure 8A). The loss of interaction between Tpz1-Ccq1 and telomerase is not due to disruption of the telomerase complex or degradation of telomerase RNA in tel1Δ rad3Δ cells, since we can pull down comparable amounts of telomerase RNA when the telomerase catalytic subunit Trt1TERT was used for IP (Figure 8B). Taken together, our data indicate that the telomerase complex (Trt1-Est1-TER1) can no longer be recruited to telomeres in the absence of Tel1ATM and Rad3ATR due to the disruption of the Pot1 sub-complex and its interaction with telomerase.
In this paper, we investigated the nature of telomere dysfunction caused by simultaneous deletion of the two major checkpoint kinases Tel1ATM and Rad3ATR in fission yeast. Results reported here support a model depicted in Figure 9A. We showed that tel1Δ rad3Δ cells accumulate longer G-tails (Figure 3A), suggesting possible defects in either protection against degradation of the C-rich strand or in coordination of leading and lagging strand synthesis at telomeres. The observed increases in recruitment of RPA, Rad51 and Rad52 to telomeres (Figure 3B–3D) further support the notion that tel1Δ rad3Δ cells are defective in protection of telomeres. Analysis of telomere complexes suggests that tel1Δ rad3Δ cells are defective in efficient accumulation of the shelterin subunits Tpz1 and Ccq1 (Figure 4). Moreover, we determined that tel1Δ rad3Δ cells were unable to recruit telomerase to telomeres due to a defect in interaction between Tpz1-Ccq1 and telomerase (Figures 7 and 8). The loss of interaction between Tpz1-Ccq1 and telomerase may be due to direct role(s) of Tel1/Rad3 in promoting this interaction, or could be indirectly caused by inefficient accumulation of Tpz1-Ccq1 at telomeres. It should also be noted that our data do not rule out the possibility that Tel1ATM and Rad3ATR phosphorylate different sets of substrates at telomeres.
Given that ccq1Δ cells were previously found to be defective in both protection of telomeres and recruitment of telomerase [7],[14], our data is consistent with the notion that all telomere defects observed in tel1Δ rad3Δ may primarily be caused by the failure to properly accumulate Ccq1 at telomeres. Ccq1 was also recently shown to be essential for suppressing Rad3ATR-dependent G2 checkpoint activation by telomeres [14]. Thus, it appears that fission yeast Tel1ATM and Rad3ATR promote accumulation of their own inhibitor Ccq1 to ensure that telomeres do not cause permanent cell cycle arrest.
The regulatory loop formed by Tel1/Rad3 and the Pot1 sub-complex (Figure 9B) ensures that telomeres that transiently become de-protected would preferentially activate Tel1/Rad3 pathways to promote recruitment of Tpz1 and Ccq1, and to re-establish proper protection of telomeres. An analogous regulatory loop appears to exist in Drosophila, where retrotransposons have replaced telomerase and neither the shelterin complex nor the Stn1-Ten1 complex exist, since ATM and ATR are redundantly required to promote recruitment of the telomere capping protein HOAP to telomeres [22],[23]. We suspect that mammalian ATM and ATR might also be involved in promotion of telomere capping by affecting the recruitment of the shelterin complex components.
Similar to budding yeast, where Tel1ATM and Mec1ATR are redundantly required to promote interaction between the G-tail binding protein Cdc13 and telomerase, our data demonstrate that fission yeast Tel1ATM and Rad3ATR are redundantly required to recruit telomerase to telomeres by promoting the interaction between the Pot1 sub-complex and telomerase. In budding yeast, phosphorylation of Cdc13 by Tel1ATM/Mec1ATR kinases promotes Cdc13-Est1 interaction to facilitate telomerase recruitment [24]. Tel1ATM/Rad3ATR kinases may also promote interaction between the Pot1 sub-complex and telomerase by phosphorylation of the Pot1 sub-complex subunits in fission yeast. Mammalian POT1-TPP1 has also been implicated in recruitment of telomerase to telomeres [6]. Thus, future studies may uncover an involvement of mammalian ATM/ATR in promoting the interaction between POT1-TPP1 and telomerase.
Fission yeast strains used in this study were constructed by standard techniques [33] and are listed in supplemental Table S1. For tel1Δ::LEU2, rad3Δ::LEU2, pku80Δ::ura4+, taz1Δ::ura4+, rap1Δ::ura4+, rif1Δ::ura4+ and rhp51Δ::ura4+, original deletion strains were described previously [11], [26], [34]–[36]. For rad11-FLAG, pku80-myc, pot1-myc, poz1-myc, stn1-myc and trt1-myc, original tagged strains were described previously [17],[37],[38]. Primers listed in Table S2 were used to construct ccq1Δ::hphMX, ccq1-myc, ccq1-FLAG, tpz1-myc, tpz1-FLAG, est1-myc and rad22-myc by PCR-based methods [39]–[41]. The plasmids pREP41H-rad3 and pREP42-myc-rad3 were used to complement tel1Δ rad3Δ strains to maintain telomeres. pREP41H-rad3 carries rad3+ under the control of the medium strength nmt1 promoter and a his3+ marker, while the pREP42-myc-rad3 carries myc-rad3+ under the control of the medium strength nmt1 promoter and an ura4+ marker [42],[43].
tel1Δ rad3Δ strains carrying either pREP41H-rad3 or pREP42-myc-rad3 were grown in YES liquid culture for 16 hours prior to plating onto YES plates in order to promote loss of plasmid. Small colonies were picked and simultaneously streaked on YES, YES+5 mM HU, and PMG UAL (-His) or HAL (-Ura) plates to verify the loss of rad3+ and the selection marker. Several colonies that were sensitive to HU and did not grow on PMG selection plates were pooled and inoculated in YES liquid culture, and grown overnight to obtain sufficient cells for subsequent experiments.
Pulsed-field gel electrophoresis of NotI-digested chromosomal DNA was performed to monitor chromosome circularization as previously described [26]. For telomere length analysis, genomic DNA samples were digested with EcoRI, separated on a 1% agarose gel, and probed with telomere probe [44] as previously described.
Native dot blot analyses were performed as described [45], with minor modifications. DNA was blotted onto Hybond-XL membrane (GE) using the BioRad Bio-Dot Microfiltration System. Hybridization was performed in Church buffer [0.25 M sodium phosphate buffer pH 7.2, 1 mM EDTA, 1% BSA, 7% SDS] at 45°C overnight with probes annealing to the G-rich strand [848: CGT GTA ACC ACG TAA CCT TGT AAC CCG ATC] or to the C-rich strand [847: GAT CGG GTT ACA AGG TTA CGT GGT TAC ACG] [46].
Cells were processed for ChIP and analyzed as previously described [17]. Monoclonal anti-myc (9B11; Cell Signaling) and anti-FLAG (M2-F1804; Sigma) antibodies and polyclonal anti-Rad51 antibody (A-92, Santa Cruz) were used. Percent precipitated DNA values (% ppt DNA) were calculated based on ΔCt between Input and IP samples after performing several independent triplicate SYBR Green-based real-time PCR (Bio-Rad) using telomere primers jk380 and jk381 [17].
Cell extracts were prepared in lysis buffer [50 mM Tris pH 8.0, 150 mM NaCl, 10% glycerol, 5 mM EDTA, 0.5% NP40, 50 mM NaF, 1 mM DTT, 1 mM PMSF, 0.2 mM APMSF, 1 mM Na3VO4, ‘Complete’ protease inhibitor cocktail] using glass beads. Extracts were preincubated with 100 µg/ml Ethidium bromide for 30 min on ice. Proteins were immunoprecipitated using either monoclonal anti-myc antibody (9B11, Cell Signaling) or monoclonal anti-FLAG antibody (M2-F1804, Sigma), and Dynabeads (Invitrogen). Immunoprecipitated proteins were analyzed by Western blot analysis.
Proteins in whole cell extract or from immunoprecipitations were analyzed by western blot using either monoclonal anti-FLAG antibody (M2-F1804) or monoclonal anti-myc antibody (9B11). Anti-Cdc2 antibody (y100.4, Abcam) was used for loading control.
Experiments were performed essentially as described [37]. Cell extracts were prepared in TMG100 buffer [10 mM Tris pH 8.0, 1 mM MgCl2, 100 mM NaCl, 10% glycerol, 1 mM EDTA, 0.1 mM DTT, 2 mM PMSF, 0.2 mM APMSF, 1 U/µl RNasin (Promega), and ‘Complete’ protease inhibitor cocktail] using glass beads. IPs were performed with 4 mg of whole cell extract in the presence of 0.5% v/v Tween20 using monoclonal anti-myc antibody (9B11) and Dynabeads (Invitrogen). Beads were subsequently washed with TMG100 buffer and treated with 0.4 mg/ml Proteinase K in [10 mM Tris pH 8.0, 100 mM NaCl, 1% SDS, 10 mM EDTA] at 37°C for 30 min. RNA was isolated using ‘Total RNA Isolation’ Kit (Clontech). RNA was reverse transcribed using M-MLV Reverse Transcriptase (Ambion) with Primer 1016 [GAT CCA TGG ATC TCA CGT AAT G], and subsequently subjected to triplicate SYBR Green-based real-time PCR analysis with primers 1015 [CAG TGT ACG TGA GTC TTC TGC CTT] and 1017 [CAA AAA TTC GTT GTG ATC TGA CAA GC]. Control reactions were also performed without reverse transcriptase to ensure that the PCR signal reflected RNA and not contaminating DNA.
In order to determine statistical significance of our data, two-tailed Student's t-tests were performed, and P values ≤0.05 were considered as statistically significant differences.
|
10.1371/journal.pgen.1003126 | RecX Facilitates Homologous Recombination by Modulating RecA Activities | The Bacillus subtilis recH342 strain, which decreases interspecies recombination without significantly affecting the frequency of transformation with homogamic DNA, carried a point mutation in the putative recX (yfhG) gene, and the mutation was renamed as recX342. We show that RecX (264 residues long), which shares partial identity with the Proteobacterial RecX (<180 residues), is a genuine recombination protein, and its primary function is to modulate the SOS response and to facilitate RecA-mediated recombinational repair and genetic recombination. RecX-YFP formed discrete foci on the nucleoid, which were coincident in time with RecF, in response to DNA damage, and on the poles and/or the nucleoid upon stochastic induction of programmed natural competence. When DNA was damaged, the RecX foci co-localized with RecA threads that persisted for a longer time in the recX context. The absence of RecX severely impaired natural transformation both with plasmid and chromosomal DNA. We show that RecX suppresses the negative effect exerted by RecA during plasmid transformation, prevents RecA mis-sensing of single-stranded DNA tracts, and modulates DNA strand exchange. RecX, by modulating the “length or packing” of a RecA filament, facilitates the initiation of recombination and increases recombination across species.
| This study describes mechanisms employed by the bacterium Bacillus subtilis to survive DNA damages by recombinational repair (RR) and to provide genetic variation via genetic recombination (GR). At the center of homologous recombination (HR) is the recombinase RecA, which forms RecA·ssDNA filaments to mediate SOS induction and to promote DNA strand exchange, a step needed for both RR and GR. Genetic data presented here highlight the complexity of the network of RecA accessory factors that regulate HR activities, with RecX counteracting the role of RecF in SOS induction. The absence of both RecA modulators, however, blocked RR and GR. Insights into the spatio-temporal recruitment of RecA to preserve genome integrity, to overcome the barriers of gene flow, and its regulation by mediators and modulators are provided. Chromosomal transformation, which declines with increasing evolutionary distance, depends on HR. Indeed, the presence of the RecX modulator decreases the genetic barrier between closely related organisms. The role of RecA mediators and modulators on the preservation of genome integrity and long-term genome evolution is discussed.
| The bacterial RecA recombinase (homologue to human RAD51 and DMC1), arranged as higher-order oligomers assembled on tracts of single-stranded (ss) DNA, is involved in the DNA strand exchange reaction to warrant genome integrity by recombinational repair (RR), and genetic diversity by genetic recombination (GR). Template-dependent RR preserves the integrity of the genetic information, re-establishes replication and ensures proper chromosomal segregation. In contrast, GR, which occurs in species that can exchange chromosomal DNA segments, is an important mechanism for natural variation among prokaryotes and plays an important role in the dissemination of important traits, such as antibiotic resistance, virulence determinants and metabolic pathways involved in adapting to environmental niches. There are three modes by which bacteria can exchange chromosomal DNA segments: viral-mediated transduction, which may be limited by the viral host range and by the host-encoded restriction system, conjugation and natural transformation. Bacillus subtilis transformation or Escherichia coli conjugation catalyze unidirectional integration of chromosomal ssDNA at a frequency that decreases exponentially with the increasing degree of DNA sequence divergence between donor and recipient reviewed in [1], [2]. In E. coli the extent of genetic isolation by HFR conjugation is determined by the activity of the mismatch repair system, and requires DNA replication and recombination functions (specifically requires overproduction of the RecA protein) [3], [4], [5]. B. subtilis natural transformation, which can take DNA of any source, is insensitive to restriction endonucleases and to mismatch repair functions, and shows no obvious requirement for extended DNA replication [2], [6]. RecA-dependent homologous recombination (HR) rather than mismatch repair seems to control the extent of genetic isolation during natural transformation [6]. Here, a specific set of recombination functions, some of which are induced by natural competence (e.g., SsbA, SsbB, DprA [Smf or CilB], RecA, CoiA), are mainly located at the cell poles (namely SsbB, DprA, RecA, CoiA and RecU) where the DNA uptake machinery is located [7]–[11]. Except for recA and dprA mutations, the B. subtilis chromosomal transformantion frequency with homogamic DNA in single rec-deficient strains, classified within the α (recF15 or ΔrecO), β (addA5), γ (recH342), δ (ΔrecN), ε (ΔruvAB, ΔrecU), ζ (ΔrecQ) or η (ΔrecG) epistatic groups, does not vary more than 3-fold relative to the rec+ value [12]–[15]. The absence of RecA blocks chromosomal transformation, and the absence of DprA results in a 50-fold reduction relative to the rec+ value [10], [16]–[18]. From those rec-deficient strains tested, the frequency of interspecies gene exchange deviated significantly from the rec+ strain only in the recH342 mutant strains [6]. The frequency of transformation with divergent donor DNA decreased >20-fold in the recH342 strain relative to the rec+ value, without affecting the frequency of transformation by closely related donors [6], suggesting that HR introduces barriers to genetic exchange, and that the “RecH342” mutation contributes to sexual isolation. Very little is known about the mutation(s) present in the recH342 (BG119) strain, but the phenotype(s) associated with it suggested that the function(s) affected in this strain might act as an accessory factor by regulating the formation of an active RecA filament [19].
Why Does RecA need accessory factors? The essential single stranded binding (SSB) protein (termed SSB in E. coli, SsbA in B. subtilis or RPA in eukaryotes), which is ubiquitous in all living organisms, is involved in multiple pathways of DNA metabolism, including DNA replication, RR and GR reviewed in [20]. The majority of naturally competent bacteria encode a second non-essential protein, termed SsbB [21]. Biochemical studies have shown that the SSBs proteins, which bind to ssDNA and remove secondary structures, limit RecA loading onto ssDNA, as a consequence of the higher affinity and faster binding kinetics, so that the net result is a SSB-coated ssDNA reviewed in [20]–[24]. Furthermore, RecA·ssDNA filament elongation is blocked by DNA secondary structures, whereas assembly of SSB proteins is not, and SSB proteins contribute to the removal of secondary structures upon ssDNA binding, hence the RecA·ssDNA filaments formed on SSB-coated ssDNA after removal of the SSB protein(s) are more efficient than those formed by RecA alone reviewed in [20]–[24]. To overcome the effect of a SSB protein on RecA nucleation onto ssDNA, and RecA filament formation, a series of RecA accessory factors regulate such stage reviewed in [22]–[25]. These factors can be divided into two broad classes: those that act before RecA nucleation by promoting assembly of RecA onto SSB-coated ssDNA (termed RecA mediators), and those that act after RecA nucleation and during homology search and strand exchange, by promoting RecA·ssDNA filament assembly and disassembly (termed RecA modulators) [25]. Genetic recombination and RR share some accessory factors, but others are specific for each event. The most ubiquitous RecA mediators are RecO and RecR, which are involved both in RR and GR. The role of RecBCD (counterpart of B. subtilis AddAB) and RecF as RecA mediators is less conserved and less well-understood in bacteria other than γ-Proteobacteria [26]–[28]. DprA is an ubiquitous RecA mediator that plays a relevant role during GR (see above). The RecA modulators RecF and RecX are widely present in bacteria, but very little is known about their in vivo role [9], [19], [22]. In vitro RecXEco destabilizes the RecAEco·ssDNA filaments and RecFEco antagonizes this effect [29], but RecAEco foci formation (nucleation onto ssDNA?) decreases in the ΔrecXEco, but increases in the recF4115Eco context [30]. The difference between the simplified in vitro system and in vivo could be related to the presence of other RecA modulators in the γ-Proteobacteria, as DinIEco and RdgCEco, whose presence in bacteria of the Firmicutes Phylum is not obvious reviewed in [22].
In B. subtilis, cytological studies have shown that RecN, RecO, RecR, RecA and RecF form a discrete focus on the nucleoid in response to DNA damage. By observing the localization and temporal order of recruitment, we learned that these proteins co-localize to a defined DNA double-strand break (DSB), with RecN localizing first, while RecO, RecR and RecA localize later, followed by RecF [31], [32, our unpublished results]. Concomitantly with RecF assembly, the RecA foci are converted onto highly dynamic filamentous structures (termed threads) across the nucleoid that are disassembled 120 min later [32]. Biochemical studies suggested that a dynamic RecA·ssDNA filament with an “effectual length” is essential for SOS induction, template-dependent RR and for programmed GR [19], [24]. Previously, it has been shown that: i) a subset of RecA functions shows optimal activity at a high ssDNA/protein ratio, which might pack less RecA per unit length of ssDNA, and requires NTP hydrolysis, whereas other catalytic activities are optimal in RecA-saturated complexes that require NTP, but do not hydrolyze it [33], and ii) RecA-mediated SOS induction requires an extended filament conformation, but no ATP hydrolysis [34], [35]. For the SOS induction an extended and saturated RecA·ssDNA filament [33], [35], is essential for LexA repressor self-cleavage [36]. The absence of LexA increases the expression of SOS genes reviewed in [37]. In E. coli, RecX inhibits the RecA coprotease activity of RecA in vitro and in vivo, but a null recX mutant (ΔrecX) strain shows no obvious phenotype [38], [39].
In E. coli and B. subtilis the SOS response is reduced and delayed in the absence of RecF, RecO and RecR [40], suggesting that these products could work as mediators and/or modulators. This is consistent with the observation that certain RecA mutant proteins act as suppressors of the recO, recR or recF defect [41], [42]. These RecA mutant variants showed the unassisted ability to displace the SSB protein [43], suggesting that specialized RecA mediators and/or modulators that regulate RecA activities are necessary to avoid the potential hazard that could be caused by miss-regulation of HR [22].
Biochemical studies with protein of E. coli origin, have shown that RecO, alone or in concert with RecR, aids RecA to overcome the limitation imposed by the SSB protein, and loads RecA onto the ssDNA [26], [44]–[47]. Then, RecX inhibits the strand exchange reaction by blocking RecA·ssDNA filament formation or facilitating RecA filament disassembly [38], [39], [48], [49], whereas RecF, which physically interacts with RecX, actively participates in the addition of RecA monomers to the nucleoprotein filament, by inhibiting the effect of RecX [29]. These proteins might also modulate the RecA/ssDNA ratios (packing) or the length of the RecA·ssDNA filament (see above).
During programmed GR in B. subtilis competent cells, the internalized ssDNA should be coated by one of the SSB proteins (SsbA or SsbB). RecO, alone or in concert with RecR, or DprA aids RecA to overcome the limitation imposed by SsbA or SsbB (or both in concert) and loads RecA onto ssDNA tracts [18], [50]. Then, RecA polymerizes on the filament (RecA threads?) and rapidly scans for a homologous dsDNA segment in the recipient that eventually binds to RecA to allow for strand exchange. Here, one strand of the recipient duplex unbinds from its partner and pairs with the internalized ssDNA. Note that henceforward in this paper, and unless stated otherwise, the indicated genes and products are of B. subtilis origin.
To gain insight into the initial state of RecA regulation, we have in vivo characterized the function(s) impaired in the B. subtilis recH342 strain. We have identified the mutation of the recH342 strain, which maps in the putative recX (yfhG) gene, so that the mutation was renamed as recX342. We have deleted the putative recX gene (ΔrecX), and investigated the in vivo role of RecX to gain insight in the regulation of RecA activity by analyzing its effect in induction of the SOS response, RR and GR. Our work reveals that the absence of RecX reduces the threshold for damage-dependent SOS response, the recF15 mutation delays it, and the effect observed in the single mutants is overcome in the ΔrecX recF15 context. We show that RecA·ssDNA filament necessary for SOS induction is not sufficient for RecA-mediated strand exchange. Here, RecX might act by increasing the stability of the joint molecule or by affecting the length of the minimum efficiently processed segment (MEPS) and indirectly removes a barrier for genetic exchange. We propose that RecA exerts a negative effect on plasmid transformation and RecX suppresses it. Our work demonstrates that RecX facilitates HR by modulating RecA activities and plasmid establishment by inhibiting RecA.
The radiation-sensitive rec342 mutant strain, which was isolated in late sixties [51], bears two separable mutations. One mutation, which leads to methyl methanesulfonate (MMS) sensitivity, was termed recH342 (BG119 strain) and classified within the γ epistatic group [12], . To identify the mutation(s) present in the recH342 strain (BG119) nucleotide sequence analysis and whole-genome comparisons (Genome Analyzer, Illumina) were performed in parallel with the isogenic rec+ (Reference strain [BG214]). The isogenic BG119 (recH342) showed 9 differences with the BG214 (rec+) strain, resulting in 5 amino acid changes, 3 intergenic mutations and 1 silent mutation (Table 1). The DNA repair phenotype observed in recH342 strain could be attributed to the substitution of Leu for Pro (L101P), in a conserved region of the YfhG protein (see Figure S1). YfhG shares a low but significant level of identity with genuine RecX proteins [53] (see below). The BG119 strain also carried a point mutation (P236S) in a variable region of the DNA translocase SftA (Table 1). SftA, which is required for coupling chromosomal segregation and cell division, assists the tyrosine recombinases in the resolution of chromosomal dimers reviewed in [54]. Since the mutations present in recH342 did not confer a significant chromosomal segregation defect [55], [56], and a plasmid-borne sftA gene failed to complement the MMS-sensitive phenotype of recH342 cells (data not shown), we assumed that the mutation in the sftA gene should not be responsible for the observed phenotype.
To test whether putative RecX is involved in RR and/or GR and if it complements the recH342 defect, a null recX (ΔrecX) mutant strain was created, and a plasmid-borne recX gene was introduced into the recH342 context. As revealed in Figure 1A, the ΔrecX or recH342 mutation rendered cells sensitive to MMS and H2O2 when compared with rec+ cells. A plasmid-borne recX gene in recH342 cells restored rec+ levels of MMS or H2O2 resistance (Figure 1A), thus the recH342 mutation was renamed as recX342. This is consistent with the observation that the physical mapping of recX gene, at 79° [57], is in good agreement with the genetic map of the recH342 mutation, in the tre - glyB region (82° interval) by PBS1 transduction [58].
To ascertain the role of RecX in RR, we transferred the ΔrecX mutation into strains lacking RecA accessory proteins (e.g., recO, recF) and assayed the ability of these strains to resist the acute exposure to MMS or H2O2. The ΔrecA strain was used as control. Upon exposure to varying concentrations of DNA damaging agents, the ΔrecX strain was moderately sensitive to both drugs when compared with the very sensitive recF15 or ΔrecO strains or the extremely sensitive ΔrecA strains (Figure 1A and 1B). The absence of both RecX and RecO or RecX and RecF, increased the sensitivity of the double mutant strains (Figure 1B) to the levels of ΔrecA (Figure 1A). It is likely that ΔrecX is not epistatic with ΔrecO or recF15 (classified within the α group), and that the three functions are essential for RecA-mediated DNA strand exchange.
Genes others than recA, which are exclusively involved in HR have been placed into seven different epistatic groups. Except recX342 (epistatic group γ) and recF and recO (α group) described above, the different epistatic groups and the genes included within them are: addA and addB (β); recN (δ); ruvA, ruvB and recU (ε); recJ, recQ and recS (ζ) and recG (η) (Figure S2) [19]. The ΔrecX mutation was moved into a representative of each epistatic group [ΔaddAB (β), ΔrecN (δ), ΔrecU (ε), ΔrecJ (ζ) or ΔrecG (η) epistatic group] (C.E.C, G. Garaulet, C. Marchisone and J.C.A., unpublished results). The single and double ΔrecX mutant strains were assayed to resist the acute exposure to MMS and H2O2 or the chronic exposure to mitomycin C (MMC) and H2O2. As previously shown for the recX342 mutation [52], [56], [59], ΔrecX was neither epistatic with ΔaddAB (β), ΔrecN (δ), ΔrecU (ε), ΔrecJ (ζ) nor with ΔrecG (η epistatic group) (C.E.C., G. Garaulet, C. Marchisone and J.C.A., unpublished results).
The recX gene shows a high ubiquity among Bacteria [53]. It is predicted to be missing only in bacteria of the Cyanobacteria and Chlamydiae Phyla and in some Classes of the Proteobacteria (e.g., α-Proteobacteria), Firmicutes (e.g., Mollicutes) or Spirochetes Phyla. The RecX orthologues (197 orthologues analyzed) showed only a limited degree of identity among Phyla, but a high degree of identity was observed between the different Classes of the same Phylum [53], [60]–[62], suggesting a high divergence or more than one possible evolutionary pathway.
The RecX orthologues were classified using a length bias criterium (Figure S1). With few exceptions (e.g., RecX of the Yersinia Genera that are significantly longer, >180-residue long polypeptide), RecX of the Proteobacteria Phylum (87 orthologues analyzed) are relatively small proteins (<170-residue long polypeptides), and share a significant degree of identity (>25%) among them [53], [62] (Figure S1). The structure of RecXEco revealed that it is a modular protein consisting of three tandem repeats of a three-helix motif (R1α1-3, R2α1-3 and R3α1-3) (Figure S1) [63]. These RecX proteins can be further divided in two subgroups represented by E. coli (Eco) and N. gonorrhoeae RecX (Ngo) (Figure S1). The recX gene of the former group is located immediately downstream of recA, forming a single transcriptional unit as in E. coli. RecX of the latter group, which is not part of the SOS response, is located elsewhere in the genome, as in Neisseria ssp. [60], [61].
RecX of the Actinobacteria Phylum (15 orthologues analyzed), represented by Mycobacterium tuberculosis (Mtu) RecX, are middle size proteins (171- to 188-residue long polypeptides) that share a significant degree of identity (>40%, ClustalW2 alignment) among them. RecXMtu shares a higher degree of identity with a large (e.g., RecX, ∼20%) than with a small RecX (e.g., RecXEco, ∼11%) protein, with RecX and RecXEco sharing a very low, ∼15%, overall identity (Figure S1).
RecX of the Firmicutes Phylum (83 orthologues analyzed), represented by B. subtilis RecX (Bsu, a 264-residue long polypeptide), are large proteins (212- to 272-residue long polypeptides) that share a significant degree of identity (>30%) among them. Inspection of the genetic organization around the Firmicutes recX gene, however, discards any conservation on the genome context, even within the closely related Classes of the Phylum. Examination of the amino acids sequence of RecX342 revealed that the conserved L101 (encircled) of the predicted α-helix 3 on repeat 1 (R1 α3, [63]) was substituted by P (L101P) (Figure S1).
From these data altogether it could be assumed that the small, middle and long proteins are distantly related classes that perform a similar function, and that longer proteins might have an additional uncharacterized function at the C-terminal region. However, secondary structure prediction of Firmicutes RecX revealed that the C-terminal region, which shares no significant identity with any domain of known activity, might fold as three tandem α-helix motifs. Examination of the amino acids sequence of the 43 C-terminal residues of RecX (residues 221–264) revealed significant identity with an internal region of few RecX orthologues (e.g., Provotella oulorum, 38% 15/40 residues). In P. oulorum RecX, this region aligned with the region of RecXEco that forms the R3α2 and R3α3 motifs (data not shown), suggesting that Firmicutes RecX, which seems to lack R1α1 and R1α2 helices when compared to RecX of different Phyla (Figure S1), might also consist of three tandems repeats of a three-helix motif.
In vivo analyses of B. subtilis cells revealed that: i) in response to DNA damage, RecA-dependent autocleavage of LexA triggers the SOS induction reviewed in [37], ii) expression of recX gene is independent of MMC-induced SOS response [64], iii) recA promoter utilization is reduced and delayed in recF15, recO16 or recR13 cells upon MMC addition [40], and iv) the interstrand crosslinks produced by MMC, as most lesions, are removed by nucleotide excision repair (NER) prior to DNA replication, but unrepaired damage induces the SOS response and then repair includes translesion synthesis and HR [65]. To determine whether a recX mutation has an effect on the levels of SOS response, the rec+, ΔrecX, recX342 or ΔlexA cells were exposed to increasing MMC concentrations, the cultures were harvested 30 min later and the levels of RecA protein, expressed from its native locus and promoter, were measured. Equivalent amounts of crude extracts proteins were separated by SDS-PAGE, transferred and blotted against polyclonal antibodies raised against RecA. Serial dilutions of purified RecA were used as concentration standard.
The absence of RecX (ΔrecX) or the presence of the RecX342 or RecF15 variants did not affect the basal level of RecA when compared with rec+ cells (estimated to be ∼4,500 monomers per cell, or ∼6 µM assuming an average cell volume of 1.2 fL) (Figure 2A and 2B). The absence of RecO, however, slightly reduced it, and the absence of LexA rendered constitutive RecA levels (Figure 2A and 2B).
The RecA protein reached its maximal level at ∼0.6 µM MMC, and maximal induction caused 4- to 6-fold increase in net RecA in the rec+ context [66], Figure 2A. As expected, in the absence of the LexA repressor, the level of RecA was comparable to levels detected upon full SOS induction (≥0.6 µM MMC) in the rec+ context (Figure 2A). In the absence of RecX, a significant net RecA accumulation (∼18,000 monomers per cell) was observed upon exposure to MMC concentrations as low as 0.07 µM. Similar results were observed in the recX342 context (Figure 2A). This reduced threshold for SOS response upon MMC addition in recX could be attributed to the lack of negative regulation of RecA. This is consistent with the observation that 0.07 µM MMC concentration neither compromised cell proliferation nor cell plating efficiency in the rec+ and ΔrecX context (data not shown). Furthermore, this SOS induction did not significantly contribute to error-prone repair by translesion synthesis (Text S1, Annex 1) in the ΔrecX context.
In vitro studies revealed that: i) RecXEco blocks RecAEco filament extension reviewed in [22], and ii) RecXEco physically interacts with RecAEco and RecFEco [29], [67]. To test the effect of the absence of the RecX and RecF functions in SOS response, the levels of RecA were measured 30 min after MMC addition. The recF15 mutation reduced net RecA accumulation when compared to rec+ cells and higher concentrations of MMC were needed to reach full induction. The absence of RecX reversed the effect of the recF15 mutation on the level of RecA, with RecA levels comparable to rec+ cells (Figure 2B). It is likely that: i) the activation of RecA as a coprotease (RecA filamented onto ssDNA), to facilitate self-cleavage of LexA, is modulated by RecX and RecF in response to MMC addition; ii) in the absence of RecX and RecF there is not net change in RecA induction (RecA·ssDNA filament formation), with one counteracting the activity of the other; and iii) the RecA filaments formed in the absence of both RecX and RecF are sufficient for SOS response (Figure 2B), but are not proficient for RR (Figure 1B) and GR (Table 2), suggesting that both modulators are necessary to avoid potential hazards that could be caused by miss-regulation of HR.
In the absence of RecO addition of MMC slightly increased (∼1.2-fold) the RecA levels when compared to the non-induced control (Figure 2B). The absence of RecX slightly increased the RecA levels in the ΔrecO context (Figure 2B). It is likely that in the absence of the RecO mediator there was a marginal increase in the nucleation of RecA protein filaments, and the RecX modulator poorly contributed, if at all, to RecA nucleation in the ΔrecO context (Figure 2B).
In response to DNA damage RecO, RecR and RecA form foci at 30 to 45 min, followed by RecF at 60 min after damage [31], [32]. To gain insight onto the mechanism by which RecX modulates RecA functions, RecX was visualized in cells grown in minimal medium at 25°C (Figure 3). A strain bearing a C-terminal fusion of RecX to YFP (RecX-YFP) was constructed (Text S1, Table S1). The RecX-YFP fusion, integrated in its native locus was fully functional. The growth rate (data not shown) and the observed survival curve, upon acute exposure to increasing concentrations of MMC for 15 min, of rec+ and recX-yfp cells were similar. As observed with other DNA damaging agents (Figure 1), the ΔrecX strain was moderately sensitive to varying concentrations of MMC, and the recF15 ΔrecX was extremely sensitive (Figure S3).
Microscopic observation of the strain in exponential growth revealed dispersed localization of RecX-YFP throughout the cells (Figure 3A), whereas this pattern of localization changed dramatically upon MMC (0.15 µM) addition. In ∼40% of the cells, RecX was concentrated into distinct foci on the nucleoid (mostly two per cell, but sometimes up to 5, 300 cells analyzed) 60 min after the addition of MMC (Figure 3B), and in 52% after 120 min (300 cells analyzed) (data not shown). Upon DNA damage, RecX was localized as distinct foci after RecA-induced foci formation, which suggests that RecX acts after RecO, RecR and RecA, and concomitant with RecF. The number of RecX foci decreased (Figure 3C) 180 min after addition of MMC, until foci were no longer detectable, and growth of cells slowly resumes.
Since biochemical [29], [38], [39] and structural analysis [63], [67] have shown that RecXEco interacts with the RecAEco·ssDNA filament, we set out experiments to visualize both proteins in living cells (Figure 4). Cells bearing a N-terminal fusion of RecA to CFP (CFP-RecA) were previously described (Text S1, Table S1) [32]. Microscopic observation of the strain in exponential growth revealed dispersed localization of RecX-YFP throughout the cells (Figure 3A and Figure 4A), and CFP-RecA throughout the nucleoid [32], Figure 4A. In response to DNA damage CFP-RecA formed foci at 30 to 45 min, followed by RecX at 60 min. Sixty min after MMC addition the RecA foci started to be more and more condensed, and then formed highly dynamic filamentous structures (Figure 4B). The formation of dynamic thread-like structures of CFP-RecA was maximal at 120 min after addition of MMC, as well as the number of cells containing RecX-YFP foci, which generally co-localized with RecA threads (Figure 4C). From 350 analyzed cells, RecX-YFP foci localized at or near the RecA signals in 41% of the cells (that is in 91% of all cells showing RecX-CFP foci), adjacent to RecA signals in 3% of the cells, or clearly did not co-localize in 1% of the cells; 55% of the cells showed CFP-RecA fluorescence, but no detectable RecX-YFP foci. Between 120 and 180 min, CFP-RecA threads became fewer in number and thus apparently disassembled, until about 180 min, repair was terminated and RecA threads were no longer visible (Figure 4D, central panels). The number of cells containing clear RecX-YFP foci also decreased in a time dependent manner up to 180 min after induction (still co-localizing with subcellular locations of high CFP-RecA signals) (Figure 4D), after which foci declined to negligible levels (0.7% of the cells showed foci, 280 analyzed cells). Growth resumed ∼180 min after the initial DNA damage in rec+ cells.
In vivo analyses revealed that: i) RecN is recruited to a defined DSB [32], [68], and ii) RecO, RecR, RecA and RecF co-localize with DNA damage-induced RecN focus [31], [32]. To test whether RecX was also recruited to a defined DSB, a strain bearing a xylose inducible promoter transcribing the HO endonuclease, an HO cleavage site and a lacO site, both integrated close to the oriC region, and the lacI-cfp cassette ectopically integrated at the threonine locus, was constructed as previously described (Text S1, Table S1) [32].
After induction of the HO endonuclease a single two-ended DNA break was induced and RecX-YFP foci were observed in about 10% of the cells (350 analyzed cells). The observed RecX foci were generally not coincident with the oriC (LacI-CFP) signal, only 1 out of 34 foci was coincident with an oriC signal (Figure 4E). These experiments show that RecX is not directly recruited to sites of DSBs.
Biochemical studies have shown that RecXEco (RecXNgo) blocks RecAEco assembly onto ssDNA tracts [29], [38], [39] or facilitates a more rapid RecANgo filament disassembly [49]. To test whether RecX affects RecA foci formation (“nucleation”) and/or thread assembly or disassembly (“filament formation”), the localization of a functional CFP-RecA fusion in rec+ and in the ΔrecX strain was monitored. For all times points, 350–400 cells were analyzed. During exponential growth, RecA localized throughout the nucleoids in both, rec+ (Figure 5A) and ΔrecX cells (data not shown). The absence of RecX neither affected the formation of RecA foci nor the assembly of RecA threads between 60 and 90 min after induction of DNA damage (data not shown), and for the first 120 min following the addition of MMC, no obvious difference in the formation of CFP-RecA threads was detectable between rec+ and ΔrecX cells, 75 to 85% of the cells contained CFP-RecA threads (Figure 5B and 5C).
The number of cells showing threads decreased in rec+ cells to less than 50% after 120 min, while 80% of all ΔrecX cells continued to contain RecA threads (Figure 5C). After 180 min only ∼4% of rec+ cells contained visible CFP-RecA threads (many contained CFP-RecA accumulations at a single cell pole), while these structures persisted in 95% of the recX mutant cells (Figure 5D). Even after 210 min, CFP-RecA threads were visible in >50% of mutant cells, while in rec+ cells, RecA was again spread uniformly on the nucleoids, and thread structures were only observed in 1.3% of the cells (Figure 5E). These data reveal that RecX is necessary for the down-regulation of RecA threads, which most likely consist of RecA·ssDNA filaments. It is likely that the balance between RecX and RecF governs the dynamics of RecA threads, but at late times, when the DNA damage signal is removed, the thread-stabilizing activity of RecF might be negatively modulated by an uncharacterized function(s), leading to net thread-destabilizing activity of RecX.
To gain insight into the chromosomal gene transfer barriers (see Introduction), the fate of RecX during GR was analyzed. At the onset of stationary phase, only 10%–20% of cells stochastically develop time-limited competence in response to specific environmental conditions. Natural competence is a genetically programmed process with a specialized membrane-associated machinery for uptake of exogenous dsDNA that subsequently processes and internalizes ssDNA into the cytosol (DNA uptake machinery) [69]. Previously it has been shown that some soluble proteins of the recombination apparatus (namely SsbB, DprA, RecA, RecU and CoiA) are located at the cell poles, where they co-localize with the DNA uptake machinery [7]–[11]. To understand the role of RecX on GR its localization was analyzed upon competence induction. Microscopic observation of RecX-YFP in cells grown to competence revealed fluorescent foci in 8 to 10% of cells (1260 cells analyzed) (Figure S4), suggesting that this is the proportion of competent cells. In the 8–10% of the cells RecX-YFP existed mainly as one focus per cell (73% of the cases), sometimes localizing to a single cell pole (∼27%), but mostly at midcell in the nucleoid (∼46%) (Figure S4). Less often, two (∼17%) foci (mostly one at the pole and one at the nucleoid), three foci (∼5%), and patched structures (∼5% of the cases) were observed.
To investigate the nature of RecX-YFP foci, we performed time-lapse microscopy, capturing images of cells grown to competence without DNA, or 30 min after addition of DNA, with 2 s time intervals (Figure S5). Irrespective of the presence or absence of DNA, RecX-YFP foci at midcell were dynamic and moved between acquisitions (note that the signal was weak, so only few frames could be captured), while foci at or near the cell pole did not move away from their position. The total number of fluorescent cells (7% of total cells in the absence of DNA) did not change markedly after addition of DNA (to 10%), but the number of cells having one discrete focus increased from 73% to 82% of the cells containing a signal 30 min after DNA addition. With increasing time upon DNA addition (0 to 30 min) the number of RecX-YFP cells with one focus at the pole decreased (from 27% to 13%) and the number of cells with one focus on the nucleoid augmented from 46 to 72% (at least 1000 cells were analyzed for each time point) (Figure S5). Upon addition of DNA, the number of cells with more than one RecX-YFP focus and with patched structures became lower to less than 1.5% of the competent cells. Figure S5, movie A, shows a polar focus that moved around the cell pole, but remained there, movie B shows a cell with many foci that moved, and movie C shows a central focus that moved. Thus, polar foci are usually static, possibly representing RecX that is associated with the DNA uptake machinery or any associated recombination protein, and non-polar foci are very dynamic.
B. subtilis competent cells can take up DNA of any source and the transfer of chromosomal genes requires HR. The frequency of appearance of met+ chromosomal transformants in single rec-deficient strains, classified within the α (recF15 or ΔrecO), β, γ (recX342), δ, ε, ζ or η epistatic groups (Figure S2), does not vary more than 3-fold relative to the rec+ value [12], [13], [15], but is blocked in the ΔrecA context (Table 2). It is likely that a certain redundancy exists and/or that the critical functions were not studied yet. The absence of RecX severely impaired chromosomal transformation (∼200-fold) with respect to that of the rec+ strain (Table 2).
To establish the potential contribution of RecO and/or RecF in the recX context, the capacity of these cells to be transformed with chromosomal DNA was measured. Chromosomal transformation was inhibited >1,000-fold in the recF15 ΔrecX or ΔrecO ΔrecX context, when compared to the absence of RecA that blocked it (>10,000-fold) (Table 2). It is likely that distinct effectual length or packing density of the RecA·ssDNA filament is necessary for chromosomal transformation (Table 2).
In the absence of homology with the recipient DNA, a linear ssDNA oligomeric plasmid molecule requires RecO, DprA and RecU for establishment [9], [10], [15], [70], [71]. Plasmid transformation, however, was only marginally affected relative to the rec+ strain in the ΔrecA context or in single rec-deficient strains classified within the α (recF15, ΔrecR), β, γ (recX432), δ, ε (ruvA2, ΔruvB), ζ or η epistatic groups (Figure S2) [9], [13], [15], [72]. Plasmid transformation was markedly reduced in the absence of RecX (Table 2). Based on results described above and the observation that the RecA·ssDNA filaments might open an unproductive avenue that is deleterious for plasmid transformation [9], [71], it is hypothesized that RecX might be required to dislodge RecA from the internalized ssDNA. If the hypothesis is correct the absence of RecA should suppress the need for RecX. To test this hypothesis the ΔrecX ΔrecA strain was constructed. The absence of RecA partially suppressed the RecX requirement for plasmid transformation, but as expected it remained blocked in chromosomal transformation (Table 2). It is likely that the RecA·ssDNA filaments might be deleterious for plasmid transformation and RecA modulators, namely RecX (Table 2) and RecU [9], [71], are required to catalyze RecA·ssDNA filament disassembly or to block filament assembly. Alternatively, in the absence of both RecA and RecX proteins, an uncharacterized recombination pathway (specific for plasmid transformation) becomes operative. To test this hypothesis, the effect of the absence of RecX and RecO, or RecX and RecF in plasmid transformation frequencies was measured after construction of the respective strains. Both chromosomal and plasmid transformation were blocked in the ΔrecX recF or ΔrecX ΔrecO context (Table 2), it was therefore considered unlikely that an uncharacterized recombination pathway exists. A similar inhibition of GR was observed in the recF15 recH342 (recX342) or ΔrecO recH342 (recX342) strain (Table 2) [15], [52].
The recH342 mutation (γ epistatic group), which leads to decreased interspecies gene transfer, maps in the putative recX gene (Table 1). The absence of RecX renders cells deficient in RR, hence recX was considered a genuine recombination gene, and the recH342 mutation was renamed as recX342. The classification of ΔrecX mutation within the epistatic group γ was confirmed by combination of the ΔrecX mutation with representative members of different epistatic groups (Figure S2, Text S1, Annex 2). As expected from data derived with recX342 [52], recX was neither epistatic with recF, recO (Figure 1) and recR (α), addA, addB (β), recN (δ), recU, ruvAB (ε), recJ, recS and recQ (ζ), recG (η epistatic group) nor with ku (a function implicated in non-homologous end-joining) (C.E.C., G. Garaulet, C. Marchisone and J.C.A., unpublished results).
In all systems with a genuine SOS response system, RecA needs to be recruited onto SSB-coated ssDNA tracts at a blocked replication fork reviewed in [19], [22]. We show that in the absence of RecX (presence of RecO and RecF), low doses of MMC, which did not significantly affect the cell doubling time, are sufficient to induce the SOS response (Figure 2A). In contrast, in the absence of RecO, high lethal concentrations of MMC were needed to marginally induce RecA synthesis when compared to rec+ cells, and in the absence of both, RecO and RecX, the synthesis of RecA was marginally facilitated when compared to ΔrecO cells (Figure 2B). It is likely that: i) RecO is essential for RecA nucleation and SOS induction, and the absence of RecX cannot override such defect; ii) RecF [31] and RecX (Figure 3 and 4) act after RecO as deduced from the cytological studies; and iii) the RR and GR deficiency in the ΔrecO ΔrecX context could be attributed to decreased RecA·ssDNA nucleation and filament dynamics. This is consistent with the in vitro observation that RecO (RecOREco) protein(s) contribute to RecA (RecAEco) filament nucleation onto SsbA- (SSBEco)-coated ssDNA and that RecOREco is unable to counteract the inhibitory effects of RecXEco on RecAEco filaments [22], [29].
In the absence of RecF (presence of RecX and RecO) high lethal concentrations of MMC were needed to marginally induce RecA synthesis when compared to rec+ cells. In the absence of both RecA modulators (RecX and RecF), however, the levels of RecA induction showed a profile similar to rec+ cells (Figure 2B). It is likely that there is a cross talk among RecX and RecF and one might antagonize the effect of the other, and that RecO and RecF contribute to RecA filament formation at different stages (Figure 2B). Conversely, in E. coli the formation of RecA filaments decreased in the ΔrecX strain, and this RecX-mediated destabilization of the RecA·ssDNA filaments was independent of RecFOR [30], [38], [73], and overexpression of RecX [38] or RecF [74] inhibited induction of the SOS response after UV irradiation.
The RecA·ssDNA filaments formed in the recF15 ΔrecX context are sufficient for SOS induction, but are not proficient for HR (Figure 1 and Table 2). It is likely, therefore, that there are different layers of regulation of RecA·ssDNA filament extension with RecF and RecX acting at different levels, and as biological antagonists. How can we rationalize this observation? It is likely that there are different types of RecA·ssDNA filaments. In the ΔrecX context, a high local RecA concentration may generate a saturated RecA·ssDNA filament, resulting in a higher affinity for LexA. In the rec+ or ΔrecX recF15 cells, sub-saturated RecA·ssDNA filaments may be formed at low MMC dose, and in these conditions RecA equilibrates among the existing DNA lattices lowering its affinity for LexA. Here, a higher MMC dose was required to induce the SOS response when compared to the ΔrecX context. Alternatively, the length of the RecA·ssDNA filament, rather than the packing, is a crucial factor in the rate-limiting step of homologous pairing [34], [49].
DNA damage-induced RecA formed discrete foci on the nucleoid ∼30 min upon MMC addition [32]. RecX formed discrete foci on the nucleoid and a RecA focus started to be converted into irregular RecA threads after 60 min of MMC addition (Figure 4). RecA thread formation was demonstrated to be independent of RecF [31], [32] and RecX (Figure 4 and Figure 5). Within 60–120 min upon DSB induction the RecX-YFP foci increased in number, and the RecA threads were dynamic filamentous extensions, to dissociate from the DNA and become undetectable at about 180 min in rec+ cells grown in minimal medium at 25°C (Figure 5). In the absence of RecX, RecA threads persisted for an extended period of time (Figure 5), revealing that RecX affected the dynamics of RecA threads (RecA·ssDNA filament disassembly) rather than RecA foci (RecA nucleation) and thread formation (RecA·ssDNA filament extension). Conversely, in response to UV irradiation the formation of E. coli RecA foci decreased in the ΔrecX or recF4115 context [30]. Unlike the RecO, RecA and RecF foci, which co-localize with RecN and with HO endonuclease-generated DSBs, the RecX foci did not co-localize with DNA DSBs (Figure 4). Consistent with its activity on RecA threads, RecX formed foci after RecA nucleation and co-localized at or near RecA threads (Figure 4). We propose that RecX has two activities: to counteract the activity of RecF, and to mediate RecA·ssDNA filament dislodging. These activities are essential for rendering a dynamic balance between RecA assembly and disassembly from ssDNA, in order to form an “active RecA·ssDNA filament” and to facilitate the inherent DNA pairing activity of the formed filaments during the search for homology.
Plasmid transformation is a RecA-independent process [72]. It was previously postulated that RecA-bound to the incoming oligomeric linear plasmid ssDNA was deleterious [71]. If the working hypothesis is correct, the RecA·ssDNA filaments should be disassembled, either by RecX (Table 2) or by RecU [71], for plasmid transformation. We have shown that RecX is necessary for plasmid transformation, and such requirement is overcome by the deletion of recA (Table 2), rather than by the opening of a new and uncharacterized recombination avenue.
During natural transformation donor DNA enters the cell as ssDNA molecules, hence end processing is not necessary. Chromosomal transformation is strictly dependent on the presence of a fully functional RecA·ssDNA filament to catalyze intermolecular recombination between the incoming ssDNA and the homologous duplex recipient DNA without significant DNA synthesis [12], [75]. Competence-induced RecX formed discrete foci at the entry pole and on the nucleoid in the absence of DNA. Upon addition of dsDNA, SsbB (in concert with SsbA) at the entry pole protects the incoming ssDNA and limits RecA loading onto SsbA- and/or SsbB-coated ssDNA [18]. RecO (or DprA) displaces SsbA and/or SsbB and recruits RecA onto SsbA- and/or SsbB-coated ssDNA leading to RecA foci formation at the entry pole [8]–[10], [18]. RecX at the entry pole and in the nucleoid should modulate RecA thread formation (RecA·ssDNA filament extension). RecA forms a filament (thread) extending from the pole to the centrally located nuclear body [8]. A RecA thread might facilitate the search for a homologous segment and mediate joint molecule formation (heteroduplex) with the resident chromosome [9]. The frequency of chromosomal DNA transformation marginally decreased in the recX342 background, but dropped to low levels (∼200-fold) in the ΔrecX context (Table 2). Studies in other bacteria are less clear, because the absence of RecXDra, results in elevated chromosomal transformation frequencies (∼2.5-fold increase) [76], whereas the absence of RecXNgo also decreases recombination frequencies, although the effect was modest (∼5-fold reduction) [60]. In vitro RecXNgo destabilizes RecANgo·ssDNA filaments by causing its local instability [49]. Although B. subtilis and N. gonorrhoeae RecX proteins might have similar functions, there are also important differences. For example, in the latter species the presence of RecF is not obvious [53], suggesting that dynamic RecA·ssDNA filament formation is modulated in this bacterium by different partners than those found in B. subtilis cells.
Cohan and coworkers [6], [77] reported that the frequency of unidirectional transfer of DNA between donor and recipient (chromosomal transformation) in B. subtilis rec+ cells decreases with increased sequence divergence. Here, each 5% increment in sequence divergence yields ∼ a 10-fold decrease in chromosomal transformation [6], [77]. The recX342 (BG119) competent cells showed a >20-fold increased sensitive to sequence divergence than when transformed with homogamic DNA [6]. However, there is no measurable effect in preventing inter-species transformation in the recF15 (BG129) or recO16 (BG107) context. Furthermore, the frequency of intra- or inter-species transformation dropped to undetectable levels in the ΔrecX recF15 or recX342 recF15 context (Table 2), leading to populations with a clonal structural potential. We propose that the RecA·ssDNA filament forms a metastable reversible intermediate, whose dynamic modulation is governed by RecF and RecX and predict that the RecX342 variant makes recombination initiation (and also termination) very sensitive to sequence divergence. A RecA·ssDNA filament to initiate recombination between donor and recipient DNA requires a MEPS of 25- to 35-base pairs to initiate recombination between donor and recipient DNA [78]. A similar extent of sequence identity was reported for viral-mediated plasmid transduction [79], [80]. It is likely that the effectual “length/packing” of the RecA·ssDNA filament and the MEPS needed for proficient inter-specific recombination are more stringent than for intra-specific recombination, and that RecX and RecF prevent the dissociation of potentially unstable heteroduplex intermediates that are essential for HR (Figure 1 and Table 2).
The RecXEco protein contributes modestly to recombinational potential reviewed in [22]. In vitro, the ability of small RecX (<180 residue long, e.g., RecXEco) to inhibit RecAEco-associated activities, i.e. ATPase, strand-exchange, and LexAEco cleavage activities [38], [63], [67], have prompted different groups to categorize RecXEco as a major negative regulator of RecAEco activities. The current model postulates that RecXEco bound to the deep helical groove of the RecAEco nucleofilament blocks RecAEco assembly onto ssDNA, leading to filament destabilization and inhibition of HR. Based on previous data reviewed in [19] and the ones presented here, we propose that RecX is necessary to avoid potential hazards that could be caused by miss-regulation of HR.
We propose that SsbA, at the ssDNA in a stalled replication fork, interacts with RecO and loads it (or RecOR) onto SsbA-coated ssDNA (Figure 6, step b) [70]. Then, RecO loads few RecAEco monomers onto ssDNA with a limited displacement of SsbA and/or SsbA·RecO (RecR) complexes (Figure 6, step c) [47]. The role of RecF in RecA nucleation is unclear, because RecF focus formation is impaired in the absence of RecO [31], and DNA damage-induced RecF foci, which co-localize with a RecA focus, are clearly detected after RecA focus formation, and concomitantly with RecA thread formation and RecX foci formation (Figure 5) [31], [32]. In E. coli cells, RecOR or RecFOR, upon interacting with SSB, loads few RecA monomers onto ssDNA with a limited displacement of SSB [46], [81], [82], [83].
Regulation of the cycle of the assembly and disassembly of RecA is achieved through the hydrolysis of (d)ATP, whereas the extension (formation of RecA threads) is controlled by RecA modulators (e.g., RecX, RecF, RecU). Based on the data presented here and the in vitro data of Lusetti and co-workers [29] we propose that there might be a crosstalk between RecF and RecX in the modulation of the RecA·ssDNA filament extension (Figure 6, step d). In the absence of RecX, RecF directly or indirectly could either facilitate RecA·ssDNA filament assembly or slow down disassembly, facilitating RecA·ssDNA filament formation even in the presence of a low DNA damage signal (Figure 6, step e). These “long” and/or saturated RecA·ssDNA filaments lead to premature induction of the SOS response at low-dose exposure of MMC (Figure 2), but these RecA·ssDNA filaments are neither proficient for RR nor for GR (Figure 1, Table 2). Conversely, in the absence of RecF, RecX directly or indirectly promotes RecA·ssDNA filament disassembly or delays filament assembly, leading to “short” and/or sub-saturated RecA·ssDNA filaments (Figure 6, step f). In the absence of both RecX and RecF, however, there is no apparent net change in the “activation” of the RecA·ssDNA filament for SOS induction, but RecA-mediated DNA strand exchange (RR and GR) is markedly impaired. We favor the view that RecX and RecF, by fine-tuning of the dynamic assembly/disassembly of RecA, facilitate the accumulation of a RecA·ssDNA filament with an effectual length or packing, as long as the DNA damage signal is on. In vitro, RecXEco destabilizes RecAEco·ssDNA filaments by either preventing growth of the filament [39] or by causing its local instability [63], and RecFEco protects RecAEco assembly by antagonizing the negative modulator RecXEco, specifically during the extension phase [29]. However, in vivo the number of RecAEco foci decreased in ΔrecXEco, but increased in the recFEco context [30].
B. subtilis strains and plasmids used in this work are listed in Table S1. All strains were isogenic to BG214 or PY79 and were grown at 37°C on LB rich medium, unless otherwise indicated. E. coli XL1-Blue cells (Stratagene) were used for routine cloning. Cells were grown in Luria Bertani (LB) medium supplemented with ampicillin (100 µg ml−1 Amp) or chloramphenicol (30 µg ml−1 Cm) as required.
All enzymes were purchased from New England Biolabs. Oligonucleotides were ordered from Sigma-Aldrich. To construct the RecX-YFP fusion, the C-terminal region of the recX (also termed yfhG) gene was amplified by PCR and cloned into a plasmid carrying a downstream yfp gene. The resulting plasmid was then transformed into a PY79 strain, where it integrated at the original gene locus by single-crossover integration. The ΔrecX strain was constructed by inserting the six-cat-six cassette at the 5′-end of the recX gene as described [59]. In short, oligonucleotides pairs CGGATATCGGATCATCTGG and GTAATCGTTAAGCCTATGGATG and CGACAGCCATTGGACATATGTC and GATAGATATCGCCATCAGCCCAAG were used to PCR amplify with Vent polymerase (New England Biolabs) fragments spanning a region 721-bp and 719-bp upstream and downstream from recX, respectively, and overlapping at the start of the recX sequence. StuI digestion of theses fragments, followed by ligation resulted in an EcoRV fragment which contains a deletion involving the 5′ end of recX, and that can be cloned into the same site of a pGEM-T vector (Promega), giving rise to pCB788 plasmid. StuI-cleaved pCB788 was joined to the six:cat:six cassette from vector pCB266 to give rise to plasmid pCB789. For the construction of strain BG1029, pCB789 was linearised with NotI and transformed into competent B. subtilis cells. Double mutants were constructed by transformation of the isogenic rec-deficient B. subtilis cells (recF15, ΔrecO) with linear pCB789 DNA with selection for CmR (Text S1, Table S1). The cat gene was deleted by β-mediated site-specific recombination to render the BG1065 strain (Text S1, Table S1). The ΔrecA mutation was introduced into BG1065 cells (ΔrecX) by SPP1 transduction [79]. B. subtilis competent cultures were obtained as described previously [12].
Genomic DNA from the Reference B. subtilis rec+ (BG214) strain and the Test recH342 (BG119) strain were sequenced by high-throughput sequence analyzer (Illumina) technology using standard sequencing libraries and filtered sequence data (BGI), of ∼1 gigabases per sample, were used to conduct paired-end nucleotide sequencing with the rec+ BG214 and the BG119 sample as described [84].
Acute survival assays were performed as previously described [66]. Briefly, B. subtilis cells were grown to an OD560 = 0.4 at 37°C in LB broth, and exposed to different concentrations of MMS, H2O2 or MMC. After 15 min (with MMS or H2O2) or 30 min (with MMC) exposition, cells were diluted and plated on LB agar plates. Colony forming units (cfu) were counted and plotted against the concentration of damaging agents, in order to obtain survival curves.
For DNA transformation experiments, B. subtilis competent cells were transformed with 100 ng of either SB19 chromosomal DNA to met+ or pUB110 plasmid DNA to kanamycin resistance (KmR). Transformants were plated on minimal medium agar plates containing tryptophan but lacking methionine or on LB agar plates containing Km (5 µg ml−1). Transformation efficiencies were normalized to the number of viable cells plated on rich medium without selection, and the values obtained were normalized against those obtained for rec+ cells.
B. subtilis strains were grown in LB to an OD560 = 0.4 at 37°C with agitation. Then, cells were treated with increasing concentrations of MMC for 30 min. The cells were centrifuged, resuspended in buffer A (50 mM Tris HCl [pH 7.5], 300 mM NaCl, 5% glycerol) and lysed by sonication. Extracts containing equal concentrations of protein from each induction experiment alongside purified RecA protein standard (10 to 500 ng) were separated on 10% sodium dodecyl sulfate-polyacrylamide gel electrophoresis.
Polyclonal rabbit antiserum was raised against purified RecA protein according to standard protocols. Immunoblots were transferred and probed with anti-RecA antibodies as described previously [66]. RecA protein bands on developed immunoblots were quantified with a scanning densitometer (Quantity One software). Purified RecA protein standards yielded a linear relationship between antibody signal and the RecA protein concentration. The amount of RecA protein in each induced sample was interpolated from the purified RecA protein standard curve.
Exponentially growing cells were obtained by inoculating overnight cultures in fresh S7 minimal media and grown to an OD560 = 0.4 at 37°C. Cells were then fixed with 2% formaldehyde, 4′,6′-diamino-2-phenylindole (DAPI) (1 µg/ml) was added for nucleoid visualization, and the cells were analyzed by fluorescence microscopy as previously described [31]. Mid-log phase cells were either untreated or exposed to 0.15 µM MMC for variable time and then fixed as described above. To further investigate the in vivo function of RecX in HR, we visualized RecX-YFP or CFP-RecA in living cells as previously described [31], [32].
|
10.1371/journal.pbio.2001045 | Gaze-informed, task-situated representation of space in primate hippocampus during virtual navigation | To elucidate how gaze informs the construction of mental space during wayfinding in visual species like primates, we jointly examined navigation behavior, visual exploration, and hippocampal activity as macaque monkeys searched a virtual reality maze for a reward. Cells sensitive to place also responded to one or more variables like head direction, point of gaze, or task context. Many cells fired at the sight (and in anticipation) of a single landmark in a viewpoint- or task-dependent manner, simultaneously encoding the animal’s logical situation within a set of actions leading to the goal. Overall, hippocampal activity was best fit by a fine-grained state space comprising current position, view, and action contexts. Our findings indicate that counterparts of rodent place cells in primates embody multidimensional, task-situated knowledge pertaining to the target of gaze, therein supporting self-awareness in the construction of space.
| In the brain of mammalian species, the hippocampus is a key structure for episodic and spatial memory and is home to neurons coding a selective location in space (“place cells”). These neurons have been mostly investigated in the rat. However, species such as rodents and primates have access to different olfactory and visual information, and it is still unclear how their hippocampal cells compare. By analyzing hippocampal activity of nonhuman primates (rhesus macaques) while they searched a virtual environment for a reward, we show that space coding is more complex than a mere position or orientation selectivity. Rather, space is represented as a combination of visually derived information and task-related knowledge. Here, we uncover how this multidimensional representation emerges from gazing at the environment at key moments of the animal’s exploration of space. We show that neurons are active for precise positions and actions related to the landmarks gazed at by the animals. Neurons were even found to anticipate the appearance of landmarks, sometimes responding to a landmark that was not yet visible. Overall, the place fields of primate hippocampal neurons appear as the projection of a multidimensional memory onto physical space.
| Place cells are the quintessential signature of hippocampal neural activity in rodents and code the animal’s position in an environment [1,2]. These neurons’ place selectivity and directionality strongly depend on the visual and/or vestibular cues, as has recently been shown in virtual reality settings in rodents [3–7]. Place cells are observed too in humans navigating virtual environments [8,9] and in other primates in real and virtual environments [10–12]. Yet, there is no consensus on how hippocampal place cells found in monkeys or humans precisely compare to place cells in rodents in the real or virtual world. Previous work [10,11] suggested that, unlike in rodents, space in the primate hippocampus may be coded in a gnostic, landmark-centered representation. Neurons in the monkey hippocampus were shown to convey much more information about the spatial view than about the place, eye position, or head direction. Although there was some modulation of spatial view responsiveness by place [13], the animal’s target of gaze (i.e., what the animal was looking at) was paramount in explaining firing rate [10]. These observations contradict other studies in macaques [14–16] and in humans in virtual reality mazes [8,9] describing robust place-coding activity. In the latter human single-cell studies [8,9], some cells were sensitive to the conjunction of place and goal or place and view, demonstrating complex task-related coding. However, as eye tracking was not feasible in these studies, neural activity was not analyzed with respect to eye movements and visual exploration. Thus, it remains unclear how active vision informs the neural construction of space at the single-cell level in the primate.
In the present study, we probed the nature of hippocampal coding in a goal-oriented task, separating goal and visual landmarks, and examined jointly how cells code for position, direction, and target of gaze. The goal-oriented setting enabled us to examine whether the task-related context of navigation modulated the activity of the cells. We thus analyzed firing in a discrete state space in which the animal’s trajectory in the maze is segmented into elemental transitions from one state in the environment to another [17–19]. Though our results are limited to virtual reality (VR), the recent use of this technique in humans [9] and in rodents [3–7,20] provides an apt comparison of the hippocampal coding in an environment in which spatial information is principally derived from visual input. In this framework, our results give a comprehensive and thorough analysis of the variables controlling the activity of hippocampal cells, bridging the gap between studies in rodents and in primates—including humans—collected in real and virtual environments. We show how hippocampal cells code for the target of gaze in an informed manner, embedding self-position with respect to elements in the environment and to action context. Thereby, we bring a new perspective on models of hippocampal spatial function by focusing on the role of the idiosyncratic visual exploration in primates in constructing a representation of the world that is highly useful to wayfinding.
We trained two rhesus macaques to navigate with a joystick in a virtual 3-D star maze (Fig 1A and 1B, Materials and Methods, and S1 and S2 Movies). The monkeys sought a hidden reward located at the end of one of the five paths, between two landmarks (by convention, the northbound path). For example, on Fig 1B and 1C, the rewarded path is located between the sunflower (northwestern landmark) and the house (northeastern landmark). On each day, new landmarks were used so that the layout was new and unfamiliar. Nothing else in the maze but the landmark layout could be used to infer the reward position because path surface and background were identical across paths. Thus, animals had to start each session without applicable information from past sessions as to the reward's location and learned to find it with respect to new landmarks by trial and error. Each session lasted for 80 trials (± 5 trials). Each trial started at the extremity of a maze path. Animals pushed the joystick to move forward and traveled towards the center (Fig 1C, first panel; a triangle symbolizes the field of view [FOV] of the animal). At the center, animals could rotate the joystick left or right to choose another path to enter (Fig 1C, second panel). The example on Fig 1C shows the monkey entering the rewarded path after a left turn (Fig 1C, third panel). When they reached the end of that path, animals received a juice reward directly in their mouth (Fig 1C, fourth panel). Finally, the animals were reallocated to a different, randomly assigned start (Fig 1C, fifth panel). This latter trajectory did not follow any maze arms, preventing the animals from retracing their steps after a correct choice. Importantly, the landmarks were positioned between the maze arms, thus dissociating the goal from the visual references. In other words, the animals could not directly associate a physical object (landmark) to the reward but had to analyze and memorize the spatial relationships between landmarks and reward. Fig 1D shows the same corresponding five steps as in Fig 1C but from the animal’s perspective (70° horizontal FOV). Overlaid on this view is a representative density heat map of the animal’s point of gaze for 500 ms at each of these steps.
We further computed the animal’s allocentric point of gaze using their virtual self-position and head direction and eye tracking data (Fig 1G). Point of gaze density maps (e.g., Fig 1H, S1 Fig, and S1 Appendix A1, Material and Methods) revealed that gaze was attracted to the rewarded path and the landmarks. During the 500 ms before each action on the joystick (Fig 1D, S1 Fig, S2 Fig, and S1 Appendix A2), gaze anticipated the direction of the subsequent movement (Wilcoxon test, p < 0.001 (S1 Fig, S2 Fig, and S1 Appendix A2). Similarly, when being relocated to a new entry, monkeys proactively gazed at the location at which the landmarks would appear (see S2 Fig and S1 Appendix A2) (Wilcoxon test, p < 0.001). These patterns of visual exploration are similar to ones described in humans when driving [21].
Animals quickly solved the task. On average, both animals learned to reach the rewarded arm in a dozen trials (S3A and S3B Fig); monkey S performed above chance after 12.3 ± 2 trials (SEM), and monkey K did so after 14.8 ± 2.4 trials. From then to when the upper confidence bound of success reached 90%, less than 10 trials were usually necessary (89% of sessions).
To more closely examine the nature of the animal’s spatial representation, we conducted probe sessions (9 sessions for monkey S and 15 sessions for monkey K). In these sessions, animals first started from only one or two entries (northeastern and southeastern entries) and were only later introduced from the new remaining entries (northwest and southwest). We hypothesized that if animals formed a cognitive map of the maze [22], they would successfully transfer knowledge acquired from the previous entries to the new entries (northwest and southwest). Monkey K was 73% correct after introducing the new entries versus 44% correct at beginning of the sessions; monkey S was 80% versus 55% correct (S3C Fig, Wilcoxon, p = 0.01). Thus, after having explored the maze from entries facing eastern landmarks, animals were able to deduce goal location when entering new paths facing western landmarks. Further, performances above the learning criterion were reached significantly faster (within 4 trials for monkey K and within 2.7 trials for monkey S, compared to 14.8 and 12.3 trials; Wilcoxon, p = 0.002).
Thus, animals attended to landmarks and used them flexibly depending on their self-position. Their exploratory behavior in a VR setting appeared similar to that described by others in real-world navigation, as it possessed essential properties of wayfinding such as reliance on landmarks and flexible trajectory planning [22].
Of the 270 cells recorded in the full extent of the right hippocampus (128 cells in monkey S, 142 in monkey K; S4 Fig), we focused on 189 cells that fired more than 100 spikes per session (approximately > 0.01 Hz; see S1 Appendix A3, S3 Fig, and S4 Fig). Only successful trials were considered for analysis.
For comparison with rodent studies, we analyzed neural activity as a function of the animal’s current location in the maze (“Position”) and the current virtual head direction (“Direction”). As vision is paramount in primates, we next examined the impact of the current allocentric point of gaze in the virtual maze (Fig 1D). Then, we considered a fourth explanatory variable by constructing a state-space representation of the maze. State spaces are often used in models of animal navigation [23] and may provide a useful framework to account for hippocampal cell activity [19]. Our state space can be thought of as a logical representation of the task as a graph (Fig 1F), with each node being a choice point (stable state of the animal—e.g., “in the center, facing the northeastern landmark”) and each link corresponding to the state change brought about by an action of the animal. Note that in our task, the animal’s actions are discrete: for example, pushing the joystick forward once is enough to make the monkey travel forward from one end (current state s(t)) to the other end (state s(t+1)) of the current maze arm (similarly, pushing the joystick leftward once is enough to rotate the monkey leftward by 72° if facing a path or by 36° if facing a landmark). Thus defined, this state space describes the resolution of the imposed navigation problem as a series of spatialized action steps. In particular, the state-space graphical representation, by combining the animal’s current view, position, and action, allows distinguishing multiple action contexts for the same position and direction at the maze center.
We chose eight cells to illustrate the diversity of responses across the population (Fig 2). Most cells appear as regular place cells (first column), some with multiple place fields. This might result from the frequent interdependence of direction and position: the strong directional sensitivity of cells 1–5 could account for these fields (second column). However, further examination shows that the strong directional firing could be explained by gazing at specific landmarks (third column). Cells 1 and 2 were recorded simultaneously and exhibited different landmark preferences. Population averages indicate that all landmarks appeared represented by the cells (S6C Fig and S1 Appendix A4). Importantly, this landmark preference was often expressed exclusively along particular segments of the animal’s trajectory. This combination is evident when cells are mapped in the state-space coordinates (fourth column). The graph shows activity on the return paths from reward position to new start as well as activity when the animal is turning in the middle of the maze and passes in front of the landmarks (central rosetta of the state-space graph). This latter activity cannot be visible in the position graph because activity for different directions cancels out when averaged in the center. In this framework, cells 1, 2, 5, and 6 exhibit selective activity for specific segments in the center associated with visible landmarks and/or the animal’s rotations (black boxes in the fourth column). The far right column shows the activity of the cells as raster plots for the corresponding trajectories highlighted either in red or in black. Although most cells showed narrow selectivity and less than three positional fields (S5 Fig), some cells (like 7 and 8) had weak but significant modulation of activity in all four coordinate sets and multiple fields (n = 3 and n = 4, respectively).
We evaluated the four coding spaces using standard measures employed in rodents: the information content per spike (IC) and a sparsity index [3,6,24]. The first quantity documents how much spatial information a spike conveys, while the second expresses spatial selectivity. To compare IC across coding spaces, we ensured that the same number of bins was used across cells and spaces and normalized IC with respect to the average IC of 999 surrogate datasets generated by randomly shuffling periods of spiking activity in time (see Materials and Methods). While this method may be conservative [25], it is widely used in rodent literature (e.g., [1–7]) as it effectively conveys information relative to the spatial distribution of the cell’s activity [26].
Statistical significance was tested by comparing measures derived from actual data with those in the 999 surrogate datasets. Significant IC (p < 0.01) was present in at least one of the spaces for 111 out of the 189 cells (59%, responsive cells). The proportion of cells responsive to each space was significantly unequal (chi-squared test, p = 0.014). The state space accounted for the largest number (84 cells, Fig 3A), and pitting state space against the other three, there were more cells responsive to state space than to point of gaze (chi-squared test, p = 0.012, Bonferroni corrected; the differences between state space and other spaces did not reach statistical significance). Sensitivity to these four variables was not mutually exclusive (Fig 3B).
Note that the simple difference of space dimensionality does not account for this IC difference (S1 Appendix, A5). Population averages of the activity maps (S6A–S6D Fig) revealed that fields across all coding spaces are inhomogeneous, this being likely related to the presence of landmarks (S1 Appendix, A4).
To disentangle direction and position, for each direction-selective cell we compared firing on active center time and peripheral paths to next start. Both IC and sparsity measures were significantly higher in the center, which is the choice point, when compared to return tracks (both p < 0.001). Direction tuning was not maintained across positions, since the correlation between direction selectivity in the center and in the return paths was significantly lower in 20 out of 56 cells (inset in Fig 3C, blue bars). This represents 36% of the population and is higher than what is expected by chance (p < 0.001, binomial test; note that correlation was only significantly higher in 3 cells, consistent with chance: p = 0.16, binomial test). Overall, these analyses confirm on a population level that head direction is rarely if ever coded alone but is combined with variables such as position, point of gaze, and action choices.
How well does a coding space account for cell activity? As a majority of cells were responsive in the state space, we took this space as a reference. Then, for each cell, we compared the difference between the normalized index computed in state space and the corresponding normalized index in another coding space, considering only cells responsive in both spaces. Normalized indexes were obtained by subtracting the average indexes obtained from the surrogate data from the raw indexes in order to prevent any bias due to space structure. The joint distribution of these differences in IC and sparsity is illustrated in Fig 3, where state space is compared to either self-position (Fig 3D), direction (Fig 3E), or point of gaze (Fig 3F). The spike information content for the state space was systematically higher when compared to position, direction, or gaze space (Wilcoxon, all p < 0.01). Sparsity was significantly higher (Wilcoxon test) in state space compared to direction space (p = 0.006) but not significantly higher compared to position space (p = 0.19), although the 15 cells for which the difference was individually significant (as assessed with permutation tests on surrogate data) were sparser in state space (p = 0.001). Conversely, sparsity was significantly higher overall in gaze space than in state space (p < 0.001), but this effect could not be confirmed when considering only 18 cells for which the difference was individually significant (p = 0.68). This pattern of results suggests that cells respond to a combination of variables rather than to one single dimension. This effect cannot be attributed to a modeling bias related to the higher number of dimensions in the state space (S1 Appendix A5 and S7 Fig). As the state space represents every trajectory within each direction in space and a joint action, the results imply that the most informative cells have a fine-grained representation of space that integrates position, direction, and point of gaze into a higher order trajectory representation.
A large part of the state-space graph corresponds to unique combinations of self-position and head direction. To distinguish a joint representation of position and direction from an actual state-space representation that includes trajectory context, we singled out the part of the state space corresponding to the maze center. There, activity corresponds to a single position and direction but differing rotations (Fig 4A), so we could define a state-space selectivity index based on the normalized differences between trajectory moves for the same head direction (see Materials and Methods). Only 4 cells showed significantly low state-space selectivity, as expected by chance (p = 0.81, binomial test), but 12 cells out of 111 (11%, p = 0.01, binomial test) showed significantly high state-space selectivity, indicating they were also sensitive to the current action of the animal (Fig 4B, red bars). The activity maps of two such context-dependent cells are shown on Fig 4C and 4D, wherein the same joystick move is present at least three times without similar accompanying firing patterns. These cells could encode an abstract representation of the maze, comprising sensory aspects of self-position and direction with respect to landmarks as well as contexts of current (and previous or upcoming) actions.
Landmarks are the only cues available to the animal to get his bearings, and as expected, population averages of the activity maps (S6C Fig) show that gazed-upon landmarks elicit strong responses. Accordingly, Fig 2 shows that some cells (cells 1, 2, 3, and 6) exhibit increased activity when the animal was looking at a specific landmark. However, the cells also seemed to exhibit a modulation of their activity from different positions with respect to the viewed landmark. To quantify systematically whether cells exhibited landmark preference and how this interacted with other variables such as distance from landmark, we compared the activity to the four different landmarks in four intervals of relative distance with respect to the landmark (4 x 4 factorial design, see Materials and Methods). In this analysis, we excluded the activity on the peripheral paths as landmark position, and aspect was not stable in the scene. Thus, we selected only the portion of trajectories on the start paths for which the relative distances were identical across each landmark. A two-way ANOVA revealed a main effect of landmark identity (30 cells out of 111 responsive cells) and relative distance (64 cells out of 111 responsive cells). The main effect of landmark identity was confirmed by a one-way ANOVA on the activity to the four landmark views at the center only (12 cells out of 111 responsive cells, which is a higher proportion than expected by chance; p = 0.01, binomial test). While the two-way ANOVA emphasizes the coding of relative distance by the cells, it also confirmed that many cells conveyed a combination of information between the identity of the landmark viewed and the position as previously unraveled by the state space. Twenty-nine cells coded a combination of these two factors with a significant interaction (14 cells) or both factors significant (15 cells). Both proportions are higher than chance (p < 0.001, binomial test). Fig 5 shows the activity of two cells recorded concurrently during the same behavioral session. The cell on the left shows a high activity for the northwestern landmark viewed from distance RD1 (F(9,756) = 4.4076, p < 0.001); the cell on the right shows a high activity for the northeastern landmark viewed from distance RD3 (F(9,756) = 2.1243, p = 0.025). In sum, the results indicate that cells are sensitive to the identity of the landmark viewed by the animal while also being modulated by the distance from the animal to the landmark. Thus, this analysis comports with the conclusions of the state-space analyses.
Some cells (e.g., cells 1, 2, and 5 in Fig 2) seem to respond similarly to one landmark from different viewpoints (trajectories highlighted in the black and red boxes). Nevertheless, the way cells relate to landmarks is usually more complex than a sensory response, as expected from the foregoing analysis. For example, in Fig 2, five of eight cells (cells 3 to 7) show activity to a landmark from only one or two viewpoints amongst several possibilities.
To investigate this viewpoint dependence at the population level, we compared activity collected for different trajectories exposing the same landmark. This was reliably possible for the two landmarks neighboring the reward that were visible in five paths (Fig 6, top row schematics). As such, we analyzed 42 cells exhibiting a significant activity to the appearance of the northwestern or northeastern landmark and significant IC for the point of gaze. Fig 6A–6C shows the activity of three cells of Fig 2 as a function of the path. To test whether the cells’ immediate response to a landmark varied with viewpoint, we computed for each landmark a path selectivity index (see Materials and Methods). The indices were computed on a 500 ms epoch beginning with the appearance of a landmark in the visual scene, with a 120 ms offset to account for visual latency. We found that only one landmark-responsive cell was significantly path invariant (Fig 6B, corresponding to cell 1 in Fig 2, 5% of the 42 cells, p = 0.37, binomial test), whereas 12/42 cells (29%, p < 0.001, binomial test) discriminated significantly amongst different viewpoints of the same landmark (Fig 6D, single examples in Fig 6A and 6C). The distributions of the actual versus surrogate selectivity indices were also significantly different (p < 0.001, Kolmogorov–Smirnov test), confirming that at the population level, viewpoint selectivity in our data is far more represented than viewpoint invariance. This viewpoint dependence would support egocentric updates of self-position in space.
How is the neural response to the landmarks related to the animal’s visual exploration of the scene? Some cells appear triggered by the first entry of a landmark in the animal’s FOV (Fig 2, cells 1–5). This interpretation is not possible for cells 6 or 7 because the firing rate increases in a start arm, well after the landmark has entered the FOV during the preceding return move. To clearly assess the relationship between the gaze on landmarks and cell activity, we aligned the activity of each cell to its best-driving landmark, either on its appearance in the FOV (landmark “onset”) or on its first foveation (Fig 7A–7D, see Materials and Methods). Then, we computed the average response for both alignments (Fig 7E). Gaze-aligned activity significantly rose until the saccade to the landmark and peaked shortly after. The distribution of the latencies of the responses to the best landmark are left skewed when aligned on gaze and right skewed when aligned on landmark onset (Fig 7F, see Materials and Methods). For a small but nonnegligible proportion of cells, the response latency even preceded the landmark’s entry in the FOV, suggesting a predictive representation of its location in space prior to becoming visible. To further examine the nature of this anticipatory activity, we analyzed the cell activity with respect to the eccentricity (distance in degrees to the fovea) of the best-driving landmark. We computed the population average, considering four different subsamples of the data: (1) all the 500 ms epochs that followed landmark appearance in the visual scene, (2) all the 500 ms epochs that preceded landmark appearance, (3) periods starting when the landmark had been visible for at least 1,000 ms, or (4) the whole dataset. For landmarks still outside the field of view (second subsample), landmark eccentricity is computed as it would appear if the field of view was complete (180° x 180°). Overall, cells showed a modulation of their activity by landmark eccentricity (Fig 7G, blue line). This modulation was increased by the recent appearance of the landmark (red line). Nevertheless, and in accordance with Fig 7E and 7F, cells could fire for a landmark still invisible, even if it were not to appear close to the fovea (purple line; activity was still higher if the monkey has previously saccaded close to the expected point of landmark appearance). This observation suggests that the animal could anticipate landmark appearance. Above all, it appears that the concept of receptive field does not apply to these cells, which is expected if they signal a context-sensitive, higher-order conjunction related to the completion of the task.
To evaluate the added value of the gaze with respect to landmark onset alone, we compared (a) the firing rate to each landmark during the 500 ms period following its appearance (excluding instances when a saccade was made at the landmark during that period) and (b) the firing rate within the 500 ms after landmark foveation. On this basis, we computed the path selectivity index (as for analysis in Fig 6) and a landmark selectivity index, which evaluates how much a cell discriminates between the different landmarks. Landmark selectivity was significantly enhanced if the landmark was foveated compared to appearing in the visual periphery (one-sided logit-Wilcoxon, p < 0.001, Fig 7G). Thus, a directed gaze correlated with a greater extraction of information from the landmarks. In contrast, path selectivity was only moderately improved by direct gaze (Fig 7H, p = 0.036), consistent with a ceiling effect whereby once the maze layout is learned, path identification would not need systematic visual checks at visible landmarks. Overall, these observations show that the firing patterns during ocular exploration are not simply triggered by low-level characteristics of the optic flow but reveal an active search for spatial information.
For the first time in the monkey we jointly examined navigation behavior, gaze fixations, and hippocampal activity to detail a comprehensive picture of the primate coding of space during a goal-oriented navigation task. We found that animals in our settings formed a representation similar to a cognitive map of the virtual maze in that they computed a trajectory from a new starting point during probe trials. The analysis of eye movement revealed that animals explored the VR environment in congruence with upcoming actions, suggesting a reliance on acquiring visual information to guide their moves.
Very importantly, this ability to compute a trajectory via awareness of self-position, direction, and action contexts was mirrored by hippocampal cells’ activity, best understood in a fine-grained state space. Indeed, many of the cells reliably responded as the animal viewed or gazed at a specific landmark but only on selective segments of the trajectory bearing a view to this landmark. Thus, rather than behaving as simple place cells or target-of-gaze cells, hippocampal neurons combined different aspects of the animal’s current sensory state with the goal-related action context. Crucially, some “state-space selective” cells discriminated between two identical position-orientations (thus, two identical sensory inputs) and even two identical prospective actions when they were part of different action contexts. This implies that cells did not code for the visual properties of the landmark or for a systematic association between a cue and an action but rather expressed self-position in an abstract, multidimensional representation of the maze.
Our findings were obtained in VR, and this technique has recently been a welcome substitution for the real world when exploring the neural basis of spatial cognition in humans and in animals [3–8,27–32], as it allows a fine control of the environment. Nevertheless, this technique raises interpretative issues as to how well navigation in VR reflects its real-world counterpart. In a virtual setup like ours or those used for human fMRI, subjects lack real inputs from motor, vestibular, and proprioceptive systems. Importantly, the optic flow activates a common neural circuitry with vestibular input that underlies self-motion encoding [33]. Our virtual setup, which primarily simulated extrafoveal territory (70° FOV), shares common properties with stimuli that are known to produce vection-like phenomena [34]. It is thus likely to produce an illusion of self-motion. The animals’ anticipatory gaze behavior, similar to human drivers [21,35], confirms this interpretation. Previous experiments using VR in monkeys had animals travel well-learned routes [27], execute repetitive motion sequences [28], or search explicit visual goals [36]. Our animals’ use of landmarks—inferring the position of a goal dissociated from the landmarks themselves as in triangulation, a hallmark of hippocampal function [22,37]—is novel, and the first replication in the monkey of aptitudes already shown in rats [31] and humans [30,32] in a VR world.
The signature of hippocampal activity in rodents is the coding by place cells of the animal’s current position [1]. Previous results in primates, including humans, showed that hippocampal cells encoded spatial views [10,11], place [9,14–16,28,38], or a mixture of place and view [8]. In a controlled VR wayfinding setting, the present findings show that hippocampal cells display a fine-grained tuning that preferentially codes one of the landmarks being viewed by the animal and its current self-position relative to the landmark. This shows that primate hippocampal cells also carry position information as rodent place cells do. Beyond that, however, we found that this tuning included trajectory-related contextual aspects and was best captured in a state space because the cells showed higher spike information content when compared to other coding spaces such as self-position, head direction, or point of gaze. Crucially, analysis of the firing rate in the center of the maze showed that the context truly gave additional information compared to a simple combination of position and direction (Fig 4). Further, as shown in Fig 5 and Fig 6, we found more cells than expected by chance that coded a combination between landmark viewed and distance to the landmark or that discriminated trajectories bearing views of the same landmark. Lastly, we observed that gazing on landmarks increases the cells’ selectivity to the path or trajectory bearing the landmark (Fig 7). In combination, these results imply that hippocampal spatial memory involves more than a simple spatial relation to the environment but rather a sensorimotor trajectory scheme in a goal-reaching context. Hippocampal cells responding to the spatial views have been previously reported [10,13]. The identity of the views represented in those cells had precedence over position in explaining their activity. While our cells bear a resemblance to the ones described by Rolls and colleagues [10,13], we found that complementary information relevant to position and action was also robustly encoded in the cellsʼ firing. This difference might be due to the task we used, in which the current viewpoint guides the next actions of the animal. Such imperative contingencies were absent in the studies by Rolls and colleagues. Our findings echo findings in rodents in which the activity of place cells depends on the goal of the trajectory [39,40] or its retrospective and prospective components [41]. In addition, our results imply that hippocampal cells in the primate represent more than the spatial view [10,11] or its associated reward value [42]. How can we interpret our results with respect to the animal’s behavior? As the same viewpoint for the same landmark, same heading, and same prospective action can be reliably discriminated for different trajectories, cells do not code the sensorimotor properties of the task only. Rather, we hypothesize that cells embody a dynamic knowledge about the self-position with respect to the landmarks in a contextual fashion, depending on the current trajectory.
In fact, this conclusion extends previous reports that the rat hippocampus shows trajectory-dependent firing [41,43,44]. To what extent these effects are due to interactions within a broader network such as the prefronto–thalamo–hippocampal circuit [44] remains to be examined in the monkey.
Our findings also underscore the power of the relationship between visual exploration and hippocampal activity. Indeed, visual exploration of the environment led by saccades and fixations is part of the primate-specific repertoire of active sensing and is supported by dedicated visual processing areas shared by human and other primates [45,46]. In VR studies like ours, directions can be primarily obtained by visual information only, and accordingly, we observed that the point of gaze importantly modulates the hippocampal activity, as already shown by Rolls et al. [11,13]. This is seen both in single examples (Fig 2) and by the high number of cells with spike information modulated by the point of gaze (Fig 3).
The temporal dynamics of cell activity with respect to the landmark-directed saccades reveals that firing mostly follows landmark appearance but precedes the eyes reaching the landmark, suggesting an anticipatory identification of the landmark (Fig 7E–7G). Although the selectivity of hippocampal cells to the identity of the fixated landmark is reminiscent of object fixation cells in the inferotemporal cortex [47–49], hippocampal cells discriminate positions from which the landmarks are gazed at (Figs 4, 5 and 6). Further, we showed the firing rate at fixation was greatly modulated by the time period during which the landmark was fixated, with activity decreasing as the landmark was fixated long after it was already visible. This pattern is coherent with a task-contextual modulation of the cells. Moreover, landmark and path selectivity tuning is enhanced through foveation, suggesting that gaze information enhances coding of self-position. In sum, our data suggest that the counterpart of place cells in primates, as compared to rodents, is expressed as activity related to point of gaze in conjunction with other variables essential to navigation, such as position and identity of visual elements at key instants of trajectory planning.
Recent rodent studies in VR provide a useful common framework to situate our findings and make cross-species comparisons [3–5]. Approximately 50% of our cells showed spatial selectivity like that in rats [3–5] and similar to that obtained in real-world settings [3,50,51]. Our results further confirm that spatial coding can be obtained in absence of vestibular and proprioceptive input. In addition, the number of fields per cell (average = 2.7) was greater than the number of fields described in the rat (about 1.5), but our maze differs from the single alleys and square or round open fields in the foregoing studies. Studies conducted in real, complex environments bearing path repetitions observe many neurons bearing more than one place field [52–54]. The nature of our task environment may thus have played a role in the multiplicity of hippocampal fields.
In rodents, direction selectivity was thought to be exclusive to one-dimensional mazes; however, it was recently demonstrated to be present in two-dimensional environments [3,7]. Our setup did not allow us to characterize direction selectivity in a strict independent way because direction and position often covaried. Thus, our results on the matter have to be taken with a note of caution. However, firing in the center of the maze displayed direction-dependent activity. Taken together, these findings contradict earlier results showing consistently direction-independent responses in 2-D environments [55,56] and further demonstrate that direction selectivity is an essential property of place cells—as recently shown in VR in the rat [3,7] and in the real world in bats [57]—and can be independent from vestibular cues. Note that direction selectivity was not preserved across spatial positions, suggesting its sensitivity to other variables such as visual cues or actions. Hence, we also observed that selectivity is stronger at choice points than at other places.
Lastly, the strength of our cells’ signals (amplitude of firing rate, number of cells recruited) is more on par with that in the rodent or bats than in previous VR monkey studies, which appeared to engage hippocampal activity rather poorly [28,38]. Our animals had to learn anew the significance of the landmark configuration during each recording session, whereas the aforementioned studies relied on a shuttle behavior between fixed reward zones, suggesting that dissimilar learning requirements account for the different firing rates.
Studies in humans of hippocampal neuronal activity during spatial navigation are rare. Two studies [8,9] provide arguments for homologies in hippocampal processing between nonhuman to human primates. In particular, Ekstrom and colleagues [8] described many cells that showed an interaction between place, goal, and view. The state space–selective cells observed in our study bear resemblances to the conjunctive cells they reported. Such a relationship could account for the interaction of place and goal found in the human hippocampus [8]. This conjunctive type of coding appears ubiquitous across mammalians, as our findings are consistent with the encoding of task-related demands by hippocampal neurons in rodents [58].
In sum, we show that space representation in primates embodies self-position with respect to the target of gaze and further carries cognitive information with respect to the current trajectory to a goal. These results clarify ambiguous results previously obtained in primates that suggested hippocampal neurons did not convey self-position but instead a spatial view [10,11] or conveyed self-position but no cognitive information [12]. They also bridge the gap between results in rats and humans by showing that place cells support self-position in a cognitive map in primates as well and clarify how this coding is constructed through visual exploration and self-movement.
For long the dominant view of hippocampal function has been influenced by the concept of place selectivity, with the hippocampus viewed as a “neural GPS,” highlighting the actual position of the animal on an internal map. In contrast, our results, clarifying and considerably extending the early advances of Rolls and colleagues, suggest a different view of the hippocampus wherein distinct elements of sensory, motor, and cognitive information are linked in order to build a memory (here, a space-related one). In this framework, place fields appear as a mere projection of this memory onto physical space. These results support a new view of hippocampal function and are of relevance for the understanding of the organization of human hippocampal function and memory.
Our study involved two nonhuman primates. All experimental procedures were approved by the animal care committee (Department of Veterinary Services, Health & Protection of Animals, permit no. 69 029 0401) and the Biology Department of the University Claude Bernard Lyon 1, in conformity with the European Community standards for the care and use of laboratory animals (European Community Council Directive No. 86–609). Further, our procedures were examined by CELYNE, the local ethics board, which approved the in vivo methods used in the laboratory. We minimized animal suffering and maintained their well-being by using anesthetics and pain management during surgeries for recording chamber implantation. During the experiments, animal’s behavior and well-being was monitored. No animal was euthanized after the experiments. Rather, both animals had their implants removed under general anesthesia. One of the monkeys was placed in a sanctuary for monkeys (Natuur Hulp Centrum, Belgium), while the second animal will be placed as soon as possible.
Animals were head restrained and placed in front of a large screen (152 x 114 cm) at a distance of 101 cm. They were further equipped with active shutter glasses (Nuvision) coupled to the computer for 3-D projection (DepthQ projector, Infocus) of a virtual world (Monkey3D, Holodia, S1–S3 Movies). The projection parameters were calibrated to render objects’ size real by calibrating disparity using the actual interpupillary distance of the monkeys (3.1 cm for monkey K and 3.0 cm for monkey S). We confirmed the animals perceived images with the depth of stereoscopic projection by measuring vergence as a small object moved from an apparent 50 cm in front of the screen to 150 cm behind the screen. To this end, two small infrared cameras were mounted above each eye and the movement of the pupils of each eye was monitored (ASL). The cameras further allowed monitoring the animal’s gaze through the task. Animals learned to navigate via the joystick towards a reward hidden at the end of one of the star maze’s arms (Fig 1B, S1–S3 Movies). The star maze had a radius of 16 m and speed of displacement was 5 m per second. This velocity was chosen to optimize the number of rewarded trials in a session and prevent the animals from getting too impatient. During a shaping period that lasted 6 mo, animals learned to find the reward targets whilst operating a joystick that controlled a sphere on the screen. Once they had mastered this task, they were introduced to a 3-D version of this task. Then, they were introduced to a simple Y maze in which they had to move the joystick to approach the sphere. Next, landmarks were introduced along the Y maze, and animals were trained with the sphere in presence of the landmarks. Then, the sphere was removed and animals were rewarded when they went toward the end of the arm where the sphere was last. To this end, they had to use the landmarks. At this point, they were introduced to the full star maze. For one animal that would not go to the end of an arm if a sphere was not there, a different strategy was adopted. We replicated the sphere five times and changed the rules such that there was a sphere at each end, but the animal had to find “the one” which would give a reward and blink when approached. Once this step was learned, the spheres were removed for him as well. Finally, animals were trained to learn new landmark arrangements every day. We used a star-shaped environment rather than using an open field to ensure multiple passes through the same trajectories and to avoid locations with too sparse data. Each day, the animals had to locate a new position of the reward with respect to new landmarks. Each trial began with the animal facing the maze from one arm end. The joystick allowed the animal to move to the center and turn left or right to choose and enter one arm. Once the animal reached the end of the arm, it was given a liquid reward only if correct and then brought to a randomly chosen new start whether the trial was correct or incorrect. Fig 1 presents the sequence of a trial from above (1C) and from the animal’s perspective (1D).
We computed the point of gaze in an allocentric frame wherein objects (landmarks) or positions in space towards which the monkey gazed were mapped relative to a top view of the maze (Fig 1G and 1H). To calculate the point of gaze, we combined the X and Z eye-tracking data with the X and Y virtual position of the animal in the maze and the orientation of the camera viewpoint. The points of regard above the horizon were mapped onto a vertical circular wall enclosing the landmarks; this wall was then flattened into an annulus in the map. This map represents where the animal is gazing in the spatial scene, not the craniocentric eye position. The coordinates obtained were then used to compute the firing rate of the cells as a function of the animal’s point of regard (gaze spike map density, Fig 2, third column).
For a period of approximately 6 mo, each animal underwent daily recording sessions during which electrodes were lowered to the target areas (see S4 Fig). Recordings began if individual cells were present on the contact electrodes, and the task was then started. Individual cells were pre-sorted online and re-sorted offline (offline sorter, Plexon Inc.), and only cells whose waveforms possessed reliable signal-to-noise ratios (two-thirds of noise) and whose activity was stable in time such as illustrated in the rasters in Fig 2 (far right panels) were included in the database. The recording sites were located in CA3 (n = 99), CA1 (n = 101), or the dentate gyrus (n = 8). See S1 Appendix A3 for a detailed description of the recording methods and S4 Fig for anatomy.
All data were analyzed with custom Matlab scripts.
See S2 Appendix for a description of the spike and behavior data files hosted on CRCNS (http://dx.doi.org/10.6080/K0R49NQV) as well as some Matlab scripts necessary to work with them.
|
10.1371/journal.pntd.0007186 | Comparing the performance of urine and copro-antigen detection in evaluating Opisthorchis viverrini infection in communities with different transmission levels in Northeast Thailand | To combat and eventually eliminate the transmission of the liver fluke Opisthorchis viverrini, an accurate and practical diagnostic test is required. A recently established urine antigen detection test using monoclonal antibody-based enzyme-linked-immunosorbent assay (mAb-ELISA) has shown promise due to its high diagnostic accuracy and the use of urine in place of fecal samples. To further test the utility of this urine assay, we performed a cross sectional study of 1,043 people in 3 opisthorchiasis endemic communities in northeast Thailand by applying urine antigen detection together with copro-antigen detection methods. The quantitative formalin-ethyl acetate concentration technique (FECT) was concurrently performed as a reference method. The prevalence of O. viverrini determined by urine antigen detection correlated well with that by copro-antigen detection and both methods showed 10–15% higher prevalence than FECT. Within the fecal negative cases by FECT, 29% and 43% were positive by urine and copro-antigen detection, respectively. The prevalence and intensity profiles determined by antigen detection and FECT showed similar patterns of increasing trends of infection with age. The concentration of antigen measured in urine showed a positive relationship with the concentration of copro-antigen, both of which were positively correlated with fecal egg counts. The data observed in this study indicate that urine antigen detection had high diagnostic accuracy and was in concordance with copro-antigen detection. Due to the ease and noninvasiveness of sample collection, the urine assay has high potential for clinical diagnosis as well as population screening in the program for the control and elimination of opisthorchiasis.
| Opisthorchiasis, caused by an infection with the liver fluke, Opisthorchis viverrini, is a neglected tropical disease endemic in Southeast Asia, particularly Thailand and Lao PDR. O. viverrini as well as Clonorchis sinensis have been classified as group I biological carcinogenic agents for cholangiocarcinoma (CCA). Due of the impact of control programs, the prevalence and worm burden in endemic communities have been reduced and this has caused the conventional fecal examination to be less sensitive and unreliable. In order to improve the diagnosis and to move towards the elimination of liver fluke to reduce CCA, we evaluated a novel urine antigen detection method by mAb-enzyme-linked immunosorbent assay for the diagnosis and screening of opisthorchiasis in endemic communities in northeast Thailand. We concurrently applied two coprological methods for antigen detection and fecal examination by formalin–ethyl acetate concentration technique, a reference method for comparison. Urine and copro-antigen detection had comparable diagnostic accuracy and both methods performed better than the fecal examination. Because of the ease and acceptance of urine specimen collection and handling, urine antigen detection has a high potential for the diagnosis and mass screening of opisthorchiasis in control and elimination programs.
| Opisthorchiasis is a neglected tropical disease caused by an infection with a small human liver fluke Opisthorchis viverrini. Based on epidemiological and experimental evidence, O. viverrini as well as a closely related species, Clonorchis sinensis, were classified as group I biological carcinogenic agents [1]. Opisthorchiasis is known to cause important public health problems in mainland Southeast Asia, particularly in Thailand, Lao PDR, Cambodia and Vietnam [2, 3]. The current transmission landscape consists of a light and uneven distribution of O. viverrini infection in endemic areas, a feature that has consequences for diagnostic accuracy, as well as the role of O. viverrini as risk factor of cholangiocarcinoma (CCA) [4, 5]. For the success of any liver fluke control and elimination program, particularly for mapping and facilitating drug treatment, an improved diagnostic method that is suitable to the current endemic conditions is needed [6, 7].
To date, definitive diagnosis of O. viverrini infection is achieved by finding parasite eggs in feces, however, such parasitological diagnosis has many drawbacks including false positivity caused by confusion with the eggs of minute intestinal flukes, or by false negativity in light infections and in biliary duct obstruction where no eggs can be detected in feces. Repeated stool examination is required to increase the reliability of the results [8–10], but the cost and requirement of an expert microscopist make this method less practical.
Previously molecular and immunological-based diagnostic methods have been developed and applied for the diagnosis of opisthorchiasis [11–15]. Although these methods have provided a better diagnostic performance compared with the parasitological method, they have several drawbacks regarding their sensitivity and specificity according to the abundance of the target genes, antigens or antibodies, and also the presence of inhibitors in clinical samples [7, 16]. An antibody-based approach for the detection of circulating antibody has limitations due to the cross reactive nature of the antigens used [17–21] and a positive result does not always indicate active infection by the parasite [19, 22, 23].
Unlike antibody detection, an antigen detection assay detects a current and viable parasite infection which better reflects the infection status in opisthorchiasis patients. In this regard, monoclonal antibody-based enzyme linked immunosorbent assays (mAb-ELISA) for detecting parasite antigen in fecal samples (copro-antigen) have been introduced and verified in clinical samples [24, 25]. The mAb-ELISA provides several advantages over conventional methods since it has a higher specificity, good reproducibility, and can be prepared in large quantities [26]. In addition to previous studies [24, 25], our group has reported an improved protocol for copro-antigen detection with high diagnostic performance [27]. Subsequently, in 2015, we reported a novel antigen detection method using urine samples for the diagnosis of opisthorchiasis [28]. Both urine and copro-antigen detection yielded a high diagnostic performance compared with standard fecal examination, but a comparison between these antigen detection methods in the same sample population has not been reported.
In this study, we aim to assess the field application of this method, comparing the diagnostic performances of urine antigen detection with that of copro-antigen detection. The formalin-ethyl acetate concentration technique (FECT) was used as a reference method. This study was conducted with the residents of 3 opisthorchiasis endemic communities in Northeast Thailand that showed varying prevalence and intensity of O. viverrini infection. We also investigate the quantitative relationships between the levels of O. viverrini antigens in urine and fecal samples and the fecal egg counts determined by the formalin-ethyl acetate concentration technique.
The human subject protocol used in this study was approved by the Human Ethics Committee of Khon Kaen University, Thailand (reference number HE561478). Written informed consent was obtained from individual subjects and those with the O. viverrini-positive examination by FECT or antigen detection methods were treated with a single oral dose of praziquantel (40 mg/kg body weight).
The experimental protocols for laboratory animal handling for monoclonal and polyclonal antibody production and also parasite antigen production were approved by the Institutional Animal Ethical Committee, Khon Kaen University (AEKKU-NELAC 26/2558). All animals were anesthetized with isoflurane inhalation before immunization. For euthanization, the animals were anesthetized by isoflurane and sacrificed by drawing of blood from the heart. No animals were demonstrated the severe health problems during in this study. The procedure was performed in strict accordance with the guidelines for the Care and Use of Laboratory Animals of the National Research Council of Thailand.
This prospective cross-sectional study began in March 2015 and ended in July 2016. The eligibility criteria of the participants were; (i) participants who were native residents in Ban Wa sub-district in Khon Kaen Province (KK), Tao Ngoi sub-district, Sakon Nakhon province (SK) and Nong Khon Thai sub-district, Chaiyaphum province (CP), Northeast Thailand; (ii) participants who agreed to provide both feces and urine on the same day for index and standard reference tests; (iii) participants who were apparently healthy and had no clinical signs or symptoms. The sample size of this study was calculated based on the average proportion of positive O. viverrini cases (28%) with the derivative set at 1.96 and corresponding to a 95% confident level ± 5%. The calculated sample size was 930 participants with a potential loss of sample submission of 20% so that the number of participants in three localities to be recruited was 1,116.
Of the 1,233 participants originally enrolling, 180 were excluded because they either failed to submit clinical specimens (n = 134) or submitted inadequate specimens (n = 56). A total of 1,043 participants fulfilled the inclusion criterion and provided complete clinical samples (Fig 1). There was no statistical difference in demographics data (age, sex, etc.) between the recruited and the excluded participants.
Feces (10 g) and samples of the first morning, mid-stream urine (5 ml) were collected from the project participants in standard plastic containers and kept in a chilled box (4°C-8°C) during transportation to the laboratory within one day of collection. At the laboratory, each feces sample was separated into 2 parts: (i) aliquoted and processed for parasite examination by the formalin ethyl-acetate concentration technique (FECT), (ii) kept in 1.5 ml tube and stored at -20°C for copro-antigen detection. The urine samples were centrifuged at 1,500 rpm at 4°C for 15 minutes and the cleared supernatants were aliquoted into new vials and stored at -20°C until required for mAb-ELISA.
The procedure for quantitative FECT was similar to the procedure detailed in the previous report [29]. In brief, 2 grams of fresh stool were fixed in 10% formalin and kept at room temperature until processing. After vigorous shaking, the specimens were centrifuged at 2,500 rpm for 5 minutes and the sediment re-suspended with NSS and strained through two layers of gauze. The tube was then centrifuged at 2,500 rpm for 5 minutes and the supernatant discarded. Seven milliliters of 0.85% saline were added to the sediment and mixed, then 3 ml of ethyl acetate were added to the sample tube, mixed thoroughly, and centrifuged at 2,500 rpm for 5 minutes. After removal of the top three layers, the sediment was re-suspended with 1 ml of 10% formalin. The final suspension was examined blind by two parasitologists and the number of O. viverrini eggs were counted in 2 drops sampled from the total suspension. The number of eggs per gram of feces (EPG) was calculated to determine the intensity of O. viverrini and other parasitic infection.
The protocols for mAb-ELISA for O. viverrini antigen measurements in urine and fecal samples were similar to those previously described [27, 28]. Polystyrene microtiter plates (NUNC, Roskilde, Denmark) were sensitized overnight at 4°C with 100 μl/well of 5 μg/ml of mAb diluted in 50 mM bicarbonate buffer pH 9.6. After coating, the plates were washed three times with normal saline containing 0.05% Tween 20 (NSST) and uncoated sites were blocked with a solution of 5% dried skimmed milk in 50 mM bicarbonate buffer pH 9.6. After incubation for 1 hour at 37°C, TCA-pretreated urine and fecal extracted samples were added and the plates (100 μl/well) were incubated at 37°C for 1 hour. Next, the plates were washed five times with NSST and purified IgG rabbit against crude O. viverrini antigen (10 μg/ml) and 2% dried skimmed milk in PBST was added and incubated at 37°C for 1 hour. After that, 100 μl/well of biotinylated goat anti-rabbit IgG conjugate (1: 4,000) in 2% dried skimmed milk in PBST were added (Invitrogen, CA, USA) and the plates were incubated at 37°C for 1 hour. After washing three times, horseradish peroxidase-conjugated streptavidin (GE Healthcare, Buckinghamshire, UK) (dilution 1: 5,000) in PBST was added and the plates were incubated at 37°C for 1 hour. Plates were washed and the substrate (o-phenylenediamine hydrochloride) solution (Sigma, St. Louis, MO, USA) was added and incubated for 20 minutes in the dark at room temperature. The enzyme reaction was stopped with 100 μl/well of 2M sulfuric acid (H2SO4) and the optical densities (OD) were read spectrophotometrically at 492 nm with an ELISA reader (Tecan Sunrise Absorbance Reader, Austria). Two well-trained laboratory staff were responsible for the simultaneous sample analysis of the index tests. During the laboratory procedure, the sample IDs were blinded and the laboratory staff had no knowledge of the sample subjects.
Known negative and positive urine samples and also fecal samples for O. viverrini infection determined by FECT were used to construct a Receiver Operation Curve (ROC). The cutoff points for diagnosis by mAb-ELISA were calculated using MedCalc software version 9.6.3 (MedCalc Software, Ostend, Belgium).
The OD values were transformed to antigen concentrations based on the standard curves for urine and feces using spiked O. viverrini-ES antigen extract with varying concentrations starting with 5,000 ng and followed by two-fold serially dilutions to produce a standard calibration curve. The separated cut off value was 19.4 ng/ml for urine and 61.2 ng/ml copro antigen detection methods. These were used to determine negative and positive tests. The sensitivity, specificity, and positive and negative predictive values with 95% confidence intervals were calculated and compared between each sample using standard parasitological methods as a reference. In addition, a composite reference standard method which combined both parasitological, urine or fecal antigen detection methods as a reference method was used to assess the diagnostic performance of each method.
In order to assess the specificity of the mAb-ELISA for O. viverrini antigen detection in both urinary and copro-antigen detection method, cross reactivity with positive samples of other parasitic infections determined by FECT was assessed.
For urinary antigen detection, the other parasite infections, i.e. Strongyloides stercoralis (n = 40), minute intestinal fluke (n = 17), hookworms (n = 10), Taenia sp. (n = 5), Echinostoma sp. (n = 8) and Trichuris trichiura (n = 5), were tested for cross reactivity. Cross reactivity of copro-antigen detection was performed with Strongyloides stercoralis (n = 40), minute intestinal fluke (n = 17), hookworms (n = 10) and Taenia sp. (n = 5).
Data recorded in the case report forms were entered into a computer using the Microsoft Excel program and analyzed using SPSS v.22 (International Business Machines, USA). Helminth species-specific fecal egg counts were transformed into number of eggs per gram of feces (EPG). Kendall’s tau-b correlation test was used to determine the correlation between urinary-antigen and copro-antigen concentration and also EPG. Performance of the test in terms of sensitivity, specificity, and predictive values was calculated as described elsewhere [31]. The OD values were transformed to a ratio between the OD of the samples and the OD of reference urine or fecal extracted samples. The reliability of urine and copro-antigen detection methods by mAb-ELISA for the diagnosis of opisthorchiasis was analyzed using odds ratios (OR) with 95% confidence intervals (CI) using logistic regression. A statistically significant level was set as p<0.05. We used the following guidelines to interpret the kappa values; ≤ 0 indicating no agreement, 0–0.2, poor agreement; 0.21–0.4; fair agreement; 0.41–06, moderate agreement; 0.61–0.8, good agreement; and 0.81–1.0, excellent agreement [32].
As shown in Table 1, based on FECT, the prevalence of O. viverrini infection in the three localities was 29–77% with an overall prevalence of 41%. The highest prevalence was in Nong Khon Thai sub district, Chiyaphum (CP), followed by Tao Ngoi subdistrict, Sakon Nakhon (SK) and the lowest was in Ban Wa sub-district, Khon Kaen (KK). The intensity of infection measured by fecal egg counts (EPG) were similar in the three localities which had mainly either no or light infection (EPG<50) with approximately 13% having an EPG>100. The average intensity of infection by locality was 25–127 EPG and the overall intensity was 54 EPG. Other parasitic infections occurred in <9% of the samples and the prevalence by order were S. stercoralis, MIFs, hookworm and Taenia spp. By urine antigen detection, the prevalence of O. viverrini was 46–77% by locality and the overall prevalence was 54%. The antigen concentration was 27–40 ng/ml of urine with an average of 31 ng/ml in all three localities. In the case of copro-antigen detection, the prevalence of O. viverrini was 48–78% with an overall value of 56%. The antigen concentration in feces was 75–90 ng/ml of feces and an average concentration of 81 ng/ml.
As shown in Fig 2, the O. viverrini infection determined by FECT showed the prevalence profile that peaked at 30–40 years in CP and SK, but in KK the prevalence increased slowly and peaked at an age >60 years. The intensity profiles increased steeply from 20–30 years and stabilized thereafter in CP and SK. In KK the intensity increased slowly with age and reached a plateau at age 50 years onwards.
The age-prevalence profiles of O. viverrini by urine and copro-antigen detection (Fig 2B and 2C) shared a similar pattern in CP and SK where they peaked at age <30 years and plateaued thereafter. In the case of KK, the prevalence increased steadily with age and became a maximum at age <50 years. The age-intensity profiles measured by urine and copro-antigen were similar to those by FECT, being highest at age <30 years and slightly less at age <50 years in CP and SK. The intensity profile for KK increased slowly with age and was highest at age <60 years.
In order to construct the standard curves for the calculation of antigen concentrations in urine and fecal samples, measurement of antigens in spiked excretory-secretory antigen of O. viverrini in urine and feces was performed using mAb-ELISA. The relationships between the concentrations of in urine and fecal samples and OD values obtained from mAb-ELISA were assessed by linear regression models. The best-fit linear regression equations for urinary and copro-antigen detection was y = 0.856x-0.831 and y = 0.653x-0.725, respectively. These equations were used to calculate the concentration of antigen in the clinical samples from the project participants (S1 Fig).
In order to determine the effect of EPG on the level of antigen detected in urine and feces, the participants from all localities were combined and then separated into 4 different intensity groups based on EPG determined by FECT (Table 2). Based on the urine and copro-antigen assays, the positive infection rates were 31% of egg-negative subjects by FECT. In egg-positive groups, the positive infection rates were 83–97% for urine and 83–98% for copro-antigen detections. Overall, urine and copro-antigen detections yielded similar positive rates and both assays yielded 14–15% higher positive rates than FECT.
The concentrations of urinary and copro-antigen showed a significant positive correlation with increasing intensity of O. viverrini (EPG) (Kendall’s tau-b, p < 0.001; Fig 3A and 3B).
Based on 1,043 participants who provided both fecal and urine samples for analysis of O. viverrini antigen concentrations, there was a significant positive correlation between urine and copro-antigen concentrations (Fig 4, R2 = 0.323, p < 0.001). The best fit linear regression equation was y = 0.635+1.283 where y = copro-antigen concentration (log-transformed value; ng/mL and x = urine antigen concentration (log-transformed value; ng/mL).
Table 3 shows the diagnostic performance of urine and copro-antigen detections determined in field-collected samples by using FECT as a gold standard. By locality, urine antigen detection exhibited a sensitivity between 86–90% and a specificity between 56–71%. The overall sensitivity in 3 localities was 89.5% (95% CI = 86.1–92.3) and specificity 71.2% (95% CI = 67.6–74.7). The diagnostic performance of copro-antigen detection was comparable to that for urine assay with the overall 90.7% sensitivity and 70% specificity.
When using combined methods as a composite standard, the performance of FECT in terms of sensitivities was 54%-85%, with the overall sensitivity of 63.6%. The specificities were 60%-98% between localities, with the overall specificity of 94.8%. Urine antigen detection showed the overall sensitivity 86.2% and was 83%-86% between localities. In the case of copro-antigen detection, the overall sensitivity was 85.3% and 85%-89% between localities (Table 4).
For ROC curve analysis, with reference to FECT, the AUC for urine antigen detection was 0.791 while it was 0.831 for copro-antigen detection (Fig 5A). Based on the composite gold standard, the AUCs for the diagnostic assay were 0.824, 0.934 and 0.957 for FECT, urine and copro-antigen, respectively (Fig 5B).
The test for agreement between FECT versus urine antigen detection and FECT versus copro-antigen detection revealed moderate agreement, Kappa values (κ) were 0.547 and 0.570, respectively. The kappa tests showed good agreement (0.770) between urine and copro-antigen detection (Table 5).
Additionally, a logistic regression analysis was performed to predict the risk of human opisthorchiasis (odds ratios) based on increasing arbitrary unit of the antigen in urine and fecal samples by mAb-ELISA. The arbitrary unit was defined as the antigen concentration of 28.050 ng/ml for urine antigen and 75.729 ng/ml that predict the odd of one in having opisthorchiasis. The analysis showed that the odds ratio values increased according to the increasing concentration of both urine and copro-antigens (Table 6).
To determine the specificity of mAb-ELISA, we applied this method to a separate set of urine and fecal extracts derived from participants with known parasite infection other than O. viverrini by the gold standard FECT.
For the urine antigen detection method, a positive result was found in 10% of subjects with Strongyloides stercoralis infection, 20% (2/10) in subjects with hookworm infection and 20% (1/5) in subjects with Trichuris trichiura infection. (Fig 6A). Copro-antigen detection showed a positive result in 7.5% (3/40) in subjects with Strongyloides stercoralis infection, 11.7% (2/17) in minute intestinal flukes infection and 10% (1/10) in hookworm infection (Fig 6B).
In order to overcome the drawbacks of conventional fecal examination, the urine assay for O. viverrini antigen detection offers not only higher diagnostic accuracy, but also the practical advantage of utilizing urine instead of feces for the diagnosis of opisthorchiasis [28]. The urine antigen detection method is based on the hypothesis that excretory-secretory antigens released by living parasites can enter the blood circulation and be excreted via the kidneys into the urine. Herein, we applied this method in a community-based epidemiological study of opisthorchiasis, together with copro-antigen detection, using a mAb-ELISA protocol. The results indicate that urine as well as copro-antigen detection have comparable diagnostic accuracy and both methods have a superior diagnostic performance to that of fecal examination by FECT. The estimated concentration of antigen in the urine as well as that in feces significantly positively correlated with the intensity of infection measured by fecal egg count (egg/gram feces). Moreover, the antigen concentration in urine positively correlated with the levels antigen in the feces and the agreement test between methods showed good concordance (kappa value; 0.770). The average antigen concentrations in both urine and feces paralleled EPG. The antigen concentration in urine was two- to three- folds less than those in feces. On average of the antigen concentration in feces was 2.6 times that of urine. Because of the qualitative and quantitative concordance between urine and copro-antigen detection, as well as that from FECT, the ease of sample collection and specimen handling strongly indicate that urine detection is desirable for the diagnosis and screening of opisthorchiasis.
There was a higher positive detection obtained by urine (46–77%) and copro-antigen detection (48–78%) compared to FECT (29–77%). The discrepancies of antigen detection versus fecal examination by FECT were seen at low to, moderate intensities of opisthorchiasis, i.e. in SK a geometric mean of 25.3 EPG, in KK a geometric mean of 59.7 EPG. By contrast, the positive rates detected by both urine (77%) and copro-antigen detections (78%) were similar to FECT (77%) in the locality with high transmission intensity (geometric mean 127.1 EPG) in CP. These findings further support the usefulness of the urine antigen detection assay for screening for opisthorchiasis with low intensities of infection. This may help in avoiding the under diagnosis of the majority of individuals who have a low worm burden. However, whether our antigen detection assays can demonstrate a single worm infection, as reported in schistosomiasis [33], is not known, but this is a critical threshold to assess the impact of parasite control and eventually elimination, and hence requires further investigation. Moreover, the occurrence of varying transmission levels observed in the three study communities reflects once again the changing landscape of infection associated with multiple factors such as mass drug administration (MDA) and public health educational efforts [2, 34]. Therefore, a detection method which is analytically sensitive yet also rapid and easy to apply is required to monitor the current endemic situation of opisthorchiasis [7].
For cases of O. viverrini egg negativity as determined by FECT (n = 620), 191 (30.8%) and 195 (31.4%) positive cases were discovered by urine and copro-antigen detection, respectively. The finding of antigen positivity in this scenario indicates the vital role of worms rather than eggs as a source of antigen in the urine and feces in opisthorchiasis. A similar situation was previously observed in an autopsy study which revealed that fecal eggs were not discovered in the majority of infected individuals with less than 20 worms [35]. Further study is required to monitor the changes in antigen profile after treatment to justify the need for chemotherapy of the antigen positive but egg-negative individuals.
As opposed to egg negative cases, in cases of positive for O. viverrini eggs (1–100 EPG), not all individuals were found positive by the antigen detection methods. Within the egg positive group (n = 423), 11.3% were negative in urine and slightly less (8.5%) were copro-antigen negative. The explanation for the under diagnosis by antigen detections is currently not known but may depend on several possibilities. First, as the antigens appearing in the urine and feces are excretory-secretory products originating from living adult worms and may not directly relate to fecal egg count. Second, the presence of urine antigen may rely on the intermittent production of secretory products from the adult worms and also the passage through kidney glomerular filtration as proteinuria [36] and motility of gastrointestinal system in the case of copro-antigen. Thus, liver fluke-induced pathology in the biliary system and abnormal glomerular filtration in the kidneys may contribute in fluctuations of urine antigen. Third, the physical properties of urine and feces such as water content and fecal mass may influence the antigen concentration and eventually detectable antigen levels. Dilution effects of urine were addressed using the concentration procedure to find the accurate antigen detection levels in urine in schistosomiasis [33]. Lastly, more antigen positive cases can be found by multiple or repeated examination of the urine and feces over consecutive days, which yield a higher positive diagnostic rate as shown in the case of schistosomiasis [37, 38]. Kidney involvement in opisthorchiasis, in terms of proteinuria and deterioration of renal function associated with immune complex, has been shown previously [39, 40]. Since the antigen detection approach in opisthorchiasis is at a rudimentary phase, further work to prove and disprove the above hypotheses is required to augment our understanding of the presence as well as the pathway of urinary antigens, and hence the improvement of diagnostic accuracy.
The urine antigen detections had a 0.79 AUC and 89% sensitivity while the copro-antigen detection had a comparable accuracy (0.83 AUC and 90% sensitivity) with respect to the conventional FECT gold standard. A similar trend but with slightly lower diagnostic values was observed when the composite diagnosis was used as a reference standard in which FECT has the lowest sensitivity (63%) but high specificity. The observed diagnostic values in the current study were comparable to our recent study for urine [28] and copro-antigen [27] but higher than that in previous reports for copro-antigen detections [24, 25]. This observation is probably due to different mAb-ELISA protocols, particularly with monoclonal antibody being used. Additionally, the diagnostic accuracy of opisthorchiasis may approach a new gold standard when antigen detection is combined with fecal examination, agreeing with an approach used in cases of clonorchiasis [41].
Although O. viverrini-specific mAb was used for antigen detection in ELISA, cross relativities with other parasite antigens in urine were observed in individuals infected with S. stercoralis, hookworms and T. trichiura. In the case of copro-antigen detection, cross reactions occurred with S. stercoralis, MIF and hookworms. These findings are similar to our previous reports [27, 28], and may be explained by the fact that these patients probably harbored a low intensity of O. viverrini infection which could not be detected by FECT. Since multiple parasite infections are common in the study areas in Thailand, further cross reaction studies on the participants from non-O. viverrini endemic areas are needed.
In addition to cross reactivity with other parasitic infections, there are drawbacks for both the urine and copro-antigen detection methods that required attention. The main drawback is that the whole process of mAb-ELISA takes approximately 5–6 hours to complete and these assays needed experienced operators and special equipment. Furthermore, samples should be kept cold (4°C or on ice) during transportation and require centrifugation and pre-treatment with TCA prior to the analyses. Therefore, a “rapid point of care and ease-of-use” platform such as a lateral flow immunochromatgraphic strip or dipstick, which can be operated at point of care setting, should be developed.
In this study, we show that O. viverrini antigen detection in urine and fecal samples has a higher diagnostic accuracy than conventional fecal examination method, FECT. Both antigen detection methods yielded comparable diagnostic accuracy and were significantly positively correlated. With high diagnostic accuracy and the ease of sample collection, the urine antigen detection method is a powerful approach for the diagnosis and population screening of opisthorchiasis in a clinical and also field setting. Pending additional study, the urine antigen detection method may serve not only the purpose diagnosis and screening, but also as a tool to follow-up opisthorchiasis patients to determine effective drug treatment in the control and elimination program.
|
10.1371/journal.pgen.1003006 | New Partners in Regulation of Gene Expression: The Enhancer of Trithorax and Polycomb Corto Interacts with Methylated Ribosomal Protein L12 Via Its Chromodomain | Chromodomains are found in many regulators of chromatin structure, and most of them recognize methylated lysines on histones. Here, we investigate the role of the Drosophila melanogaster protein Corto's chromodomain. The Enhancer of Trithorax and Polycomb Corto is involved in both silencing and activation of gene expression. Over-expression of the Corto chromodomain (CortoCD) in transgenic flies shows that it is a chromatin-targeting module, critical for Corto function. Unexpectedly, mass spectrometry analysis reveals that polypeptides pulled down by CortoCD from nuclear extracts correspond to ribosomal proteins. Furthermore, real-time interaction analyses demonstrate that CortoCD binds with high affinity RPL12 tri-methylated on lysine 3. Corto and RPL12 co-localize with active epigenetic marks on polytene chromosomes, suggesting that both are involved in fine-tuning transcription of genes in open chromatin. RNA–seq based transcriptomes of wing imaginal discs over-expressing either CortoCD or RPL12 reveal that both factors deregulate large sets of common genes, which are enriched in heat-response and ribosomal protein genes, suggesting that they could be implicated in dynamic coordination of ribosome biogenesis. Chromatin immunoprecipitation experiments show that Corto and RPL12 bind hsp70 and are similarly recruited on gene body after heat shock. Hence, Corto and RPL12 could be involved together in regulation of gene transcription. We discuss whether pseudo-ribosomal complexes composed of various ribosomal proteins might participate in regulation of gene expression in connection with chromatin regulators.
| Chromatin, the combination of DNA and histones, strongly impacts transcriptional regulation of genes. This is achieved thanks to various protein complexes that bind chromatin and remodel its structure. These complexes bind specific motifs, also called epigenetic marks, through specific protein domains. Among these domains, chromodomains are well known to bind methylated histones. Investigating the chromodomain of the Drosophila melanogaster chromatin factor Corto, we found that it interacts with methylated ribosomal protein L12 rather than with methylated histones. This is the first time that such an interaction is shown. Moreover, Corto and RPL12 co-localize with active epigenetic marks on polytene chromosomes, suggesting that both are involved in fine-tuning transcription of genes. Our results represent a major breakthrough in the understanding of mechanisms by which ribosomal proteins achieve extra-ribosomal functions such as transcriptional regulation. Genome-wide analysis of larval tissue transcripts reveals that Corto and RPL12 deregulate large sets of common genes, which are enriched in ribosomal protein genes, suggesting that both proteins are implicated in dynamic coordination of ribosome biogenesis.
| Chromatin structure strongly impacts on regulation of gene expression. Indeed, post-translational histone modifications (methylations, acetylations, phosphorylations etc…) called epigenetic marks, are recognized by protein complexes that shape chromatin (reviewed in [1]). A number of protein domains specifically interact with these modifications, thus inducing recruitment of chromatin remodeling or transcriptional complexes. Bromodomains recognize acetylated histones (reviewed in [2]) whereas 14-3-3 domains recognize phosphorylated histones (reviewed in [3]). Methylated histones are recognized by chromodomains (chromatin organization modifier) [4], which therefore belong to the Royal family of domains, known for their methylated lysine or arginine binding activity (reviewed in [5]). Chromodomains share a common structure encompassing a folded three-stranded anti-parallel ß-sheet supported by an α-helix that runs across the sheet. This structure contains two to four well-conserved aromatic residues that form a cage around the methylated ligand [5], [6].
Chromodomains were first identified in Polycomb (PC) and Heterochromatin Protein 1 (HP1) [4]. They are found in many other chromatin-associated proteins that belong to three classes according to their global structure: (1) PC/CBX family proteins harbor a single N-terminal chromodomain, (2) HP1 family proteins have an N-terminal chromodomain followed by a region termed a chromoshadow domain, and (3) CHD (Chromodomain/Helicase/DNA-binding domain) family proteins present two tandem chromodomains (reviewed in [5]). Most chromodomains specifically recognize particular methylated residues on histones. For instance, the chromodomain of PC, which is a subunit of the PRC1 complex (Polycomb Responsive Complex 1), binds specifically H3K27me3 [7], [8]. Once recruited, PRC1 prevents RNA Polymerase II recruitment or transcriptional elongation and therefore mediates gene silencing (reviewed in [9]). The chromodomain of HP1 binds H3K9me2 and H3K9me3, which are epigenetic marks characteristic of heterochromatin, and thus participates in heterochromatin shaping [10], [11]. Very few cases of non-histone chromodomain substrates are known [12]. For example, the HP1 chromodomain also recognizes an autocatalytically methylated residue of the G9a histone H3 methyl-transferase [13].
The D. melanogaster corto gene encodes an Enhancer of Trithorax and Polycomb (ETP), i.e. a Polycomb (PcG) and Trithorax (TrxG) complex co-factor, involved in both silencing and activation of gene expression [14], [15]. Indeed, Corto participates in transcriptional regulation of several homeotic genes together with these complexes and other ETPs [16], [17]. Corto binds chromatin and contains in its N-terminal part a single structured domain identified by hydrophobic cluster analysis and structural comparison as a chromodomain [18]. Hence, Corto would be closer to CBX proteins of the PcG class [5]. However, its chromodomain is rather divergent, since only two aromatic residues are conserved among the four that make a cage around the methylated residue. How Corto anchors to chromatin and more specifically, whether the chromodomain addresses Corto to chromatin, is not known. Here, we address this question by expressing a tagged Corto chromodomain in flies or in S2 cells. We show that the Corto chromodomain is a functional chromatin-targeting module. Surprisingly, peptide pull-down, mass spectrometry and Biacore show that the Corto chromodomain interacts with nuclear ribosomal proteins, and notably binds with high affinity RPL12 tri-methylated on lysine 3 (RPL12K3me3). Co-localization of Corto and RPL12 with active transcriptional epigenetic marks on polytene chromosomes suggests that both proteins are involved in fine-tuning transcription of genes located in open chromatin. Investigation of Corto and RPL12 transcriptional targets by RNA-seq reveals that many are shared by both factors. Analysis of hsp70 occupancy by chromatin immunoprecipitation suggests that Corto and RPL12 cooperate in transcriptional regulation. Interestingly, the potential common targets of Corto and RPL12 are enriched in genes involved in heat response and ribosomal biogenesis.
To address the role of the Corto chromodomain in vivo, we used germline transformation and the binary UAS/Gal4 system to produce transgenic flies. These lines expressed either FLAG and HA double-tagged cortoCD fused to a nuclear localization signal coding sequence to force its entry into nuclei (FH-cortoCD), FLAG and HA double-tagged corto deleted of the chromodomain sequence (FH-cortoΔCD), or corto full-length (both FH-CortoΔCD and Corto full-length spontaneously enter the nucleus although no nuclear localization signal was detected, data not shown). Whereas transgenic flies ubiquitously over-expressing cortoΔCD [using either Actin5C (Act::Gal4>UAS::FH-cortoΔCD) or daughterless (da::Gal4>UAS::FH-cortoΔCD) drivers] were perfectly viable and had no visible phenotype, over-expression of corto using the same drivers was 100% lethal. Over-expression of cortoCD using again these drivers also induced high lethality at all developmental stages (from 63% to 100% depending on the transgenic line and the driver, Table S1). Escaper flies displayed rotated genitalia and duplicated macrochaetae as well as very penetrant homeotic phenotypes (Figure 1). Many flies presented a partial transformation of arista into leg, a homeotic phenotype called Aristapedia that could reflect down-regulation of the spineless-aristapedia gene [19]. Similar phenotypes were observed when over-expressing full-length corto using the weaker ubiquitous driver armadillo (arm::Gal4) (Table S1). Males over-expressing cortoCD also displayed smaller sex combs, a phenotype opposed to that of corto mutant males who have ectopic sex combs [15], [20], and which could reflect reduced expression of the homeotic gene Sex combs reduced (Scr) [21]. Taken together, these results suggest that the chromodomain is critical for Corto function.
Corto binds polytene chromosomes of third instar larva salivary glands at many sites [18]. To test the role of Corto chromodomain in chromatin binding, we immunostained polytene chromosomes of larvae over-expressing cortoCD in salivary glands [escargot Gal4 driver, (esg::Gal4>UAS::FH-cortoCD)] with anti-FLAG antibodies. FH-CortoCD bound polytene chromosomes at many discrete sites (Figure 2A). Like endogenous Corto, FH-CortoCD preferentially bound DAPI interbands and puffs, i.e. regions corresponding to open or actively transcribed chromatin. Comparison of endogenous Corto binding in wild-type larvae and FH-CortoCD binding in esg::Gal4>UAS::FH-cortoCD larvae at the tip of chromosome 3L showed that these proteins shared most of their binding sites (Figure 2B).
These results indicate that FH-CortoCD mimics Corto binding on polytene chromosomes and that the Corto chromodomain is a genuine chromatin-addressing module.
These results prompted us to identify the anchor(s) of Corto chromodomain on chromatin. We incubated GST-CortoCD covalently bound on agarose beads with nuclear or cytoplasmic extracts from embryos and resolved retained polypeptides by SDS-PAGE. Four bands between 30 and 15 kDa (P30, P21, P20 and P15) were consistently retained by GST-CortoCD and were enriched in peptide pull-down experiments performed with nuclear extracts versus cytoplasmic extracts (Figure 3A). The contents of the bands were identified by mass spectrometry. Surprisingly, all four bands contained ribosomal proteins (RPs): RPL7 for P30, RPS11 for P21, RPS10, RPL12 and RPL27 for P20, and RPS14 for P15 (Table S2). Although RPs are usually considered as contaminants, their consistent enrichment after incubation with nuclear extracts as well as the previously shown association of RPS11, RPL12 and RPS14 with polytene chromosomes [22] prompted us to consider their binding to CortoCD. These proteins might then interact with Corto directly on chromatin. To verify the interaction between RPs and CortoCD, we generated vectors to produce FLAG-tagged CortoCD supplied with a nuclear localization signal and Myc-tagged RPs in Drosophila S2 cells. Co-immunoprecipitations were performed on cell extracts from transfected cells, using either anti-FLAG or anti-Myc antibodies. No co-immunoprecipitation was observed between CortoCD and RPL7, RPS10 or RPS14 (Figure S1). However, anti-FLAG co-immunoprecipitated Myc-RPL12 with FLAG-CortoCD whereas anti-Myc co-immunoprecipitated FLAG-CortoCD with Myc-RPL12 (Figure 3B). In a similar experiment using FLAG-tagged full-length Corto, co-immunoprecipitation was again observed in both directions (Figure 3C). However, no co-immunoprecipitation was observed between FLAG-tagged CortoΔCD and Myc-tagged RPL12 (Figure 3D). These experiments demonstrate that RPL12 and Corto interact and that the Corto chromodomain is necessary and sufficient for this interaction. The identification of other RPs among the pulled-down polypeptides suggests that CortoCD interacts with a complex of RPs via a direct interaction with RPL12.
Since chromodomains typically recognize methylated lysines, we asked whether Corto chromodomain could bind a methylated form of RPL12. D. melanogaster RPL12 was aligned with RPL12 from several other species to identify conserved residues described to be methylated in some of them [23]–[25] (Figure 4A). Lysines 3, 10, 39 and 83, as well as arginine 67 fulfilled these criteria. Using site-directed mutagenesis, we replaced their codons with alanine codons in the Drosophila RPL12 cDNA, thus generating a series of mutants (RPL12K3A, RPL12K10A, RPL12K39A, RPL12R67A and RPL12K83A). These mutant cDNAs were introduced into a plasmid allowing their expression as mRFP-tagged proteins in Drosophila S2 cells. Similarly, the cortoCD cDNA, supplied with a nuclear localization signal, was introduced into a plasmid allowing its expression as an EGFP-tagged protein in S2 cells. When expressed in these cells, EGFP-CortoCD artificially entered the nucleus where it exhibited a punctuated pattern that recalled Polycomb bodies (Figure 4B) [26]. A similar nuclear pattern was observed after immunostaining untransfected S2 cells with anti-Corto antibodies. However, these “Corto bodies” did not overlap with Polyhomeotic (PH), a component of the PRC1 complex, but with RNA Polymerase II suggesting that they were transcriptional factories rather than Polycomb bodies (Figure S2). RPL12-mRFP expressed alone was present in the cytoplasm and the nucleus, where it appeared slightly punctuated (Figure 4B). Interestingly, when co-expressed with EGFP-CortoCD, all RPL12-mRFP localized in the nucleus (Figure 4C). Both proteins perfectly colocalized in a punctuated nuclear pattern, corroborating the interaction between CortoCD and RPL12 and suggesting that Corto could drive RPL12 in the nucleus. Similar experiments were carried out using the RPL12 mutant forms. Whereas RPL12K10A, RPL12K39A, RPL12R67A and RPL12K83A co-localized with CortoCD, RPL12K3A did not, strongly suggesting that RPL12 lysine 3 is required for Corto chromodomain-RPL12 interaction (Figure 4C).
To test whether Corto directly interacted with RPL12 lysine 3, we measured real-time binding between CortoCD and several RPL12 peptides using Biacore. GST-CortoCD and GST were immobilized on a CM5 sensor chip. Then, several RPL12 peptides [unmodified (RPL12um), methylated on lysine 3 (RPL12K3me2, RPL12K3me3), methylated on lysine 10 (RPL12K10me3) or lysine 3 mutated (RPL12K3A)] were assayed for their binding to GST-CortoCD or GST (Figure 5, Figure S3). None of these peptides bound GST. Furthermore, unmodified RPL12, RPL12K3me2, RPL12K10me3 and RPL12K3A peptides did not interact with CortoCD (no binding or unspecific binding i.e. KD>200 µM; Figure 5C). Only RPL12K3me3 interacted with high specificity with CortoCD (KD = 8 µM).
To investigate whether RPL12K3me3 could bind to other chromodomains, we repeated these experiments using that of HP1 (HP1CD). GST-HP1CD was immobilized on the sensor chip and binding of either RPL12, RPL12K3me3, RPL12K10me3 or RPL12K3A was tested. None of these peptides specifically interacted with HP1CD (KD>200 µM) (Figure 5C, Figure S3). Although no histones were revealed among peptides pulled down by CortoCD, we monitored binding of several histone H3 peptides to CortoCD. No binding of unmodified H3, H3K27me3, H3K9me3 or H3K4me3 peptides was observed (Figure 5D) while, as expected, the H3K9me3 peptide bound HP1CD with high affinity (KD = 0.4 µM). Surprisingly, the H3K27me3 peptide bound HP1CD with a similar affinity (KD = 0.7 µM), probably because sequences adjacent to the chromodomain (i.e. the hinge region) are required for selective targeting [27].
Altogether these data demonstrate that the Corto chromodomain specifically recognizes RPL12 trimethylated on lysine 3 (RPL12K3me3).
RPL12, along with 19 other ribosomal proteins, is known to bind polytene chromosomes of Drosophila larval salivary glands where it specifically associates with sites of transcription [22]. To investigate the potential role of the Corto-RPL12 interaction in gene expression regulation, we first analyzed the binding of these proteins on polytene chromosomes. For this, we generated Myc-tagged RPL12 transgenic fly lines (UAS::RpL12-Myc). Unlike corto or cortoCD, RpL12-Myc over-expression using ubiquitous Gal4 drivers (da::Gal4>UAS::RpL12-Myc or Act::Gal4>UAS::RpL12-Myc) induced no lethality and adult flies presented no visible phenotype except a shortened development (data not shown). RpL12-Myc was then expressed in salivary glands with the esg driver (esg::Gal4>UAS::RpL12-Myc) to test its binding to polytene chromosomes. RPL12-Myc bound polytene chromosomes at numerous sites, preferentially at DAPI interbands and puffs, suggesting that it mimics the binding of endogenous RPL12 [22] (Figure 6A). Co-immunostaining of RPL12 and the endogenous Corto protein showed that about 40% of the Corto sites were bound by RPL12 (Figure 6A, 6C). Simultaneous over-expression of FH-CortoCD and RPL12-Myc (esg::Gal4>UAS::FH-cortoCD,UAS::RpL12-Myc) established that CortoCD co-localized with RPL12 on a similar number of sites (Figure 6B).
Chromatin environment of Corto and RPL12 was further analyzed using antibodies against epigenetic marks (H3K27me3, H3K4me3) and RNA Polymerase II (paused, i.e. phosphorylated on serine 5: RNAPolIIS5p; elongating i.e. phosphorylated on serine 2: RNAPolIIS2p) (Figure 7 and Figure 8). In agreement with our Biacore analyses, Corto did not bind centromeric heterochromatin – marked by H3K9me3 – and did not overlap with H3K27me3 (except at the tip of chromosome X) (Figure 7A). Similarly, very few co-localizations with H3K27me3 were observed for RPL12-Myc (Figure 8A). Corto, as well as RPL12-Myc, partially co-localized with H3K4me3 (Figure 7B, Figure 8B). However, whereas Corto showed preferential co-localization with RNAPolIIS5p versus RNAPolIIS2p (Figure 7), RPL12 shared but few sites with RNAPolIIS5p and strongly co-localized with RNAPolIIS2p (Figure 8), as previously described [22].
Taken together, these data suggest that Corto and RPL12 mostly bind open, transcriptionally permissive chromatin.
To address the role of Corto and RPL12 in transcriptional regulation, we deep-sequenced transcripts from wing imaginal discs of third instar larvae over-expressing either FH-cortoCD or RpL12-Myc under control of the wing-specific scalloped::Gal4 driver (sd::Gal4>UAS::FH-cortoCD or sd::Gal4>UAS::RpL12-Myc) (hereafter called assays). Total RNA from the assays, the sd::Gal4/+ control or a w1118 reference line were isolated from pools of wing imaginal discs and subjected to RNA-seq on an Illumina high throughput sequencer. Sequence reads were aligned with the D. melanogaster genome to generate global gene expression profiles. Sequence reads of the assays were compared to sequence reads of the sd::Gal4/+ control. Differential analyses were performed to obtain adjusted P-values associated to expression changes for the assays compared to the sd::Gal4/+ control. In addition, sequence reads from the w1118 reference line were compared to sequence reads of the sd::Gal4/+ control. This reference was used to fix the threshold of the adjusted P-value to get only 1% of transcripts as differentially expressed in this control experiment (false discovery rate). By doing so, we obtained an adjusted P-value cutoff of 4.10−18. Using this threshold, we retrieved the highest expression variations from the two assays [with absolute log2(assay/control)>1]. 463 genes were upregulated when over-expressing cortoCD (Table S3). Among them, 314 were also upregulated when over-expressing RpL12, representing 75% of all genes upregulated by RpL12 over-expression (Table S4). Furthermore, 211 genes were down-regulated when over-expressing cortoCD (Table S5). Among them, 197 were also down-regulated when over-expressing RpL12, representing 67% of all genes down-regulated by RpL12 over-expression (Table S6). These results are summarized on Figure 9 and Table S7. They suggest that Corto and RPL12 share many transcriptional targets. Strikingly, analysis of Gene Ontology (GO) revealed that common upregulated genes were enriched in the “translation” (54.4% for Corto and 38.3% for RPL12) and “response to heat” (11.9% for Corto and 9.8% for RPL12) categories (Figure 10 and Tables S8, S9, S10, S11).
The high correlation between genes deregulated when over-expressing either cortoCD or RpL12 (R2 = 0.634) (Figure 9) as well as the numerous co-localizations of CortoCD and RPL12 on polytene chromosomes suggest that some deregulated genes were direct targets of Corto and RpL12. To test this hypothesis and to get insight in the functional interaction between Corto and RPL12, we focused on hsp70 that was one of the shared upregulated genes (Figure S4). We analyzed binding of CortoCD and RPL12 by chromatin immunoprecipitation before and after heat shock in wing imaginal discs (Figure 11). qPCR analyses were performed using a set of primers that cover the promoter and gene body of hsp70 [28].
At 25°C, in control w1118 discs, higher RNAPolII occupancy of the promoter as compared to the gene body suggests that hsp70 was paused, corroborating previous results [28]. In wing imaginal discs overexpressing either cortoCD or RpL12 (sd::gal4>UAS-FH-CortoCD or sd::gal4>UAS-RpL12-Myc), CortoCD and RPL12 bound hsp70 indicating that this gene was a direct target of both proteins. Simultaneously, RNAPolII binding was increased but kept the same profile suggesting that the gene was still paused but more loaded with RNAPolII. This could explain why more transcripts were generated. Thus, these data suggest that Corto, as well as RPL12, favors recruitment of RNAPolII on hsp70 in absence of heat shock.
After a short heat shock (5 minutes), CortoCD, as well as RPL12, were massively recruited on hsp70. Interestingly, CortoCD and RPL12 displayed the same binding profile i.e. increased binding from 5′ to 3′ of the gene body. CortoCD and RPL12 recruitment followed hsp70 transcription as revealed by enhancement of RNAPolII on gene body. Strikingly, recruitment of RNAPolII was higher in wing discs expressing cortoCD or RpL12 than in control wing discs, suggesting that Corto and RPL12 control transcriptional activation of hsp70.
Chromodomains play a critical role in addressing transcriptional regulators to chromatin. Investigating the role of the ETP Corto's chromodomain, we found that it is a typical chromodomain, acting as a chromatin-targeting module. Surprisingly, the Corto chromodomain does not bind methylated histones, as most known chromodomains do, but Ribosomal Protein L12 trimethylated on lysine 3 (RPL12K3me3). In agreement, RPL12 and Corto share many sites on polytene chromosomes. Transcriptomic analyses of wing imaginal tissues in which either cortoCD or RpL12 were over-expressed reveal that a large fraction of deregulated genes are common. Chromatin immunoprecipitation experiments reveal that CortoCD and RPL12 similarly bind one of these shared upregulated genes, hsp70, and are massively loaded on the promoter and gene body after heat shock. Hence, the ETP Corto and RPL12 might indeed be partners in regulation of some transcriptional targets.
The ETP Corto is a partner of Polycomb and Trithorax complexes and participates in epigenetic maintenance of gene expression, notably of homeotic genes [15],[17]. Multiple Corto binding sites on polytene chromosomes as well as pleiotropic phenotypes of corto mutants show that Corto transcriptional targets are numerous and involved in many developmental pathways. The interaction reported here between Corto and RPL12 raises the interesting possibility of a connection between RPs and epigenetic regulation of gene expression. Our previous investigations into Corto partners have highlighted its interaction with several PcG proteins, leading to the conclusion that Corto might regulate PRC1 and PRC2 functions [18]. Strikingly, RPs also co-purify with PRC1 [29]. Moreover, the ETP DSP1, that binds Corto, directly interacts with RPS11 [30]. Another ETP, ASXL1, belongs to the repressor complex H1.2 that also contains RPs [31]. Presence of RPs in the direct environment of chromatin binding factors, notably ETP, seems then to be a widespread situation. However, the role of RPs in these cases could be related to structure preservation and not to transcriptional regulation per se.
Apart from protein synthesis, RPs are involved in many cellular functions referred to as “extra-ribosomal” (reviewed in [32]). The first report on an RP's role in transcriptional regulation came from E. coli where RPS10 is involved in anti-termination of transcription [33]. Many eukaryotic RPs, notably RPL12, regulate their own transcription, basically by regulating their own splicing (reviewed in [34]).
For more than 40 years, many genetic screens to isolate new Polycomb (PcG) and trithorax (trxG) genes in flies have identified Minute mutants as PcG and trxG modifiers [14]. Indeed, Minute mutations suppress the ectopic sex comb phenotype of Polycomb or polyhomeotic mutants [35], [36]. D. melanogaster Minute loci are disseminated throughout the genome and many correspond to RP genes ([37] and references therein). Minute mutations might indirectly suppress phenotypes of PcG mutants by lengthening development, thus globally counteracting homeosis. However, Minute mutants can exhibit PcG mutant phenotypes, which is at variance with this assumption. For example, mutants in stubarista that encodes RP40 exhibit transformation of arista into legs [38].
Quasi-systematic presence of RPs at sites of transcription on Drosophila polytene chromosomes [22] as well as direct interaction between several RPs and histone H1 in transcriptional repression [39], suggest that RPs could actively participate in transcription modulation. Massive recruitment of Corto and RPL12 on hsp70 upon transcriptional activation as well as similarity between their occupancy profiles and the one of RNA polymerase II suggest that these two proteins could travel along the gene body together with the transcriptional machinery. Interestingly, BRM, the catalytic subunit of the SWI/SNF TrxG complex, associates with components of the spliceosome [40] that contains several RPs including RPL12 [41]. Overall, these findings lead us to favor the hypothesis of an active involvement of RPs in regulation of gene expression.
Whether individual RPs regulate transcription independently of other RPs or in the context of a ribosome-like complex is an interesting and much debated question (reviewed in [42]). Many data point to a collaborative role of RPs in transcription. In D. melanogaster, at least 20 RPs as well as rRNAs are present at transcription sites on polytene chromosomes, suggesting that they could be components of ribosome-like subunits [22]. Genome-wide ChIP-on-chip analyses of RPL7, L11 and L25 in S. pombe reveal a striking similarity of their binding sites, suggesting that they might bind chromatin as complexes [43]. Along the same line, mass spectrometry of Corto partners identified not only RPL12 but also RPL7, L27, S10, S11 and S14, indicating that Corto might interact via RPL12 with several RPs that could form a complex. Interestingly, RPL12 and L7 form a flexible protruding stalk in ribosomes that acts as a recruitment platform for translation factors [44]. Our results might point to the existence of pseudo-ribosomes composed of several RPs on chromatin. The role of RPs in nuclear translation has been very much debated and whether these pseudo-ribosomes are involved in translation is still unknown [45]. However, this possibility seems unlikely in view of the numerous data showing lack of translation factors in nuclei as well as association on chromatin between RPs and both nascent coding and non-coding RNAs [46]. Overall, these data suggest that pseudo-ribosomal complexes composed of various RPs are associated on chromatin and could thus participate in transcriptional regulation.
Like histones, RPs are subjected to a plethora of post-translational modifications including ubiquitinylations, phosphorylations, acetylations and methylations ([47] and references therein). We show here that the Corto chromodomain binds RPL12K3me3. Strikingly, the chromodomain protein CBX1, a human homolog of Drosophila HP1β, also interacts with RPL12 [48], suggesting that chromodomain binding to methylated RPL12 might be conserved. It is tempting to speculate about a role for RPL12 methylation in chromodomain protein recruitment to chromatin. This mechanism might be analogous to the one by which histone methylation marks, such as H3K27me3, recruit the PRC1 complex, i.e. by binding of the Polycomb chromodomain to methyl groups. Under this hypothesis, RPL12K3me3 might recruit Corto to chromatin. In yeast and A. thaliana, RPL12 can be trimethylated on lysine 3 by methyl-transferase SET11/Rkm2 [25], [47], [49]. Rkm2 is conserved in Drosophila and abundantly transcribed in S2 cells as well as all along development [50]. It would be interesting to determine whether this enzyme is an RPL12K3 methyl-tranferase in Drosophila.
Based on the existence of a panel of ribosomes composed of diverse RPs bearing various post-translational modifications, it was proposed that selective mRNA translation might depend on a ribosome code similar to the histone code [51]. Our results lead us to suggest that such a ribosome code might also concern regulation of gene transcription.
Surprisingly, GO analysis of RPL12 and Corto upregulated genes reveals that the “translation” and “structural component of ribosomes” categories are over-represented. Interestingly, the expression of RP genes decreases in RPL12A mutants in yeast [51]. Our finding that over-expression of Drosophila RpL12 increased RP gene expression reinforces the idea that RPL12 can activate RPs at the transcriptional level. Moreover, up-regulation of ribosome related genes is also observed in mutants of ash2 that encodes a TrxG protein, and that genetically interacts with corto [15], [52]. Hence RPL12, Corto and chromatin regulators of the TrxG family might all participate in dynamic coordination of ribosome biogenesis thus controlling cell growth. Intriguingly, we have recently shown that Corto interacts with an atypical cyclin, namely Cyclin G that also binds chromatin. This cyclin is suspected to control transcription of many genes, and controls cell growth [17], [53], [54]. These combined findings provide new avenues of research concerning transcriptional regulation of tissue growth homeostasis. Global regulation of genes involved in ribosome biogenesis could be a way to maintain this homeostasis. Co-regulation of genes involved in a given function has already been documented in eukaryotes. In Drosophila, housekeeping genes are co-regulated by the NSL complex and, in yeast, RPL12 coordinates transcription of genes involved in phosphate assimilation as well as RP genes [51], [55], [56]. As regulation of ribosome biogenesis is essential for cellular health and growth homeostasis [57], such a transcriptional co-regulation of RP genes might have evolved to insure that the cell's protein synthesis capacity can be rapidly adjusted to changing environmental conditions.
Clones and site-directed mutagenesis are described in Text S1. Primers are described in Table S12.
D. melanogaster stocks and crosses were kept on standard medium at 25°C. UAS::FH-cortoCD, UAS::FH-cortoΔCD and UAS::RpL12-Myc transgenic lines were established by standard P-element mediated transformation. Over-expression was carried out using Gal4 drivers either ubiquitous [daughterless (da::Gal4); Actin5C (Act::Gal4)], expressed in salivary glands [escargot (esg::Gal4)], or wing-specific [scalloped (sd::Gal4)]. Five females bearing the Gal4 driver were crossed with three males bearing the UAS transgene or w1118 as a control. Crosses were transferred to new vials every third day. The sd::Gal4, UAS::FH-cortoCD and UAS::RpL12-Myc lines were isogenized for six rounds with the isogenic w1118 line, prior to deep-sequencing, as described [58]. Lethality was calculated as described [59].
Cytoplasmic and nuclear extracts were prepared from 0–16 h embryos as described in [60]. GST or GST-CortoCD were covalently linked on agarose beads using the GST orientation kit (Pierce) following the manufacturer's instructions. 1 mg of protein extract was incubated with 200 µg of purified GST or GST-CortoCD in binding buffer [0.5 mM DTT, 0.1 mM EDTA, 4 mM MgCl2, 0.05% Igepal, 20 mM Hepes, 300 mM KCl, 10% glycerol, protease inhibitor cocktail (Roche)] for 1 h at 25°C. After 5 washes in binding buffer, bound polypeptides were resolved on a large 15% SDS–polyacrylamide gel and stained either with EZblue (Sigma) or with SilverQuest staining kit (Invitrogen). Bands were excised from the gel and were analyzed by LC-MS/MS mass spectrometry.
S2 cells were cultured at 25°C in Schneider's Drosophila medium (Lonza) supplemented with 10% heat inactivated fetal bovine serum and 100 units.mL−1 of penicillin and streptomycin. Cells were transfected using Effecten (Qiagen) as described [61].
Co-immunoprecipitations were performed as described [61] using anti-FLAG (Sigma F-3165) or anti-Myc (Santa Cruz, sc-40).
S2 cells were harvested 24 h after transfection and treated as described [62]. For each transfection, 30 to 60 nuclei were analyzed with an SP5 confocal microscope (Leica microsystems) using LAS (Image Analysis Software).
GST or GST fusion proteins were dialyzed using a Slide-A-Lyser cassette (Thermo Scientific) in running buffer (10 mM Hepes pH7.4, 150 mM NaCl, 3 mM EDTA, 0.005% P20 surfactant) (GE Healthcare). Real-time protein interaction assays were performed using a Biacore 3000. Kinetics and binding tests were first performed on empty surfaces. Data presented here result from substraction of empty surface RU (1 to 5 depending on the experiment) from active surface RU. GST was covalently coupled to a CM5 sensor chip (GE Healthcare) via its N-terminal amino acid. The carboxymethylated dextran surface was activated by injecting a mixture of 0.2 M 1-ethyl-3-(3-dimethylaminopropyl) and 0.05 M N-hydroxysuccinimide. GST was immobilized on the chip by injecting a 30 µg.mL−1 solution in NaAc pH5 buffer. GST-CortoCD and GST-HP1CD were immobilized by injecting a 100 µg.mL−1 solution in the same buffer. Binding tests were performed by injecting peptides at 1 or 10 µM in running buffer at a flow rate of 5 µL.min−1 during 5 min. Kinetic assays were performed only when the binding test was positive. Real-time monitoring was displayed in a sensorgram as the optical response (RU) versus time (s). To calculate association constants, peptides were diluted in series from 1 to 10 µM in running buffer and dilutions were injected sequentially at a flow rate of 5 µL.mn−1 during 5 min. Dissociation kinetics were then run during 10 min to calculate dissociation constants. Between assays, the chip was regenerated with 10 mM glycine pH2.0. Kinetic constants were calculated with BIAevaluation Software (Biacore) using the Fit kinetic simultaneous ka/kd (1∶1 binding; Langmuir algorithm). RPL12 peptides were synthesized at the IFR83 Peptide synthesis facility (Table S13). Histone peptides were provided by Diagenode (H3K27me3: sp-069-050; H3K4me3: sp-003-050; H3K9me3: sp-056-050; H3K4/K9um: sp-999-050; H3K27um: sp-998-050).
Polytene chromosome immunostainings were performed as described [63] for all antigens except RNA Pol II, for which we used experimental conditions described in [64]. Mouse anti-FLAG (1∶20) (Sigma, F-3165), mouse anti-Myc (1∶20) (Santa Cruz, sc-40), rabbit anti-H3K4me3 (1∶40) (Diagenode, pAB-003), rabbit anti-H3K27me3 (1∶70) (Diagenode, pAB-069), rabbit anti-RNA Pol II Ser2p (1∶200) (Abcam, an5095), rabbit anti-RNA Pol II Ser5p (1∶40) (Covance, MMS-134R) and rabbit anti-Corto (1∶30) [18] were used as primary antibodies. Secondary antibodies [Alexa Fluor 488 goat anti-rabbit IgG (Molecular Probes, A-11008), Alexa Fluor 594 goat anti-mouse IgG (Molecular probes, A-11005) and Alexa Fluor 488 goat anti-mouse IgG, IgA and IgM (Molecular Probes, A-10667)] were used at a 1∶1000 dilution.
Wing imaginal discs of third instar female larvae (one disc per larva) were dissected by batches of 50 in ice-cold PBS and frozen in liquid nitrogen. 300 discs (6 batches) were pooled and homogenized in lysis buffer using a FastPrep-24 during 20 s at 4 m.s−1 (MP Biomedicals, Lysing Matrix D). Total RNA were extracted using RNeasy kit (Qiagen). Library preparation and Illumina sequencing (multiplexed 50 bp paired-end sequencing on HiSeq 2000) were performed at the BC Cancer Agency Genome Sciences Center (Canada). Messenger (polyA+) RNAs were purified from 4 µg of total RNA with oligo(dT). Libraries were prepared using the bi-directional RNA-Seq library preparation kit (Illumina). A mean of 46±11 million reads was obtained for each of the 4 samples (w1118, sd::Gal4/+; sd::Gal4>UAS::cortoCD; sd::Gal4>UAS::RpL12). Detailed informations on Paired-End read counts at each step of the analysis workflow are available in Table S14. Before mapping, poly N read tails were trimmed, reads ≤11 bases were removed, and reads with quality mean ≤12 were discarded. Reads were then aligned against the D. melanogaster genome (dm3 genome assembly, BDGP Release 5.38) using Bowtie mapper (version 0.12.7) [65]. Alignments from reads matching more than once on the reference genome were removed using Java version of samtools. To compute gene expression, D. melanogaster GFF3 genome annotation from FlyBase (version 5.38) was used. All overlapping regions between alignments and referenced exons were counted.
Technical replicates coming from paired-end reads were first summed. Then, all samples were normalized together. Data were normalized according to the scaling normalization proposed by Robinson and Oshlack and implemented in the edgeR package version 1.6.10 [66]. A Fisher's Exact Test was then performed using the sage.test function of the statmod package version 1.4.6. Finally, a Benjamini and Hochberg (BH) P-value adjustment was made. The RNA-Seq gene expression data and raw fastq files are available at the GEO repository (www.ncbi.nlm.nih.gov/geo/) under accession number: GSE38435.
Wing imaginal discs of third instar larvae were dissected by batches of 100 in serum-free Schneider medium at room temperature. They were fixed in 500 µL of paraformaldehyde 1% in PBS for 10 minutes at room temperature under gentle agitation. Cross-link reaction was stopped by adding 50 µL of glycine 1.25 M. Fixed wing discs were washed 3 times with PBS, dried, flash-freezed in liquid nitrogen and stored at −80°C. Cell lysis was performed by adding 100 µL of lysis buffer (140 mM NaCl, 10 mM Tris-HCl pH8.0, 1 mM EDTA, 1% Triton X-100, 0.1% sodium deoxycholate, Roche complete EDTA-free protease inhibitor cocktail) complemented with 1% SDS, and sonicated in a Bioruptor sonifier (Diagenode). Conditions were established to obtain chromatin fragments from 200 to 1000 bp in length (30″ ON 30″ OFF, high power, 15 cycles). Pooled chromatin was centrifuged for 10 min at 13000 g at 4°C. The supernatant (soluble chromatin) was recovered and 5 µL were kept as input sample. For each IP, 10 µl of 50% (v/v) protein A or G coated paramagnetic beads (Diagenode) were washed once in lysis buffer, 1 µg of antibody was added, and beads were incubated for 2 h at 4°C on a rotating wheel. After washing, antibody coated beads were resuspended in 450 µL of lysis buffer and 50 µl of chromatin were added. After incubation on a rotating wheel overnight at 4°C, beads were washed at 4°C five times for 10 min each in lysis buffer, once in LiCl buffer (Tris-HCl 10 mM pH8, LiCl 0.25 M, 0.5% NP-40, 0.5% sodium deoxycholate, 1 mM EDTA) and twice in TE (10 mM Tris-HCl, pH 8.0, 1 mM EDTA). Immunoprecipitated as well as input DNAs were purified with the IPure kit following the manufacturer's instructions (Diagenode). Elution was performed twice with 35 µl of water. 2 µl of DNA were used per PCR. Real-time PCR data were normalized against Input sample and depicted as percentage of Input (see Table S12 for primers).
Antibodies used for chromatin immunoprecipitation were anti-RNA Polymerase II S2p (Abcam, ab5095), anti-HA tag (Abcam, ab9110) and anti-Myc tag (Abcam, ab9132). Mouse IgGs were used as a negative control (Mock, Diagenode).
Heat shock treatments were performed as previously described [28]. Briefly, wing discs were subjected to instantaneous heat shock by addition of an equal volume of 48°C pre-heated Schneider medium. After keeping tubes at 37°C for 5 minutes, discs were immediately cooled down by addition of 1/3 total volume of 4°C medium.
|
10.1371/journal.pntd.0004841 | Allovahlkampfia spelaea Causing Keratitis in Humans | Free-living amoebae are present worldwide. They can survive in different environment causing human diseases in some instances. Acanthamoeba sp. is known for causing sight-threatening keratitis in humans. Free-living amoeba keratitis is more common in developing countries. Amoebae of family Vahlkampfiidae are rarely reported to cause such affections. A new genus, Allovahlkampfia spelaea was recently identified from caves with no data about pathogenicity in humans. We tried to identify the causative free-living amoeba in a case of keratitis in an Egyptian patient using morphological and molecular techniques.
Pathogenic amoebae were culture using monoxenic culture system. Identification through morphological features and 18S ribosomal RNA subunit DNA amplification and sequencing was done. Pathogenicity to laboratory rabbits and ability to produce keratitis were assessed experimentally.
Allovahlkampfia spelaea was identified as a cause of human keratitis. Whole sequence of 18S ribosomal subunit DNA was sequenced and assembled. The Egyptian strain was closely related to SK1 strain isolated in Slovenia. The ability to induce keratitis was confirmed using animal model.
This the first time to report Allovahlkampfia spelaea as a human pathogen. Combining both molecular and morphological identification is critical to correctly diagnose amoebae causing keratitis in humans. Use of different pairs of primers and sequencing amplified DNA is needed to prevent misdiagnosis.
| Free-living amoebae are present worldwide. Some species are known to cause chronic keratitis in human. Amoebic chronic keratitis is sight-threatening disease occurring in both developing and well-developed countries. Allovahlkampfia spelaea is a newly discovered free-living amoeba. We report the first human case of chronic keratitis due to that amoeba. For correct identification, both morphological and molecular techniques should be combined.
| Free-living amoebae (FLA) are present in different environments worldwide. They can survive in soil, surface water, other aquatic environments and even desert[1,2]. Some members of family Acanthamoebidae and Vahlkampfiidae are amphizoic, occurring as human parasite. The most commonly known genera are Acanthamoeba and Naegleria causing keratitis and primary amoebic meningoencephalitis[3,4]. Members of both families have vegetative form, the trophozoite, and quiescent form, the cyst, both can be used for morphological identification of different genera [5–7]. Cysts can survive for many years in environment as a potential source of infection[8].
Amoebic keratitis is an uncommon corneal disease that could eventually lead to loss of vision. It is usually associated with contact lens wearing or trauma[9,10]. The most common cause is the genus Acanthamoeba [4,11]. In India, Acanthamoeba keratitis was up to 2.5% of cases of non-viral keratitis and was more prevalent in rural poor areas[12,13]. Vahlkampfia was reported to cause keratitis with co-infection with Acanthamoeba or Candida [11,14,15]. Correct diagnosis is essential for treatment and prevention of vision loss. Diagnosis depends mainly on culture from corneal scraping [9,10] and molecular identification using polymerase chain reaction (PCR). Primers designed to amplify 18S ribosomal subunit are widely used[16–18].
In 2009, Allovahlkampfia spelaea was identified as a new genus and new species. It was reported as a FLA inhabiting cave in Slovenia [19]. Until now, there is no data about the ability of such amoeba to produce disease in human beings.
In this work, we aimed to identify FLA causing keratitis in an Egyptian patient using morphological and molecular approaches.
Corneal scrapings from patient presented with keratitis were cultured on 1.5% non-nutrient agar made with Page’s saline and seeded with Escherichia coli kept in incubator at 30°C for 7 days. Cultures were examined using inverted microscope for presence of FLA every day and if FLA was detected, sub-culture was done every 10 to 14 days by inverting a slice on a new agar plate was done[9,20,21]. Morphology of trophozoites and cysts (non-stained and Giemsa’s Stained) was identified using light microscope and inverted microscope according to Smirnov and Goodkov (1999) and Smirnov and Brown (2004) [6,7]. Trial to axenize isolated FLA was done using Trypticase Soy Broth with Yeast Extract (TSY). Medium was prepared using BD Bacto Tryptic Soy broth (ref 211825) 30 grams, BD Bacto Yeast Extract (ref 212750) 10 grams and distilled water up to 1000 ml, pH adjust to 7.3 and autoclaved at 121°C for 15 minutes. To start axenic culture, cysts were collected using phosphate buffered saline (PBS) containing 0.01N HCL for 15 minutes, then washed 3 times in PBS by centrifugation at 600xg for 4 minutes. Cyst then were suspended in 10 ml TSY and kept in 25cm2 Falcon culture flasks (both vented and air tight flasks were tested).
Cysts were collected and treated as previous in axenic trial. Cysts were monitored under inverted microscope for excystation and attachment of trophozoites to flask wall, then medium was decanted and replaced for 3 times to ensure only attached trophozoites were present. Preheated cell lysis solution (80°C) from Gentra Puregene Yeast/Bact. Kit B (Qiagen) was added to flask to ensure rapid lysis of trophozoites. Then DNA was extracted according to kit protocol.
We used 3 different pairs of primers to identify the FLA; JDP1 (5'GGCCCAGATCGTTTACCGTGAA) and JDP2 (5'TCTCACAAGCTGCTAGGGAGTCA)[16] as standard known primers to identify Acanthamoeba. Universal primers F-566 (5'CAG CAG CCG CGG TAA TTC C) and R-1200 (5' CCC GTG TTG AGT CAA ATT AAG C) as general primers for 18S ribosomal subunit [22], and Naeg-F (5'GAACCTGCGTAGGGATCATTT) and Naeg-R (5'TTTCTTTTCCTCCCCTTATTA) as general primers for ribosomal internal transcribed spacers (ITS) [3,20]. PCR reactions were done using Thermo Scientific Phusion High-Fidelity PCR Master Mix (ref F-531L) according to product manual for 35–40 cycles. PCR products were identified using 1% Agarose gel stained with ethidium bromide and were purified for sequencing using Qiaquick spin columns (Qiagen). Sequencing was done using Applied Biosystems 3730xl DNA Analyzer. Sequence data were retrieved and blasted using NCBI Blastn engine.
After identification of isolated FLA as Allovahlkampfia spelaea, 3 pair of specific primers were designed using NCBI/Primer-BLAST using A. spelaea strain SK1 [GenBank:EU696948] as template (shown in Table 1). Amplification and sequencing were done as previously described.
Phylogenetic trees were created using Mega6 platform [23], choosing MUSCLE [24] for multiple alignment, and maximum likelihood tree using Tamura-Nei model for generation of tree with gaps removal. Sequences for other amoebae were downloaded from GenBank [accession numbers for 18S ribosomal RNA gene, GenBank: JQ271723, M98052, AJ224887, M18732, FJ169185, GU230754, U94740, AF251938, EU696948, AY425009, AY029409, DQ388520 and for ITS, GenBank: KC820644, AB330071, AJ698838, FJ169186, AJ698839, V00003, AJ132032, K00471, EU696949, KF547910].
In order to confirm the ability of A. spelaea to produce keratitis, cysts were collected and treated as previously described in axenic trials then suspended in sterile Page’s saline. They were allowed to excyst in culture flask and supernatant was discarded and replaced. The flask was chilled on ice for 3 minutes and shaken to collect trophozoite. Counting was done using hemocytometer and volume was adjusted to have 1x105 trophozoites/ml. Three laboratory rabbits weighing 1500–2000 gms about 3 months old were used for induction of keratitis. Each rabbit was anesthetized using ether, the left cornea was scratched using 27 gauge sterile needle and 10 μl were instilled in its eye. It was kept under anesthesia for 30 minutes to allow adherence of trophozoites. Right eye was only scratched with 27gauge sterile needle. Eyes were examined for the presence of keratitis grossly and with slit lamp.
Corneal scrapings were obtained as routine investigation from patient with chronic keratitis. Patient made written consent for using his samples for both diagnostic and research purposes. Faculty of Medicine research ethics committee, Assiut University, approved this study. Animal experiments were done in Animal House, Faculty of Medicine, Assiut University. Animal House ethical committee, Faculty of Medicine, Assiut University, approved them. Animal handling protocols meet the standard international guidelines by the National Institutes of Health guide for the care and use of Laboratory animals and guideline used in other Egyptian universities and research centers.
A case of a middle-aged patient with history of trauma to eye and resistant keratitis showed amoeboid trophozoites, which did not resemble Acanthamoeba morphologically, on third day of culture. They were pleomorphic with single vesicular nucleus. Movement tends to be unidirectional but not always. Cysts are round with single cell wall and single nucleus with clear perinuclear ring. They tend to aggregate in groups. Once cysts were moved to water, they rapidly excysted and some of the emerging trophozoites showed filopodia (Fig 1). These morphological characters are similar of vahlkamphid amoeba as previously described by Smirnov and Goodkov (1999), Pélandakis and Pernin (2002) Smirnov and Brown (2004), Walochnik and Mulec (2009) and González-Robles et.al. (2012) [3,6,7,15,19]. Trials to grow isolated FLA on TSY medium axenically were done with no success to continuously maintain them.
JDP1 and JDP2 primers were used as standard diagnostic primers for Acanthamoeba, but we got a faint band at expected band size (about 515bp). As the morphology of isolated FLA was similar to vahlkamphid amoebae, we tried another 2 pairs of primers. We had good amplification band at universal primers F-566 and R-1200 primers (about 750 bp) and a main 550bp band for Naeg-F and Naeg-R primers with auxiliary faint band (about 750bp) (Fig 2A). Such auxiliary band is indicative of long ITS variant gene[3].
Bands except auxiliary band for ITS gene were excised from gel. DNA was purified and sequenced except for JDP1and JDP2 faint band where another run of amplification was used before sequencing to increase DNA yield. Sequence obtained from JDP1and JDP2 was of low quality indicating non-specific amplification and blast result showed query cover of 59% and identity of 93% for with E value 0 with Alcaligenes faecalis strain ZD02 (CP013119.1), a free-living gram-negative bacteria that could co-exist with FLA amoebae. JDP1and JDP2 are used as specific primers for Acanthamoeba as suggested by Schroeder et.al. (2001) [16]. This finding confirms that no co-infection with Acanthamoeba occurred as the obtained sequence showed no match. It also should raise awareness that some cases of claimed Acanthamoeba keratitis diagnosed by PCR technique only could be due to other FLA. The use of JDP1and JDP2 primers is common in clinical practice to diagnose amoebic keratitis without culture. The appearance of such faint band could occur due to binding of primers to non-target sequence leading to weak signal; this is confirmed by sequencing results that showed no significant matches.
Blasting the sequence we had using universal primers F-566 and R-1200 for highly similar sequences, the sequence matched mainly Allovahlkampfia with a query cover of 96% and identity of 96% for A. spelaea strain SK1 and 71–96% for other strain of Allovahlkampfia (S1 Fig). Also, blasting the sequence we had using Naeg-F and Naeg-R primers main band, the sequence matched mainly Allovahlkampfia with a query cover of 95% and identity of 87% for A. spelaea strain SK1 with E value 7e-141 and 71–95% for other strain of Allovahlkampfia (S2 Fig).
Using the sequence of A. spelaea strain SK1 18S ribosomal RNA gene (accession number EU696948), we designed 3 pairs of primers to amplify the whole length of 18S ribosomal gene. Amplification was done successfully and DNA was extracted and sequenced (Fig 2B). Resulting sequences were merged using EMBOSS merger (http://emboss.bioinformatics.nl/cgi-bin/emboss/merger) and it was blasted. The merged sequence matched that of A. spelaea strain SK1 with a query cover of 97% and identity of 97% with E value of zero. The sequence showed only matching to members of family Vahlkampfidae. A. spelaea were first described in 2009 as FLA of caves[19]. This the first time to describe this FLA in Egypt and it is the first time to report it as human pathogen.
Using sequences for ITS and complete 18S ribosomal RNA sequence to create maximum likelihood phylogenetic trees showed that the Egyptian eye strain is closely related to A. spelaea strain SK1 (Fig 3). This indicates that A. spelaea can be more widely distributed in the world as only limited differences are present between SK1 strain and our strain. All sequences can be found in S1–S9 Files.
To evaluate the pathogenicity of A. spelaea, rabbit model was chosen as the size of the rabbit’s eye allows easier detection of lesions and easier examination with slit lamp. After 24 hours post inoculation (pi), right eyes of rabbits showed normal cornea while left eyes showed mild keratitis with redness of conjunctiva. On the 3rd day pi, evident corneal ulcers were detected using battery light with or without staining with methylene blue or fluorescein. Rabbits were irritable once light focused on their head. On 5th day pi, ulcer became very large and the degree of conjunctival congestion was not proportional to the size of the ulcer (Fig 4). On 7th day pi, ulcer destroyed most of corneal epithelium in two rabbits and one rabbit showed total loss of epithelium leaving only the basal membrane. These findings prove that A. spelaea Egyptian eye strain is capable of inducing keratitis in contrary to what was previously speculated with closely related Vahlkampfia sp. in which keratitis was thought to be due to mixed infection with Acanthamoeba, Hartmannella or Candida[11,14,25].
Further evaluation is needed to judge the ability of Allovahlkampfia sp. to produce keratitis in totally healthy cornea with no history of trauma or friction keratitis as Demirci et.al. (2006) reported Acanthamoeba keratitis in a 5 years old child with no history of trauma or contact lens wearing[10]. The corneal changes appear to be similar to what happens in human case of Acanthamoeba keratitis where early in infection only herpetic-like lesions appear which progress in few days to larger ulcer.[10]
Allovahlkampfia spelaea can cause keratitis in humans. This is the 1st time to report such parasite as a human parasite. The presence of FLA in coexistence with each other and with bacteria and fungi makes it necessary to combine both culture and molecular methods for correct diagnosis. For correct molecular diagnosis, use of different primers and sequencing of amplified DNA are important for correct identification of parasite. The close genetic relation between strain isolated in Slovenia and Egypt suggests that the genome of Allovahlkampfia spelaea is not much evolutionary separated but further analysis using full genome sequence is needed.
|
10.1371/journal.pgen.0030202 | Developmental Role and Regulation of cortex, a Meiosis-Specific Anaphase-Promoting Complex/Cyclosome Activator | During oogenesis in metazoans, the meiotic divisions must be coordinated with development of the oocyte to ensure successful fertilization and subsequent embryogenesis. The ways in which the mitotic machinery is specialized for meiosis are not fully understood. cortex, which encodes a putative female meiosis-specific anaphase-promoting complex/cyclosome (APC/C) activator, is required for proper meiosis in Drosophila. We demonstrate that CORT physically associates with core subunits of the APC/C in ovaries. APC/CCORT targets Cyclin A for degradation prior to the metaphase I arrest, while Cyclins B and B3 are not targeted until after egg activation. We investigate the regulation of CORT and find that CORT protein is specifically expressed during the meiotic divisions in the oocyte. Polyadenylation of cort mRNA is correlated with appearance of CORT protein at oocyte maturation, while deadenylation of cort mRNA occurs in the early embryo. CORT protein is targeted for degradation by the APC/C following egg activation, and this degradation is dependent on an intact D-box in the C terminus of CORT. Our studies reveal the mechanism for developmental regulation of an APC/C activator and suggest it is one strategy for control of the female meiotic cell cycle in a multicellular organism.
| Meiosis is a modified cell cycle that generates four gametes, each containing half the genetic content of the parent cell, through a reductional division followed by an equational division without an intervening DNA synthesis phase. During oogenesis of multicellular organisms, proper coordination of the meiotic divisions with the development of the oocyte is crucial for successful fertilization and the initiation of zygotic development. Very little is known about how general cell-cycle regulators as well as meiosis-specific regulators contribute to this coordination. In this study we describe the role and developmental regulation of cortex, a meiosis-specific activator of the anaphase-promoting complex/cyclosome (APC/C). CORT protein physically associates with the APC/C and triggers the sequential degradation of mitotic cyclins in meiosis. We find that cortex is subject to both post-transcriptional and post-translational regulatory mechanisms, which result in expression of CORT protein being restricted to the meiotic divisions. This developmental regulation may be important for proper meiosis as well as the transition from the completion of meiosis to mitotic divisions in the early embryo.
| Developmental regulation of meiosis is crucial for generating viable eggs and sperm and, thus, a successful fertilization event. Meiosis is a modified cell cycle in which segregation of homologous chromosomes is followed by segregation of sister chromatids without an intervening S phase. These unique divisions are controlled by general mitotic cell-cycle regulators as well as specialized meiotic proteins [1]. During oogenesis in multicellular organisms, meiosis presents a particular regulatory challenge. The meiotic divisions must be coordinated tightly with growth and development of the oocyte to allow for oocyte differentiation and to ensure that the completion of meiosis is coordinated with fertilization. To achieve this coordination, oocytes arrest at prophase I and again at metaphase I or metaphase II and are released from these arrests through processes called oocyte maturation and activation, respectively [2,3]. Furthermore, additional specialized cell-cycle regulation is required for the transition between meiosis and restart of the cell cycle in embryogenesis. In Drosophila, meiosis is completed without cytokinesis in the same common cytoplasm in which the rapid mitotic divisions of embryogenesis begin. Upon fertilization, the oocyte must quickly inactivate meiotic regulators to prevent interference with embryonic mitotic cycles. The ways in which general mitotic proteins act together with meiosis-specific proteins to meet the multiple regulatory challenges of meiosis in metazoans are not well understood.
The anaphase-promoting complex/cyclosome (APC/C) plays a critical role in mitosis, but much remains to be understood about its function in meiosis. The APC/C is a large E3 ubiquitin ligase composed of at least 12 core subunits, which targets specific substrate proteins for degradation by the 26S proteasome [4]. In mitosis, the APC/C is crucial for proper cell division through targeting of key substrates. Securin, an inhibitor of separase, must be degraded to allow for cleavage of cohesin and subsequent segregation of sister chromatids, and mitotic cyclins must be degraded to allow for the metaphase to anaphase transition and events associated with mitotic exit [5–8]. In addition, the APC/C targets many other proteins for degradation including proteins involved in spindle function and regulators of DNA replication [9,10].
Substrate specificity is conferred to the APC/C by activator proteins Cdc20/Fizzy and Cdh1/Fizzy-related, which recognize substrate proteins containing D-box or KEN box motifs [11–15]. Regulation of these specificity factors is one crucial way by which APC/C activity is modulated. Cdc20 is transcribed and translated during S phase and G2, phosphorylated in mitosis, and degraded in an APC/CCdh1-dependent manner in G1 [13,16–19]. Phosphorylation of several APC/C subunits in mitosis facilitates the ability of Cdc20 to bind to and activate the APC/C [18,20–23]. Levels of Cdh1 are constant in mitosis and lowered in late G1 and S, but inhibitory phosphorylation of Cdh1 prevents its association with APC/C during S, G2, and M phases [16,18,24]. Thus, differential regulation of Cdc20 and Cdh1 directs their transient association with the APC/C at different times during the cell cycle to target specific subsets of proteins for degradation.
In meiosis, the role of the APC/C and its regulation is less clear. An appealing hypothesis is that the meiotic divisions are driven in part by the degradation of specific meiotic APC/C substrates, and thus, the APC/C must require unique regulation during these divisions. In yeast, it is known that disjunction of homologous chromosomes in meiosis I and sister chromatids in meiosis II requires APC-mediated destruction of Pds1/securin to release separase for cleavage of cohesin [25–27]. In multicellular organisms, however, a requirement for the APC/C in meiosis has been more difficult to demonstrate. Mutations in or RNA interference against APC/C subunits in Caenorhabditis elegans result in a metaphase I arrest [28,29]. In Drosophila, mutations in fzy cause both meiosis I and meiosis II arrests [30]. Several studies in mouse oocytes have shown that APC/C-mediated protein degradation is required for homolog disjunction and polar body extrusion [31–34]. However, inhibiting APC/C subunits by depletion or antibody injections in Xenopus laevis does not prevent the metaphase I to anaphase I transition but does cause arrest in metaphase II [35,36]. The reasons behind this apparent inconsistency in Xenopus remain unknown.
One way in which the APC/C may be regulated uniquely in meiosis is through its association with meiosis-specific activators. Cdc20/FZY family members that are expressed exclusively in meiosis have been identified in yeast. In Saccharomyces cerevisiae, Ama1 activates the APC/C after meiosis I and is required for spore wall assembly [37–39]. Similarly, in S. pombe, fzr1/mfr1 is required for proper spore formation after the completion of the meiotic divisions [40,41]. The identification and study of meiosis-specific APC/C activators are starting points from which to understand the unique regulation of the APC/C during meiosis as well as to identify meiosis-specific substrates of the APC/C.
Drosophila provides the best candidate for a meiosis-specific APC/C activator in metazoans. cortex (cort) encodes a distant member of the Cdc20/FZY family and is expressed exclusively in oogenesis and early embryogenesis [42]. cort is required for proper female meiosis. Eggs laid by cort mutant mothers display aberrant chromosome segregation in meiosis I and arrest terminally in metaphase II [43,44]. In addition, cort mutant eggs contain elevated levels of mitotic cyclins, and misexpression of cort causes a decrease in levels of mitotic cyclins [30]. cort presents a unique opportunity for understanding the function and the developmental regulation of the APC/C during meiosis in a multicellular organism. Drosophila is an ideal system for studying female meiosis because different meiotic stages can be distinguished easily by cytology and isolated by microdissection. In this study, we demonstrate that CORT interacts biochemically with the APC/C during female meiosis and reveal a mechanism for developmental regulation of cort through both post-transcriptional and post-translational processes.
A recent study suggested that the cortex gene encodes a functional activator of the APC/C [30]. This assertion was based on the ability of cort to affect cyclin protein levels. Levels of mitotic cyclins are elevated in cort mutants, and misexpression of cort causes decreased levels of cyclins in wing imaginal discs. While these data are strong evidence of cort's function as an activator of the APC/C, demonstration of a physical association between CORT and the APC/C during oogenesis is lacking. We looked for an association between CORT and the APC/C by co-immunoprecipitation. We made transgenic lines with a MYC-tagged form of cort under control of the UAS response element and drove MYC-CORT expression in the female germline using nos-gal4. MYC-CORT is functional, because expression of this transgene rescued the meiotic arrest in progeny of cortRH65 mutant females (Table S1).
We immunoprecipitated MYC-CORT from ovary extracts using a MYC antibody. Cdc27 and Cdc16, tetratricopeptide repeat core subunits of the APC/C, co-immunoprecipitate with MYC-CORT but not with a control mouse immunoglobulin G (IgG) (Figure 1A) (unpublished data). Morula (MR), the APC2 homolog in Drosophila, does not co-immunoprecipitate with MYC-CORT (Figure 1A). Data from S. cerevisiae suggest that the APC/C exists in two distinct subcomplexes bridged together by Apc1. One subcomplex contains Apc2 and Apc11, while the other contains the tetratricopeptide proteins Cdc27, Cdc16, and Cdc23 [45]. Failure of MR to co-immunoprecipitate with MYC-CORT may be explained by a tighter and more direct association of CORT with the tetratricopeptide protein-containing subcomplex but not with the Apc2-containing subcomplex. Furthermore, buffer conditions may be causing CORT to dissociate from MR, as extensive high-salt washes of human APC cause dissociation of Apc2 and Apc11 from the rest of the complex [46]. On the basis of previous studies of APC/C, we still think it is likely that CORT acts together with MR and Cdc27 in one complex. As an additional control, we immunoprecipitated an unrelated MYC-tagged protein, PLU-MYC, from ovary extracts to confirm that the associations of Cdc27 and Cdc16 with CORT are specific. Neither Cdc27 nor Cdc16 co-immunoprecipitate with PLU-MYC, indicating that they are associating with CORT and not with the MYC tag (Figure 1A) (unpublished data).
In the reciprocal experiment, we immunoprecipitated Cdc27 from ovary extracts (Figure 1B). MYC-CORT co-immunoprecipitates with Cdc27 but not with a control rabbit IgG, suggesting that there is a strong physical interaction between CORT and Cdc27.
In addition to demonstrating the physical association between CORT and the APC/C, we identified motifs in the CORT protein sequence that have been shown to contribute to binding of APC/C activator proteins to the APC/C (Figure 1C). CORT contains an internal motif called the C-box that is important for binding to the APC/C and is conserved in Cdc20 and Cdh1 proteins throughout many species [47,48]. In addition, CORT contains the C-terminal IR (isoleucine-arginine) tail motif that is present in all APC/C activators as well as Doc1 and has been shown to mediate a direct interaction between Cdh1 and Cdc27 [46,49]. Together, the co-immunoprecipitation of CORT with core subunits of the APC/C in oogenesis and the presence of conserved motifs in the CORT sequence confirm CORT's identity as a meiosis-specific APC/C activator.
An additional way to determine if cort functions as an APC/C activator is to ask whether it can provide the function of another APC/C activator when over- or misexpressed. To investigate whether cort acts similarly to fzy, we asked if cort can functionally substitute for fzy in the early embryo. fzy6/fzy7 females lay eggs that arrest in metaphase after a few rounds of mitotic divisions [50]. If cort can provide fzy function, we would expect these embryos to progress further in embryogenesis.
We overexpressed myc-cort in the germline of fzy6/fzy7 females, collected embryos from these females, visualized the DNA and the spindles by immunofluorescence, and counted the number of mitotic spindles. Embryos overexpressing cort did not contain more mitotic spindles compared to fzy alone, thus we did not observe any rescue of the fzy phenotype. In contrast, overexpression of cort slightly worsened the fzy phenotype (presented below). This result suggests either that cort does not function as an APC/C activator, or, more likely, that CORT confers significantly different substrate specificity to the APC/C than FZY and, therefore, cannot provide its function.
In mitosis, cyclins are degraded sequentially by the APC/C, with Cyclin A being degraded prior to Cyclin B and Cyclin B3 [51,52]. For a more detailed analysis of CORT function in meiosis, we examined levels of putative APC/CCORT substrates at different time points during the meiotic divisions. We can isolate different meiotic stages by dissecting egg chambers from ovaries: immature ovaries with egg chamber stages 1–12 contain oocytes arrested in prophase I, and mature stage 14 oocytes are arrested in metaphase I.
We performed western blots on extracts from immature egg chamber stages and stage 14 oocytes to assay levels of Cyclin A at these meiotic time points in cort mutants. We found that Cyclin A levels are slightly reduced in cort mutant immature ovaries (stages 1–11), although the significance of this effect is not clear as we have not observed any defects in these stages of cort mutant ovaries. Cyclin A levels are elevated in mutant stage 14 oocytes compared to a heterozygous control (Figure 2A). The heterozygous control blots indicate that Cyclin A is normally degraded at some point between release of the prophase I arrest and establishment of the metaphase I arrest. In cort mutants, the failure to degrade Cyclin A by the metaphase I arrest indicates that APC/CCORT is required for Cyclin A degradation at this time. In contrast, levels of Cyclin B, Cyclin B3, and PIMPLES (PIM), the securin homolog in Drosophila, are not elevated in cort mutant stage 14 oocytes compared to heterozygous controls, suggesting that these proteins are not subject to degradation by APC/CCORT at this developmental stage (Figure 2B).
Upon egg activation in Drosophila, the metaphase I arrest is released, and meiosis is rapidly completed as the egg is ovulated and laid [53]. Meiosis is completed regardless of whether the oocyte is fertilized. Thus, unfertilized eggs represent a time point just after the completion of meiosis. We examined cyclin levels in eggs laid by cort mutant females, which do not complete meiosis (Figure 2C). As a control we used heterozygous unfertilized eggs. We found that all three mitotic cyclins, as well as PIM, are elevated in cort mutant eggs, which suggests that APC/CCORT targets all of these substrates for degradation after release of the metaphase I arrest. Complementary results for cyclin levels have been previously observed [30]. This degradation may take place at the metaphase I to anaphase I transition, the metaphase II to anaphase II transition, or during both transitions. Cyclin B, at least, is likely degraded at both transitions as expression of a nondegradable form of Cyclin B in the female germline results in both meiosis I and meiosis II arrests [30].
The sequential CORT-dependent degradation of cyclins we observe in Drosophila female meiosis is parallel to observations made in mitosis that degradation of Cyclin A begins just after nuclear envelope breakdown in prometaphase, while degradation of Cyclin B is initiated at the beginning of metaphase [52,54–56]. Nuclear envelope breakdown occurs in Drosophila female meiosis in stage 13, and just after this stage is when we see an increase of Cyclin A but not Cyclins B or B3 in cort mutants compared to heterozygous controls. We see CORT-dependent degradation of all three cyclins in eggs, consistent with degradation of Cyclin B and Cyclin B3 not occurring until the metaphase I to anaphase I transition. To our knowledge, this is the first observation of sequential cyclin degradation during meiosis in a metazoan.
Given the difference in timing of CORT-dependent degradation of cyclins, we examined the protein expression pattern of CORT during meiosis to see if differential protein regulation of CORT correlates with differential cyclin degradation. Eggs laid by cort mutant mothers arrest in metaphase II and never complete meiosis [44]. This strong arrest phenotype indicates a critical role for cort specifically in meiosis. However, cort mRNA is present throughout oogenesis and early embryogenesis, suggesting a much broader developmental role [42]. We determined the timing of CORT protein expression to define better the scope of its activity.
To investigate the developmental control of CORT protein expression, we made a polyclonal antibody against a glutathione S-transferase (GST)-tagged N-terminal fragment of CORT. Anti-CORT serum recognizes a band of approximately 47 kDa in wild-type oocyte extracts (Figure 3A). To test for antibody specificity, we probed oocyte extracts from cortRH65 mutants that contain a cort allele with a premature stop codon [42]. The serum does not recognize a band of the same size in these mutants. We also did not detect an N-terminal fragment in this mutant extract. In addition, we probed extracts from grauzone (grau) mutant oocytes. grau encodes a transcription factor that activates expression of cort [57]. The CORT band of 47 kDa is reduced in grau mutants, confirming the specificity of our antibody.
CORT expression is specific to oogenesis, as we detected a CORT band in whole female fly extracts but not in female fly extracts from which the ovaries were removed (Figure 3A). We also did not detect CORT in whole male fly extracts. We performed developmental western analysis on different stages of oogenesis to determine specifically when CORT protein is expressed (Figure 3B). CORT is undetectable in early stage 1–8 egg chambers, and very low levels are detectable in stages 9–10B egg chambers. CORT levels increase dramatically in stage 12–13 egg chambers and remain high in mature stage 14 oocytes. The appearance of CORT protein occurs at the same time that Cyclin A degradation is triggered (Figure 2A), indicating that APC/CCORT targets Cyclin A as soon as CORT protein is expressed, while simultaneously being prevented from targeting Cyclins B and B3 and PIM until after release of the metaphase I arrest.
The timing of appearance of CORT protein correlates with timing of the unmasking of maternal mRNA by cytoplasmic polyadenylation. Many organisms utilize cytoplasmic polyadenylation as a strategy to turn on the translation of specific transcripts at specific developmental time points [58]. Elongation of the poly(A) tail of these transcripts is thought to allow for a stable closed-loop conformation of the translational machinery and thus to activate translation. This process occurs during oocyte maturation when oocytes are released from prophase I arrest to reenter the meiotic cell cycle [2]. In Drosophila, oocyte maturation takes place in stage 13 of oogenesis [59]. Given the correlation of the appearance of CORT protein with the timing of oocyte maturation, we investigated the lengths of cort poly(A) tails at different developmental time points.
We conducted ligase-mediated poly(A) tail (PAT) assays on immature egg chambers and mature stage 14 oocytes to determine if the poly(A) tail length of cort changes upon oocyte maturation [60]. We observed an elongation of the poly(A) tail in stage 14 oocytes compared with stage 1–11 egg chambers (Figure 3C). Poly(A) tails are approximately 20 As in immature stages and elongate to approximately 70 As in mature oocytes. As a positive control, we measured the poly(A) tail length of cyclin B mRNA in these stages, because cyclin B mRNA is known to be polyadenylated upon oocyte maturation (Figure 3C) [61]. We observed a similar increase in cyclin B mRNA poly(A) tail lengths as has been previously shown. The appearance of CORT protein in stage 13 of oogenesis when oocyte maturation occurs together with the elongation of cort's poly(A) tail in mature oocytes suggests that cort translation is developmentally regulated by cytoplasmic polyadenylation.
If APC/CCORT activity is necessary for meiosis but dispensable for mitosis, cort may be inactivated in the early embryo to prevent its association with a mitotic APC/C complex. Early Drosophila embryos are transcriptionally quiescent, so post-transcriptional control is essential for regulating the activity of maternal gene products. In many organisms, egg activation triggers destabilization of a subset of maternal transcripts [58]. As deadenylation is often the rate-limiting step in mRNA decay, we investigated the polyadenylation status of cort mRNA after the completion of meiosis.
We performed PAT assays to measure the poly(A) tail length of cort mRNA in mature stage 14 oocytes and 0–1-h embryos. We found that cort mRNA is deadenylated in early embryos compared to mature oocytes (Figure 4A). The tail decreases from a length of approximately 70 As to 20 As. We used cyclin B mRNA as a positive control that, in contrast, is further polyadenylated upon egg activation [61]. CCR4 is the main catalytic subunit of the Ccr4-Pop2-Not deadenylase complex in S. cerevisiae [62]. A CCR4 homolog exists in Drosophila and has deadenylase activity [63]. We measured the poly(A) tail length of cort mRNA in embryos from ccr4 mutant mothers and found that cort's poly(A) tail length is elongated in the mutant (Figure 4B). Thus, deadenylation of cort in the early embryo is dependent on the conserved CCR4-NOT deadenylase complex.
We performed western analysis on CORT after the completion of meiosis to determine when CORT protein is expressed in the early embryo (Figure 5A). Surprisingly, we found that CORT protein levels drop dramatically by the time meiosis is completed in unfertilized eggs. We detect CORT at very low levels for up to 1.5 h of embryogenesis before it disappears. Given the rapid timing of CORT degradation by the end of the meiotic divisions, we hypothesized that CORT itself may be a target of the APC/C.
To test whether the APC/C plays a role in CORT degradation, we looked at CORT levels in mr/APC2 mutants (Figure 5B). We found that CORT levels are unaffected in mr mutant ovaries, but CORT levels increase strongly in eggs from mr mutant females. As a positive control, we probed for Cyclin B in these samples and found that it is also elevated in mr mutant eggs (unpublished data). These results parallel the timing of changes in CORT protein levels in a wild-type background; CORT levels normally drop by the time that meiosis is completed, and, similarly, the dependence of CORT degradation on mr is only apparent in unfertilized eggs, in which meiosis has been completed. These results strongly suggest that CORT is targeted for degradation by the APC/C at some point after the release of the metaphase I arrest and by the time that the meiotic divisions are completed. The specific timing of CORT degradation suggests that it is critical for development of the embryo that CORT protein levels be greatly reduced by the time meiosis is completed.
The APC/C targets proteins for degradation through recognition of specific motifs in its substrates. The two most common motifs are D-boxes (R-X-X-L-X-X-X-X-N) and KEN boxes (K-E-N-X-X-X-E/D/N), although additional motifs have been identified [4,14,64]. We identified a putative D-box in the C terminus of CORT (residues 424–432) but no KEN box (Figure 5C).
To determine whether the putative D-box in CORT is functional, we constructed a D-box mutant form and asked whether protein stability is affected in an embryo injection experiment. We mutated all of the conserved residues in CORT's D-box to alanine (Figure 5B) and tagged it with MYC to distinguish the protein from endogenous CORT. We know that a MYC-tagged form of CORT is regulated in a similar way to endogenous CORT, because transgenic MYC-tagged CORT is degraded with similar developmental timing in embryos to endogenous CORT in vivo (Figure S1). MYC-tagged D-box mutant cort and MYC-tagged wild-type cort were transcribed in vitro. The RNA was microinjected into 0–30-min post-deposition wild-type embryos. After incubating the embryos to allow for translation of the RNA and post-translational modifications of the proteins, extracts were made and analyzed by western blot (Figure 5D). We found that D-box mutant CORT is stabilized compared to wild-type CORT. Thus, the D-box motif in CORT is required for its degradation in early embryos.
Given our previous observation that cort mRNA is deadenylated in early embryos, we wondered if this regulatory mechanism also contributes to the drop in CORT protein levels after the completion of meiosis. To determine whether Ccr4-mediated deadenylation of cort mRNA is required for low levels of CORT in early embryos, we looked at CORT protein levels in ccr4 mutants by western blot (Figure 5E). We found that CORT protein levels are unchanged in both ccr4/Df stage 14 oocytes and 0–1-h embryos when compared to heterozygous sibling controls. This result suggests that although ccr4 is required for cort deadenylation, it is not required for a subsequent decrease in protein levels. Thus, APC/C-mediated degradation of CORT is the primary mechanism by which CORT protein levels are lowered in early embryos. It is likely that deadenylation serves as a backup mechanism to block future synthesis of CORT after fertilization.
Given the dependence of CORT degradation on mr and an intact D-box, we wanted to determine which APC/C activator is responsible for CORT's destruction. FZR protein is undetectable in 0–2-h embryos and, thus, is not a good candidate [65]. FZY, however, is present in 0–2-h embryos and is the most likely activator of APC/C-mediated degradation of CORT [65].
To test this hypothesis, we first looked at CORT levels in eggs laid by fzy mutant females. We were unable to detect an increase of CORT protein in these embryos, but these alleles are hypomorphic and may not show an effect on CORT (unpublished data). Next, we looked for genetic interactions between cort and fzy. If CORT is a substrate of APC/CFZY in the early embryo, we would expect them to antagonize each other in a genetic pathway. Reducing the level of cort expression in fzy mutants should suppress the fzy phenotype, and increasing the amount of cort expression in fzy mutants should enhance the fzy phenotype.
We carried out these genetic tests using fzy female-sterile mutants in which embryos arrest in metaphase after a few mitotic divisions [50]. Reducing the gene copy number of cort by one in a fzy mutant background causes a modest suppression of the fzy phenotype. Over 75% of embryos laid by these double mutant females arrest with three or more mitotic spindles, whereas only 33% of embryos from fzy single mutants develop this far (Figure 6A). Conversely, overexpressing cort in the germline of fzy mutant females slightly enhances the fzy phenotype. In this case, fzy embryos containing excess cort arrest with fewer mitotic spindles compared to fzy alone (Figure 6B). The results of these genetic interaction tests are consistent with CORT being a substrate of APC/CFZY and suggest that the arrest phenotype of fzy embryos is due in part to the presence of excess CORT protein.
In this study, we investigated the function of cort and its developmental regulation in Drosophila female meiosis. We found that CORT protein physically associates with the APC/C in vivo and confirmed its function as an APC/C activator. We looked at levels of mitotic APC/C substrates in cort mutants and found that Cyclin A is targeted for destruction by APC/CCORT in mature metaphase I arrested oocytes while Cyclin B, Cyclin B3, and PIMPLES/Securin are not targeted until egg activation. Developmental analysis of CORT protein showed that it is translated at oocyte maturation, and appearance of the protein correlates with polyadenylation of cort mRNA. Finally, we found that cort is regulated post-transcriptionally and post-translationally after the completion of meiosis. cort mRNA is deadenylated in early embryos, and CORT protein is degraded after egg activation in an APC/C-dependent manner. CORT contains a conserved D-box motif that is required for the efficiency of its degradation. Our results shed light on the mechanism for the regulation of a meiosis-specific APC/C activator, which results in restriction of its expression to a specific developmental time point.
In this study, our demonstration of a physical interaction between CORT and the APC/C strengthens and confirms previous suggestions that cort encodes a functional meiosis-specific APC/C activator. A strong metaphase II arrest phenotype in cort mutant eggs and distant homology to the Cdc20/FZY protein family initially suggested that CORT might function as an APC/C activator [42,44]. More recently, cort was shown to negatively regulate levels of mitotic cyclin proteins, which is consistent with a role for CORT in activating the APC/C [30]. However, biochemical evidence linking CORT to the APC/C in vivo is crucial for this argument. We have shown that CORT physically associates with core subunits of the APC/C in ovaries, strongly supporting CORT's role as an APC/C activator.
Coordination of the meiotic divisions with oogenesis and the transition from meiosis to restart of the mitotic cell cycle in embryogenesis present unique regulatory challenges for the organism. Our studies of cortex in Drosophila suggest that developmental control of levels of a meiosis-specific APC/C activator is one way in which meiosis is developmentally regulated, which has not been previously observed in a multicellular organism. This strategy exploits ongoing regulatory mechanisms occurring during meiosis and embryogenesis: cytoplasmic polyadenylation during oocyte maturation, deadenylation after egg activation, and APC/C-dependent degradation in the early embryo.
Cytoplasmic polyadenylation upon oocyte maturation has been shown to translationally activate maternal transcripts of genes that are required for meiotic entry, transition between meiosis I and meiosis II, and metaphase II arrest in vertebrates [58]. We have shown that cort mRNA is polyadenylated at oocyte maturation, which adds an APC/C subunit to this group of transcripts that are translationally unmasked for entry into the meiotic divisions. What is the signal for polyadenylation of cort? Masked transcripts contain a cis-acting cytoplasmic polyadenylation element (CPE) to which CPE binding protein (CPEB) is bound. Phosphorylation of CPEB upon oocyte maturation triggers elongation of the poly(A) tail and activation of translation [66]. We have not yet identified a CPE in the 3′ UTR of cort, although CPE sequences are quite variable. In addition, we have tested but have not observed a dependence of cort polyadenylation on orb, the CPEB homolog in Drosophila (unpublished data). Because the orb alleles we used are hypomorphic, we cannot fully rule out the possibility that polyadenylation of cort is orb/CPEB-dependent.
Egg activation triggers maternal transcript destabilization in several organisms, some of which occurs through ccr4-dependent deadenylation, and this is likely to be important for localization of maternal transcripts in the embryo and proper zygotic development [58]. We showed in this study that cort mRNA is deadenylated in the early embryo in a ccr4-dependent manner, but this deadenylation is not required for lowering CORT protein levels. However, we may not be able to detect a difference in protein levels because of the rapid APC/C-dependent degradation of CORT protein that occurs after release of the metaphase I arrest. Deadenylation could serve as a backup mechanism to ensure that CORT protein levels remain low in the early embryo by destabilizing cort mRNA.
The APC/C drives degradation of Cyclin B and other substrates during the rapid syncytial mitotic divisions of early embryogenesis in Drosophila [50,65,67,68]. We found that CORT is targeted for APC/C-dependent degradation by the completion of meiosis in the early embryo. The targeting of an APC/C activator for degradation by another form of APC/C is not unprecedented, as APC/CCdh1 targets Cdc20 for degradation in G1 [16,17,19].
Our data support the conclusion that CORT is targeted by APC/CFZY. First, FZY is thought to be the only activator present in early embryos [65]. Second, we show here that cort and fzy interact genetically in a way that is consistent with cort being a negative downstream target of fzy in embryos. Third, in our embryo injection experiments, we showed that exogenous MYC-CORT is degraded in a D-box–dependent manner in injected embryos. Because the only APC/C activator in early embryos is FZY, degradation of MYC-CORT is likely to occur through APC/CFZY in this assay.
It is also possible the APC/CCORT regulates itself in a negative feedback loop by targeting CORT for degradation when levels of CORT reach a certain threshold at the end of meiosis. To address this possibility, we looked at the degradation of CORT in a homozygous cortQW55 background in which there is no functional CORT protein. CORTQW55 mutant protein is not degraded at the transition from mature stage 14 oocytes to unfertilized eggs, unlike in a heterozygous control background (unpublished data). These results suggest that CORT could be targeted by itself, but it remains a possibility that the lesion in the cortQW55 allele prevents an interaction between CORTQW55 mutant protein and the APC/C machinery. The lesion does not disrupt the D-box, but it could affect proper folding and structure of the protein. In summary, we conclude that CORT is targeted for degradation by the APC/C. It is most likely that FZY is the participating APC/C activator, but CORT may also contribute to targeting itself for degradation.
Recent work has shown that both cort and fzy are required for the meiotic divisions in Drosophila female meiosis. Mutant analysis suggests that cort and fzy act redundantly to control the metaphase I to anaphase I transition, whereas they seem to act with different temporal and spatial specificity in targeting Cyclin B for destruction along the meiosis II spindles [30]. We showed in this study that cort cannot functionally substitute for fzy in the early embryo, suggesting that they target nonredundant sets of substrates. However, in this experiment, we cannot rule out the possibility that MYC-CORT was not present in sufficient levels in early embryos for rescue because of low expression levels or protein instability (Figure S1). Although MYC-CORT is expressed at high levels in stage 14 oocytes, it appears to be subject to degradation after the completion of meiosis, like the endogenous CORT protein.
Furthermore, homozygous cort mutants alone exhibit a strong metaphase II arrest, indicating that the wild-type levels of fzy in this background are not able to act in place of cort to control passage through metaphase II [44]. Finally, we have observed that FZY is expressed at a uniform level during oogenesis and embryogenesis (Figure S2) (unpublished data), which is in contrast to our results in this study showing that CORT expression is specifically upregulated during the meiotic divisions. On the basis of all of these observations, we think it is likely that in addition to the mitotic cyclins, APC/CCORT targets a unique set of substrates in meiosis that are not recognized by APC/CFZY. The identification of these meiotic substrates will be crucial for understanding how the meiotic divisions are controlled in the oocyte.
The study of meiotic control of the APC/C is especially intriguing in Drosophila, because in addition to cort, a female meiosis-specific activator, the genome contains fizzy-related 2 (fzr2), another member of the Cdc20/FZY family. fzr2 is expressed exclusively in testes and may act as a male meiosis-specific activator [69]. Further study of both cort and fzr2 will be important for understanding differential developmental regulation of the APC/C during meiosis in females versus males.
In mitosis, cyclins are targeted sequentially for destruction by the APC/C. Degradation of Cyclin A begins just after nuclear envelope breakdown in prometaphase, while degradation of Cyclin B does not occur until the metaphase to anaphase transition [52,54–56]. Sequential degradation of Cyclin A, Cyclin B, and, finally, Cyclin B3 in Drosophila triggers a series of distinct events leading to exit from mitosis [51,70]. We have found that a similar situation exists in Drosophila female meiosis, in which degradation of Cyclin A by APC/CCORT initiates upon nuclear envelope breakdown, but degradation of Cyclin B and Cyclin B3 does not occur until after the metaphase I to anaphase I transition.
The difference in timing of Cyclin A and Cyclin B degradation in mitosis is due to regulation of the APC/C by the spindle assembly checkpoint. The spindle assembly checkpoint inhibits APC/CCdc20 from initiating anaphase until all chromosomes are bioriented on the spindle, in part through direct binding of Cdc20 to Mad2 and BubR1 [71]. Spindle assembly checkpoint proteins specifically inhibit APC/C-dependent ubiquitination of Cyclin B but not of Cyclin A [52,55,56]. APC/CCORT may be regulated in a similar manner during meiosis I. Indeed, the spindle assembly checkpoint is likely to function during meiosis I in Drosophila, as the conserved spindle checkpoint kinase Mps1 is required for delaying entry into anaphase I to allow for proper segregation of achiasmate homologs and maintenance of chiasmate homolog connections in Drosophila oocytes [72,73]. Furthermore, a functional Mad2-dependent checkpoint exists during meiosis I in mouse oocytes, and spindle checkpoint components have been shown to regulate the APC/C during meiosis I in C. elegans [33,74,75].
To determine whether APC/CCORT is regulated by the spindle checkpoint, we asked if BubR1 or Mad2 physically associate with CORT in stage 14-enriched ovaries. We were unable to detect an association with BubR1 or Mad2 (unpublished data). Although this negative result does not rule out the possibility of regulation of APC/CCORT by the spindle checkpoint, it suggests that APC/CCORT may be subject to other types of regulation that inhibit it from targeting Cyclin B and Cyclin B3 for degradation until after the metaphase I arrest.
In conclusion, through the investigation of cortex, a meiosis-specific APC/C activator, we have found one way in which the meiotic cell cycle may be developmentally controlled during oogenesis. cort is developmentally regulated by existing post-transcriptional and post-translational mechanisms, resulting in expression of CORT protein being restricted to the meiotic divisions. Further study of APC/CCORT will continue to elucidate the ways in which developmental control of the APC/C contributes to proper female meiosis in a metazoan.
Crosses were performed, and flies were maintained between 22 °C and 25 °C using standard techniques [76]. The wild-type stocks used were Oregon R and yw. The cortRH65 and cortQW55 alleles have been described [42,44,77]. To obtain ccr4 mutant flies, ccr4KG877, a ccr4 allele generated by the Berkeley Drosophila Genome Project (http://www.fruitfly.org), was placed in trans to Df(3R)crb-F89-4, a large deficiency that deletes the ccr4 locus [63]. Female-sterile alleles of morula, mr1 and mr2, were originally isolated from natural populations and have been described [78–80]. Female-sterile alleles of fizzy, fzy6 and fzy7, have been described [50]. UASp myc-cort was made by PCR amplification of cort cDNA (LD43270) and subcloning into pUASp with a 6xMYC tag at the N terminus. The LD43270 clone is missing coding sequence for nine amino acids on the 5′ end that we added during PCR amplification. Expression of 6xmyc-cort was driven in the female germline with the nanos-Gal4-VP16 driver [81]. The plu-myc transgenic line has been described [82]. To generate unfertilized eggs, we crossed virgin females to sterile males, which do not produce sperm but are able to stimulate females to lay eggs. The sterile males are from strain T(Y;2)#11cn bwD mr2/b cn mr1 bs2/SM6A, a gift from B. Reed.
To prepare ovary extracts for immunoprecipitations, whole ovaries were dissected in Grace's insect medium (Gibco) from 32 females fattened 3 d on wet yeast at 25 °C. Ovaries were homogenized in 3× volume homogenization buffer (25 mM HEPES [pH 7.5], 0.4 M NaCl, 0.1 mM EDTA, 0.1 mM EGTA, 1 mM PMSF, 10% glycerol, complete mini EDTA-free protease inhibitors, 1 tablet/10 ml [Roche]), snap frozen in liquid nitrogen, and stored at −80 °C. A total of 30 μl Protein A Sepharose beads (Amersham) were precoupled to antibodies for 1 h at 4 °C. For precoupling, antibodies were as follows: 2 μl mouse IgG (Sigma I5381); 12 μl mouse monoclonal anti-myc, 9E10 (Covance); 2 μl rabbit IgG (Sigma I5006); or 10 μl affinity-purified rabbit anti-Cdc27 [68]. After removing an aliquot for input, 70 μl ovary extract was added to antibody-bound beads and incubated for 2–4 h at 4 °C. Beads were washed once in ice-cold IP buffer (25 mM HEPES [pH 7.5], 100 mM NaCl, 1 mM EGTA, 0.1% Triton X-100, 10% glycerol, complete mini EDTA-free protease inhibitors, 1 tablet/10 ml [Roche]), once in IP buffer plus 0.5 M NaCl, and 4 times in IP buffer. Inputs, immunocomplexes, and supernatants were resolved by SDS-PAGE and analyzed by immunoblot as described below.
A fusion between GST and 152 amino acids from the N terminus of CORT was used to produce antibodies in guinea pigs. The construct encoding GST-CORT_N was made by PCR amplification of cort cDNA (clone number LD43270) as described above, followed by subcloning into pGEX-4T-1 expression vector (Pharmacia). GST-CORT_N was expressed in TOP10 E. coli cells by IPTG induction. The majority of GST-CORT_N was insoluble so it was gel purified from the insoluble material after cell lysis. Crude lysate was clarified, and the insoluble pellet resuspended in 5× Sample Buffer (60 mM 1 M Tris-HCl [pH 6.8], 25% glycerol, 2% SDS, 14.4 mM 2-mercaptoethanol, 0.1% bromophenol blue). Sample was resolved by SDS-PAGE on a preparative 10% 150:1 (30% acrylamide/2% bis-acrylamide) gel. Vertical strips from either side of the gel were stained with GelCode Blue (Pierce) and used as a guide to cut out the unstained band of GST-CORT_N. Gel slice was pulverized with cold 1× SDS Electrophoresis Buffer (25 mM Tris base, 192 mM glycine, 0.1% SDS) through a 10 ml syringe and gently rocked for 30 min at 4 °C to elute protein. Gel slice mixture was filtered through 125 μm nylon mesh (Tetko), and the eluate concentrated in Amicon Centricon YM-10 (Millipore). Concentrated protein was injected into guinea pigs for antibody production (Covance). The anti-CORT antibody recognizes a band of approximately 47 kDa that is the CORT protein.
Protein extracts were made by homogenizing staged egg chambers, whole ovaries, unfertilized eggs, or embryos in 3:1 Urea Sample Buffer (8 M urea, 2% SDS, 100 mM Tris [pH 7.5], 5% Ficoll)/embryo (vol/vol). Unfertilized eggs were collected for 0–2 h. Whole fly extracts were made by homogenizing flies directly in 5× Sample Buffer. Protein extracts were resolved by SDS-PAGE and transferred to Immobilon-P membranes (Millipore). We used 10.5%–14% acrylamide gels for immunoprecipitations (Figure 1) and substrate blots (Figure 2). We used 10% acrylamide gels for all CORT blots and RNA injection assays (Figures 3 and 5). Equal amounts of protein were loaded per lane and confirmed by anti-α-Tubulin blotting. Blots were probed with the following antibodies: mouse monoclonal anti-MYC, 9E10 (1:1000, Covance); affinity-purified rabbit anti-Cdc27 (1:500 [68]); affinity-purified rabbit anti-MR (1:200 [83]); guinea pig anti-CORT serum (1:2000); rat monoclonal anti-α-Tubulin, YL1/2 and YOL1/34 (1:200, Harlan Sera-lab); mouse monoclonal anti-Cyclin A, A19 (1:50, gift of P. O'Farrell); mouse monoclonal anti-Cyclin B, F2F4 (1:200 [84]); rabbit anti-Cyclin B3 serum (1:4000 [51]); rabbit anti-PIM serum (1:10,000 [85]); and affinity-purified rabbit anti-FZY (1:1000 [65]). Alkaline phosphatase- or horseradish peroxidase-conjugated secondary antibodies were used to detect bound primary antibodies. Protein was detected using ECL Plus (Amersham).
Ovary or embryo mRNA was isolated using the PolyATtract System 1000 (Promega). LM-PAT assays were performed using 100 ng mRNA as described [60]. cDNA was made using the Reverse Transcription System (Promega). PCRs were performed with message-specific primers, and a fraction of the PCR product was tested on a gel to permit approximately equal loading of the PCR product for the experiment. PCR products were separated on a 2% MetaPhore agarose gel and stained with ethidium bromide.
Point mutations were introduced into cort cDNA using the Phusion Site-Directed Mutagenesis Kit (Finnzymes). Wild-type or mutated D-box cort cDNA was subcloned into pCS2 containing a 6xMYC tag at the N terminus. Capped mRNAs were synthesized from these vectors using the SP6 mMessage mMachine Kit (Ambion). mRNA was purified using the MEGAclear Kit (Ambion). yw embryos that were 0–30 min postdeposition were dechorionated and prepared for injection. Samples were prepared containing 250 ng/μl wild-type or mutant cort RNA in injection buffer (5 mM KCl, 0.1 M K2HPO4 [pH 7.8]). A no RNA control contained injection buffer alone. Each sample was injected into 150 embryos. After 40 min at room temperature, the embryos were harvested in heptane, washed 2 times in embryo wash (0.4% NaCl, 0.03% Triton X-100), and homogenized in 20 μl USB. Extracts were resolved by SDS-PAGE and analyzed by immunoblotting as described above. The experiment was repeated 5 independent times to confirm results.
Eggs were collected for 0–3 h for Figure 6A and for 2–4 h for Figure 6B, dechorionated in 50% bleach, devitellinized in methanol and heptane, and fixed in methanol for 3 h at room temperature or overnight at 4 °C. Eggs were stained for DNA with Propidium Iodide and for Tubulin with rat monoclonal anti-α-Tubulin, YL 1/2, and YOL 1/34 (1:20, Harlan Sera-lab). Antibodies were detected using fluorescent secondary antibodies (Jackson Immunoresearch). Imaging was performed using a Zeiss Axioskop.
The FlyBase (http://flybase.bio.indiana.edu/search/) accession numbers for genes and gene products discussed in this paper are bubR1 (FBgn0025458), ccr4 (FBgn0011725), cdc16 (FBgn0025781), cdc27 (FBgn0012058), cort (FBgn0000351), cycA (FBgn0000404), cycB (FBgn0000405), cycB3 (FBgn0015625), fzr (FBgn0003200), fzr2 (FBgn0034937), fzy (FBgn0001086), grau (FBgn0001133), mad2 (FBgn0035640), mr (FBgn0002791), orb (FBgn0004882), and pim (FBgn0003087). |
10.1371/journal.pntd.0003859 | A Large-Scale Community-Based Outbreak of Paratyphoid Fever Caused by Hospital-Derived Transmission in Southern China | Since the 1990s, paratyphoid fever caused by Salmonella Paratyphi A has emerged in Southeast Asia and China. In 2010, a large-scale outbreak involving 601 cases of paratyphoid fever occurred in the whole of Yuanjiang county in China. Epidemiological and laboratory investigations were conducted to determine the etiology, source and transmission factors of the outbreak.
A case-control study was performed to identify the risk factors for this paratyphoid outbreak. Cases were identified as patients with blood culture–confirmed S. Paratyphi A infection. Controls were healthy persons without fever within the past month and matched to cases by age, gender and geography. Pulsed-field gel electrophoresis and whole-genome sequencing of the S. Paratyphi A strains isolated from patients and environmental sources were performed to facilitate transmission analysis and source tracking. We found that farmers and young adults were the populations mainly affected in this outbreak, and the consumption of raw vegetables was the main risk factor associated with paratyphoid fever. Molecular subtyping and genome sequencing of S. Paratyphi A isolates recovered from improperly disinfected hospital wastewater showed indistinguishable patterns matching most of the isolates from the cases. An investigation showed that hospital wastewater mixed with surface water was used for crop irrigation, promoting a cycle of contamination. After prohibition of the planting of vegetables in contaminated fields and the thorough disinfection of hospital wastewater, the outbreak subsided. Further analysis of the isolates indicated that the origin of the outbreak was most likely from patients outside Yuanjiang county.
This outbreak is an example of the combined effect of social behaviors, prevailing ecological conditions and improper disinfection of hospital wastewater on facilitating a sustained epidemic of paratyphoid fever. This study underscores the critical need for strict treatment measures of hospital wastewater and the maintenance of independent agricultural irrigation systems in rural areas.
| Typhoid and paratyphoid fever remain public health concerns for developing countries. From May 2010 to June 2011, a large-scale outbreak involving 601 cases of paratyphoid fever occurred in China. Epidemiological and laboratory investigations were conducted to determine the etiology, source and transmission factors of the outbreak. Farmers and young adults were the populations mainly affected in this outbreak, and the consumption of raw vegetables was the main risk factor associated with paratyphoid fever. We found that hospital wastewater mixed with surface water was used for vegetable irrigation. The contaminated water from hospitals combined with the regional habit of eating uncooked vegetables lead to the massive outbreak of paratyphoid. After prohibition of the planting of vegetables in contaminated fields and the thorough disinfection of hospital wastewater, the outbreak subsided. Molecular subtyping and whole-genome sequencing of S. Paratyphi A isolates recovered from improperly disinfected hospital wastewater showed indistinguishable patterns matching most of the isolates from the cases. Further analysis of the isolates indicated that the origin of the outbreak was most likely from patients outside Yuanjiang county. This study underscores the critical need for strict treatment measures of hospital wastewater and the maintenance of independent agricultural irrigation systems in rural areas.
| Typhoid and paratyphoid (enteric) fever caused by Salmonella enterica serovar Typhi and Paratyphi A remain significant public health problems for developing countries. Globally, 13.5 million cases are estimated to occur annually and are associated with 190,000 deaths in 2010 [1]. Before the 1990s, typhoid cases were more prevalent than paratyphoid cases in Southeast Asia; however, the latter have been steadily increasing [2]. Paratyphoid fever has also been increasingly reported in China since 1998 and has resulted in localized outbreaks in some provinces [3, 4].
Understanding the risk factors of enteric fever is important for prevention and control and to provide tools for interrupting disease transmission in the early phases of an outbreak. In low-prevalence areas, travel and immigration from endemic areas are major risk factors [5]. In high-prevalence countries, the main risk factors include the consumption of unsafe drinking water and contaminated foods and close contact with active cases or carriers [6, 7]. In China, the risk factors for enteric fever differ between urban and rural areas, as well as between coastal cities and inland regions [8, 9].
During epidemic investigations, pulsed-field gel electrophoresis (PFGE) is quite useful for identifying outbreak-associated isolates and source tracing [10]. Recently, whole-genome sequencing has provided increased sensitivity for microbial evolution and molecular epidemiology studies, improving the understanding of disease transmission [11–14].
From May 2010 to June 2011, an outbreak causing 601 cases of paratyphoid fever was documented in Yuanjiang county, Yunnan Province, China. In this study, we report a risk factor analysis coupled with the laboratory-based characterization of outbreak-associated isolates, with the goal of determining the source of the outbreak, implementing control measures and assessing their effectiveness.
This study was reviewed and approved by the ethics committee of National Institute for Communicable Disease Control and Prevention, China CDC, according to the medical research regulations of the Ministry of Health, China (ICDC-2014008).
A suspect enteric fever case was defined as persistent fever (≥ 37.5°C for more than three days) accompanied by headache and body ache, without obvious upper respiratory or urinary tract infections, trauma or other diagnosed causes of fever. Both suspect cases and laboratory-confirmed cases (culture positive) were reported daily to the Chinese Center for Disease Control and Prevention through an internet-based disease reporting system [15]. Other data were extracted from patient records (e.g., age, sex, home address, work place, occupation, date of presentation to the hospital, suspected or laboratory-confirmed diagnosis). A case-control study with 1:1 matching of controls to cases was initiated and included blood culture–confirmed cases of S. Paratyphi A infection and healthy persons without fever during the 1 month prior to the study. The cases and controls were matched for age (no more than 5 years between cases and controls), gender and residential location. Controls were enrolled from households next-door to the households of case subjects. If the case subject lived in a bungalow or stand-alone house, the household to the right of the “case-household” was approached by an epidemiological expert within 1 week of enrollment of the case subject. If the person in that house refused to join the study or failed to meet the enrollment criteria (same gender, within 5 years of the age of the case subject and no fever in 1 month prior to the administration of the questionnaire), the household to the left of the “case-household” was approached, followed by the house parallel to the “case-household” across the street. If the case subject lived in a multistory building, the household to the right of the “case-household” on the same floor was approached. If persons in that house refused to join the study or failed to meet the enrollment criteria, the household to the left of the “case-household” was approached, followed by the household one story above or below the “case-household”. Standard questionnaires administered by local epidemiological experts through face-to-face interview to gather information on a four-week time window of exposure history, including close contact with S. Paratyphi A cases, recent history of travel and food and water consumption for all cases and controls. All controls were asked about the time period that coincided with period of exposure history of the matched case subject. All data entry and analyses were performed using SPSS 17.1 for Windows (SPSS Inc., Chicago, IL, USA). The odds ratios (ORs) and exact 95% confidence intervals (CIs) were calculated via a matched multivariate analysis through conditional logistic regression to examine relationships between exposures and illness. To avoid confounding effects, the data of 60% subjects who drank both bottled and municipal water was discarded from the analysis of categorical variables to prevent statistical bias. Statistical significance was designated as a p value <0.05.
Environmental samples of untreated surface water, treated township water, drinking water, hospital waste and suspected food and vegetables were collected and cultured for bacteria. Standard methods for bacterial isolation and identification were utilized [16].
During the outbreak period, samples of blood, stool and urine were collected for the suspect cases. An 8-mL sample of whole blood was injected into a single Bactec culture bottle (Biomerieux, Durham, NC, USA) and incubated at 37°C. Approximately 3 g stool from each patient was used to inoculate selenite broth for enrichment. Urine was centrifuged, and the pellets were suspended in selenite broth and incubated overnight at 37°C. To confirm S. Paratyphi A, a serological agglutination reaction was performed using Salmonella antisera (S&A Reagents Lab, Bangkok, Thailand) and biochemical tests (Microbact, Medvet Diagnostics, Adelaide, Australia). Antimicrobial susceptibility testing was performed following the Clinical Laboratory Standards Institute (CLSI) broth microdilution method [17]. Antimicrobial agents including ampicillin, amoxicillin/clavulanic acid, ceftriaxone, cefotaxime, ceftazidime, nalidixic acid, ciprofloxacin, chloramphenicol, gentamicin, kanamycin, streptomycin, sulfisoxazole, tetracycline, azithromycin and trimethoprim/sulfamethoxazole were evaluated using CLSI minimum inhibitory concentration (MIC) interpretive standards for Enterobacteriaceae to calculate resistance thresholds [18].
PFGE was performed according to the modified PulseNet protocol for S. Paratyphi A [19] using a CHEF DRIII system (Bio-Rad, Hercules, CA, USA). Band similarities were analyzed using BioNumerics software (version 2.5, Applied Maths, Kortrjk, Belgium) by the unweighted pair-group method with arithmetic means method to produce a dendrogram with 1.5% position tolerance. The outbreak strains from Yuanjiang county and the epidemic isolates from Yuxi city were conducted by PFGE analysis and molecular comparison.
Genomic DNA was prepared from overnight cultures using a Wizard Genomic DNA Purification Kit (Promega, Madison, WI, USA) according to the manufacturer’s instructions. Whole-genome sequencing (WGS) was performed on 17 S. Paratyphi A isolates using an Illumina HiSeq 2000 with 500-bp paired-end libraries in 8-fold multiplexes (Beijing Genome Institute, Shenzhen, China). The sample set represents geographically dispersed isolates collected over an array of time points (14 months from Feb, 2010 to April, 2011) and multiple PFGE pattern combinations, including 12 outbreak isolates from patients distributed in Yuanjiang county, 2 isolates from wastewater, one contemporaneous isolate from other county, and 2 isolates circulating in Hongta district of Yuxi representing the top 2 PFGE patterns of paratyphoid A. Details of the genomic sequencing of each isolate are summarized in Supplementary S1 Table, including the accession codes of the reads in European Nucleotide Archive (www.ebi.ac.uk/ena).
The WGS reads were assembled de novo using SOAPdenovo2 [20] with optimal K-mer and minimal coverage parameters. Intra-scaffold gaps were filled using GapCloser v1.12 (SOAP package). SOAPaligner v2.21 [21] was used to remap the reads to the assembled scaffolds for validation of the quality of each base that was called in the assembly. The sites were filtered with a quality score of < 20 or a read coverage of < 10 with Bowtie and SAMtools [22].
A core genome of 4,585,284 bp was obtained by comparing the 17 S. Paratyphi A genomes from this study with 6 previously sequenced genomes (S1 Table). The non-repetitive core genome was found to contain 513 SNPs. We identified 270 SNPs after removing SNPs in recombinant regions, as suggested by ClonalFrame [23]. A maximum likelihood tree was inferred with these 270 SNPs in MEGA v5 [24]. The genomic sequence of ATCC9150 was determined as the root because it was most distantly related to the outbreak isolates.
All sequence are available at http://www.ebi.ac.uk/ena/data/view/PRJEB9577 and refer to [25, 26]. The accession numbers are: PA1477, ERS747619 SAMEA3451357 PA1477; PA1478, ERS747620 SAMEA3451358 PA1478; PA1815, ERS747621 SAMEA3451359 PA1815; PA1822, ERS747622 SAMEA3451360 PA1822; PA1850, ERS747623 SAMEA3451361 PA1850; PA1886, ERS747624 SAMEA3451362 PA1886; PA1909, ERS747625 SAMEA3451363 PA1909; PA2183, ERS747626 SAMEA3451364 PA2183; PA2184, ERS747627 SAMEA3451365 PA2184; PA2199, ERS747628 SAMEA3451366 PA2199; PA2207, ERS747629 SAMEA3451367 PA2207; PA2216, ERS747630 SAMEA3451368 PA2216; PA2243, ERS747631 SAMEA3451369 PA2243; PA2635, ERS747632 SAMEA3451370 PA2635; PA2641, ERS747633 SAMEA3451371 PA2641; PA2161, ERS747634 SAMEA3451372 PA2161; PA2191, ERS747635 SAMEA3451373 PA2191.
From May 2010, a sharp increase in paratyphoid fever A cases was noted in Yuanjiang county (Figs 1 and S1). In August, 23 cases were reported, suggesting that an outbreak of paratyphoid fever was in progress. By the end of 2010, 519 cases of enteric fever were reported (Fig 1); 503 were lab-confirmed via the isolation of S. Paratyphi A from blood culture. Since May 2011, no more than 10 cases per month have been reported (Fig 1); only 13 cases were reported in 2012, and three cases in 2013.
Based on the case reporting, the outbreak caused a total of 601 cases from May 2010 to June 2011, involving 10 towns in the county (S2 Fig); most cases (347) were from Lijiang. The incidence in Lijiang reached 708.4/100,000 compared with the average incidence of 45.4/100,000 (1,004 cases) in the Yuxi region and 4.75/100,000 (2,171 cases) in Yunnan Province in 2010. The ratio of male to female cases was 1.3 (1.3:1), and most cases (81%) were in the 20 to 49 [year] age group. Most cases, nearly 61%, were farmers (7.2% (14532/201857) of total population is made up of farmers).
A random sample of 106 paratyphoid cases was reviewed. All patients had fever; other prominently reported signs and symptoms included headache (84%), chills (74%), fatigue (43%), body aches (32%), cough (26%) and dizziness (19%). Most patients presented to the local hospital after one day of fever; 98 cases had a body temperature >37.5°C but less than 39°C, and eight patients had a fever >40°C. The mean fever duration of the cases was three days (range 1–15 days). All S. Paratyphi A isolates had the same antibiotic susceptibility testing profile: they were sensitive to cephalosporins and azithromycin (MIC = 8–16 mg/mL) but resistant to nalidixic acid and had decreased susceptibility to ciprofloxacin (MIC = 0.5 mg/mL). Ninety percent of the hospitalized patients were given fluoroquinolones, including oral norfloxacin (0.2 g, three times per day) or intravenous levofloxacin (0.2 g, once a day) combined with ceftriaxone (2.5 g, intravenously, twice per day) for two weeks.
A case-control study was conducted that included 109 pairs of laboratory-confirmed cases and controls matched by age, gender and residential location. As the number of enrolled culture–confirmed cases of paratyphoid fever was more than 100, we chosen the simple 1:1 matching model to conduct the risk association analysis routinely. Eighty (73.40%) cases and their matched controls lived in urban districts of Lijiang town, and the remainder resided in its surrounding areas. Using a multivariate conditional logistic regression analysis, the consumption of raw vegetables (OR = 65.3, 95% CI: 8.3–511.6, P<0.001, Table 1), especially the consumption of raw vegetables not prepared at home (OR = 6.1, 95% CI: 3.1–11.9, P<0.001), was significantly associated with paratyphoid fever. The direct consumption of bottled water (approximately 30% in either group only drank bottled water (unboiled water)) was not associated with the disease (OR = 2.2, P = 0.18). Only 2 cases and 1 control admitted to drinking unboiled municipal water, and this consumption was not shown to have an effect on disease exposure. In addition, no fecal or thermotolerant coliforms were detected in the municipal water of Lijiang. The rates of consumption of uncooked vegetables purchased from Ximen Farm Product Market, small restaurants, supermarkets, canteens and street stalls in cases and controls were also analyzed through conditional logistic regression analysis. The results showed a significant association between paratyphoid fever and the consumption of uncooked vegetables obtained from street stalls (OR = 6.4, 95% CI: 1.9–21.6, P = 0.003), the Ximen Farm Product Market (OR = 18.3, 95% CI: 3.6–93.0, P<0.001), and small restaurants (OR = 29.6, 95% CI: 6.9–127.1, P<0.001).
Raw vegetables are a favorite food of the local people. Ximen Farm Product Market is the largest vegetable market and supplies farm products for the entire town of Lijiang (Yuanjiang county). The primary source of the cold-served vegetables served at stalls in Ximen Farm Product Market was a vegetable patch located on the east side of Lijiang (Fig 2); this location is also close to County People's Hospital. A gutter from the urban drainage crosses the vegetable patch (Fig 2). Additionally, an open gutter for wastewater discharged directly from the hospital also crossed the land used for growing vegetables, connecting to the urban gutter (Fig 2). The vegetables cultivated in this patch included coriander, mint, green onions and lettuce, all of which were used for the preparation of the cold dishes served in the market. Prior to 2010, farmers used water from a spring near the vegetable patch (Fig 2) to irrigate the vegetables. However, from 2009 to 2010 a severe drought affected most of Yunnan Province, including Yuanjiang. The spring dried up, and beginning in May 2010, the farmers used the wastewater to irrigate vegetables. As mentioned above, wastewater from County People's Hospital, where most enteric fever cases were treated, with 6 isolates recovered from the hospital’s wastewater, was also mixed with the surface water used to cultivate the vegetables, creating a potential for an on-going cycle of contamination.
In total, 630 S. Paratyphi A isolates were obtained from the blood, stool and urine of patients during the outbreak. In total, 160 isolates recovered at different times and places during the outbreak, including 10 isolates recovered from wastewater, were analyzed by PFGE. Seven unique strain patterns were identified; 138 isolates, including the 10 from the sewage samples, clustered with the dominant pattern JKPX01.CN0001/JKPS18.CN0001 (Figs 3 and S3). The PFGE pattern similarity for 97% isolates was greater than 98% and included three distinct but similar patterns: JKPX01.CN0001/JKPS18.CN0001, JKPX01.CN0001/JKPS18.CN0109 and JKPX01.CN0001/JKPS18.CN0003. In addition, the dominant pattern, JKPX01.CN0001/JKPS18.CN0001, was indistinguishable from that attributed to S. Paratyphi A isolates from the city of Yuxi during the period of 2008 to 2009 (Figs 3 and S3). These data suggest that the outbreak in Yuanjiang may have originated from a previous outbreak in Yuxi.
Laboratory testing of 13 vegetable samples from Ximen Farm Product Market failed to yield any S. Paratyphi A; however, abundant fecal coliforms (≥24,000 MPN/100 g) were recovered, well in excess of the limit of 100 MPN/100 g according to the standard of Microbiological Examination of Food hygiene in China. High coliform counts were detected from mint and lettuce samples, demonstrating fecal contamination of these vegetables.
In October and November 2010, 6 isolates of S. Paratyphi A were recovered from 10 wastewater outfall samples obtained from County People’s Hospital (three different collection times). Additionally, four S. Paratyphi A isolates were recovered from eight wastewater samples collected from two additional hospitals, Minority Hospital and Chinese Medicine Hospital, both of which had treated enteric fever cases. The wastewater from these hospitals was not treated, and both discharged their wastewater directly into the covered urban drainage system (Fig 2). No S. Paratyphi A was detected from the effluent of the three main urban sewage outfalls of the county. Nonetheless, the amount of fecal coliforms (≥16,000 MPN/L) exceeded the threshold of 10,000 MPN/L according to the discharge standards for urban sewage treatment. PGFE was performed on the ten isolates from the wastewater, and only one pattern combination (JKPX01.CN0001/JKPS18.CN0001) was obtained, which was indistinguishable from the predominant pattern of isolates from the patients of the Yuanjiang epidemic (Figs 3 and S3).
The S. Paratyphi A isolates (n = 286) recovered from patients living in Hongta district in 2008 and 2009 were analyzed by PFGE. Although a diversity of XbaI and SpeI pattern combinations were found, one dominant pattern, JKPX01.CN0001/JKPS18.CN0001, was indistinguishable from the dominant pattern of the Yuanjiang outbreak. Other minor PFGE pattern combinations found in the Yuanjiang outbreak, including JKPX01.CN0001/JKPS18.CN0002 and JKPX01.CN0001/JKPS18.CN0029, were also present as common patterns in the Hongta district epidemic. Two patients that represent probable index cases in the Yuanjiang outbreak worked in the Hongta district and had returned to Yuanjiang county for the treatment of enteric fever in Yuanjiang County People's Hospital, providing a potential mechanism of paratyphoidal fever transmission from the Yuxi urban area. The combinations JKPX01.CN0001/JKPS18.CN0003, JKPX01.CN0001/JKPS18.CN0109 and JKPX01.CN0001/JKPS18.CN0110 were not found in the Hongta district epidemic isolates, suggesting the possible mutation of some isolates during the outbreak in Yuanjiang.
We randomly selected 14 isolates from the Yuanjiang outbreak and 3 epidemic isolates from Hongta district and Huaning county based on PFGE subtyping differences for the whole-genome sequencing. Regarding the phylogeny inferred from the core genome SNPs, 13 of the 14 isolates from Yuanjiang formed one tight clade (Fig 4). This clade showed that a predominant clone circulated in Yuanjiang, though mutations were also observed during transmission among the patients in the outbreak, which was supported by the PFGE analysis (Figs 3 and S3). A distinct strain (PA1815) from Yuanjiang, which may represent a different clone, was also found (Fig 4). Thirteen of the 14 Yuanjiang isolates are closely related to PA1477 from the Hongta district. These 13 Yuanjiang isolates may have been transmitted from Hongta at least once, with a pathogen that is similar to PA1477. Moreover, we observed a close relationship between the epidemic case isolates and the hospital wastewater isolates; no SNP was found between one case isolate and one wastewater isolate. However, one Yuanjiang case isolate (PA1815) differed from the other Yuanjiang outbreak isolates by 14 SNPs. We speculate that this isolate had either been resident in Yuanjiang for a long time or was transmitted by a lineage in Hongta that was not observed by the sequence analysis.
The epidemiological and laboratory investigations suggested roles for both vegetables and hospital wastewater in the outbreak. On October 2nd, 2010, immediately after the detection of S. Paratyphi A isolates in the untreated hospital effluent, measures were taken to control the outbreak and to interrupt the transmission of paratyphoid fever in Yuanjiang county. Public health education, additional sampling and the chlorination of municipal water were performed. Additionally, wastewater disinfection and patient waste management control were strengthened in the hospitals. Individual patient waste was sterilized with 20% sodium hypochlorite for approximately 30 minutes prior to the communal collection of waste. Additionally, calcium hypochlorite powder was automatically (final concentration of 20%) added into the pooled sewage from the hospitals and treated for eight hours before discharge. In addition, the disinfected wastewater was cultured for S. Paratyphi A and fecal coliform bacteria every week by the local CDC to ensure the effectiveness of the measures. On October 8th, 2010, the vegetable farmers were notified to stop planting on the parcel of land near County People's Hospital, and the selling of raw vegetables to restaurants was prohibited. The number of cases decreased rapidly in the ensuing months (Fig 1).
In March 2011, the number of reported paratyphoid fever cases increased slightly (Fig 1). Four months after the public health measures were relaxed, some restaurants, farmers’ markets and street vendors resumed selling raw vegetables cultivated from the patch near the hospital. In early April 2011, the local government issued a stricter injunction to prohibit the planting of vegetables used for cold dishes and destroyed all the vegetables from the plot near County People’s Hospital. After May, the incidence of enteric fever cases dropped to 11 cases within 7 months, similar to the pre-outbreak incidence baseline.
The Yuanjiang paratyphoid fever outbreak of 2010–2011 is a classic example of a paratyphoid fever outbreak in a developing economy. The outbreak occurred in an area with low sanitary protection, poor wastewater disposal and drought. Though people are the only natural host and reservoir for S. Typhi and S. Paratyphi A, the two pathogens could survive in variant environments outside of humans, such as water and foods. S. Typhi can survive for days in groundwater, pondwater, or seawater, and for months in contaminated eggs and frozen oysters [27–30]. For S. Paratyphi A, it could survive 3 days in river water or polluted water, more than 10 days in contaminated razor calms [31]. Surveillance showed that waterborne causes accounted for 53% of all outbreaks of typhoid and paratyphoid fever during 2004–2007 in China [8]. Raw vegetable-associated enteric fever outbreaks have been reported in other countries [6, 32, 33]. Although S. Paratyphi A isolates were not obtained directly from the vegetables examined, the abnormal fecal coliform counts provided indirect evidence of the contamination of the vegetables. When the consumption of these vegetables was prohibited, along with the imposition of strict hospital sewage treatment policies, the number of cases of this disease was radically reduced. Although changes in temperature could not be excluded as a factor in reducing disease during the paratyphoid fever outbreak, we emphasize that a second wave of cases were reported when raw vegetables from the same garden plot were again sold. Indeed, only when this source of vegetables was completely prohibited did the case number drop back to background. Considering the continuation of paratyphoid fever epidemic in the Yuxi region in 2011 and 2012, while the cases in Yuanjiang county decreased to the baseline before the outbreak, it is clear that the comprehensive interventions, largely by the government, played a very important role in interrupting the transmission of paratyphoid fever in developing areas.
In this outbreak, an uncommon but critical role of untreated hospital wastewater was revealed. From January to November 2010, County People's Hospital admitted more than 300 paratyphoid fever patients while the wastewater was not treated appropriately. The outbreak was most likely sustained by the transmission of the pathogen from the hospital to the community. The associations between paratyphoid fever and the consumption of uncooked vegetables obtained from street stalls, the Ximen Farm Product Market and small restaurants indicate that the consumption of raw vegetables, especially dining at a location with poor sanitation may increase the risk of exposure to S. Paratyphi A. The vegetables supplied by the local farmers’ market to street vendors and restaurants and subsequently served in cold dishes as seasoning were continually recontaminated from a source where an urban drain mixed with effluent from County Peoples’ Hospital. Our findings highlight the need for healthcare institutions to take preventative action to mitigate outbreaks or epidemics in the populations they serve.
Molecular subtyping combined with genome sequencing and epidemiological studies strongly support a transmission link between an S. Paratyphi A clone from a recent Yuxi epidemic and the Yuanjiang outbreak. These data show that when S. Paratyphi A is introduced into a population by the use of contaminated ground water for irrigation due to drought, the continued cycle of contaminated water from hospitals combined with the regional habit of eating uncooked vegetables lead to ongoing cases.
S. Typhi and S. Paratyphi isolates with decreased susceptibility to ciprofloxacin could result in more frequent treatment failure than susceptible strains among patients with typhoid or paratyphoid fever [34–36]. In this study, it is possible that patients infected with S. Parayphi A experienced poor response to therapy or treatment failure, because all isolates from the outbreak displayed decreased susceptibility to ciprofloxacin in vitro. Therefore, the combinational treatment with fluoroquinolones and cephalosporins was administrated since all isolates were sensitive to ceftriaxone. Unfortunately, the carriage of S. Parayphi A was not confirmed in patients before they were discharged and few medical records were obtained simultaneously. The data from this study suggests that it is very important to design treatment procedure based on the bacterial resistant profile to avoid treatment failure and necessary to confirm the carriage of bacterial pathogen before the discharge of patients from hospital.
There were some limitations of this study. First, recall bias of cases and controls could not be avoided, especially the lack of details of the eaten vegetables concerning which vegetables were associated with illness. The other is, we didn’t recover any S. Paratyphi A isolates from raw vegetables. Though isolation of S. Paratyphi A from environmental water and food is quite difficult, to get S. Paratyphi A isolates from vegetables will present powerful support for its spread. However in this study, elimination of risk factors (i.e., raw vegetables) arrested the outbreak of paratyphoid fever will provide evidence for supporting the identification of the risk factor in this outbreak. Another lesson from this outbreak control study is the earlier investigation and following by intervention response will reduce the scale of the outbreak.
Safe water and good hygiene are critical for prevention of fecal-oral transmission of infectious diseases. In this case, lack of well water for irrigation, a result of prevailing drought conditions, played an important role, resulting in the use of wastewater to irrigate vegetable fields. It is necessary to establish separate agricultural irrigation water and waste water systems, and it is also important to enhance the management and treatment of patients, to strengthen the monitoring of medical waste.
|
10.1371/journal.pcbi.1005298 | Feedback, Mass Conservation and Reaction Kinetics Impact the Robustness of Cellular Oscillations | Oscillations occur in a wide variety of cellular processes, for example in calcium and p53 signaling responses, in metabolic pathways or within gene-regulatory networks, e.g. the circadian system. Since it is of central importance to understand the influence of perturbations on the dynamics of these systems a number of experimental and theoretical studies have examined their robustness. The period of circadian oscillations has been found to be very robust and to provide reliable timing. For intracellular calcium oscillations the period has been shown to be very sensitive and to allow for frequency-encoded signaling. We here apply a comprehensive computational approach to study the robustness of period and amplitude of oscillatory systems. We employ different prototype oscillator models and a large number of parameter sets obtained by random sampling. This framework is used to examine the effect of three design principles on the sensitivities towards perturbations of the kinetic parameters. We find that a prototype oscillator with negative feedback has lower period sensitivities than a prototype oscillator relying on positive feedback, but on average higher amplitude sensitivities. For both oscillator types, the use of Michaelis-Menten instead of mass action kinetics in all degradation and conversion reactions leads to an increase in period as well as amplitude sensitivities. We observe moderate changes in sensitivities if replacing mass conversion reactions by purely regulatory reactions. These insights are validated for a set of established models of various cellular rhythms. Overall, our work highlights the importance of reaction kinetics and feedback type for the variability of period and amplitude and therefore for the establishment of predictive models.
| Rhythmic behavior is omnipresent in biology and has many crucial functions. In cells the activation levels and abundances of signaling molecules such as NF-κB, p53, EGFR or calcium repeatedly increase and decrease in response to stimuli. Such a dynamic behavior can also be observed monitoring the concentrations of mRNAs and proteins in the circadian clock and the cell cycle. Period and amplitude which are the time span between peaks and the peak height, respectively, as well as their variabilities are important features of oscillations. The circadian period is very stable allowing for proper time keeping, whereas in calcium signaling the period is very variable encoding different stimulation strengths. Our goal is to examine the origin of differences in sensitivities of periods and amplitudes using a computational approach. We use prototype oscillators and demonstrate that they can be used to derive general principles that explain the degree of robustness in period and amplitude for a set of commonly used models of cellular oscillators. Our findings imply that the robustness of oscillating systems can be influenced by feedback type and kinetic properties to which special attention should be paid when designing mathematical models of cellular rhythms.
| Various self-sustained autonomous oscillations are found at the cellular level. Prominent examples are calcium, p53 and NF-κB oscillations in signaling systems, circadian and cell cycle oscillations in genetic networks and oxidation-reduction cycles in metabolism [1,2,3,4]. A central question is in how far these systems are able to maintain their dynamical characteristics facing environmental changes, a feature that has been termed robustness [5,6]. Mathematical models have been proposed for many oscillatory processes and the examination of their robustness is considered to give valuable indications on the organization and functioning of the respective underlying biological processes. A number of studies have focused on the size and shape of the parameter space that allows for oscillatory dynamics [6,7,8,9,10]. Yet, also the period and amplitude of oscillations may be differently robust to changes in the environment. For example, circadian oscillations endue a time-keeping function. It has been shown that their period of approximately 24 hours is temperature compensated and does not change significantly with varying pH or nutritional conditions [11,12,13,14]. In contrast, the period of intracellular calcium oscillations varies from seconds to minutes and is highly responsive to changes in temperature and agonist concentrations [15,16]. The latter is a phenomenon referred to as frequency encoding of the stimulus [17,18]. Furthermore, a robust amplitude has been shown to be important for the proper function of the cell cycle [19]. In this system, an amplitude reduction has been reported to result in disordered cell cycle events.
Mathematical models have been intensively used to analyze the period and amplitude sensitivities with respect to parameter perturbations. There have been mainly three computational approaches: (i) the viable region approach which examines the size of the parameter region of a certain period or amplitude [20,21]; (ii) the determination of the tunability of period or amplitude which captures the extent of their changes upon altering a parameter over a large range [22,23,24]; and (iii) sensitivity analyses which assess how strongly the period or amplitude changes upon small parameter perturbations, e.g. [20,25,26,27,28,29].
So far, the main goals of computational investigations have been to compare different model designs for a particular biological process [20,27], or to determine which parameters or types of parameters are the most sensitive for an oscillatory model [25,27,28]. It is, however, of particular interest which structural properties of a model render the period and the amplitude robust or sensitive. Such a knowledge is important to understand evolutionary mechanisms in multitasking systems: If certain structural properties already favor low or high period or amplitude sensitivities, the values of the parameters could be adapted during evolution with respect to other criteria, e.g. the capability of fast entrainment or specific phase relationships. Likewise, if certain structural properties are known to preferentially result in specific period or amplitude sensitivities, this knowledge could be used in the design of synthetic oscillators with requested characteristics.
A systematic analysis can be fostered by the analysis of prototype oscillators. Generally, biological oscillators have been classified into negative feedback oscillators, substrate-depletion oscillators and inhibitor-activator oscillators [30]. The period and amplitude sensitivities of a number of prototype oscillators within these classes have been investigated using the viable region approach (i) [21], the tunability assessment with respect to one particular kinetic parameter (ii) [22,24], or sensitivity analyses (iii) [26,29]. These studies have been mostly focused on the influence of the feedback type on sensitivity properties. While the viable region approach and the tunability assessment have been performed for multiple parameter sets for each model, the sensitivity analyses have been so far restricted to single parameter sets.
Here, we combined Monte-Carlo random sampling in the parameter space and sensitivity analyses with the aim to systematically investigate period and amplitude robustness for a large number of parameter sets. We characterized robustness by sensitivity measures summarizing the effect of single parameter perturbations. First, we asked whether particular period and amplitude sensitivities are inherent properties of the model and how strongly they can vary if other parameter sets are considered. To this end, we first investigated the sensitivities of one representative model for circadian and calcium oscillations each. Subsequently, we utilized a set of prototype oscillator models to determine which of the three following structural properties have an impact on period and amplitude robustness: the type of feedback; the reaction kinetics, in particular the impact of saturating interaction functions versus mass action kinetics; and the mass conservation properties, that is the impact of interactions that are governed by mass exchange, such as in metabolic conversion reactions, versus information transfer processes, as e.g. occurring in transcription and translation.
As prototype oscillators we employed chain models of length four with either a negative or a positive feedback. They represent two of three prototype oscillator classes mentioned above [30]. The chain model with a negative feedback constitutes a negative feedback oscillator. It resembles the Goodwin model [31] which has been extensively used in a number of studies, e.g. in [32,33,34]. The chain model with a positive feedback represents a substrate-depletion oscillator. The structures of our two prototype oscillators are very similar with respect to the number of species and the position of the exerted feedback. This enables a direct comparison of the impact of the structural properties of the models on the sensitivities.
Our analyses demonstrate that not only the feedback type, but also the considered kinetics and mass conservation properties have an impact on the period and amplitude sensitivities obtained in oscillating models. We validated the applicability of our results concerning feedbacks and kinetics using a set of established models of circadian and calcium oscillations. Moreover, we confirmed the effects of the kinetics for additional oscillatory models.
Biological oscillators have been studied in great detail and described by mathematical models that capture the main characteristics of the underlying processes. Here, we are interested to which extent the response towards perturbations depends on the model properties. Therefore, we examined the robustness of mathematical models with a focus on sustained oscillations and the sensitivities of period and amplitude. Environmental changes are represented by perturbations of the kinetic parameters of the models. We performed sensitivity analyses using measures which enable the comparison of models with different topologies, number of parameters, periods, and amplitudes. The non-dimensional period and amplitude sensitivities (Eqs 1 and 2, respectively) are given by
σT=1r∑l=1r(RlT)2
(Eq. 1)
and
σA=1r∑l=1r(RlA)2,
(Eq. 2)
with r being the number of perturbed parameters [35,36]. In the equations above, the sensitivity coefficients
RlT=ΔT/TΔparl/parl
(Eq. 3)
and
RlA=ΔA/AΔparl/parl
(Eq. 4)
quantify relative changes in period T and amplitude A upon changes in a parameter parl (see Methods). Here, we use the mean of the amplitude of all model variables as amplitude A. Thus, the sensitivities σT and σA measure how strongly period and amplitude of the oscillations are affected by parameter perturbations.
The question arises to which extent the calculated sensitivities vary for different parameter sets of the model. To address this question we employed a random sampling approach to obtain multiple parameter sets and combined it with sensitivity analyses. Specifically, we applied a bottom-up sampling in which the steady state concentrations and reaction flows are sampled directly over seven orders of magnitude. The rate coefficients are then calculated from the sampled values. Details are provided in the Methods section. For each model, parameter sets are sampled until the period and amplitude sensitivities for 2 500 different parameter sets yielding sustained oscillations for each parameter perturbation could be analyzed. Depending on the specific model between 5.9·104 and 2.4·107 parameter sets had to be sampled (numbers given in the S1 File).
The resulting data are depicted in scatter plots where each dot represents the period sensitivity and the amplitude sensitivity obtained for one particular parameter set. The median values of the distributions indicate the value around which the sensitivities are centered. They give an estimate of the sensitivities that can be expected for the model in general. The widths of the distributions (as measured by the 90% data range) show how the choice of a parameter set can alter the observed sensitivities and are considered as measures of variabilities. The scatter plots are accompanied by box-plots which capture important characteristics of the sensitivity distributions (see Methods).
With this analysis work-flow, we first studied the sensitivities of a representative model of a circadian oscillator and of a representative model of a calcium oscillator. We chose the mammalian circadian rhythm model proposed by Becker-Weimann and co-workers [37] and the phenomenological model of calcium oscillations by Goldbeter and colleagues [38], respectively (model schemes in Fig 1A and 1B, model descriptions provided in the S1 Supplementary Information).
For these models, we find significant differences when comparing their period and amplitude sensitivity distributions (Fig 1C, p-values are zero, Tables B, C in the S1 File). The analysis reveals that the period sensitivity of the circadian model has a very narrow distribution, which implies that for every sampled parameter set the period sensitivity values are very similar (Fig 1C). In contrast, the period sensitivities in the calcium model vary broadly depending on the parameter set. For both models the amplitude sensitivities are variable (Fig 1C). The same findings are observed if the sensitivities for 75 000 instead of 2 500 parameter sets are determined with almost identical statistical characteristics of the sensitivity distributions (S1A Fig, Table N in the S1 File). Comparing both models, the period of the oscillations is systematically more sensitive to parameter perturbations in the calcium model than in the circadian model (Fig 1C and Tables B, C in the S1 File). In the calcium model, the median for the period sensitivities is eight-fold higher than in the circadian model and no overlap between the period sensitivity values of the two models exists regarding the level of 90% data ranges. Likewise, the amplitude sensitivities of the models are different (Fig 1C and Tables B, C in the S1 File). The calcium model exhibits a median amplitude sensitivity which is twice as large as that of the circadian model. In terms of the quartiles, there is no overlap of the amplitude sensitivities of both models (Table B in the S1 File). Similar results are obtained for altered sensitivity measures, i.e. if considering perturbations only in a specific subset of parameters, or if considering only the three largest sensitive coefficients for the overall sensitivity (S2 Fig).
Taken together, the analysis shows that the robustness of the period and the amplitude in both models is not exclusively determined by the choice of the kinetic parameter values. We find that the sensitivity distributions of the two examined models differ considerably. This indicates that the robustness is strongly affected also by the model structure.
Indeed, the biological processes and consequently the models of circadian and calcium oscillations (reviewed in [39,40,41,42]) differ in their core feedback type and their mass conservation properties. The core of the circadian oscillator is formed by a transcription-translation negative feedback loop (Fig 1A, line ending in T-shape with minus sign) whereas calcium oscillations rely on a positive feedback installed by calcium-induced calcium release (Fig 1B, arrow with plus sign). Circadian oscillation models are based on processes of transcription and translation leading to regulatory interactions without mass flow (dashed arrows in Fig 1A). In contrast, the transport processes that predominate in the calcium oscillator constitute mass conversions (solid arrows in Fig 1B). In addition to the two properties mentioned above the type of reaction kinetics applied in the models is of interest since it has already been shown that kinetics influences the steady state sensitivities [43] and the oscillation probability [7,32,34,44]. Therefore, we compare two major types of reaction kinetics which are used in both calcium and circadian models. First, linear reaction kinetics are described by the law of mass action, and second, saturating reaction kinetics are represented by Michaelis-Menten expressions. In the following, we examined the impact of these three structural properties on period and amplitude sensitivities. In order to foster a systematic analysis, we employed minimal prototype oscillator models.
First, we investigated the influence of the feedback type on the robustness of an oscillatory system. To this end, we considered prototype models of positive and negative feedback oscillators [29]. The models employed here describe a linear chain of four species where the last species exerts an inhibiting or activating feedback (Fig 2A or 2B, model equations provided in the S1 Supplementary Information).
A prerequisite for limit cycle oscillations is the existence of an unstable steady state. We therefore performed linear stability analyses (Methods) to examine under which conditions the chain models can exhibit sustained oscillations. We depict the percentage of unstable steady states in dependence on parameters characterizing the feedback (Fig 2C). The feedback terms fb (Fig 2A and 2B) are characterized by the Hill coefficient n and the ratio S4/kn1 which incorporates the concentration of species S4 and the inhibition or activation constant kn1. Unstable steady states occur for a Hill coefficient n ≥ 9 for the negative feedback model (confirming results for the Goodwin oscillator [45]), and for n ≥ 2 for the positive feedback model (Fig 2C). For both types of feedback, the percentage of parameter sets with unstable steady states is increased for larger Hill coefficients and S4/kn1 ≥ 0.7. In the following, we therefore set the Hill coefficients to n = 9 and n = 2 for the negative and positive feedback chain models, respectively. We validated our sensitivity results for Hill coefficients n = 9 in both models (S3 Fig).
For both chain models, we performed sensitivity analyses following the established work-flow (see Methods). The sensitivities of the positive and negative feedback model segregate into two populations (Fig 2D). The negative feedback chain model exhibits consistently low period sensitivities in a narrow range (median period sensitivity 0.19, 90% data range 0.05, Table D in the S1 File) but quite variable amplitude sensitivities (median amplitude sensitivity 0.66, 90% data range 1.49, Table D in the S1 File). The positive feedback chain model displays significantly higher period sensitivities in a broader range (median period sensitivity 0.68, 90% data range 1.99, p-value is zero, Tables D, E in the S1 File). The amplitude sensitivity is significantly lower in the positive feedback chain model than in the negative feedback chain model (decrease of the median value by 16%, p-value <10−5, Tables D, E in the S1 File) but covers a broader range (90% data range 4.48). Interestingly, the positive feedback chain model yields a subset of parameters characterized by a very low amplitude sensitivity (interquartile range of 0.06 around the median of 0.57 for the amplitude sensitivity distribution, Table D in the S1 File) while the period sensitivity is variable.
We next analyzed the underlying reasons for the observed sensitivity differences between the negative and positive feedback chain model. We chose to track the dynamical behavior of both models for selected parameter sets by bifurcation analyses. Such analyses allow the study of changes in the period and amplitude for a continuous alteration of one selected parameter, the so-called bifurcation parameter. We selected the parameter which most strongly affected the period (i.e. the parameter with highest |RT|) to be the bifurcation parameter. For the negative feedback chain model, the three bifurcation diagrams (S4A–S4C Fig, according to marks A-C in S4D Fig) show only slight variations of the period and smooth variations of the amplitude. For the positive feedback chain model, for each of the four parameter sets (S5A–S5D Fig, according to marks A-D in S5E Fig) the variability of the period strongly changes with the value of the bifurcation parameter. The strongest variations in the amplitudes are found for bifurcation parameter values close to the bifurcation points, while the amplitude changes considerably less elsewhere. Additionally, the bifurcation analysis reveals the possibility of transitions between limit cycles with different periods and amplitudes for bifurcation parameter changes (S5A–S5C Fig) which opens the possibility for the occurrence of high sensitivities as realized in S5C Fig, but not in S5A and S5B Fig.
Overall, although we chose parameter sets with different sensitivity values for the bifurcation analyses, the bifurcation diagrams are similar among those for the negative feedback chain model and among those for the positive feedback chain model. However, comparing the models, the bifurcation diagrams differ considerably emphasizing the importance of the feedback type for the variability in the dynamical behavior and thus also for the sensitivities of the models.
Altogether, investigating the chain models we observe that a negative feedback leads to low period sensitivities whereas a positive feedback enables higher and more variable period sensitivities. Amplitude sensitivities are highly variable for the negative feedback chain model, but less variable in terms of the interquartile range and overall lower for the positive feedback chain model.
Having assessed the influence of the feedback type on the sensitivities, we then focused on the effect of the type of reaction kinetics as further important feature of models of biological networks. We here considered mass action kinetics with reaction rates ν = k · S, and Michaelis-Menten kinetics with reaction rates ν = V · S/(S + KM).
To study their effect on period and amplitude sensitivity, we adapted the chain models with positive and negative feedback which have been described with mass action kinetics so far by introducing Michaelis-Menten kinetics (marked in gray in Fig 3A–3F, equations given in the S1 Supplementary Information). We considered three different scenarios: Replacing exclusively the degradation reactions 3, 5, 7, 8 (Fig 3A and 3D), replacing exclusively the conversion reactions 2, 4, 6 (Fig 3B and 3E), and replacing all reactions but the production rate of species S1 (Fig 3C and 3F).
The results of the sensitivity analyses for these modified chain models are compared among each other and to those for the corresponding models with mass action kinetics (Fig 3G–3J). In the majority of cases, we observe larger medians and distribution ranges for the period and amplitude sensitivities for the chain models with Michaelis-Menten kinetics than with mass action kinetics for both types of feedback. In particular, the median period sensitivities are significantly increased compared to the corresponding chain model with mass action kinetics (between 1.2- and 3.5-fold, Fig 3G, p-values <10−5, Tables F, G in the S1 File), except for the positive feedback model with Michaelis-Menten kinetics only in the degradation reactions for which no significant change is revealed (p-value 0.29, Table G in the S1 File). Whether the median amplitude sensitivities increase or decrease depends on the type of reaction that employs Michaelis-Menten instead of mass action kinetics. Compared to the respective model with mass action kinetics, amplitude sensitivities are slightly decreased by 1.1-fold for both feedback types if exclusively the degradation reactions obey Michaelis-Menten kinetics (Fig 3H, p-values <6·10−3, Tables F, G in the S1 File). If conversion reactions obey Michaelis-Menten kinetics (Fig 3B, 3C, 3E and 3F), the amplitude sensitivities are increased compared to the corresponding model with mass action kinetics (between 1.3- and 2-fold, Fig 3H, p-values <10−5, Tables F, G in the S1 File). The distribution widths of the period and amplitude sensitivities are enlarged if one compares the models with mass action kinetics to those employing Michaelis-Menten kinetics (between 2.5- and 126-fold enlarged 90% data ranges, Fig 3G and 3H, Table F in the S1 File). The only exception is made by the positive feedback chain model with Michaelis-Menten kinetics only in the degradation reactions for which a 4-fold reduction of the width of the amplitude sensitivity is observed (Fig 3D and 3H, Table F in the S1 File). The comparison of the effect of the change to Michaelis-Menten kinetics in the different reaction types shows that for the negative feedback chain model, the period sensitivities are more strongly affected if altering the kinetics in the degradation reactions than in the conversion reactions. The opposite is observed for the amplitude sensitivities. For the positive feedback chain model, altering kinetics in conversion reactions has a dominant increasing effect for both period and amplitude sensitivities.
Employing Michaelis-Menten kinetics instead of mass action kinetics leads to the introduction of additional parameters, the KM-values, in a model. An analysis of the sensitivity coefficients shed light on the contribution of each single parameter to the overall sensitivity in the chain model with positive or negative feedback and either mass action or Michaelis-Menten kinetics in all degradation and conversion reactions (S6 Fig). Larger sensitivity coefficients indicate stronger influence of the corresponding parameter. For the models with Michaelis-Menten kinetics, we observe that the KM-values have a smaller impact on the periods and amplitudes than their corresponding rate coefficients (S6 Fig, compare the last two box-plots for each triple). Comparing the sensitivity coefficients of the rate coefficients between models with different kinetics, we observe an increase for the models employing Michaelis-Menten kinetics (S6 Fig, compare the first two box-plots for each triple). Hence, the increase in the sensitivities in the models with Michaelis-Menten kinetics does not solely result from the introduction of the KM-values but we also observe an increase in the sensitivities of the rate coefficients.
In total, using Michaelis-Menten instead of mass action kinetics in all reactions leads to increased period and amplitude sensitivities for the negative as well as positive feedback chain model. In addition, increased ranges of the sensitivity distributions in models with Michaelis-Menten kinetics depict a stronger dependence of the sensitivities on the choice of the parameter set which are found in all models except for the amplitude sensitivity in the model from Fig 3D. Employing Michaelis-Menten kinetics only in degradation or conversion reactions mostly leads to a weaker increase in the sensitivities than employing this kinetics in both.
We further investigated whether the mass conservation properties in a model affect its robustness with respect to period and amplitude. The two qualitatively different types of reactions are mass conversions and regulatory interactions without mass flow, referred to as regulated production rates (Fig 4A). A mass conversion mediates a mass flow from the source species of the reaction to the product species. Thereby, the source species are consumed via the reaction. Examples are metabolic processes, transport processes and modifications like phosphorylations or methylations, in which the unmodified species’ concentrations are decreased while the concentrations of the modified species increase. In contrast, regulated productions imply an information transfer, but no mass conversion, between the regulating and regulated species. Examples are translation of mRNA into protein, or kinases mediating a phosphorylation without changing their own concentration.
So far, the chain models employed mass conversions for reactions 2, 4 and 6. In the negative feedback chain model, we replaced these three mass conversions with regulated production rates (Fig 4B, dashed arrows indicate regulated production rates, model equations given in the S1 Supplementary Information). In the positive feedback chain model, only reactions 4 and 6 were replaced by regulated productions (Fig 4C, dashed arrows indicate regulated production rates). If reaction 2 is considered to be a regulated production rate in this model, it is not a substrate-depletion oscillator anymore and sustained oscillations do not occur.
For the two modified models we performed sensitivity analyses following the established work-flow (see Methods). Substituting mass conversions by regulated production rates renders the period as well as amplitude more sensitive to parameter perturbations for both feedback types (Fig 4D and 4E). For both models median period and amplitude sensitivities are increased between 1.05- and 1.37-fold (p-values <10−5, Tables H, I in the S1 File). The effects of employing regulated productions instead of mass conversions on the level of the individual sensitivity coefficients differ for the chain models with negative and positive feedback (S7 Fig). The ranges of the period and amplitude sensitivities become smaller for the model with regulated productions for the negative feedback, but larger in case of the positive feedback (compare values of the 90% data ranges in Table H in the S1 File).
Differences in the sensitivities introduced by employing regulated production rates instead of mass conversions remain significant but small compared to the differences originating from the different types of feedbacks or kinetics (Figs 2D and 3).
The sensitivities of the chain models including the combinations of feedback type, kinetics and mass conservation properties are summarized in S8 Fig. We considered only the cases of mass action kinetics or Michaelis-Menten kinetics in all reactions. The data for models with negative feedback N1-N3 and positive feedback P1-P3 correspond to those presented in Figs 2D, 3I, 3J, 4D and 4E. They allow for the comparison between chain models with negative and with positive feedback (N1, P1), reactions with mass action and with Michaelis-Menten kinetics in models with both feedback types (N1 to N2 and P1 to P2), or with mass conversions and with regulated production rates in negative and positive feedback models (N1 to N3 and P1 to P3). Additionally, we also analyzed the effect of employing Michaelis-Menten kinetics in combination with regulated production rates in both feedback models (N4, P4 in S8A Fig). The comparison of N3 to N4 and P3 to P4 shows a strong increase in the sensitivities if using Michaelis-Menten instead of mass action kinetics. Hence, the impact of the type of kinetics on the sensitivities is strong and appears independent of the considered assumption on mass conservations. Comparing N2 to N4 and P2 to P4 demonstrates that if employing Michaelis-Menten kinetics, the introduction of regulated production rates instead of mass conversions has rather moderate effects on the period sensitivities. The amplitude sensitivities are influenced, however, less than if changing the reaction kinetics (compare N1 to N2, P1 to P2, N3 to N4, P3 to P4).
In order to test whether our insights for the chain models are valid beyond prototype oscillators, we analyzed established models of circadian and calcium oscillations in addition to the two models that have been examined initially (Fig 1). For circadian oscillations, we selected the model for Drosophila melanogaster proposed by Goldbeter and colleagues [46] and the model for Arabidopsis thaliana by Locke and co-workers [47]. For calcium oscillations, we examined the models published by Sneyd and co-workers [48] and by De Young and Keizer [49] in order to represent an open cell model and a closed cell model (equations of all four models are provided in the S1 Supplementary Information).
First, the sensitivities of the circadian oscillation models are compared (Fig 5A–5C, 5G and 5H). We observe that only the mammalian circadian rhythm model (Fig 5A, see also Fig 1C) shows a low period sensitivity for almost all parameter sets. The circadian models for D. melanogaster (Fig 5B) and A. thaliana (Fig 5C) exhibit larger median period sensitivities and decisively larger distribution widths of the period sensitivities (Tables J, K in the S1 File). Similar results are obtained for the amplitude sensitivities. Hence, in those models, the sensitivities can vary depending on the chosen parameter set. Interestingly, the reference parameter sets published together with the models give rise to low period sensitivities (white symbols in Fig 5B and 5C).
We addressed the question whether the increased and more variable period and amplitude sensitivities can be explained applying the insights obtained from the analyses of the chain models. Therefore, we investigated the model structure of the circadian rhythm models with respect to the feedback type and reaction kinetics (Fig 6). We did not examine the mass conservation properties due to the rather moderate effect of regulated productions versus mass conversions on the sensitivities observed for the chain models (Fig 4D and 4E).
The sensitivity characteristics of the mammalian circadian rhythm model resemble that of the chain model with negative feedback and mass action kinetics (compare Fig 5A to the orange stars in Fig 2D). Indeed, the circadian model employs a negative feedback (negative regulation on reaction 1, Fig 6A) and mass action kinetics for all reactions except for the regulatory terms. In addition, a positive feedback acts on production reaction 1. In that case, the positive feedback does not establish a substrate-depletion mechanism which would lead to an increase of the sensitivities according to our chain model analysis (Fig 2D). Here, however, the positive feedback seems not to influence the sensitivities.
The model for circadian oscillations in D. melanogaster relies on a negative feedback (Fig 6B) and frequently employs Michaelis-Menten kinetics in both conversion as well as degradation reactions (gray arrows in Fig 6B). Thus, the model is comparable to the chain model with negative feedback applying Michaelis-Menten kinetics in all reactions (compare Fig 5B to the dark red dots in Fig 3I). From these model structure properties one would expect considerably larger median period and amplitude sensitivity values and ranges for the D. melanogaster model compared to the mammalian circadian model which are in fact obtained (compare Fig 5B to Fig 5A, p-values <10−5, Tables J, K in the S1 File). The circadian model for A. thaliana encompasses positive and negative regulations which result in only negative feedbacks (Fig 6C), and all degradation reactions obey Michaelis-Menten kinetics (gray arrows in Fig 6C). Thus, it resembles the negative feedback chain model with Michaelis-Menten kinetics in the degradation reactions (Fig 3A, 3G and 3H). One would expect enlarged median period sensitivities and distribution ranges and slightly decreased amplitude sensitivities compared to the mammalian circadian model. We observe the former, but also increased amplitude sensitivities (compare Fig 5C to Fig 5A, p-values <10−5, Tables J, K in the S1 File), which could result from the overall different structures of the two models (Fig 6A and 6C). Taken together, in most cases we can explain the sensitivity distributions of the examined circadian models [37,46,47] using insights gained from the analyses of the chain models. Like in the prototype oscillator models the use of Michaelis-Menten kinetics influences the sensitivity distributions of circadian oscillation models.
The median period sensitivities of the calcium oscillation models (Fig 5D–5G) are high compared to those of the mammalian circadian model (Fig 5A, p-values are 0, Table K in the S1 File). Among the three calcium models, the initially chosen model (Fig 5D) shows the highest median period and amplitude sensitivities (Fig 5D, 5G and 5H, p-values are 0, Table K in the S1 File). The median period sensitivities of the open and closed cell calcium models (Fig 5E and 5F) are similar to that of the circadian model of D. melanogaster (Fig 5B and 5G, p-values 0.13 and 0.20, Table K in the S1 File) while their median amplitude sensitivities are lower than for all other examined models (Fig 5H, p-values <3·10−5, Table K in the S1 File). The reference parameter sets published with the three calcium models have not been selected for high or low period sensitivity and lead to values above as well as below the median values (compare white and black symbols in Fig 5D–5F).
We analyzed the structures of the calcium models in order to elucidate the underlying reasons for the observed sensitivity distributions. The phenomenological calcium model (Fig 6D, Fig 5D) includes a positive regulation acting on a conversion reaction thus establishing a substrate-depletion motif. This motif has been shown to result in increased period sensitivities (Fig 2D) giving an explanation for the high period sensitivities in the calcium model compared to the mammalian circadian model (Fig 5A).
The other two calcium oscillation models exhibit regulatory structures very different from those of the chain models (Fig 6E and 6F). Most variables of the open cell and the closed cell model represent opening probabilities of receptor states leading to highly connected models. Cytosolic calcium occurs in 17 out of 24 reactions for the open cell model and in 11 out of 27 reactions for the closed cell model. Consequently, explaining the obtained sensitivities based on the examinations of the lowly connected chain models is difficult. Indeed, the period and amplitude sensitivity distributions of these calcium models do not resemble any of the patterns observed in the chain models (Fig 5E and 5F).
Hence, insights obtained from the chain models can be used to explain the calculated sensitivity distributions of the circadian models and the phenomenological calcium model. In contrast, conclusions for the distributions computed for the open and closed cell calcium models cannot be drawn because their structures strongly differ from those of the analyzed chain models.
For the chain models we have demonstrated that employing Michaelis-Menten instead of mass action kinetics can result in both increased medians and increased distribution widths of the period and amplitude sensitivities (Fig 3), and made similar observations for models of circadian oscillations (Fig 5). Here, we investigated whether this also holds beyond the chain models and the models of circadian oscillations. Therefore, we chose further models of different biological oscillations and analyzed their sensitivities using the originally published type of reaction kinetics. In addition, we replaced all kinetics with the respective other type and compared the resulting sensitivity distributions. We selected the repressilator model [50] (sensitivities in Fig 7A) and a model of MAPK signaling [51] (sensitivities in Fig 7B) representing negative feedback oscillators and models describing the glycolysis [52,53] (sensitivities in Fig 7C) and the cell cycle [54] (sensitivities in Fig 7D) as representatives for substrate-depletion oscillators, model structures are given in S9 Fig, equations of the models are provided in the S1 Supplementary Information. In the original publications, the MAPK model employs Michaelis-Menten, all other three models consider mass action kinetics.
Also in these models of biological oscillations, the employed type of kinetics has an impact on the medians and the distribution widths of the period and amplitude sensitivities. If Michaelis-Menten instead of mass action kinetics are considered, median period sensitivities increased in all four oscillator models (between 1.1- and 9.3-fold, Fig 7A–7D, p-values <10−5, Tables L, M in the S1 File). Median amplitude sensitivities increased 3.3-fold for the repressilator (Fig 7A) and 1.3-fold for the cell cycle model (Fig 7D) but slightly decreased for the MAPK and the glycolysis models (1.1- and 1.02-fold, respectively, p-values <3·10−4, Tables L, M in the S1 File). For the repressilator, the MAPK and the cell cycle model, the distribution widths of both the period and the amplitude sensitivities enlarged if Michaelis-Menten instead of mass action kinetics are used (between 2.1- and 70-fold increased 90% data ranges, Table L in the S1 File). Regarding the glycolysis model, the distribution width of the period sensitivity is decreased 1.04-fold while the width of the amplitude sensitivity distribution is increased 1.4-fold (Table L in the S1 File).
Hence, for the four considered models of biological oscillations the conclusion holds that employing Michaelis-Menten instead of mass action kinetics leads to increased median period sensitivities, and to increased period sensitivity distribution widths (except for the glycolysis model). Also, the widths of the amplitude sensitivity distributions enlarged for all four models which agrees with the corresponding observations in the chain models. Regarding the amplitude sensitivities, the analyses of the chain models have already revealed that they are differently affected depending on the type of reaction whose mass action kinetics is replaced by Michaelis-Menten kinetics (Fig 3). We also found ambiguous effects for the four additional models of which two show increased median amplitude sensitivities (Fig 7A and 7D), two show decreased median amplitude sensitivities (Fig 7B and 7C). All in all, also in the models of various biological oscillations, the employed type of reaction kinetics strongly influences the sensitivity distributions.
In this work, we studied the robustness of period and amplitude of cellular oscillations by investigating a variety of model systems known to exhibit sustained oscillations. In order to distinguish properties that are inherent to the model structure from those resulting from the choice of parameter values, we examined the sensitivities for 2 500 randomly sampled parameter sets for each of the models.
To investigate the impact of structural characteristics on robustness, we used prototype oscillator models which allow for a dissection of the influence of feedback type, reaction kinetics and mass conservation properties. In many studies, prototype models have successfully been used to analyze the impact of design principles on system characteristics [24,55]. Very often, minimal models are used as a first approach to describe new biological observations (e.g. [31,52,56]) implying that our insights can be directly used to design models with specific robustness properties. Using our prototype models, we found the feedback type and the reaction kinetics to be of major importance for the period and amplitude sensitivities, illustrated in Fig 8. Employing positive feedback instead of negative feedback leads to an increase in the period sensitivities. If Michaelis-Menten kinetics instead of mass action kinetics are considered in all conversion and degradation reactions, the period and amplitude sensitivities are increased. Thereby, the different types of reactions exert different impacts. The period sensitivities are influenced moderately by the mass conservation properties, whereas the amplitude sensitivities are altered to a similar extent as by feedback types and reaction kinetics. We found these conclusions on the effect of feedback type and kinetics to be valid when extending the analysis to several representative models of circadian and calcium oscillations as long as the model structure bears some resemblances to the prototype models. The effect of the type of the kinetics on the period sensitivities was also confirmed for further models of various biological oscillations.
Previous computational approaches have demonstrated that saturating kinetics in degradation reactions can increase the oscillation probability, especially in negative feedback oscillators [7,34,44]. Taking this effect of saturating kinetics together with its herein observed effect of increasing the period sensitivities might imply a robustness trade-off between different characteristics of an oscillating system: the higher robustness of periods for negative feedback models with linear reaction kinetics may come at cost of a smaller region in the parameter space in that the systems exhibit oscillations.
Our results for the period and amplitude sensitivity can be compared to those obtained for the tunability of these characteristics in oscillating models relying on a negative feedback with or without an additional positive feedback [24]. In the former study, an individual parameter has been varied over a broad range and changes in period and amplitude have been followed. It has been found that the negative feedback model delivers stable periods and tunable amplitudes while adding a positive feedback results in tunable periods and stable amplitudes [24]. Despite the different feedback implementations and robustness measures, those results and the results of our study are in agreement (compare Fig 2D). Since our approach examines small changes in all parameters, it avoids bias towards the choice of varied parameters. As both approaches result in comparable observations the conclusion is emphasized that indeed the model structure significantly affects period and amplitude sensitivities.
While we observe overall low amplitude sensitivities in the positive feedback chain model, a detailed inspection of Fig 2D also reveals parameter sets with high amplitude sensitivities. Interestingly, in the previously published study [24] only low amplitude sensitivities have been found in the model including a positive feedback. This might result from the fact that therein the maximal amplitude has been defined as a parameter which has been set to unity. During the sampling procedure it has not been varied which might lead to an underestimation of the variation in the amplitude in their investigation.
In order to select parameter sets for each of the examined models, we used a bottom-up sampling approach. Thereby the steady state concentrations and flows were sampled and the rate coefficients were calculated from them. This method allows a very fast detection of steady state stability and the exclusion of parameter sets not yielding sustained oscillations. In terms of computational costs, it therefore outperforms top-down approaches in which kinetic parameters are sampled and steady state stability is not determined (for details see Methods). Moreover, in systems in which steady state concentrations and flows of participating species are more reliably characterized than reaction rate coefficients, it is more straightforward to sample these values in the determined intervals. For the examination of the period and amplitude sensitivities, the results remained unaffected by the specific sampling interval due to the consideration of relative changes (see Methods). We here assumed that the sampling intervals for concentrations and KM-values or inhibition and activation constants are similar which enables the occurrence of all possible regulatory modes of non-linear regulatory terms (see Methods).
Overall, our results extend an earlier study in which only the feedback type has been held responsible for the sensitivities of period and amplitude [24]. We here demonstrate that both the feedback type and the reaction kinetics influence the range of the sensitivity values considerably. Consequently, the model structure determines which influence the choice of the parameter set can have on the sensitivities. In the light of evolution this might be important for biological rhythms being subject to different constraints with respect to their input-output characteristics. For example, the negative feedback oscillator structure, together with linear reaction kinetics, enables the occurrence of low period sensitivities regardless of the choice of the parameter set. Thus, the parameter values of the model representing reaction velocities and binding constants can be tuned such that the oscillator can exhibit other important characteristics, as e.g. entrainability by light cues in the case of circadian rhythms. In contrast, substrate-depletion oscillators and oscillators with saturating kinetics display wide period and amplitude sensitivity distributions. This opens the possibility that these oscillators show different period and/or amplitude sensitivities and thus satisfy cell-type-, tissue- or organism-specific demands by only adapting reaction velocities while keeping the wiring of the participating species unchanged.
Our approach can be extended in multiple directions. First, it might be of interest to elucidate the impact of additional design principles of the model structure on the sensitivities. Earlier studies have shown that also the delay affects the occurrence of oscillations as well as the obtained amplitudes and periods [34,57,58,59]. The role of delay on the sensitivity of the period for parameter perturbations has been examined for one parameter set in the chain model [29] for which increasing delay (i.e. increasing chain length) decreased period sensitivities. It would be interesting to investigate whether the effects of delay hold for the models for many randomly sampled parameter sets. Second, the work-flow is not restricted to the analysis of models based on the description of biochemical reactions but can also be applied to models developed to generate oscillations with specific properties. This allows dissecting the influence of characteristics not captured in biochemical-based models or to expand the analyses towards more theoretical aspects determining the specific oscillatory behavior. As proof of concept, we computed the sensitivities of the FitzHugh-Nagumo model of neural dynamics [60] and of a specific λ-ω system [61]. The FitzHugh-Nagumo model delivers high period and low amplitude sensitivities, the selected λ-ω system shows variable period sensitivities and nearly constant, low amplitude sensitivities (S10 Fig).
Furthermore, in the last years, new measurement techniques have elucidated oscillations on the level of single cells which underlie intrinsic fluctuations, stressing the importance to analyze stochastic models. The effect of noise on the robustness of oscillations has been computationally studied so far for a model of calcium oscillations showing that the sensitivity of the period towards external signals can be reduced by intrinsic noise [62]. Moreover, an enhancing effect of intrinsic fluctuations on the occurrence of oscillations has been demonstrated for circadian models and the NF-κB system [63,64,65].
The work-flow of the analysis is shown in Fig 9 using the model of Goldbeter et al. [38] as an example. Details of each part are given in the following.
For each examined oscillatory process a mathematical model given by an ordinary differential equation (ODE) system was chosen. Other model approaches such as network analysis, stoichiometric analysis and structural kinetic modeling do not provide enough detail with respect to the oscillatory properties [66].
In the ODE approach the concentration Si of the ith intermediate is determined by
dSidt=∑j=1Mηijνj
(Eq. 5)
for i = 1,…, m, with m the number of variables, ηij the stoichiometric coefficients and M the number of flows νj. The flows νj are always positive. Their magnitude is determined by the kinetic parameters and intermediate concentrations. We here distinguished between three types of parameters: (i) Each flow has a corresponding rate coefficient that enters linearly, e.g. maximal reaction velocities. The flows can also depend non-linearly on the intermediate concentrations and kinetic parameters. Those parameters can be either (ii) cooperativity parameters, e.g. a Hill-coefficient, defining the non-linearity of the rate laws, or (iii) kinetic parameters such as Michaelis-Menten constants, which are called nl-parameters.
All model parameters are considered dimensionless since for each model system an appropriate non-dimensionalization is used.
In order to allow a comparison of models of biological processes acting on various time scales and in different concentration ranges an unbiased parameter sampling approach is necessary. During the sampling procedure we did not vary the characteristics of the mechanistic interactions between species, that is, stoichiometry, cooperativity parameters, and functional form of the rate laws are not altered.
In order to derive sets of kinetic parameters we employed a bottom-up approach: We performed a random sampling of steady state concentrations, flows and nl-parameters. Each of the sampled sets characterizes a unique state of the model. In principle, the random sampling procedure allows to obtain all possible states. From each set of sampled steady state concentrations, flows, and nl-parameters the according set of rate coefficients was directly calculated.
In detail, we sampled the steady state concentrations in the range (10−3, 103) such that the decadic logarithms are uniformly distributed (referred to as log10-uniform distribution in the following). We used the log10-uniform instead of a uniform distribution to take into account that intermediate concentrations within one system may occur at very different orders of magnitude [67]. The log10-uniform distribution allows for an equal representation of each of the seven orders of magnitude for the intermediate concentrations. For variables denoting probabilities in the calcium models [48,49], we adapted the interval to (10−3, 1).
Next, we chose the steady state flows such that each flow is in the interval (10−3, 103) and the steady state condition
0=∑j=1Mηijνj
(Eq. 6)
is fulfilled for each variable i = 1,…, m. In all investigated models, the number of flows, M, is higher than the number of variables, m. Therefore, the system of m equations and M unknowns given by the steady state condition (Eq 6) yields (M-m) linearly independent flows. Those were chosen in a random manner from the M flows and sampled log10-uniformly in the interval (10−3, 103), details given in S11 Fig. From the (M-m) sampled flows we computed the m depending flows. If necessary, the sampling was repeated until all of the computed flows reached a value in the interval (10−3, 103).
Finally, having determined steady state concentrations and flows, we sampled all nl-parameters log10-uniformly in the range (10−3, 103).
From the sampled values we calculated the rate coefficients using the equations defining the flows (for example, see the equations for the calcium model by Goldbeter and colleagues [38] in the S1 Supplementary Information). The set of rate coefficients can be derived as a unique solution since each rate coefficient depends linearly on its corresponding flow.
Note that while our approach ensures a log10-uniform distribution for all possible cell states with respect to the steady state concentrations, flows and nl-parameters, it does not necessarily imply that the obtained rate coefficients follow the same distribution or lie within a specific interval.
In the following the chosen sampling intervals are discussed briefly. Since we are interested in relative changes of the oscillatory properties for relative perturbations of the parameters, the results are independent of the choice of the sampling interval. This can be shown numerically by setting the sampling region to (10−1, 105) for all quantities. In that case the sensitivity statistics remain unaltered (S12 Fig), even though the obtained amplitudes of the intermediates change.
Moreover, we assumed that the sampling intervals for concentrations and nl-parameters are equal. Generally, nl-parameters characterize the affinity of an enzyme or channel for its substrate or regulatory intermediate. Sampling the affinity from the same interval as the intermediate concentrations allows covering all possible regulatory modes, e.g. values close to 0 and close to 1 for Michaelis-Menten or Hill terms. As an example, we show that the median value of a sampled Michaelis-Menten term is 0.5 (S13 Fig), implying that high and low values of those terms are represented equally well in the sampling process.
We found the bottom-up approach described so far advantageous over a top-down approach in which all kinetic parameters are directly randomly sampled (e.g. as applied in [24]). A top-down approach requires either simulating the ODE system for all sampled parameter sets or solving a set of non-linear equations to determine the steady state which leads to problems in cases of multi-stationarity. Our approach circumvents this issue by directly sampling the steady state concentrations. We compared the number of parameter sets for which the ODE system had to be simulated and the sensitivity results obtained for the proposed bottom-up sampling with those for a top-down sampling without previous examination of the steady state stability (Table A in the S1 File, S14 Fig). Our bottom-up approach requires simulating considerably less frequently (Table A in the S1 File) and thus yields lower computational costs than the top-down sampling approach. The obtained sensitivity results are similar for both sampling methods (S14 Fig).
For each parameter set sampled as described in Parameter sampling we calculated the Jacobian matrix and computed its eigenvalues to assess the stability of the sampled steady state. It is stable if all eigenvalues have negative real parts, if there is at least one eigenvalue with a positive real part the steady state is unstable. We only continued the analysis for parameter sets yielding an unstable steady state.
For the numerical integration, starting values S(0) close to the steady state vector (S0) were used, specifically S(0) = 0.95·S0. Exceptions were models which include conserved moieties (model [49]) in which the conservation conditions had to be taken into account.
For the simulations we used MATLAB [Release 2010b, The MathWorks, Inc., Natick, Massachusetts, United States] and therein the integration methods for ordinary differential equations ode45 (general solver) and ode15s (designated for stiff systems). For each parameter set the model integration was performed for 2 seconds using both methods. The method which delivered the solution to a longer integration interval was selected for further computation. The absolute and relative integration error tolerances were set to 10−8 and 10−6, respectively. The integration was performed until regular oscillations were detected (see next paragraph). Otherwise, integration was terminated if either 20 000 time units or 3 minutes (for systems with ≤10 variables) or 5 minutes (for systems with >10 variables) of calculation were reached without detecting regular oscillations. In the latter case the parameter set was discarded.
We analyzed the numerical solution with respect to occurring maxima and minima. If the intermediate with largest difference between maximum and minimum had five consecutive equal maxima (relative precision of 10−6) we studied the time intervals between them. If they were equal (relative precision of 10−4) we classified a solution as regular oscillation and amplitude and period were calculated as described below. The system was integrated again with the end-point of the previous calculation as a starting point, using 100-fold reduced calculation tolerances until 100-fold of the calculated period was reached. The period and amplitudes of this more precise solution were determined. If the oscillatory properties of initial and precise simulations agreed, the parameter set was used for sensitivity analysis, if not, it was discarded.
The period T is defined as the time interval between two maxima with identical values. The amplitude Ai of an intermediate is the concentration difference between the largest and the lowest concentration value of the species during one period. Since each of the analyzed oscillatory models consisted of more than one species, we examined the arithmetic mean value of the amplitudes Ai of all species of a model as amplitude A.
In the sensitivity analysis we are interested in the changes in period (ΔT = Tperturbed - Tunperturbed) and amplitude (ΔA = Aperturbed - Aunperturbed) due to changes in relevant kinetic parameters (Δpar = parperturbed - parunperturbed). Therefore, for each sampled parameter set the rate coefficient and nl-parameters, which are maximal reaction velocities, inhibition or activation constants and Michaelis-Menten constants, were individually increased by 2%. The model with the parameter set including one perturbed parameter was again integrated and period and amplitude were determined as described above. This allowed for computing the sensitivity coefficients according to Eqs (3) and (4) (Fig 9, box 7).
The overall sensitivities per sampled parameter set were calculated from the sensitivity coefficients RA and RT according to Eqs (1) and (2). It was sampled until the sensitivity values for 2 500 parameter sets could be calculated. The number of parameter sets required to be sampled for that analysis is indicated for each model in the S1 File.
The period and amplitude sensitivities are depicted in scatter plots, with each dot representing the sensitivities of one parameter set (Fig 9, box 8). The median values of the corresponding sensitivity distributions are given (white circle). Moreover, the characteristics of the sensitivity distributions are captured in box-plots giving the median (line as central value), the 95% confidence interval of the median (notch, [68]), the first and third quartiles (box), the 5th and 95th percentiles (end of whiskers). In addition, the utmost 10% of the data are given as crosses, either according to their actual values if they are between the end of the whiskers and the broken line, or as arbitrary values if beyond that line. The distance between the 5th and 95th percentiles is called 90% (data) range and is used as the measure of variability.
Comparing the results of the sensitivity analysis for 2 500 to 75 000 parameter sets for two models reveals almost identical characteristics with respect to median, interquartile range, and 90% data range (S1A Fig, Table N in the S1 File). Moreover, we tracked the sensitivity characteristics for increasing sample size and found decreasing variabilities of median values and 90% data range limits (S1B Fig). This illustrates that 2 500 parameter sets are sufficient to get adequately precise information on the statistical characteristics.
In addition, we checked whether the magnitude of the parameter perturbations had an impact on the sensitivities. We obtained similar results for other small parameter perturbations (1%, 5%, 10% and -2%, S15 Fig).
The described work-flow was applied to all examined models.
For all comparisons of sensitivity distributions between different models the Mann-Whitney-U test was used.
For the analysis of steady state stability of the chain models we examined the eigenvalues of the Jacobian matrix. We sampled until 10 000 parameter sets with S4/kn1 in (10a, 10a+0.1) for each a = -6, -5.9,…, 6 were obtained for every integer Hill coefficient n in the interval [1,25].
|
10.1371/journal.pbio.2003892 | Identification of osmoadaptive strategies in the halophile, heterotrophic ciliate Schmidingerothrix salinarum | Hypersaline environments pose major challenges to their microbial residents. Microorganisms have to cope with increased osmotic pressure and low water activity and therefore require specific adaptation mechanisms. Although mechanisms have already been thoroughly investigated in the green alga Dunaliella salina and some halophilic yeasts, strategies for osmoadaptation in other protistan groups (especially heterotrophs) are neither as well known nor as deeply investigated as for their prokaryotic counterpart. This is not only due to the recent awareness of the high protistan diversity and ecological relevance in hypersaline systems, but also due to methodological shortcomings. We provide the first experimental study on haloadaptation in heterotrophic microeukaryotes, using the halophilic ciliate Schmidingerothrix salinarum as a model organism. We established three approaches to investigate fundamental adaptation strategies known from prokaryotes. First, proton nuclear magnetic resonance (1H-NMR) spectroscopy was used for the detection, identification, and quantification of intracellular compatible solutes. Second, ion-imaging with cation-specific fluorescent dyes was employed to analyze changes in the relative ion concentrations in intact cells. Third, the effect of salt concentrations on the catalytic performance of S. salinarum malate dehydrogenase (MDH) and isocitrate dehydrogenase (ICDH) was determined. 1H-NMR spectroscopy identified glycine betaine (GB) and ectoine (Ect) as the main compatible solutes in S. salinarum. Moreover, a significant positive correlation of intracellular GB and Ect concentrations and external salinity was observed. The addition of exogenous GB, Ect, and choline (Ch) stimulated the cell growth notably, indicating that S. salinarum accumulates the solutes from the external medium. Addition of external 13C2-Ch resulted in conversion to 13C2-GB, indicating biosynthesis of GB from Ch. An increase of external salinity up to 21% did not result in an increase in cytoplasmic sodium concentration in S. salinarum. This, together with the decrease in the catalytic activities of MDH and ICDH at high salt concentration, demonstrates that S. salinarum employs the salt-out strategy for haloadaptation.
| Salinity is one of the strongest abiotic factors in nature and can have harmful effects on organisms. When a cell is placed in a high-salt environment, osmotic pressure pulls unbound water molecules out of the cell, leaving behind a highly concentrated cytoplasm, which impairs cell function and can cause cell death. Organisms living in high-salt environments therefore need to have specific adaptation mechanisms. For prokaryotes and fungi, such adaptations have been examined in detail. Protists—unicellular eukaryotes that constitute most of the eukaryotic diversity—are less well studied, mainly because of the lack of adequate methodological approaches. The scope of our study therefore includes the establishment of techniques to investigate fundamental salt adaptation strategies in protists and the application of these techniques to a halophilic, heterotroph model ciliate protist to reveal its specific haloadaptations. We provide the first experimental evidence for the biosynthesis and accumulation of two compatible solutes that the ciliate uses to combat increasing salt concentrations. One of these, glycine betaine, is widely distributed across kingdoms, while the other, ectoine, has to date only been identified in prokaryotes.
| Salinity is a decisive environmental determinant of microbial community composition and dispersal, exerting a high evolutionary selection pressure [1]. Shifts in salinity represent one of the most difficult environmental barriers to cross for macro- [2,3] and microorganisms alike [4,5]. Reasons are mainly the energetic costs of osmoregulation and the requirement of specific adaptive mechanisms to cope with high salt concentrations [6]. Salinity has detrimental effects on organisms when not counteracted. Hypertonic conditions induce an osmotic gradient, which lowers the relative water content of a cell, leaving behind a highly concentrated cytoplasm. In such a cell environment, nucleic acids, proteins, and other macromolecules denature and lose their functions [7]. Proteins form aggregates, precipitate [8], and disrupt their tertiary structure. Enzymes lose their flexibility and, consequently, their catalytic activity [9]. Nevertheless, flourishing communities of archaea, bacteria, and microbial eukaryotes are found in various hypersaline environments on Earth [10].
Due to the only recent awareness of the ecological importance of microbial eukaryotes in hypersaline systems [11,12], most knowledge about haloadaptation strategies derives from prokaryotic research, in which in-depth investigations already revealed a broad repertoire of different haloadaptation mechanisms. We, here, describe the two main and fundamentally different processes as adaptations to high-salt environments (reviewed in detail by [13]). The salt-in strategy involves the intracellular accumulation of molar concentrations of chloride ions (Cl−) and potassium ions (K+). Because proteins need to retain their functional conformation and activity at high intracellular salt concentrations, massive adaptations of the enzymatic machinery are mandatory [14]. Salt-in strategists, therefore, frequently possess highly acidic proteomes. Additionally, halophilic proteins are enriched with hydrophilic amino acids [14], which prevent overly rigidly folded conformations under high-salt conditions. Consequently, hypoosmotic conditions cannot easily be counterbalanced and lead to destabilization and malfunction of proteins. Thus, the salt-in strategy is used by few groups of halophiles only, such as the aerobic members of Halobacteriaceae (Archaea), Salinibacter (Bacteria), and the anaerobic Halanaerobiales (Bacteria). Alternatively, the more common salt-out strategy involves the exclusion of salt ions from the cytoplasm, simultaneously synthesizing or accumulating high concentrations of compatible solutes. These are uncharged or zwitterionic low molecular mass organic compounds, which do not interfere with the cell metabolism at high cytoplasmatic concentrations, provide osmotic balance, and further protect (nonsalt-adapted) enzymes and other macromolecular structures against inactivation, inhibition, and denaturation [15].
Knowledge about microeukaryotic salt adaptation strategies is restricted to some genome-sequenced fungi (halotolerant Hortea werneckii, halophilic Wallemia ichtyophaga, and halophilic Eurotium rubrum), and the green alga Dunaliella. Even though all three fungal taxa share some common haloadaptive traits, the differences in their genetic repertoire suggest multiple molecular pathways for haloadaptation and a high diversity in adaptive strategies [16]. All three fungi accumulate glycerol as their main compatible solute [16–18] and have a significantly higher proportion of acidic residues in predicted proteins [16]. Genome and transcriptome analyses of H. werneckii, which maintains low intracellular K+ and sodium ion (Na+) concentrations, revealed considerable expansion of gene families encoding membrane-bound metal cation transporters [17,19,20]. Interestingly, the halophiles W. ichthyophaga and E. rubrum have a substantially lower number of such genes [18–20]. This indicates that in contrast to H. werneckii, elevated intracellular Na+ concentrations are not toxic for W. ichthyophaga, further documenting the different salt-combatting strategies in these two model fungi. Intracellular accumulation of Na+ to promote growth under salt stress was also reported for the halotolerant yeast Debaryomyces hansenii [21]. Transcriptome data of D. salina identified an active signaling system, which enables it to quickly adapt to a wide variety of salt concentrations as well as a significantly enriched photosynthetic glycerol pathway as adaptation to osmotic changes [22]. This corroborates earlier findings of low intracellular Na+-concentrations [23] and an electron transport-coupled Na+-extrusion [24] in D. salina. Only recently Harding et al. [25] published a transcriptome study of the halophilic flagellates Halocafeteria seosinensis and Pharyngomonas kirbyi to investigate salt adaptation strategies in heterotroph protists. Both species do not possess an acidic proteome characteristic for salt-in strategists, but display an increased expression of enzymes involved in the synthesis and transport of the compatible solutes 5-hydroxyectoine and myo-inositol.
However, up to now there is no experimental evidence for intracellular ion concentrations and the accumulation and synthesis of compatible solutes with increasing salinity in heterotrophic protists. This is mainly due to methodological reasons, such as the lack of axenic cultures, disturbance of measurements of intracellular contents by ingested food bacteria, or the lack of adequate sensitive techniques. Methods, which measure intracellular concentrations of inorganic ions and other compounds, were so far mainly applied to bacteria (e.g., [26–29]). These methods rely on the measurement of intracellular and extracellular water volumes, which are used to determine the concentration of cell-associated ions in centrifuged cell pellets. As reviewed in [30] and [31], those methods are error-prone and even small mistakes in the determination of the intracellular and extracellular water volumes result in large deviations in the respective ion concentrations. Another approach to measure intracellular concentrations of Na+, K+, and Cl− is based on X-ray microanalysis, in which ion content and cell volume are microscopically estimated [32]. Both approaches could not be successfully applied to heterotrophic protists and generally led to cell rupture and distortion of cell shapes, which impeded calculations of the cell volume. More realistic estimates of intracellular ion concentrations were obtained for the green alga Dunaliella, due to the physical separation of cells from the extracellular medium by passing them through cation-exchange minicolumns [33,34] and an approach based on 23Na-nuclear magnetic resonance (NMR) spectroscopy using a dysprosium tripolyphosphate complex as a Na+-shift reagent to discriminate between intracellular and extracellular Na+ [35]. Those studies, however, were based on cell assemblages requiring reliable determinations of cell numbers.
A promising approach for the identification of compatible solutes is high-resolution NMR spectroscopy, which is routinely used in prokaryotic research to investigate the salt-out strategy, but was rarely applied to salt-related microeukaryotic research in this respect. One study using this technique focused on the effect of UV-induced stress on the metabolome of the moderate halophile ciliate Fabrea salina. Results suggested the accumulation of ectoine (Ect) and glycine betaine (GB) as osmoprotectans in a late reaction to UV radiation [36].
We here provide, to our knowledge, the first experimental study that investigates the salt adaptation strategy in the halophilic heterotroph ciliate S. salinarum [37]. Ciliates belong to the dominating protistan groups in hypersaline environments [38–41] and are therefore of high ecological relevance, e. g., as grazers exerting a top-down control on bacterial biomass. The heterotroph ciliate S. salinarum shows a wide geographic distribution and has a broad growth range between 1% and 21% salinity, with optimal growth at 9%.
Using S. salinarum as a model organism, the goals of our study were to establish experimental methods to (1) detect, identify, and quantify intracellular compatible solutes in salt-adapted heterotroph protists using proton (1H)-NMR spectroscopy and (2) determine relative intracellular ion concentrations in living single cells using ion imaging. Additionally, we tested and compared the activity of two enzymes of S. salinarum and Escherichia coli under high- and low-salt conditions. We found that S. salinarum uses the compatible solutes GB and Ect rather than accumulating inorganic ions to counterbalance osmotic stress. Intracellular concentrations of GB and Ect increased significantly with an increasing external salinity. Addition of exogenous GB, Ect, and choline (Ch; precursor of GB) stimulated the growth of S. salinarum notably, indicating that the organism is both able to accumulate the solutes from the external medium and also, apparently, to synthesize them. Enzyme activity assays support the salt-out strategy, as enzyme activity was reduced at high salt conditions, but not affected by GB at molar concentrations.
To investigate if S. salinarum implements the salt-out strategy, 1H-NMR spectroscopy was applied to detect and identify potential compatible solutes. Chemical shifts and coupling constants of all obtained 1H-NMR spectra from S. salinarum were compared to those obtained from authentic compatible solute spectra recorded under identical conditions, revealing the presence of GB and Ect (Fig 1). Separate addition of authentic GB and Ect to S. salinarum samples confirmed our results (S1 and S2 Figs). 1H-NMR spectra of ethanolic cell extracts gave NMR spectra as described above (S3 Fig), indicating that no cell material was lost during 1H-NMR sample preparations. To avoid spurious ethanol peaks, only nonethanolic extracts were analyzed.
Because our model organism S. salinarum is a bacterivore ciliate, it cannot be cultivated in axenic cultures. Our cultures, therefore, included a mixture of known indigenous bacteria (S1 Table) from the original sampling site serving as a food source. The main purpose of 1H-NMR controls was to exclude the possibility of detected compatible solutes originating from these bacteria. Our control experiments verified that the detected compatible solutes derive from S. salinarum cells and not from residual food bacteria. Neither osmolyte peaks could be detected in the bacterial 1H-NMR spectra (S4A Fig) nor any food bacteria in the external medium (S4B Fig) or in the food vacuoles of starved S. salinarum cells (S4C Fig).
The compatible solutes GB and Ect were quantified by integrating the peak areas with respect to an internal standard (trimethylsilylpropanoic acid [TMSP]) along a salinity gradient (0.85 mol/l–3.59 mol/l NaCl). Average intracellular concentrations of GB (c(GB)) increased with increasing external salinity from 0.35 mol/l (± 0.2) at a salinity of 0.85 mol/l NaCl to up to 6.35 mol/l (± 1.0) at a salinity of 3.59 mol/l NaCl (Fig 2) (S5A Fig). Intracellular Ect concentration (c(Ect)) was as low as 0.09 mol/l (± 0.1) at an external NaCl concentration of 0.85 mol/l and increased to 2 mol/l (± 0.6) at a salinity of 3.59 mol/l NaCl (Fig 2) (S5B Fig). Between 3.25 mol/l NaCl and 3.59 mol/l NaCl, intracellular GB and Ect concentrations changed by a factor of three (otherwise: < 2). External NaCl concentration and average intracellular concentrations of GB and Ect, respectively, revealed a significant linear relationship (GB: R2 = 0.85, p = 0.001; Ect: R2 = 0.78, p = 0.002; Fig 2) between 0.85 mol/l NaCl and 3.25 mol/l NaCl. Total intracellular osmolyte concentration (sum of GB and Ect) and external NaCl concentration showed an even stronger linearity (R2 = 0.92, p = 0.000). Including the intracellular osmolyte concentrations measured at an external NaCl concentration of 3.59 mol/l lowered the coefficient of determination (GB: R2 = 0.46, p = 0.027; Ect: R2 = 0.64, p = 0.006; total: R2 = 0.51, p = 0.018).
As revealed by Welch’s t test (S5A Fig), significant changes in c(GB) occurred between 5% and 7% salinity (c(GB)7% = 0.72 mol/l [± 0.3]; p = 0.032), between 11% and 13% salinity (c(GB)11% = 0.85 mol/l 7 [± 0.3], c(GB)13% = 1.23 mol/l [± 0.3]; p = 0.032), between 17% and 19% salinity (c(GB)17% = 1.21 mol/l [± 0.5], c(GB)19% = 1.94 mol/l [± 0.4]; p = 0.000), and between 19% and 21% salinity (c(GB)21% = 6.35 mol/l [±1.04]; p = 0.000). Significant changes in c(Ect) along the salinity gradient appeared between 11% and 13% salinity (c(Ect)11% = 0.23 mol/l [± 0.1], c(Ect)13% = 0.45 mol/l (± 0.1); p = 0.007) and 19% and 21% salinity (c(Ect)19% = 0.64 mol/l [± 0.2], (c(Ect)21% = 2 mol/l (± 0.6); p = 0.002) (S5B Fig).
In total, c(GB) were at least two times higher than c(Ect) at all salinities (S6 Fig).
To examine whether S. salinarum is able to accumulate compatible solutes from the external medium, cultures were incubated with GB, Ect, and Ch, whereas Ch acts as a precursor in the synthesis pathway of GB. After nine-d growth, ciliate cell abundance in the GB-enriched culture medium was increased compared to the control cultures (Fig 3). Control cultures included on average 280 cells ml−1 (± 181), whereas the GB-enriched culture included 983 cells ml−1 (± 32). The increase of cell abundance was pronounced but not significant (p = 0.85). In the Ch-enriched cultures cell abundances (3,247 cells ml−1 ± 1557) were significantly increased compared to the control (p = 0.004) and compared to the GB-enriched cultures (p = 0.027). Exogenously provided Ect increased the cell density up to 1,380 cells ml−1 (± 28). The increase of cell abundances was pronounced but not significant compared to the control and the GB treatment (p > 0.05).
The dominant pathway to synthesize GB is by a two-step oxidation of Ch via betaine aldehyde. This oxidative pathway can occur with different enzymes, e.g., in plants, Ch is oxidized to betaine aldehyde by a ferredoxin-dependent Ch monooxygenase [42] and betaine aldehyde is mediated by a soluble, nicotinamide adenine dinucleotide (NAD+)-linked betaine aldehyde dehydrogenase into GB [43]. However, the enhanced growth of S. salinarum with exogenously supplied Ch indicated a preference for the GB-precursor Ch over GB itself (see growth experiments above). Because this would point to a potential GB synthesis based on the conversion of Ch to GB, we conducted an uptake experiment with labelled 13C2-Ch. In a first step, S. salinarum cultures were starved to exclude the possibility of bacterial contaminations. After three d of starvation, DAPI staining revealed bacteria-free cultures (Fig 4A). 1H-NMR spectroscopy was conducted to examine if S. salinarum is able to take up and oxidize exogenously provided 13C2-Ch into 13C2-GB. 1H-NMR spectroscopy revealed that the 1,2-13C2-labeled Ch was internalized and converted to 1,2-13C2-GB in all replicate cultures (S7 Fig). The ratio of GB:1,2-13C2-GB:Ect showed no significant differences between the individual replicates and was on average 3:8:1 (Fig 4B).
To study the intracellular Na+ and K+ concentrations of S. salinarum, we established a microscopy-based approach, which relies on ion-specific fluorescent dyes (ion imaging). For the relative fluorescence measures of the cytoplasm, only the cytoplasm was integrated as a region of interest, not the vacuole-like structures. Results are given as normalized corrected relative fluorescence intensities (CRF).
Na+ imaging of S. salinarum cells grown in the respective salinities (long-time adapted cells) revealed stable Na+ concentrations in the cytoplasm of S. salinarum at external salinities between 3.8% and 17% (CRF4% = 46.70 ± 12; CRF9% = 52.03 ± 14, CRF13% = 49.70 ± 10, CRF17% = 47.30 ± 15), and a significant Na+ increase at 21% salinity (CRF21% = 88.70 ± 15; Fig 5A). Additionally, vacuole-like structures containing elevated Na+ concentrations were observed in the S. salinarum cells exposed to 21% salinity (Fig 5A). K+ imaging of long-time adapted S. salinarum cells revealed steady cytoplasmic K+ concentrations with increasing external salinity (CRF4% = 17.17 ± 3; CRF10% = 17.04 ± 4; CRF21% = 18.40 ± 4; p > 0.05) (Fig 5B).
To test whether the observed increase in Na+ concentration with increasing salinity in long-time adapted cells was a result of passive ion influx or rather a product of active regulated uptake (i.e., presence of specific membrane transporters), short-time adaptation experiments were conducted with single S. salinarum cells exposed to increasing salt concentrations. Na+ imaging with short-time adapted cells showed no significant changes in Na+ concentrations in the cytoplasm. The relative fluorescence intensity was on average 12.48 ± 2. After a salinity increase up to 18%, the cells busted due to the high osmotic pressure (S8 Fig). Increasing the adaptation time to 10 min to allow for a longer equilibration resulted in slightly lower average relative fluorescence intensity (8.62 ± 1; S8B Fig), but did also not show any significant increase in Na+ concentration.
High intracellular Na+ concentrations can inhibit enzymes if the proteomes of the cells are not sufficiently adapted. In a third experimental approach, we investigated the kinetic performance of two cytosolic enzymes under different salt and osmolyte conditions and compared their protein charge with respective proteins of E. coli to get a hint of whether S. salinarum might possess an acidic proteome.
The first enzyme, malate dehydrogenase (MDH) catalyses the interconversion of L-malate to oxaloacetate using the reduction of NAD+ to NADH (nicotinamide adenine dinucleotide). Whereas prokaryotes contain only a single MDH, eukaryotes have several MDHs, including a cytoplasmic form. Cytosolic MDH assists the malate-aspartate shuttle so that reducing equivalents can pass through the mitochondrial or peroxisomal membrane for cellular processes. Our phylogenetic analysis (using the dataset from [44]) shows that MDH from S. salinarum and the highly similar MDHs from the ciliates Stylonychia lemnae (79% identity) and Oxytricha trifallax (77% identity) belong to the clade I subset of cytosolic and plastidial MDHs, including the human cytosolic MDH (S9 Fig). MDH from S. salinarum only shows a weak sequence similarity with the human mitochondrial MDH (23% identity) and E. coli MDH (21% identity). Inspection of the DNA sequence 3ʹ of the stop codon of the S. salinarum MDH gene did not reveal evidence for a read-through stop codon followed by a peroxisomal serine-lysine-leucine targeting sequence [45]. Nevertheless, MDH enzymes have similar kinetic properties and share a common catalytic mechanism, which is reflected by a high degree of structural similarity [46] between the various MDH clades. S. salinarum MDH has all highly conserved amino acid residues involved in substrate binding (R116, R122, R186, N131) of the active site (H187) and for NAD+ binding site (N129), which leaves no doubt about the assignment as genuine MDH enzyme.
The second enzyme investigated, isocitrate dehydrogenase (ICDH), catalyses the oxidative decarboxylation of isocitrate to 2-oxoglutarate and requires NAD+ or nicotinamide adenine dinucleotide phosphate (NADP+), producing NADH and NADPH, respectively. Whereas E. coli has only one form, humans contain three classes of ICDH isoenzymes: mitochondrial NAD+-dependent, mitochondrial NADP+-dependent, and cytosolic NADP+-dependent ICDH. ICDH from S. salinarum and the ciliates S. lemnae and O. trifallax ICDHs (both 78% identity) are phylogenetically related to homodimeric NADP+-dependent ICDHs [47] such as the human cytoplasmic enzyme (68% identity, S10 Fig), which has a significant role in NADPH production. The cytosolic form from many eukaryotes can also be localized to the peroxisomes through a C-terminal peroxisomal targeting sequence. Analysis of the S. salinarum DNA sequence neither identified C-terminal and N-terminal peroxisome-targeting sequences for peroxisomal import, nor a clear mitochondrial targeting sequence, supporting cytosolic localization.
ICDH and MDHs were overexpressed in E. coli cells and purified using nickel-nitrilotriacetic acid (Ni-NTA) chromatography. Most of these proteins appeared in the soluble fraction upon fractionation. After sodium dodecyl sulfatepolyacrylamide gel electrophoresis (SDS-PAGE), clear bands corresponding to calculated molecular masses (MDHE.coli: 34.2 kDa (kilodalton), MDHS.salinarum: 40.7 kDa, ICDHE.coli: 48.2 kDa, ICDHS.salinarum: 48.7 kDa) were observed in the purified proteins (estimated purity 95%–98%, S11 Fig). The effect of NaCl at low (0.05 M) and high salt conditions (1.2 and 2.4 M NaCl) on the enzymatic activity was analysed by determination of the Michaelis Menten constants (Table 1). MDHE.coli at 0.05 M NaCl presented a hyperbolic curve, with specific activity (Vmax of 1,486 U mg−1) and Km value for oxaloacetate (0.022 mM), values typical for this class of enzyme [48,49] (Fig 6A and Table 1). At a low salt concentration, MDHS.salinarum presents an even higher specific activity (Vmax = 2,062 U mg−1) than the E. coli enzyme, with a comparable Km value for oxaloacetate (0.038 mM) to the microbial enzyme (Fig 6B). When the salt concentration in the buffer was elevated to 1.2 M, activity of the MDH enzymes from both organisms strongly decreased and the Km values increased almost tenfold (Table 1). At 2.4 M NaCl, only a residual MDH activity of 5% and 7% for the E. coli and S. salinarum enzyme, respectively, remained. It is known that E. coli can accumulate K+ up to 0.9 M in the initial phase of osmoadaptation, followed by intracellular accumulation of compatible solutes, e.g., GB [50]. Our results show that addition of GB at high concentration (2.5 M) to the buffer impaired MDHE.coli activity (18% residual activity) but to a much lower extent than with 2.4 M NaCl. Under the same condition, MDHS.salinarum retained more activity (25%). GB had no remarkable effect on the Km values.
A qualitatively similar picture was observed for ICDH (Fig 6C and 6D and Table 1). At low salt conditions, ICDHE.coli had a Vmax of 46 U mg−1 and a Km for DL-isocitrate of 0.021 mM, values comparable to those found for many ICDHs [51]. Interestingly, ICDHS.salinarum had a similar Km (0.010 mM) but was less active (Vmax of 6.8 U mg−1), like some eukaryotic ICDHs [52]. It is possible that an allosteric activation mechanism regulates the S. salinarum enzyme. As observed for all ICDHs, ICDHS.salinarum also required magnesium (Mg2+) ions to be active. ICDH from E. coli and from S. salinarum were much less active upon increasing the salt concentration. At 1.2 M NaCl, 27% and 23% of the Vmax was observed for E. coli and S. salinarum enzymes, whereas at 2.4 M NaCl residual activities of 10% and 5% were found. ICDHS.salinarum had a better performance in 2.5 M GB than ICDHE.coli, as the enzyme retained about 35% of its activity in comparison with 7% for the E. coli protein. Again, GB had no remarkable effect on the Km values (Table 1). These findings contrast with the kinetic properties of ICDHs purified from the halophiles Haloferax volcanii and Halobacterium salinarum, which are salt-activated enzymes reaching a Vmax comparable to the E. coli enzyme [53,54].
Taken together, the in vitro activity measurements support the salt-out strategy as indicated by the 1H-NMR experiments of S. salinarum.
Using 1H-NMR spectroscopy, we were able to provide the first experimental detection of compatible solutes and their accumulation with increasing salinity in heterotrophic protists. Based on our results, we can conclude that S. salinarum utilizes two major compatible solutes, GB and Ect (Fig 1), to combat high salt concentrations in its environment. Furthermore, our approach revealed an accumulation of both compatible solutes with increasing salinity. Thus, the concentration of both seems to be regulated in dependence of different salt concentrations (Fig 2) (S5 Fig).
The dominant compatible solute throughout all tested salinities was GB (N,N,N-trimethylglycine), which is one of the most common osmoprotectants in halophilic bacteria and archaea of diverse phylogenetic affiliation [55], halotolerant plants, as well as algae [56,57], and was also detected in halotolerant heterokont thraustochytrids [58]. Usually, organisms accumulate GB actively from culture media because many of the regular components, like yeast extract, contain this organic solute [57]. Nevertheless, there are two generally different pathways describing how GB can be synthesized (Fig 7A and 7B). In almost all cases, it is synthesized by a two-step oxidation of Ch via betaine aldehyde. This oxidative pathway can occur with different enzymes. In plants, Ch is oxidized to betaine aldehyde by a ferredoxin-dependent Ch monooxygenase [42] and betaine aldehyde is mediated by a soluble, NAD+-linked betaine aldehyde dehydrogenase into GB [43]. Both enzymes are located in plant chloroplasts [59]. In bacteria, including E. coli, a membrane-bound electron transfer-linked Ch dehydrogenase oxidizes Ch to betaine aldehyde, which is subsequently oxidized to GB by betaine aldehyde dehydrogenase [60]. Alternatively, some Arthrobacter spp. have a soluble Ch oxidase (COX) that carries out both oxidations [61]. In a completely different pathway, GB is synthesized via consecutive methylation of glycine to GB. However, as far as is known, this pathway is utilized by a few extremely halophilic bacteria only ([62], and references therein).
The second identified compatible solute Ect (1,4,5,6-tetrahydro-2-methyl-4-pyrimidinecarboxylic acid) is an almost equally wide distributed compatible solute throughout different halophilic and halotolerant bacteria [64–71]. Among the so far investigated compatible solutes, Ect displays the most powerful stabilizing properties in a variety of different enzymatic processes [72]. It features distinct protective effects against the stress factors heat, cold, and high salt concentrations in diverse prokaryotes [72–74] and is known to be accumulated by the moderate halophilic ciliate F. salina to protect the organism against UV radiation [36]. Ect can be both, accumulated from the environment or synthesized de novo [75]. The biosynthesis of Ect has been intensively studied in the bacterium Halomonas elongata [64] (Fig 7C). In a first step, aspartate-β-semialdehyde is converted to L-2,4-diaminobutyric acid catalyzed by diaminobutyric acid aminotransferase. Subsequently, L-2,4-diaminobutyric acid is converted by diaminobutyric acid acetyltransferase into Nγ-acetyl-L-2,4-diaminobutyrate. Ultimately, N-acetyl-L-2,4-diaminobutyrate is converted into Ect via Ect synthase [75]. For a longer period of time, Ect synthesis was thought to be restricted to bacteria and a few archaea [76], but only recently expression patterns of genes potentially involved in the synthesis of Ect and 5-hydroxyectoine (a derivative of Ect) were recognized in the predicted proteomes of the two halophilic flagellates H. seosinensis and P. kirbyi [25].
The accumulation of different solutes, instead of only one, is known from most halotolerant and halophilic prokaryotes (e.g., [77,78]). Occasionally it is a combination of zwitterions—like GB and Ect—but more frequently of several solutes with the same net charge (as reviewed in [78]). At first glance, our results suggest the exclusive accumulation of GB and Ect, which has been observed in a few halophilic and halotolerant prokaryotes only [66,79,80]. If this would be true, we would expect that the sum of the osmolyte concentrations is osmotically equivalent to the external NaCl concentration. For S. salinarum, however, we measured an internal osmolyte concentration of ca. 2 M at an external NaCl concentration of 3 M (Fig 2). This disparity might be explained by the following reason: calculations of the cell volumes of S. salinarum were based on size measurements of individual cells from examined organisms using scanning electron microscopy (SEM) pictures. The estimation of an exact cell volume, however, is heavily dependent on the applied mathematics, as discussed previously by Sun and Liu [81]. Even the most adequate model cannot completely mirror the exact cell shape and profile and thus one should be aware that calculations based on cell width, length, and height are only approximations. Additionally, those mathematic models do not consider “dead spaces” occupied by e.g., oil bodies and/or starch granules, which can be detected in S. salinarum cells [37]. Those structures can account for a notable proportion of the total cell volume. We therefore attribute the difference in estimated internal and known external osmolality to the uncertainty in the calculation of the cell volume.
Our growth experiments with exogenously provided GB and Ect showed an increase of S. salinarum densities (Fig 3). Interestingly, the addition of the GB precursor Ch resulted in a 12 times higher cell number when compared to the control culture (without any exogenous solutes). This points to a possible GB synthesis by oxidation of Ch to GB. It should be noted that the growth experiments were influenced by the number of food bacteria in the cultures, which could have benefited from exogenously provided solutes. Therefore, a direct approach was conducted by adding labeled 1,2-13C2-Ch to the medium of starved, bacteria-free S. salinarum cultures (Fig 4). Using 1H-NMR spectroscopy, indeed, 1,2-13C2-GB was detected, probably resulting from the oxidation of 1,2-13C2-Ch to 1,2-13C2-GB (Fig 4) (S7 Fig). Our results demonstrate for the first time that Ch oxidation might well occur in heterotrophic protozoa. Analysis of the S. salinarum transcriptome could shed light on the enzymes involved in the conversion. Although the Ect biosynthesis was not experimentally investigated in this study, we observed an increased Ect concentration and therefore importance at higher salinities (S6 Fig). This is supported by the recent transcriptome data on two halophilic flagellates, emphasizing a de novo synthesis of Ect in halophilic protists [25].
Although even low concentrations of Na+ are toxic for most living cells [82], a few halophilic microorganisms, in particular some Halobacteria [83] and the yeast D. hansenii [84], require high intracellular concentrations of Na+ for normal enzymatic activities. For S. salinarum, ion imaging revealed no significant differences in the relative intracellular cytoplasmatic Na+ concentration (Fig 5A) at lower salinities (4%–17%). However, Baxter and Gibbons [85] pointed out that a constant low intracellular ionic environment can only be achieved by energy-dependent mechanisms. As high extracellular NaCl concentrations lead to a constant influx of Na+ ions because of a not completely impermeable membrane and the additional influx of Na+ during cotransport of certain substrates or amino acids [86], it is to assume that S. salinarum requires some Na+ extrusion mechanisms in its cell membrane. So far, there are two major mechanisms for several halotolerant and halophilic prokaryotes (e.g., [86–88]), plants (e.g., [89–91]), and halotolerant yeasts (e.g., [21,82,92–94]) described: the activity of Na+/H+ antiporter and the presence of a primary respiration-driven Na+ pump. Short-time ion imaging experiments revealed that S. salinarum was not able to adapt to salinities higher than 16% in a time span of 5 to 10 min. It might be possible, that the observed Na+ exclusion is not achieved by increased activity of certain transport mechanisms, but rather by an increased number of transporters, which requires time-consuming de novo synthesis. An overexpression of genes encoding for a Na+ extrusion pump triggered by high NaCl concentrations was, for instance, found in the salt-loving yeast D. hansenii [21]. However, this hypothesis should be proofed, e.g., by a comparative transcriptome analysis of S. salinarum at low and high external salt concentration.
S. salinarum, shifted to higher salinities (21%), showed a significant increase in Na+ concentration in the cytoplasm (Fig 5A). Because S. salinarum does not tolerate external salinities higher than 21%–23%, this increase of intracellular Na+ may be explained by the failure to actively exclude the high concentration of salt ions. Also, the strong increase of osmolytes (Fig 2) (S5 Fig) at this salinity may be a result of a penultimate attempt to counterbalance the elevated influx of salt ions. Instead, S. salinarum seems to accumulate the excess of Na+ in vacuole-like structures (Fig 5A, indicated by white arrows) to maintain a chemically balanced cytoplasm. These structures may be a mechanism to sequestrate intracellular Na+. As a fact, the Na+/H+ exchanger (NHX1), which is known from yeast [95] and higher plants [96], transports Na+ into vacuolar and/or prevacuolar compartments. In plants, the vacuolar sequestration of Na+ not only lowers Na+ concentrations in the cytoplasm but also contributes to osmotic adjustment to maintain water uptake from saline solutions [97]. Besides vacuoles, even plastids and mitochondria accumulate Na+ and thereby additionally contribute to the overall subcellular compartmentalization of Na+. Furthermore, the overexpression of specific Na+/H+ transporters localized, e.g., in tonoplast membranes and the increased expression of vacuolar H+/ATPase components due to osmotic stress were reported in various plants and are known to substantially enhance plant salt tolerance [96,98–102].
K+ concentrations did not change with increasing salinities, neither in the cytoplasm nor in the vacuoles (Fig 5B). This, together with steady intracellular Na+ concentrations between 4%–17% external salinity, perfectly fits the assumption that S. salinarum is a salt-out strategist (using compatible solutes) to balance its osmotic potential. Salt-out strategists need to maintain a stable high intracellular K+/Na+ ratio for several physiological functions like osmotic regulation, protein synthesis, enzyme activation or the maintenance of the plasma membrane potential [103]. Unfortunately, the ion imaging approach only informs about relative internal ion concentration ratios between different external salinities. A focused analysis of ion homeostasis in the transcriptome of S. salinarum will provide more information.
Species implementing the salt-in strategy are frequently characterized by an excess of acidic amino acids such as glutamate and aspartate, a low content of alkaline amino acids such as lysine and arginine, and a low content of hydrophobic amino acid residues [104]. Acting in concert, these proteome modifications help to prevent aggregation and to stabilize proteins, and also to maintain a hydration shell and the solubility of proteins [14,105,106]. Our results indicate that this peculiarity is not shared by the here investigated ciliate MDH and ICDH proteins. As judged by the calculated isoelectric point (pI) values (MDH = 6.08; ICDH = 8.49), the proteins from S. salinarum are neutral or slightly positively charged at physiological pH. Halotolerant eukaryotes, e.g., Tamarix tetragyna, have MDH with decreased activity at increasing NaCl concentration [107]. As S. salinarum, these organisms use organic compatible solutes that allow flexibility with respect to the range of salt concentrations tolerated and do not require a high degree of adaptation of the intracellular enzymes. Biochemically, these enzymes do not greatly differ from those of nonhalophilic prokaryotes [108]. Taken together, our data clearly show that S. salinarum MDH and ICDH do not require high salt concentrations for their activity; contrarily, they are much less active at a NaCl concentration corresponding to optimal growth salinity. Compatible with our findings of molar concentrations of the compatible solute GB in S. salinarum cells, substrate affinity and activity of MDH and ICDH are unaffected or only mildly affected by 2.5 M GB.
This is, to our knowledge, the first experimental study on haloadaptation strategies of a halophilic heterotrophic protist, for which three approaches were established to investigate haloadaptation strategies, previously known only from archaea and bacteria. Our results suggest that the heterotroph ciliate S. salinarum is a salt-out strategist. This is based on the direct measurement of two compatible solutes, namely GB and Ect, the active exclusion of Na+ and K+ ions out of the internal cell environment and a reduced protein activity of two key enzymes at high salt conditions in S. salinarum. This study is pioneering the route towards a better understanding of haloadaptations in heterotrophic protists and lays the foundation for future research and scientific studies in this field.
The halophile ciliate S. salinarum [37] was isolated from a solar saltern pond in Ses Salines, Mallorca, Spain (Salinas d`Es Trenc, N 39° 21’ 16.11” E 3° 0’ 42.16”) with a salinity of 12%. Single ciliate cells were handpicked in a 5 μl volume and transferred into 2 ml of sterilized salt medium (artificial seawater, ASW; Instant Ocean, Aquarium Systems, Cleveland, Ohio; for chemical composition see [109]) to establish initial cell cultures. This way, a proportion of indigenous bacteria, required by the ciliates as food source, were also transferred into the initial culture. When ciliate cell density reached about 100 cells/ml, the initial culture was diluted with sterile salt medium to a volume of 15 ml. Subsequent cultures were established by dividing the initial culture, therefore always transferring S. salinarum cells as well as indigenous food bacteria into the new cultures. Ciliate cells were cultivated in different salt concentrations ranging from 3.8%–21% at room temperature (RT) with a 12 h light-dark cycle. All cultures were amended with one sterile wheat grain per 15 ml culture to support growth of indigenous bacteria. Further information about the indigenous food bacteria is supplied in S1 Table.
1H-NMR spectroscopy was applied to detect and identify possible compatible solutes in S. salinarum. For this, cells were concentrated by centrifugation in a pear-shaped ASTM centrifuge tube (Lenz Laborglas GmbH & Co. KG, Wertheim, Germany) for 5 min at 400 g (Roto Silenta III, Hettich, Germany), followed by removal of the supernatant. Cells were checked for positive vitality under a stereo microscope (SZ-PT, Olympus, Tokyo, Japan) and then starved in sterile ASW medium for three d to reduce the total number of food bacteria. After starvation, ciliate abundance was determined by cell counting and samples were filtered immediately onto Isopore membranes (25 mm; 5 μm; Millipore, Schwalbach am Taunus, Germany). Cells were broken by shear forces in a Concentrator 5301 (Eppendorf AG, Hamburg, Germany) at maximum speed at 60 °C for 4 h. The samples were resuspended in 500 μl D2O, vortexed, and centrifuged for 20 min at 13.000 rpm (Hermle Z 233 MK, Hermle-Labortechnik, Wehingen, Germany). The supernatant was transferred into a new tube, the filter membrane was suspended with 400 μl D2O, and centrifuged again for 10 min at 13.000 rpm. Supernatants were mixed and centrifuged an additional time for 10 min. Finally, 600 μl sample was transferred to an NMR tube and mixed with TMSP and phosphate buffer (final concentration 6 mM K2HPO4; 4 mM KH2PO4). TMSP was added at a concentration of 500 mM as an internal chemical shift reference and for the quantification of organic osmolytes [110]. Measurements were done for six culture replicates growing in 5, 7, 9, 11, 13, 15, 17, 19, and 21% salinity.
To avoid losing any cellular substances by the extraction protocol described above, an ethanol extraction following the protocol of [64] but with minor modifications was also performed. Briefly, after filtration of cultures, cells were washed with ASW, resuspended in 20 ml of 80% (v/v) ethanol, vortexed for 5 min at RT, incubated for 2 h to extract the intracellular solutes and centrifuged again for 10 min at 13.000 rpm. The supernatant was retained for determination of compatible solutes. Cells were re-extracted as described above with 10 ml 80% ethanol. Cell extracts were combined and evaporated to dryness at 60 °C. Sample preparation for 1H-NMR spectroscopy followed the protocol described above.
One-dimensional 1H-NMR spectra were recorded on a 400-MHz Bruker Advance III spectrometer (Bruker BioSpin GmbH, Rheinstetten, Germany), operated using a 30° pulse, a spectral width of 8223.56 Hz, acquisition time of 3.98 s, and a repetition delay of 1 s. Analyses of 1D-spectra were performed with the software ACD/NMR Processor Academic Edition (version 12.01; Advanced Chemistry Development, Inc., Toronto, Canada). Solutes were identified by comparing experimental spectra with reference spectra of known compatible solutes [78]. The chemical shifts of the peaks used for the integration and quantification were 3.27 ppm −N+(CH3)3 for GB and 2.25 ppm −CH3 for Ect. At least six culture replicates per salinity were measured and results are given as the mean ± SD/cell volume. Quantification of compatible solutes was done by integrating the peak areas relative to the internal standard TMSP. The calculation of the cell volume of S. salinarum was based on generic SEM pictures generated as described by [111]. Width, length, and height of different individual cells were measured and used to calculate average values. The cell shape of S. salinarum was estimated according to [81] with the following formula:
Vcylinder=π*(1/2averagewidth)2*averagelength
V2cones=2*(1/3*π*(1/2averagewidth)2*averagelength)
The measured amount of intracellular compatible solute per cell was set in relation to the calculated cell volume (average cell volume: 8.01 pl) to estimate the intracellular concentration of the compatible solutes.
Statistical analyses were performed in R Studio (version 3.3.1, http://r-project.org) using the basic package. First, the different data sets were tested for normal distribution using the Shapiro-Wilk test. Second, the Bartlett test for homogeneity of variance was used to test if the variances of the different data sets have been homogenous or not. Because the data sets have been normally distributed and have shown unequal variances, significant differences for the intracellular concentration of GB and Ect in different salinities were calculated by using Welch’s t test. A correlation between the extracellular NaCl concentration and the intracellular concentration of compatible solutes was achieved by using a Pearson product-moment correlation.
To ensure that the starvation step was successful and that possibly remaining food bacteria did not disturb 1H-NMR measurements, three control experiments were conducted. SEM of filter-fixed cells prepared for NMR analyses was done to screen for potential bacterial contaminations. SEM followed the protocol of [111]. DAPI (4ʹ,6-Diamidin-2-phenylindol; Sigma-Aldrich, Taufkirchen, Germany) staining was conducted to visualize remaining ingested bacteria in the starved ciliates as described in [111]. 1H-NMR spectra of the flow-through resulting from the filtration step during 1H-NMR sample preparation were generated to assess whether the remaining bacteria in the cultivation medium show the same osmolyte NMR peaks as S. salinarum.
To investigate whether S. salinarum is able to accumulate compatible solutes from the external medium, cultures, grown at 9% salinity, were incubated with GB and Ect (final concentration: 1 mM; Sigma-Aldrich), respectively. Additionally, 1 mM Ch, a precursor of GB, was added to the medium to reveal a potential GB synthesis pathway in S. salinarum. During the growth experiments, cell abundances were counted in triplicates after 3, 6, and 9 d. Pure cultures of S. salinarum served as control.
In general, dynamic metabolic fluxes can be examined by the combination of NMR spectroscopy with the introduction of 13C-labeled substrates, such as 13C2-Ch. As the substrate is metabolized, the 13C-label is transferred downstream to metabolic products. Studies and reviews on the principles and applications of dynamic studies with NMR spectroscopy are readily available [112–114]. The 13C-label turnover can be detected by direct 13C-NMR spectroscopy or, as it was done in our study, indirectly by the coupling of the 13C-nucleus with protons as detected by 1H-NMR. The advantage of 1H-NMR spectroscopy is that it detects both the protons attached to 12C-nuclei as well as to 13C-nuclei. To assess whether S. salinarum can synthesize GB by the conversion of Ch, cell cultures with a salinity of 9% were starved for three d, salt shocked by increasing the salinity up to 13% and incubated with 13C2-Ch for another three d at RT (final concentration: 1 mM; Sigma-Aldrich). Afterwards, the 13C-label turnover was detected by 1H-NMR. The sample preparation and the peak identification followed the protocol described above. Measurements were applied to three culture replicates.
Ion imaging was performed with Na+- and K+-specific fluorescence markers to investigate whether S. salinarum uses the salt-in strategy to counterbalance osmotic stress. Therefore, cells were concentrated by centrifugation in a pear-shaped ASTM centrifuge tube (Lenz Laborgals GmbH & Co.) for 5 min at 400 g (Roto Silenta III). For Na+ imaging, concentrated cells were resuspended in CoroNa Green loading buffer (DMSO mixed with CoroNa Green, AM Sodium Indicator, Molecular Probes, final concentration: 0.5 μM) and incubated in the dark at 37 °C for 10 min. Incubation time was determined by preliminary experiments with incubation times at 10, 15, 20, 25, 30, 35, 40, and 45 min. Because no significant differences in relative fluorescence intensity at the different time points were observed, the minimum incubation time of 10 min was used for further experiments. Cells were washed three times and resuspended in ASW to remove excess fluorescence dyes. For K+ imaging, cells were resuspended in Asante Potassium loading buffer (DMSO mixed with Asante Potassium, TEFLABS, final concentration: 0.5 μM) and incubated in the dark at RT for 30 min. Stained cells were attached to poly-L-lysine-coated glass slides and visualized with a Carl Zeiss Axiostar plus epifluorescence microscope equipped with a specific filter set (Abs/Em for CoroNa Green/Na+ complex: 492/516 nm; Abs/Em for Asante Potassium/K+ complex: 488 to 517/540 nm). Images were taken with a cooled black and white high resolution CCDMicrocam (Intas, Germany) and the QImager software. During Na+ imaging and K+ imaging, exposure times for each time frame were 500 ms and 700 ms, respectively, the camera settings for all microphotographs were identical. For Na+ imaging, pictures were taken at 4, 9, 13, 17, and 21% salinity. Because no increase in relative fluorescence intensity was observed during K+ imaging, pictures were only taken at 4, 10, and 21% salinity.
Analysis of fluorescence intensity was done using the open source software FIJI [115]. Average fluorescence intensity (F) of individual cells was determined by defining three areas of the cell by intensity thresholds: background (IB), cytoplasm (IC), and vacuoles (IV), where I was pixel intensity (IB < IC < IV) and IB the mean value of four different background regions. The CRF was calculated as
CRFc=IC−(ACxIB),
where AC was the cytoplasm area. CRF was normalized to the measured area as
CRFnorm=CRFC/AC.
Measurements were carried out for at least 10 cells per salinity.
Statistical analyses were performed in R Studio (version 3.3.1, http://r-project.org) as described above. Because the data sets have been normally distributed and homogeneity of variance existed, significant differences for the different salinities were calculated by ANOVA. Because ANOVA only informs about overall differences between the tested groups, a post-hoc Tukey-honest significant difference (HSD) test was run to identify which specific groups differed.
To test if the observed increase of intracellular Na+ concentration with increasing external salinity of long-term adapted S. salinarum cells is a result of passive ion-influx or rather a product of active regulated uptake (i.e., presence of specific membrane transporters), short-term experiments were conducted. Therefore, a customized poly-L-lysine-coated microscopic slide was used to enable a quick exchange of the medium by gravity flow. Single S. salinarum cells adapted to a salinity of 4% were loaded with CoroNa Green fluorescence indicator as described above and attached to the customized slide. During the experiment the salinity was stepwise increased from 4 to 6, 8, 10, 12, 14, 16, and 21%. After each increase in salinity the investigated cell was allowed to rest for 5 min. Then, fluorescence intensity was recorded using epifluorescence microscopy and analysed as outlined above. Measurements were applied to three replicate cells.
To experimentally investigate the properties of S. salinarum proteins, the two well-conserved proteins MDH and isocitrate ICDH were tested for their activity under different salt and osmolyte conditions and their protein charge in comparison to respective proteins of E. coli. Sequence information for both proteins were obtained by genome sequencing of S. salinarums macronuclear DNA.
|
10.1371/journal.pgen.1003550 | Deregulation of the Protocadherin Gene FAT1 Alters Muscle Shapes: Implications for the Pathogenesis of Facioscapulohumeral Dystrophy | Generation of skeletal muscles with forms adapted to their function is essential for normal movement. Muscle shape is patterned by the coordinated polarity of collectively migrating myoblasts. Constitutive inactivation of the protocadherin gene Fat1 uncoupled individual myoblast polarity within chains, altering the shape of selective groups of muscles in the shoulder and face. These shape abnormalities were followed by early onset regionalised muscle defects in adult Fat1-deficient mice. Tissue-specific ablation of Fat1 driven by Pax3-cre reproduced muscle shape defects in limb but not face muscles, indicating a cell-autonomous contribution of Fat1 in migrating muscle precursors. Strikingly, the topography of muscle abnormalities caused by Fat1 loss-of-function resembles that of human patients with facioscapulohumeral dystrophy (FSHD). FAT1 lies near the critical locus involved in causing FSHD, and Fat1 mutant mice also show retinal vasculopathy, mimicking another symptom of FSHD, and showed abnormal inner ear patterning, predictive of deafness, reminiscent of another burden of FSHD. Muscle-specific reduction of FAT1 expression and promoter silencing was observed in foetal FSHD1 cases. CGH array-based studies identified deletion polymorphisms within a putative regulatory enhancer of FAT1, predictive of tissue-specific depletion of FAT1 expression, which preferentially segregate with FSHD. Our study identifies FAT1 as a critical determinant of muscle form, misregulation of which associates with FSHD.
| Facioscapulohumeral muscular dystrophy (FSHD) is a hereditary human myopathy affecting groups of skeletal muscles in the face and shoulders. Despite recent advances on the molecular cascade initiated by its main genetic cause, with identification of DUX4 as the main pathogenic agent, how this leads to the specific clinical picture is still poorly understood. Here, we investigated the role of the FAT1 protocadherin gene, located near the FSHD locus, which was repressed by DUX4 in human muscle cells. Disruption of the mouse Fat1 gene causes muscular and non-muscular phenotypes highly reminiscent of FSHD symptoms. We show that Fat1 is required in migrating muscle precursors, and that the altered muscle shapes caused by Fat1 mutations are predictive of early onset defects in muscle integrity in adult mutants, with a topography matching the map of muscles affected in FSHD. In humans, we observed FAT1 lowering in muscle but not brain of foetal cases with canonical FSHD1, and identified deletions of conserved elements in the FAT1 locus predictive of changes in FAT1 expression, that were enriched among FSHD patients. Thus, deregulating Fat1 in FSHD-related tissues provides a unique means to mimic FSHD symptoms in mice and learn about pathogenesis of this complex disease.
| Developmental genetics has provided considerable insight into the regulatory networks controlling overall skeletal muscle development. Perturbation of these common mechanisms is associated with congenital abnormalities of the muscle lineage as well as with later-onset muscle pathologies [1]. In contrast, less is known about the mechanisms of functional diversification within the muscle lineage. Such diversification may be either metabolic - fast versus slow fibres, for example - or morphological, such as the position and shape of individual muscles. Genes controlling diversification too are likely to be of clinical significance [2]–[4], since several human muscular dystrophies do not affect all muscles evenly, but specifically target regionalized groups [5]. This is true for limb girdle muscular dystrophy (LGMD), oculopharyngeal muscular dystrophy (OPMD), myotonic dystrophies with oculomotor involvement, distal myopathies, scapuloperoneal dystrophy, and facioscapulohumeral dystrophy (FSHD) [5]–[6]. In no case, however, is the rationale for this geographic specificity currently understood.
One characteristic example of focal myopathies is FSHD, which affects subsets of muscles in the facial and shoulder areas [6]. The main form of FSHD - FSHD1 - is an autosomal dominant disorder associated with the contraction of an array of 3.3 Kb macrosatellite repeats (D4Z4), located at the subtelomeric 4q35 locus [6]. The mechanism by which the D4Z4 contraction triggers the disease represents one of the most enigmatic conundrums for human geneticists and remains incompletely understood. The D4Z4 array has been suggested to act as an insulator between telomeres and subtelomeric genes [7]–[8], such that its contraction might result in regulatory changes in neighbouring genes that could in turn alter muscle physiology [6], [9]–[11]. Despite intense focus on deregulated 4q35 genes, including one of the close neighbours, FRG1 [12], and despite numerous large-scale investigations aimed at uncovering additional relevant candidates, none of the genes reported accounts for all aspects of FSHD, and additional players are still actively sought [6], [9], [13]. An emerging model is that the pathogenic effect of D4Z4 contraction in FSHD1 is mediated in part by DUX4, a retrogene present within D4Z4 repeats themselves encoding a homeobox containing transcription factor that is normally silent in muscle [14]–[15]. In FSHD1 patients, the contraction of the D4Z4 repeat array leads to a change in chromatin structure that facilitates DUX4 expression [16]. Furthermore, the pathogenicity of the D4Z4 contraction requires polymorphisms distal to the last D4Z4 repeat, that create a polyadenylation signal and thereby stabilize DUX4 mRNA [17]. This stabilized RNA thus leads to increased expression levels in FSHD muscles of a pathogenic isoform of DUX4, which activity is thought to be toxic for muscles through transcriptional activation of various target genes including Pitx1 and p53 [18]–[21]. Another less frequent form of FSHD, clinically identical to FSHD1, is observed in absence of D4Z4 contraction. These cases, referred to as contraction-independent FSHD, include cases called FSHD2, that were shown to exhibit hypomethylated D4Z4 repeats, recently shown to be caused by mutations in the SMCHD1 gene [22]. FSHD2 is caused by the combination of such SMCHD1 mutations with a DUX4 permissive (polyA) context, and also leads to DUX4 overexpression [22]. While FSHD2 cases represent so far the majority of contraction-independent cases, rare cases of contraction-independent FSHD with typical symptoms may also occur without hypomethylation, and be caused by yet unidentified pathogenic contexts. Neither the specificity of SMCHD1 or of DUX4 expression nor of its target genes identified so far [18]–[21], [23]–[24], provide sufficient account for the specificity of the muscle map and the non-muscular symptoms that characterize FSHD.
The regional specificity in the map of muscles affected in FSHD suggests that the causal abnormality interferes with a muscle subtype-specific developmental process. A gene involved in functional diversification during muscle development would thus provide a logical candidate to fill this gap. We focused on the cell adhesion molecule FAT1 because Fat-like protocadherins are known modulators of the planar cell polarity (PCP) pathway [25]–[27], a genetic cascade involved in coordinating tissue polarity, morphogenetic movements, and polarized cell flow [28]–[30]. Fat1 has been reported to be expressed in developing muscles and tendons [31] and to be regulated by muscle developmental genes such as Pax3, Lbx1, or Met [32]–[34]. Thus, FAT1 may control muscle shape through PCP-like mechanisms analogous to those involved in polarized migration of vascular endothelial smooth muscle cells [35].
Here, we report the unexpected finding that Fat1-deficient mice reproduce the highly selective muscular and non-muscular aspects of the clinical picture of FSHD. We show that Fat1 is required during development to shape specific groups of shoulder and facial muscles by modulating the polarity of myoblast migration. While constitutive inactivation of Fat1 leads to neonatal lethality due to defects in kidney development [36], Fat1 hypomorphic mice exhibit defects of muscle integrity with a topography prefiguring the map of muscles affected in FSHD. Furthermore, conditional mutagenesis suggests that a cell-autonomous function of Fat1 in migrating muscle cells may account for a significant part of its muscle shaping function. The human FAT1 gene is located only 3.6 Mb from the critical FSHD genomic region at 4q35, and emerges as a potential transcriptional target of DUX4 or p53 [18], [37]–[38]. We present evidence of altered FAT1 levels in some foetal FSHD1 cases, in muscle, but not brain, accompanied with epigenetic modifications characteristic of silenced chromatin. Finally, we identified genetic variants deleting variable lengths of a putative cis-regulatory enhancer in the FAT1 locus, which segregate with FSHD. Thus, either in presence or absence of D4Z4 contractions, mechanisms leading to tissue-specific deregulation of FAT1 expression are associated with FSHD and may contribute to causing regional-specific muscle shape abnormalities that prefigure muscle degeneration in the adult.
In search of mechanisms that control muscle position and form, we studied Fat1 expression at stages of muscle morphogenesis. We chose first to study a muscle with a characteristic fan-shaped form, the subcutaneous muscle cutaneous maximus (CM). During embryogenesis, following delamination from the dermomyotomal lip at forelimb levels, CM precursors, identified through their specific expression of GDNF, reach the base of the limb, turn, and spread under the skin in a radial manner [39]–[40] (Figure 1A). This migration pattern reflects collective and polarized cell migration, visible owing to expression of the MLC3F2E reporter line or of the muscle fate marker MyoD, through the formation of chains of myoblasts aligned in radial directions (Figure 1B and 1E top right panel). At the stages of CM migration, whole mount X-gal staining in embryos carrying a LacZ reporter gene-trap insertion in the mouse Fat1 gene revealed a hot-spot of Fat1 expression highlighting the migration area (Figure 1C, Figure S1). We found that CM myoblasts express Fat1 RNA and appear to be positioned in a subcutaneous layer which itself expresses Fat1 RNA, this surrounding subcutaneous tissue displaying a rostrocaudal gradient of intensity, with highest intensity caudal to the extremity of the CM (Figure 1C; D). Thus, CM myoblasts express Fat1 and appear to migrate along an increasing gradient of Fat1 expression.
We therefore asked whether Fat1 was required for CM location and/or form. We first took advantage of a mouse model carrying a gene-trap insertion in the mouse Fat1 gene [41]–[42] (allele referred to as Fat1LacZ). Initial differentiation along the muscle lineage was unaffected in Fat1LacZ/LacZ embryos since CM myoblasts retained expression of broadly-expressed markers such as MyoD (n = 6), and markers of subsets of myoblasts (such as Six1 (n = 2), gdnf (n = 2), and Lbx1 (n = 2); data not shown). This allowed us to use MyoD expression to monitor precursor migration in Fat1 mutants. In E12.5 Fat1LacZ/LacZ embryos, we observed 1) an aberrant morphology of the CM muscle, reduced in size, and with ill-defined anterior limits (Figure 1E), 2) a dispersion of migrating myoblasts not only within the CM but also in ectopic areas traditionally devoid of muscle cells. In the CM, higher magnification observations revealed that migration myoblasts failed to show a preferential alignment of their nuclei into migratory chains (Figure 1E–H). This phenotype was associated with morphological changes in individual myoblasts, such as the loss of long cytoplasmic protrusions extending from the leading edge and rounded morphology of some nuclei within the chains (Figure 1G, H). In further support of a role for Fat1 in migration polarity, numerous clusters of ectopic myoblasts or disoriented single myoblasts were found in the shoulder region of E12.5 mutants, either in ectopic places, or within additional shoulder muscles such as the spinotrapezius muscle (Figure 1E orange arrowheads in orange dotted area; Figure S2, red arrows).
Further genetic evidence of such a function of FAT1 in control of muscle shape was obtained with another targeted allele of the Fat1 locus, which we engineered by flanking two exons, 24 and 25, the latter containing the transmembrane domain, with LoxP sites (Figure S3A, targeted allele referred to as Fat1Fln). Crossing of mice carrying the conditional Fat1Fln allele with a ubiquitous CRE-expressing mouse line produced, by germline excision of the floxed exons, a constitutively recombined allele, Fat1ΔTM, which encodes FAT1 protein isoforms lacking the corresponding transmembrane domain (Figure 2A,B). Analysis of myogenic differentiation by in situ hybridization with a myoD probe indicated that Fat1ΔTM/ΔTM embryos exhibited phenotypes identical to those seen in Fat1LacZ/LacZ embryos (data not shown). This new allele also allowed studying later steps of muscle differentiation by crossing Fat1ΔTM mice with a transgenic line in which nls-LacZ reporter activity is driven by an enhancer from the mlc3f gene (MLC3F-2E) [43]. Expression of this transgene (MLC3F-2E:LacZ) is detected slightly later than myoD expression as it reflects differentiation in myocytes and sarcomere assembly [43], hence it allows visualising muscle shapes, but not migrating myoblasts. MLC3f-2E expression in Fat1ΔTM/ΔTM embryos revealed again the altered morphology of the CM muscle, with missoriented chains of myocytes in the ventral/pectoral half of the CM and shoulder belt muscles (Figure 2D, and Figure S3B). Furthermore, Fat1ΔTM/ΔTM embryos were found to exhibit an extra muscle ectopically located in the shoulder area (Figure 2D). Finally, we also visualized multinucleated myofibres owing to the nuclear β-galactosidase staining at late gestation stages, and confirmed the persistence of misoriented myofibers in the mature CM muscle of Fat1ΔTM/ΔTM E18.5/P0 embryos (Figure 2D). Taken together, our data show that Fat1 is required to control the shape and position of subsets of migratory muscles in the developing embryo, by controlling coordinated polarity of collectively migrating myoblasts.
We next wished to extend our description of the map of Fat1-dependent muscles by exploring the phenotypes exhibited by Fat1ΔTM/ΔTM embryos carrying the MLC3F-2E transgene at later developmental stages (E14.5 and E15.5), when migration has been completed and muscle shapes are determined. In the scapulohumeral area of all Fat1ΔTM/ΔTM;MLC3F-2E embryos examined, we consistently observed an extra muscle in a stereotyped ectopic position, systematically attached between the spinodeltoid muscle and the triceps brachii muscles (Figure 3A,B). Just dorsal to the spinodeltoid, we found a subcutaneous portion of the spinotrapezius muscle (SpTS) to be drastically reduced in Fat1ΔTM/ΔTM;MLC3F-2E embryos (Figure 3A, orange arrows). Observation from a dorsal point of view reveals that midline junction of the CM muscle and of Rhomboid muscles (Rh) is delayed, so that a large gap is seen in the back of an E14.5 Fat1ΔTM/ΔTM embryo (Figure 3B, orange line). Numerous mispositionned myofibres create ectopic bridges between the acromiotrapezius and spinotrapezius muscles in Fat1ΔTM/ΔTM;MLC3F-2E embryos (Figure 3B; read arrows in top and middle picture). Analysis of muscles in the face at E14.5, E15.5, and at P0, reveals abnormalities in shape, myofibre orientation, and density in several subcutaneous muscles in the facial skin (Figure 3C, red arrows) that occupy positions reminiscent of the position of human muscles of facial expression. The flat structure of these subcutaneous muscles is analogous to that of the CM muscle, and the alterations observed in Fat1ΔTM/ΔTM neonates also include random orientation of multinucleated myofibres (Figure 3C). In contrast, deeper muscles such as the masseters display normal shape in Fat1ΔTM/ΔTM mutants (see Figure 3C and data not shown). Of notice, although muscle shape defects were found in stereotyped places, their severity was variable, and Fat1ΔTM/ΔTM embryos were frequently asymmetrically affected (Figure S4, see also Figure S12A). As previously observed in Fat1LacZ/LacZ mutants, examination of muscle development at E14.5 and E15.5 in Fat1ΔTM/ΔTM embryos confirmed that Fat1 loss of function selectively affects muscles of the facial and scapulohumeral ares, and that Fat1 is not required to shape other migratory muscles such as the diaphragm or hindlimb muscles, which were identical between wild type and Fat1ΔTM/ΔTM embryos (Figure S4 and data not shown). Overall, in addition to the abnormal shape of the cutaneous maximus muscle, we found that Fat1 was required to shape selective and stereotyped groups of muscles in the scapulohumeral interface, as well as subcutaneous muscles of the face.
We next asked what the consequences of these muscle shape abnormalities were at postnatal stages. Constitutive deletion of Fat1 was initially shown to lead to neonatal lethality most likely due to defects in kidney filtration [36], [42]. Likewise, constitutive deletion of the transmembrane domain (Fat1ΔTM/ΔTM mice) also leads to more than 50% lethality at birth, with only a small proportion of mutants surviving to adulthood (Figure S3C). We chose to examine adult Fat1LacZ/LacZ mutants, since the hypomorphic Fat1LacZ allele, which results from an insertion of a gene-trap construct in an intron, not deleting any functional domain, allows expression of variable amounts of residual Fat1 RNA and FAT1 protein in Fat1LacZ/LacZ mutants (Figure 2E, Figures S5, and S13). This hypomorphic allele, in the genetic background we used, allowed bypassing the neonatal lethality in Fat1LacZ/LacZ mutants, with more than half the mutant mice surviving after 3 months (Figure 4C), and enabled us to study the postnatal consequences of reduced Fat1 levels. The variable amounts of residual Fat1 correlates with the variability in the severity of phenotypes and in the age of death of Fat1LacZ/LacZ mice. A fraction of these adult phenotypes, in particular the lethality, is likely to result from systemic consequences of kidney phenotype. Indeed, analysis of kidney morphology in the subset of Fat1LacZ/LacZ mice that exhibited severe weight loss revealed features characteristic of polycystic kidneys, such as cysts formed of enlarged tubules in the cortical renal area (data not shown). Therefore, to score with an objective criterion the progression through adult phenotype stages, body weight was measured for each individual and compared to its own maximal weight [44]. We arbitrarily set the moment a Fat1LacZ/LacZ mutant mouse has lost 10% of its weight as the visible onset of symptoms associated with kidney malfunction or with other phenotypes likely to have systemic consequences. Mutant mice showing more than 10% loss at the stage of analysis were defined as “symptomatic” (related to generalized symptoms, and not to muscles only), and the degree of severity was recorded as percentage weight loss, while Fat1LacZ/LacZ mutant mice that did not exhibit any weight loss yet were defined as presymptomatic. Although this threshold of 10% weight loss was defined arbitrarily, and even though we cannot exclude that kidney phenotypes also have systemic consequences earlier than this limit, it is difficult, during symptomatic phase, to attribute a primary cause to the symptoms observed. We therefore focused on the presymptomatic phase for most of our studies of adult muscle, and also chose to exclude from our adult studies mutant mice with an impaired growth curve. While Fat1LacZ/LacZ mice at symptomatic stages (with 20–30% body weight loss) displayed generalized muscle mass reduction (Figures S6B–C, presymptomatic mutant mice showed scapular winging, whereas lumbar posture and hindlimb function appeared unaffected (Figure 4A). Postural abnormalities affecting the shoulder area, indicating weakness of the muscles involved in scapular movements, can be seen when presymptomatic mice move on a cage grid, especially in situations in which they challenge the shoulder girdle muscles by transferring bodyweight rostrally on their forelimbs. These postural abnormalities were accompanied by functional motor defects evidenced in rotarod assays at presymptomatic stages (Figure 4E). Early symptomatic mice (around the 10% threshold) also showed kyphosis, a curvature of the spine known as a hallmark of muscle wasting in the shoulder girdle (Figure 4D, F), without displaying skeletal abnormalities (Figure 4B, X-ray). Similar observations were made in the small proportion of Fat1ΔTM/ΔTM mice that survived to adult stages.
We next investigated the pathological basis for the selective postural abnormality of the scapulae at presymptomatic stages. Dissection of individual muscles in presymptomatic Fat1LacZ/LacZ mice revealed a significant mass reduction for both rhomboid muscles when compared to controls (Figure 4D). As expected from the embryonic defect, a severe reduction in thickness of the CM muscle was also observed, although its subcutaneous location made accurate dissection and therefore mass measurement unfeasible. Defects in myofibre orientation similar to those observed at late embryonic stages were confirmed in CM (Figure S6D and data not shown) and in rhomboid muscles (Figure 4G) at all stages examined. In contrast, masses of muscles with unaltered shape when examined during development (i.e hindlimb muscles such as gastrocnemius or soleus) were also not significantly reduced at presymptomatic stages (Figure 4D, Figure S6B, S7). This argues that persistence in mature muscles of misoriented myofibres resulting from fusion of depolarized myoblasts contributes to the shoulder muscle phenotype in presymptomatic mice, although it does not rule out an additional direct function of Fat1 in muscle, whose loss may also cause muscle degeneration. Lastly, another consequence of developmental dysgenesis that is likely to contribute to focal muscle wasting is the persistence of ectopic muscles (Figure S7). Such ectopic muscles were found to share tendon attachment sites with existing muscles (typically two ipsilateral muscles) including shoulder belt muscles (trapezius, LD, pectoral muscles), and the humeral muscle triceps brachii (Figure S7). This association correlated with a unilateral reduction of the corresponding muscle mass, reduction that nevertheless did not result significant until early symptomatic stages (Figure 4D and data not shown).
The phenotypes resulting from developmental dysgenesis were not restricted to muscle shape and mass. Histological analyses revealed that a significant reduction in fibre diameter was detectable already at early symptomatic stages in those muscles in which we detected developmental defects, including the CM, Rhomboids (Figure 4G, superior and profundis), and Trapezius muscle (Figure 5C, pooled analysis). This was also true for Fat1ΔTM/ΔTM mice analysed at presymptomatic stages (Figure S8). In contrast, at presymptomatic stages, analysis of myofiber diameters in muscles whose shape was unaffected at developmental stages (such as gastrocnemius or soleus, and also diaphragm) revealed no significant abnormality as compared to control mice (Figure 4D, Figure S6B, and data not shown). In affected muscles (trapezius, rhomboid, Pectoralis Major, LD, and CM), we observed a range of additional abnormalities including inflammatory infiltrations between myofibres, most frequently perivascular, in both presymptomatic Fat1LacZ/LacZ and Fat1ΔTM/ΔTM mice (Figure S6D and Figure S7). Fibre necrosis was also observed at more advanced symptomatic stages (beyond 10% weight loss, Figure S7L and data not shown), but as mentioned earlier, it is impossible to distinguish whether any abnormality at symptomatic stage is strictly related to muscle defects, or reflects systemic consequences of unrelated phenotypes. Finally, observation of myofibre structure in affected muscles (trapezius, rhomboid, Pectoralis Major, LD, and CM) revealed progressive disruption of higher level organization, with appearance at presymptomatic stages of multiple faults disrupting the regular alignment of sarcomeric structures (Figure 5A, D), and the detachment of the sarcolemma from the contractile apparatus (Figure 5D). Overall, alterations of muscle integrity at pre-symptomatic stages were only detected in those muscles in which we reported fully penetrant myoblast or myofibre orientation defects (CM, Rhomboids, and Tapezius). Analysis of neuromuscular junctions in affected shoulder muscles also revealed a proportion of junctions showing fragmentation (Figure 5B), denervation, and atrophy (Figure S9). Such defects did not reflect a primary failure of NMJ innervations, as all neuromuscular junctions observed at early postnatal stages (P3) were indistinguishable from wild type (data not shown). Nevertheless, although the muscles that were spared during development and at presymptomatic stages (e.g gastrocnemius, soleus, masseters) were seen to harbour histological signs of muscle atrophy (evenly reduced myofiber diameter) at advanced symptomatic stages (Figure S6B), we did not observe muscle degeneration, inflammation, necrosis, or fragmentation of the contractile apparatus (data not shown). These results are consistent with the possibility that the developmental abnormalities of muscle shape constitute a topographic frame in which muscles might be predisposed to undergo early onset muscle wasting, prior to the appearance of systemic consequences of non-muscle phenotypes and the concomitant generalization of muscle wasting. These findings do not exclude however the possibility that Fat1 may play additional roles during muscle biology other than controlling shape during development.
We next asked if the function of Fat1 in shaping facioscapulohumeral muscles was exerted cell-autonomously in migrating muscle precursors. In order to perform tissue-specific ablation of Fat1 in muscles at a stage compatible with migration, we reasoned that transgenic lines in which CRE expression would reproduce that of genes of the muscle differentiation cascade, such as myoD or Myf5, would occur too late to have an impact on the migration itself. Therefore, to ablate Fat1 exons 24 and 25 in premigratory myoblasts, we took advantage of the Pax3-cre knock-in line [45] (Figure S10). Our conditional allele of Fat1 (Fat1Fln) initially includes the neo cassette that was used to engineer the mouse model. Although presence the neo cassette caused mild lowering of Fat1 expression levels (Figure S11), this only resulted in subtle, although statistically significant, morphological defects in Fat1Fln/Fln embryos/mice compared to controls (Figure 6 and Figure S12). This allowed using the Fat1Fln/Fln mutants for conditional studies with tissue-specific CRE lines, without requiring Flp/FRT recombination to further ablate the neo cassette. We therefore compared muscle development in Fat1Fln/Fln;Pax3cre/+ and Fat1Fln/Fln embryos, taking advantage of the MLC3F-2E transgene 1) to visualize the shape of every muscle and 2) to quantify the number of muscle cells dispersed in ectopic areas. We followed muscles belonging to Pax3-derived territories in the scapulohumeral area, where ablation of Fat1 leads to measurable phenotypes in Fat1ΔTM/ΔTM;MLC3F-2E+ embryos (Figure 6A). First, we found significantly higher numbers of dispersed myocytes in the forelimb of Fat1Fln/Fln;Pax3cre/+ embryos than in Fat1Fln/Fln embryos (Figure 6A, B). Second, an ectopic muscle similar to the one found in Fat1ΔTM/ΔTM embryos could be measured in Fat1Fln/Fln;Pax3cre/+ embryos, and its surface was significantly larger than in Fat1Fln/Fln embryos (Figure 6A, C). At later developmental stages, in addition to confirming the persistence and position of this ectopic muscle in Fat1Fln/Fln;Pax3cre/+ embryos, as in Fat1ΔTM/ΔTM; MLC3F-2E+ embryo. Furthermore we also detected a reduced density of myofibers in the CM muscle and in the subcutaneous part of the spinotrapezoid muscle (Figure S12). As the Pax3cre/+ line is a CRE knock-in, but also a knock-out of the endogenous Pax3 locus, the resulting loss of one copy of Pax3 may be in itself sufficient to enhance FAT1-dependent phenotypes. To rule this out, we have evaluated the effect of combining a Pax3cre/+ context to the recombined Fat1ΔTM allele, and found no enhanced phenotype in either Fat1ΔTM/+:Pax3cre/+ or Fat1ΔTM/ΔTM:Pax3cre/+ embryos compared to Fat1ΔTM/+ or Fat1ΔTM/ΔTM embryos, respectively (data not shown). Finally, Fat1Fln/Fln;Pax3cre/+ embryos did not display significantly more abnormalities in the subcutaneous facial muscles or in the spinotrapezius muscle than the mild phenotypes observed in Fat1Fln/Fln embryos (Figure S12), consistent with the fact that facial muscles do not belong to the Pax3-CRE lineage [46]. Furthermore, if ablation in facial neural crest cells, driven by Pax3-CRE activity, had been responsible for altering muscle shape, it would have done so as efficiently in facial muscles as in trunk muscles. The lack of enhancement of facial muscle phenotypes in Fat1Fln/Fln;Pax3cre/+ compared to Fat1Fln/Fln embryos thereby also excludes a contributing role of Fat1 expression in neural crest-derived cells. Thus ablating Fat1 in Pax3-derived cells is sufficient to partially reproduce the defects observed in scapulohumeral muscles of the constitutive Fat1 mutants, indicating that Fat1 is required cell-autonomously in migrating myoblasts to control the polarity of their migration.
As we asked whether in addition to the control of muscle migration, Fat1 may play additional roles in mature muscle, we noticed that in mouse, Fat1 is also expressed in differentiated muscle fibres after migration stages. This expression can be detected through the pattern of β-galactosidase expression in Fat1LacZ/+ embryos, and by in situ hybridization (Figure 7A). Furthermore antibodies against FAT1 C-terminal cytoplasmic tail detected a protein localized in stripes within muscle fibres (Figure 7B–D), on either side of alpha-actinin-positive sarcomere boundaries (so called Z-bands, Figure 7B). In adult mouse muscle, the stripes of FAT1 protein are closely juxtaposed with DHPR, a calcium channel present in transverse (t)-tubules [47] (Figure 7B). Such localization is consistent with Fat1 also playing a direct role in muscle biology, distinct from its early function in orienting myoblast polarity. Consistent with previous reports showing that cytoplasmic variants in FAT1 proteins exhibit distinct subcellular localisation [48], and that the cytoplasmic domain can translocate in the nucleus [49], another antibody directed against the cytoplasmic domain (FAT1-1465 antibody) also detected FAT1 protein in significant proportion of nuclei in adult mouse muscle fibres (data not shown). Western blot analyses indicated that a full length FAT1 protein is only detected in whole embryo extracts (at E12.5, Figure 2B) or in isolated brain tissue, but not in muscle tissue, where the most abundant bands detected with anti-FAT1-ICD antibodies were smaller molecular weight proteins (Figure S13), which production is spared by the genetic alterations in both Fat1LacZ/LacZ and Fat1ΔTM/ΔTM mutants (Figure 7C,D, Figure S5, S11, S13 and data not shown). While some of these smaller isoforms might be cleavage products of full length FAT1 [50]–[52], additional short isoforms are also consistent with gene products resulting from transcript initiation at alternative downstream promoters, as proposed by genome browsers (Ensembl, UCSC; Figure S5A, with EST-based genes referenced in NCBIM37 mouse genome and in GRCh37 human genome assemblies). Neither the gene trap insertion after the first exon (this study), nor the removal of the entire first exon (in the published knockout allele [36]), suppress such gene products. Deletion of the transmembrane domain in Fat1ΔTM/ΔTM mutants also allowed expression of protein products with unchanged size (Figure S13), although it nevertheless led to a more severe phenotype with drastic neonatal lethality (compare Figure S3C and Figure 4C). Quantitative RT-PCR confirmed the presence of significant amounts of Fat1 RNA containing the last exons (26 to 28) in Fat1ΔTM/ΔTM mutants, albeit at reduced levels when compared to wild types (Figure S11). Thus, in the case of all mutant alleles, the remaining smaller isoforms might still carry out Fat1 functions at least partially, resulting in hypomorphic phenotypes with variable severity. Consistently, in immunohistochemistry experiments on muscle sections, residual FAT1 staining is also observed in myofibres of Fat1ΔTM/ΔTM mutants and Fat1LacZ/LacZ mice, and staining intensity in Fat1LacZ/LacZ mice that survived to adulthood inversely correlated with phenotype severity at the level of individual myofibers (Figure 7C,D and data not shown). Presence of unchanged smaller FAT1 isoforms in muscles of Fat1ΔTM/ΔTM mutants precludes using this mouse line to investigate their function. However, it indicates that the phenotype of muscle migration is not the consequence of their deletion, but results from ablation (constitutive or driven by Pax3-cre) of the transmembrane domain in full length FAT1 proteins that are abundant at developmental stages (Figure 2B).
Strikingly, the topography of selective alterations in muscle shape that we observed during development in Fat1 mutant mice closely resembles the map of muscles affected in early phases of human FSHD. Muscle shape abnormalities such as those seen in facial subcutaneous muscles, in trapezius, or in rhomboid muscles are expected to result in lack of facial skin mobility and scapular winging, two symptoms that are frequently the first clinical manifestations of FSHD. The selective muscle weakness observed in presymptomatic Fat1 mutants in muscles belonging to the developmental map was also reminiscent of the early phase of FSHD. Even at the scale of EM observations, defects in myofibre structure, such as sarcolemma detachment (Figure 5D), included aspects similar to those reported in FSHD biopsies [53]. Finally, asymmetry of muscle symptoms is an important aspect of FSHD symptoms. Asymmetries in muscle shape abnormalities were observed not only in the robust phenotypes displayed by Fat1ΔTM/ΔTM embryos, but also in the very subtle phenotypes associated with by mild lowering of FAT1 expression in Fat1Fln/Fln embryos (Figure 6, Figure S12A). In this context, it was interesting to note that the human FAT1 gene is located at 4q35.2, 3.6 Mb proximal to the D4Z4 array whose contraction is associated with FSHD (Figure 8A). We therefore asked whether in addition to muscle phenotypes, Fat1-deficient mice may also share similarities with non muscular symptoms of FSHD. Besides muscular abnormalities, the phenotypic spectrum of FSHD patients also includes vision defects linked to vascular abnormalities [6], [54]–[55]. As previously reported, constitutive FAT1 loss-of-function causes abnormalities in eye development, with variable severity and penetrance [36]. The Fat1LacZ/LacZ mice surviving as adults carried milder phenotypes ranging from residual patterning defects (aniridia, small eye, Figure 8B) to perfectly shaped eyes and retina, in which analysis of vasculature with IB4 or PECAM staining revealed numerous areas with intraretinal telangiectasia, microvascular lesions, micro-aneurysms, and frequent retinal detachments (Figure 8C). Additional non-muscular symptoms associated with FSHD also include high frequency hearing loss, although the cause of these deficits remains underexplored. Fat1-deficiency was recently reported [56] to cause mild morphological defects in the inner ear, such as reduced cochlear elongation, and to exacerbate the appearance of ectopic sensory hair caused by loss of FAT4, another FAT-like protocadherin, reflecting their cooperation during in elongation and sensory hair cell patterning in the cochlea [26], [56]–[58]. Furthermore, owing to expression of the MLC3f-2E transgene during inner ear development [59], we observed shortening of the endolymphatic duct and endolymphatic sac in Fat1ΔTM/ΔTM embryos at E12.5 (7 affected sides out of 12), this shortening being frequently asymmetric (Figure 8D, E). These phenotypes are expected to influence audition. Thus, in addition to the similarity of muscle abnormalities, adult Fat1 mutant mice also show non-muscular defects reminiscent of clinical symptoms of FSHD. Nevertheless, the severity scale of these phenotypes includes phenotypes more dramatic than those seen in FSHD, and Fat1-deficiency also leads to phenotypes such as the previously reported kidney abnormalities, that have no equivalent in FSHD.
Considering the gene location and the provocative similarities between Fat1-deficiency in mouse and FSHD, we therefore asked whether alterations in Fat1 expression might be an essential step in the molecular mechanism leading to FSHD pathology in human. As in spite of the essential role of Fat1 in kidney development, FSHD is not known to be associated with kidney abnormalities, if a mechanism linking FSHD to Fat1 exists, it is expected to involve partial functional alterations only, such as tissue-specific deregulation of FAT1 during development. We thus first asked whether in addition to the previously reported gene expression changes [9]–[11], [60], any deregulation of FAT1 expression levels could be detected in the classical context of FSHD1, in which the pathology is due to the presence of a contracted D4Z4 array on a permissive/pathogenic DUX4-activating context (4qA haplotype) [17]. This possibility was reinforced by the finding that FAT1 appears to be downregulated by DUX4-fl, but not by DUX4-short in human myoblasts [18]. This result was further validated by qPCR, after lentiviral infection of human myoblasts with DUX4-fl as compared with GFP control (Figure S15D), indicating that DUX4 overexpression is capable of lowering FAT1 expression in cultured muscle cells. As our results in mice point to the crucial role of FAT1 deregulation during development, we aimed to analyse FAT1 expression in rare cases of biopsies from foetuses with a prenatal diagnosis of FSHD1, in spite of the fact that stages of myoblast migration were not accessible to experimentation in this context. Nevertheless, the observation that FAT1 protein is a component of differentiated muscle fibres, enriched in the t-tubule system, is consistent with additional later functions of FAT1 necessary for muscle integrity.
Possible alterations of FAT1 expression were therefore assessed in muscle biopsies of human FSHD1 cases at foetal stages through a series of independent approaches. Human FAT1 protein was detected by immunohistochemistry in human muscle biopsies from control foetuses of various stages with antibodies against FAT1 C-terminal cytoplasmic tail, with a striped pattern similar to that seen in mice (Figure 9A, Figure S15). We thus first studied FAT1 expression levels in tissues from an FSHD1 human foetus carrying a pathogenic 4qA allele harbouring 1.5 D4Z4 copies, expected from previous family history to lead to severe infantile FSHD (Figure S14). Immunocytochemistry with anti-FAT1 antibodies on sections from the quadriceps muscle revealed an overall decrease in FAT1 protein levels compared to quadriceps biospies from control foetuses (Figure 9A), with an irregularly stripped pattern of FAT1 in myofibres that otherwise show a normal distribution of other muscle proteins, such as DHPR. To assess this FAT1 lowering quantitatively, mRNA expression levels were then followed by qRT-PCR in muscle biopsies from 4 FSHD human foetuses carrying pathogenic 4qA alleles harbouring 1.5, 4.3, and 7 D4Z4 copies (referred to as F1, to F4, respectively; Figure S14A). In F1 foetus, FAT1 levels were reduced 5-fold in the deltoid (a muscle belonging to the FSHD map) and 3-fold in the quadriceps muscles (a muscle traditionally affected only at late stages in the human disease; Figure 9B). This was also confirmed by Western Blot with anti-FAT1-ICD antibodies (Figure S15A). Additional regulatory changes were detected (Figure S15B), such as an increased level of MURF1 or dysferlin RNAs, while RNA of other muscle components, such as DHPR or γ-Sarcoglycan, were unchanged, ruling out secondary effects of loss of muscle integrity at this stage or quality of the biopsy. In contrast, no significant difference in FAT1 mRNA levels could be observed in brain when comparing FSHD and control samples from the same foetuses (Figure 9B). Reduction of FAT1 mRNA levels, albeit to a lesser extent (25% reduction; Figure 9B), and aberrant protein localisation (Figure S15C) were observed in the quadriceps of a second FSHD foetus harbouring 4.3 D4Z4 repeats (F2), from an independent family with previous FSHD history (Figure S14). Finally, no significant quantitative changes were observed in muscle biopsies of twin FSHD foetuses with 7 D4Z4 repeats (Figure 9B), although accumulation of FAT1 protein could be observed in some myofibre nuclei (data not shown), a localization never observed in age matched control biopsies, but reminiscent of adult mouse muscles. In contrast to foetal stages, analysis of FAT1 mRNA levels in a series of adult FSHD1 biopsies or FSHD-derived myoblasts did not reveal any significant change compared to control biopsies or myoblasts (data not shown), a result consistent with published data [10], [60], or with data available on GEO NCBI. Overall, these results indicate that 1) a reduction of FAT1 levels in differentiated muscles can be observed is some FSHD1 cases but is not common to all FSHD1 cases at the stages examined; 2) the observed changes in FAT1 expression levels in FSHD1 occur only during development.
We next asked whether the changes we observed were accompanied with alterations in chromatin state around regulatory sequences of the FAT1 locus. We thus performed chromatin immunoprecipitations (ChIP) on muscle biopsies derived from these same FSHD1 and control foetuses (Figure 9C), looking for potential changes in the levels of two widely studied chromatin marks: H3K4me3 (trimethylation of histone H3 on lysine 4), a mark of active promoters, and H3K27me3 (trimethylation of histone H3 on lysine 27), which marks transcriptionally silent chromatin [61]–[62]. Consistent with RT-PCR data, we observed a significant decrease in the level of H3K4me3 decorating the FAT1 promoter region in the two FSHDs foetuses with less than 5 repeats, but not in the foetuses with 7 repeats, as compared to 4 control muscle biopsies of similar age range (Figure 9C right). However, all 4 FSHD1 foetuses nevertheless showed a significant increase in H3K27me3 levels (Figure 9C left). These data are consistent with a switch in chromatin conformation towards the silenced state in the same FSHD1 samples in which RNA levels were reduced, a switch that has the potential to account for a large part of the observed decrease in FAT1 levels.
FAT1 deregulation is not the only gene expression change reported to be associated with the D4Z4 contraction causing FSHD1. As we also wished to determine to what extent the changes we found were relevant to the specific clinical phenotype, rather than a silent consequence of the D4Z4 contraction, we therefore extended our investigation to contraction-independent FSHD cases. Such patients have typical FSHD symptoms, but are not genetically associated to a pathogenic contraction of the D4Z4 array on chromosome 4. A large fraction of these contraction-independent FSHD cases is now known as FSHD2, in which hypomethylated D4Z4 repeats are combined with with a normal sized D4Z4 array on chromosome 4 permissive for DUX4 expression [22], [63]–[64]. Besides, other rare cases of contraction-independent FSHD cases remains unexplained, and represent interesting candidates to test whether alterations of the FAT1 locus might be directly associated with FSHD. To identify such alterations of the FAT1 locus, we performed an array-based comparative genomic hybridization screen (CHG [65]), a method used to uncover copy number variants. The custom-designed CGH array we employed covered the whole FAT1 genomic region, including non-coding sequences. In our CGH survey of 29 FSHD cases, including 10 FSHD1 cases and 19 contraction-independent cases (5 of which at least not showing D4Z4 hypomethylation, see Table S1 for clinical and genetic characterization of patients), we detected 5 cases exhibiting loss of portions of the intron 17 (between exons 17 and 18), or intron 16 of the FAT1 gene (Figure 10A,B, Figure S16). Besides the overlap with exon 17, we noticed that these deletions mapped near or within a hot spot of H3K4me1 methylation, a hallmark of cis-regulatory enhancers [61], spanning across intron 16 and part of intron 17 (Figure 10A, and Encode high throughput data, available on the UCSC browser [66]). According to the ENCODE ChIP seq data set [67], this element appears labeled as having strong enhancer activity in a human skeletal muscle myoblast line (HSMM) but not in 8 other non-muscle cell lines (Figure S16B). Examining the chromatin status at this locus by ChIP experiments, we consistently found that in control foetal muscle biopsies, intron 16 but also intron 17 were decorated by high levels of the enhancer signature H3K4me1 and negligible amounts of H3K4me3 (promoter signature) (Figure 10D, blue lanes, and data not shown), providing further in vivo support to the possibility that this sequence might indeed act as regulatory element in vivo.
To determine whether loss of functional portions of the putative enhancer were associated with FSHD, we analyzed copy number variants (CNVs) in a set of 40 healthy controls, 19 contraction-independent FSHD cases, and 10 FSHD1 cases. As the sensitivity of the CGH method might not allow detecting all cases with accurate precision, we applied a more precise qPCR method, and evaluated relative copy numbers by comparing 3 positions within and around the putative enhancer to a control spot on another chromosome (Figure 10A, C; 3 additional positions shown in Figure S16). Having set the threshold for considering a genome as carrying reduced copy numbers (loss) to 75% of the value in a healthy control used as reference genome, we found some healthy controls that exhibited reduced copy numbers of genomic regions at the core of the H3K4me1 hotspot in intron 16 (5% of controls) or in either surrounding exons (10% of control cases in both cases). This finding is consistent with a study, available through public datablases, that identified cases with loss of similar genomic segments at this locus in a group of 90 healthy individuals [68]. Thus, such deletions/copy number reductions are not sufficient on their own to cause FSHD symptoms, when occurring on only one allele of FAT1. However, in all three positions, the proportion of FSHD cases (all cases included) who exhibited loss was significantly higher than the proportion of healthy controls carrying reduced copy numbers at the same spot (Figure 10C,D; X2 test, p values<0.016; <0.00075; and <0.00041, for exon 17; enhancer; and exon 16, respectively). Cases with a deletion spanning the whole region were also significantly more frequent in the FSHD group than among controls. When considering only contraction-independent FSHD cases, as much as 47% carried the CNV including the putative enhancer, as compared to 5% of controls, and up to 68% carried a CNV encompassing at least one of the three considered positions, as opposed to 20% of the controls (Figure 10C,D, Fischer test, p<0.0004 and p<0.0001 for enhancer and exon 16, respectively). Conversely, when considering the distribution of cases with increased copy numbers (gain, above a threshold of 1.25× over the average control value) we found that there were significantly less FSHD cases with gain-CNVs than among the control group (X2 test, p<0.017 and p<0.014 when considering all FSHD cases or contraction-independenty cases only, respectively). Finally, we also analyzed the methylation status at D4Z4 repeats on chromosome 4 on a subset of our group of contraction-independent FSHD patients (5 out of 19), and found no indication of hypomethylation (at the CpoI site, Table S1) on the proximal D4Z4 unit [64]. This does not exclude that others patients in our c.i-FSHD group would be diagnosed as FSHD2, but indicates that FSHD can occur in non-contracted patients independently of the hypomethylation, known FSHD2 hallmark [22], [64]. Together, these results indicate that partial or complete deletions of FAT1 intron 16/17 putative enhancer represent a polymorphism not sufficient to cause FSHD by itself when present on one allele only of chromosome 4, but which segregates with FSHD. Therefore, this CNV can be combined with pathogenic or sub-pathogenic contexts, and may act as a novel disease modifier in FSHD.
FAT-like cadherins play various roles in tissue morphogenesis, by modulating cell polarity, adhesion and tissue growth. Here we show that during development, FAT1 controls the shape of subsets of muscles in the facial and scapulohumeral regions, and does so by modulating the polarity of collective myoblast migration, a function in accordance with the emerging link between planar cell polarity and collective directional migration events [29]–[30], [69]. These muscle shape abnormalities are predictive of early onset muscle wasting, as observed in Fat1-deficient mice that bypassed neonatal lethality. Using Pax3-cre for conditional ablation of Fat1 functions in premigratory myoblasts, we show that a cell autonomous requirement for Fat1 function in the migrating myoblasts accounts for a significant component of this role in shaping muscles. Taken together, the location of the human FAT1 gene next to the critical FSHD locus at 4q35, the similarity between the Fat1-dependent muscles and those affected in FSHD, and the appearance in Fat1 mutants of non-muscle features of FSHD, suggest a possible role of FAT1 in the pathophysiology of this disease. In our human studies, we found two ways by which altered FAT1 regulation underlies a link with FSHD: 1- we observed muscle-specific lowering in foetal FSHD1 biopsies; 2- we identified a polymorphism deleting a putative cis-regulatory enhancer in the FAT1 locus, which significantly segregated with FSHD. Together, these results strongly support the idea that tissue-specific de-regulation of FAT1 expression/function might play a critical role in FSHD pathophysiology.
The altered myoblast migration polarity caused by loss of Fat1 functions leads to selective developmental dysgenesis of scapulo-humeral and subsets of subcutaneous muscles of the face. Understanding how Fat1 controls muscle shape required first determining which part of its expression domain accounts for this function. In addition to the muscles, Fat1 is expressed in several of the cell types that interact with migrating muscle cells. The highest expression was seen in non-muscle cells, such as the subcutaneous layer towards which CM myoblasts migrate (Figure 1). This muscle-skin interface is analogous to the bone-muscle interfaces (tendons, joints) of skeletal muscles, where Fat1 also accumulates at later stages (Figure 7A). Here, however, we show that ablating the floxed transmembrane domain of FAT1 with a Pax3-cre knock-in line leads to efficient excision in premigratory muscles of the limb but not the face, and reproduces at least partially the migration phenotype observed in constitutive Fat1 knockouts in the scapulohumeral region. Pax3-cre excision does not occur in motor neurons, hence ablation in this cell type does not contribute to the phenotype observed in Fat1Fln/Fln;Pax3 cre/+ embryos. No significant muscle shape defects were caused by Pax3-cre -mediated Fat1 ablation in subcutaneous muscles of the face. This is not surprising, as muscles in the face do not derive from Pax3-expressing precursors but were previously shown to derive from a subset of islet1-expressing pharyngeal mesoderm cells [46], [70]. In addition to trunk migrating myoblasts, Pax3-cre-mediated excision occurs in dorsal neural tube and neural crest. Although Fat1 expression is detected in Schwann cells (neural crest-derived) along the nerves at P0, we did not detect such an expression at the stage of muscle migration (E12.5, see Figure S11C), making it unlikely to for Fat1 to control migration polarity by acting in neural crest derivatives. Furthermore, as Pax3-cre-derived neural crest amply colonizes the developing face, the lack of enhanced muscle phenotype in the face of Fat1Fln/Fln;Pax3 cre/+ embryos disqualifies the neural crest component of Fat1 expression from playing a major contribution in muscle shaping, and strongly suggests that Fat1 is required cell-autonomously in migrating myoblasts to control the polarity of their migration. As however, the muscle phenotype of Fat1Fln/Fln;Pax3 cre/+ embryos is significantly weaker than the phenotype of constitutive mutants, it leaves the possibility that other component of Fat1 expression domain may also contribute to its function in muscle patterning.
The rationale for why such a selective group of muscles is affected by Fat1 loss of function is still unclear. This group of muscle includes subsets of migratory muscles of the face and shoulder area. In the face, defects are restricted to branchiomeric muscles derived from the second brachial arch (subcutaneous muscles of the skin, Figure 3), while first branchial arch derived muscles (masseters and temporalis), as well as extraocular muscles, are unaffected (Figure 5 and data not shown) [70]–[72]. The scapulohumeral region can be divided in two components: 1) the CM, as well as humeral muscles (triceps, deltoid, or muscles which pattern is affected by the supernumerary muscle) derive from somitic Pax3-driven hypaxial migratory precursors (Figure S10); 2) In contrast, some of the shoulder muscles such as the acromiotrapezius and spinotrapezius, or the rhomboids, belong to the cucullaris group and were previously shown to derive from non-somitic, occipital lateral plate mesoderm [46], [72]–[73]. Such specificity is in apparent contrast with the broader expression domain of Fat1 in muscles as observed at E12.5 and later (Figure 1, 7, and S1), although clear differences in expression levels between muscles can be distinguished (Figure 7A). Given that distinct regulatory programs govern the development of these muscle groups [2], [74], the selective impact of Fat1 on muscle shapes could be determined by its interaction with some of the selective myogenic regulators.
Advanced symptomatic stages in Fat1-deficient mice are likely systemic consequences of such non-muscle phenotypes. Nevertheless, the muscle wasting and dystrophic features measured at presymptomatic stages were detectable selectively in those muscles that exhibited myofiber orientation defects, even in cases with no other detectable phenotypes. Despite the important variability in postnatal phenotype strengths observed with the Fat1LacZ allele, myofibre orientation defects and dystrophic features in the CM and shoulder muscles (Rhomboids, Trapeze) were observed in all mutant cases examined, not only of embryos, but also at adult stages, even in cases of Fat1LacZ/LacZ mice surviving to old ages with no other detectable phenotype. This specificity argues against the idea that restricted topography of muscle defects would be a consequence of renal problems or of other non-muscular defects. Furthermore, the observed match between the topography of the developmental phenotype and the specific map of muscles that undergo wasting at presymptomatic stages in adult Fat1LacZ/LacZ mice supports the idea that the selective muscle degeneration might occur as a consequence of the altered muscle shape. Future experiments will be necessary to determine whether the limited defects observed in Pax3-cre/Fat1 embryos are sufficient to predispose muscles to early onset degeneration, and whether additional triggers might be required for degeneration to occur in adult life. Among phenotypes observed in adult Fat1-deficient muscles, it will also be interesting to distinguish secondary consequence of the altered muscle shapes, from phenotypes reflecting additional, independent functions of Fat1, whether exerted in muscles too or in other cell types.
The spatial distribution of muscles mis-shaped as a result of Fat1 loss of function as seen at E14.5/E15.5 (Figure 3) appears to overlap very closely with, and thus to predict, the map of muscles affected at early stages in FSHD. Furthermore, the observation of non-muscle phenotypes such as defects in retinal vascularisation or inner ear patterning also bears some similarities with symptoms observed in FSHD patients. Despite this strong concordance between the phenotype of Fat1-deficient mice and FSHD symptoms, the selectivity of the shared phenotypes raises a paradox. Fat1 expression during development is not restricted to FSHD-relevant tissues, and constitutive deletion of Fat1 leads to pronounced renal defects and neonatal lethality. Even the Fat1 hypomorphic phenotypes presented above cannot be considered as an exact phenocopy of FSHD. Overall this mouse model is also more severe than FSHD, and 50% of the mice die within 3 months, likely of milder versions of the kidney phenotype (such as polycystic kidney). In contrast, FSHD is not known as a lethal disease, and has no reported association with kidney problems. Absence of renal dysfunction in FSHD is a strong indication that FSHD cannot simply be considered a “FAT1 knockout”. Thus, cases of patients with severe FAT1 loss of functions and kidney failure might be fatal before onset of muscle dystrophy and might thus fail to be classified as FSHD. In support of this hypothesis, a rare case of a 5-year-old girl carrying a duplication of the D4Z4 array and showing vascular retinopathy and sensorineural deafness was also reported to have focal glomerulosclerosis of the kidney [75]. Instead, lack of association between FSHD and renal dysfunction indicates that any FSHD mechanism involving FAT1 alterations must necessarily preserve FAT1 expression/function in kidney (at least). Our results with mice suggest that such selective alterations of FAT1 function/expression may matter during development, in muscle precursors, at a stage when their migration occurs, for which FSHD human material was not available so far - and can ethically not be sought. FAT1 levels may not be changed to an equal extent in all tissues and times, consistent with our observation that FAT1 levels were reduced in disease-relevant muscles but not in brain, and at foetal but not adult stages. Thus, an engineered mouse model in which Fat1 functions are specifically ablated in muscles and preserved in the renal system, even though lacking effects of other DUX4 target genes, may represent a more suitable tool to study consequences of the muscle abnormalities in adult, and a better model reflecting the tissue-specific FAT1 depletion that we propose might be occurring in FSHD.
The finding that human cases of contraction-independent FSHD, with such a characteristic and restricted set of clinical symptoms, segregate with the deletion of a putative regulatory genomic element in the FAT1 locus instead of the traditional D4Z4 contraction, strongly supports the idea that altered FAT1 regulation plays a key role in the pathology. The putative cis-regulatory enhancer reported in this study, which deletion segregates with FSHD in contraction-independent cases is likely to carry tissue-specificity information driving FAT1 expression in FSHD-relevant cell types, and future experiments are required to demonstrate such activity. The finding that healthy controls can exhibit heterozygous loss of this fragment of the FAT1 locus, containing two exons and an enhancer, is consistent with the observation that heterozygous loss of Fat1 functions in mice does not have major consequence of life span, health, and muscle integrity. However, we did observe a significant degree of haploinsufficiency in Fat1ΔTM/+ embryos, evidenced by the presence of subtle muscle shape defects (Figure 6B,C, see indicated p values), the most frequent position being between the acromyotrapezius and spinotrapezius muscles (as in Figure 3B). Such phenotypes were also consistently detected in Fat1Fln/Fln embryos (Figure 6 and Figure S10D), in which expression levels were similar to those measured in Fat1ΔTM/+ embryos (Figure S10B, C), suggesting that muscles in the shoulder area are highly sensitive to Fat1 dosage. While copy number variants outside of the putative enhancer might occur without causing any regulation change, we reasoned that the further such deletions would extend into the ENCODE predicted enhancer, the more functional transcription factor binding sites they may remove, hence increasingly interfering with FAT1 regulation on the deleted allele, thereby sensitizing the locus to additional contexts that may additionally impact on FAT1 expression.
Interestingly, two of the FSHD1 cases presented here were monozygotic twins, both carrying a contracted 4q35 allele with 3 D4Z4 units, one of the twins being asymptomatic while the other twin had been diagnosed with a classical FSHD. We found that the twin with FSHD symptoms displayed reduced copy numbers throughout the length of the studied area, encompassing both exons 16 and 17 and the intron 16 putative enhancer, while the asymptomatic twin exhibited reduced copy numbers only at the distal-most region towards exon 16, this difference possibly representing a de novo somatic mutation (Figure 10 and Table S1). Although this correlation does not constitute a demonstration of causality, it provides support to the hypothesis that this lowered copy numbers (heterozygous) of FAT1 exons 17/16 and of portions of the putative FAT1 enhancer portions have the potential to worsen FSHD symptoms when combined to a pathogenic context. However, obtaining a formal demonstration of this hypothesis will require studying phenotypes/genotype correlations on a large cohort of patients, and knowing in each case if the FSHD-causing genetic context is FSHD1, FSHD2, or other un-identified contraction-independent contexts. Overall, deregulation of the FAT1 gene is associated with FSHD, either as a consequence of DUX4 overexpression, and/or epigenetically encoded in FSHD1 and FSHD2, or through the deletion of a putative enhancer that segregates with contraction-independent FSHD patients.
Among possible products of the Fat1-gene, our results in mice indicate that the control of migration polarity and muscle shape requires a Fat1 RNA containing a transmembrane domain encoded by the floxed exons and deleted in the Fat1ΔTM allele. In contrast, other functions can be executed by incomplete Fat1 isoforms. Residual RNAs containing 3′ Fat1 exons can rescue (to an extent correlating with RNA levels) kidney defects and their consequences, but not muscle dysgenesis. Interestingly, however, both mouse models retain the capacity to produce FAT1 protein isoforms containing an intracellular domain, albeit at reduced levels quantified by qPCRs (Figure S10B,C), ruling out a major contribution of these isoforms to the muscle shape phenotypes observed in both mouse models. In muscle fibres, FAT1 is a novel component of t-tubules. Does Fat1 expression in differentiating and mature muscle reflect additional functions in muscle biology? The presence of FAT1 protein in close association with the contractile apparatus, as soon as differentiation starts, may reflect a role in sarcomere assembly. These FAT1-enriched stripes are maintained in mature muscle fibres, tightly juxtaposed with the t-tubule system (Figure 7B). This may indicate a further involvement in excitation-contraction coupling, an essential process required throughout adult life for muscle function and maintenance. However, this striped pattern is established as early as the contractile apparatus assembles (Figure 7C), before the alignment and docking of T-tubules to the contractile apparatus takes place, the latter phenomenon occuring postnatally in mice [76]. This indicates that in muscle, FAT1 isoforms are not inserted in the t-tubule compartment itself, but may be located at an interface juxtaposing t-tubules and the contractile units, possibly reflecting a new function for Fat1 for example during assembly of the t-tubule network. As myoblast migration precedes differentiation and sarcomere assembly, the accumulation of these FAT1 protein isoforms in the contractile apparatus occurs too late to be accountable of the function in migration polarity.
FAT-like proteins were previously reported to be subject to various cleavage events by Furin convertase or by α- or γ-secretases [50]–[52]. Furthermore, alternative splicing events in the cytoplasmic exons were reported to influence subcellular targeting of FAT1 proteins [48]. Our work in mice unexpectedly indicated that in addition to producing a large transmembrane protein and its cleavage products, the Fat1 gene also produces small molecular weight protein products which appear not to contain a transmembrane domain, and synthesis of which is largely preserved in both Fat1-deficient mouse models, although at reduced levels. Bioinformatic scans and existing ESTs reported on all genomic browsers are indeed consistent with the possibility that short isoforms may result from transcript initiation at alternative downstream promoters, and may code for protein products devoid of leader peptide and transmembrane domain and potentially produced in the cytosol (lacking a leader sequence). Thus, understanding the roles played by the isoforms of FAT1 produced in muscles will require first characterizing the exact exon and domain composition of the Fat1 RNA and protein isoforms produced in muscle (wild type and Fat1ΔTM/ΔTM), and second designing novel strategies to ablate them independently of the transmembrane domain containing isoforms.
Interestingly, residual expression of such muscle-specific isoforms is genetic background dependent and its levels in Fat1LacZ/LacZ mice inversely correlated with phenotype severity. Furthermore, reduced expression levels and abnormal sub-cellular localization were observed in muscle of human foetal cases with expected severe and early onset FSHD1 (as predicted by the degree of D4Z4 contraction and family history), while no significant changes in RNA levels were detected in adult FSHD1 muscles compared to controls. These observations are consistent with the idea that deregulated FAT1 expression in differentiated muscle may be predictive of early (infantile) onset and severe dystrophy. These data suggest that the causes of the early phase, common to all FSHD patients and restricted to muscles of the face and shoulder, might be uncoupled from the causes of later phases of the disease - which spreads to other muscles, a condition that occurs in a subset of FSHD patients with childhood onset, the latter ending up wheel-chair bound [6].
Recent studies have brought to light several possible molecular pathways by which the D4Z4 contraction on a 4qA allele may exert its pathological effect in FSHD1. Among those, stabilization of DUX4-fl mRNAs by polyA-creating polymorphisms was shown to enable expression of a toxic form of DUX4, the latter causing muscle dystrophy through altered regulation of numerous target genes, including Pitx1, p53, and other germline-specific genes or myogenic regulators [17]–[21], [23]–[24]. Another mechanism involves production by the contracted region of DBE-T, a chromatin-associated long-non-coding RNA that causes de-repression of several 4q35 genes [77], including FRG1, whose overexpression was previously proposed to contribute to causing muscle degeneration too [11]–[12]. Other mechanisms also influencing 4q35 gene expression include a telomeric position effect, according to which propagation across 4q35 of changes in methylation or chromatin conformation might be due to the loss of the CTCF barrier function of the D4Z4 array [8], [13], [78]. The relative contribution of DUX4-mediated gene regulation and of mechanisms leading to altered 4q35 gene expression is controversial [9], [17] and may reflect an underestimated diversity in the clinical expression of FSHD1 [79]–[80]. Understanding which of these mechanisms, or what combination, contributes to modifying tissue-specific distribution of FAT1 will require developing cellular or animal models adequately reproducing FSHD mechanisms and mimicking in vitro key steps of muscle shape development. This will also allow defining whether there are differences in the sensitivity to a contracted allele between developmental stages and adult muscle, but also between FAT1 isoforms. DUX4 can repress FAT1 expression in human myoblasts ([18] and Figure S15D). Such regulatory influence could involve some DUX4 target genes such as p53 [37]–[38], or myogenic transcription factors. Our data suggest that irrespective of whether FAT1 is regulated by DUX4, by DBE-T, or by anyone of their respective downstream or upstream targets, this regulation must occur primarily during development, in the cell type in which FAT1 is required to control migration polarity. This model does not exclude the possibility that the pathogenic 4q35 allele may further contribute to directly triggering muscular dystrophy in adult muscle, through additional mechanisms independent of FAT1 de-regulation.
A number of clinical features of FSHD, including non-muscular symptoms such as hearing loss and retinal vasculopathy [81]–[82], carry the signature of defects in the Wnt/PCP pathway [26], a cascade of tissue polarity regulating genes, involving non-canonical Wnt/Frizzled signalling (core PCP genes) and modulated by the protocadherins FAT and Dachsous [25], [27]. Sensory hair cell polarity in the cochlea is the best mammalian PCP paradigm, and deafness has become a traditional hallmark of altered PCP signalling [26], [57]–[58]. Even through the anatomical nature of auditory abnormalities in FSHD is not known, it will be relevant to explore whether it carries further characteristics in common with altered PCP. Furthermore, vascular abnormalities in the retina, also known as Coats disease, are phenotypically similar to familial exudative vitroretinopathy (FEVR), recently linked to mutations in the Wnt receptor Frizzled4 (FZD4) and its ligand Norrin [83]–[85]. Moreover, the Wnt/PCP pathway is also known to play key roles in muscle biology. PCP-activating Wnts, such as Wnt11 or Wnt7a act as instructive signals for myofibre orientation during muscle morphogenesis [86], for muscle satellite cell expansion through symmetric division [87], and for neuromuscular synapse development [88]. Thus, altered regulation of FAT1 may in turn de-regulate the function or expression of its genetic partners, such as other components of the planar cell polarity cascade but also of the Hippo pathway. Mutations in other components of these genetic cascades may also play a causal role in a subset of the FSHD patients lacking the D4Z4 contraction. Overall, by linking FSHD to FAT1, our work opens new avenues for the exploration and treatment of this and other neuromuscular disorders.
Animals were maintained and sacrificed in accordance with institutional guidelines. Adult mice were either sacrificed for experiments through anaesthesia, or euthanized by cervical dislocation. Efforts were made to minimize the number of adult Fat1-deficient mutant mice examined after more than 25% weight loss.
Human DNAs were obtained from FSHD and control cases at La Timone Hospital (Marseille, France). The protocol for their collection was approved by the Université de la Méditerranée (Marseille, France) Committee on Human Research and an agreement of informed consent authorizing scientific experiments was signed by each individual patients. Human Tissues samples were obtained from abortus cases at La Timone Hospital (Marseille, France) and at AP-HP (Assistance Publique-Hopitaux de Paris, France). The protocol for their collection was approved by the Université de la Méditerranée (Marseille, France) Committee on Human Research and an agreement authorizing scientific experiments was signed by the parents. Termination of pregnancy (performed at the stages corresponding to individual cases) was decided after late prenatal diagnosis.
Human Tissues samples were obtained from abortus cases (see ethics statement) after termination of pregnancy, decided after late prenatal diagnosis (of FSHD1 or of non muscular medical symptoms for control cases). The cases used, and their respective stages are described in Figure S8A. Four cases of foetuses diagnosed with FSHD were used (Figure S8A) referred to as F1, F2, F3 and F4, respectively. Family history included in the F1 case early-onset and severe FSHD phenotypes in a sibship carrying the same haplotype (family tree shown in Figure S8D). In the she second (F2) and third (F3 and F4, twin foetuses) cases one parent had FSHD. FSHD diagnosis was characterized by standard procedures involving southern blotting using a combination of restriction enzymes and probes, to characterize contraction status, 10 versus 4 chromosome, and haplotype. The p13E-11 probe was used on genomic DNA digested with EcoRI alone or with EcoRI and BlnI, hence determining D4Z4 array length and distinguishing and 4q contractions from 10q contractions [93]. Molecular combing is then performed with a combination of probes (including those for D4Z4, the p13E-11, qA and qB-specific probes, and 10q versus 4q specific) allowing to distinguish simultaneously 10q from 4q as well as qA from qB haplotypes and the degree of contraction [94] (see also simplified probe set in Figure S14B–E). Control biopsies (Figure S14A) were also obtained from abortus cases, for which termination of pregnancy was performed on the basis of medical diagnosis different from FSHD or other muscle related diseases. Detailed information on clinical and genetic diagnostic for the patients used for CGH and qPCR studies is provided in the Table S1.
X-gal staining was performed using classical procedures on embryos or postnatal tissues previously fixed in paraformaldehyde (PFA) 4% (time depending on strength of lacZ expression), rinsed in PBS, and incubated in X-gal in combination with potassium ferri- and ferro-cyanide (FeCN). Staining was terminated by rinsing in PBS, and post-fixing in PFA4%. Embryos were transferred in 100% glycerol for imaging and counting dispersed myoblasts.
For adult murine tissues, anaesthetized mice were perfused with PFA 4% in phosphate buffer saline (PBS) prior to dissection. Shoulder belt muscles were carefully dissected under a stereomicroscope, rinsed in PBS, shortly incubated with fluorescent alpha-Bungarotoxin to visualise neuromuscular junctions. When necessary, observation under fluorescence was used to visualise and sub-dissect zones enriched in neuromuscular junctions. Samples were cryoprotected in 25% sucrose (in PBS), embedded in a mix with 7.5% gelatine and 15% sucrose in PBS, and frozen for cryostat sections.
Immunofluorescence was performed using primary antibodies to neurofilament (NF-M, Ab1789, Chemicon), tau (AbCAM), laminin (Sigma), alpha-actinin (Clone EA-53, Sigma), Ryanodine Receptor RyR1 (MA3-925, Thermo scientific), Dyhydropyridine Receptor alpha 1S (MA3-920, Thermo scientific), rabbit anti-GFP (invitrogen). Antibodies against FAT1 were the following: two rabbit polyclonal antibodies raised against human FAT1, HPA001869, HPA023882 from SIGMA (epitopes described in the human protein Atlas (http://www.proteinatlas.org) recognised two regions of the extracellular domain of FAT1 indicated FAT1-1869 and FAT1-23882, respectively, in Figure 2A, B. Two antibodies against the intracellular domain of mouse FAT1 were used: Fat1-ICD from ref [35], and an additional anti-Fat1 rabbit antisera (Rb1465) we raised against a GST-fusion protein encompassing the intracellular domain of mouse FAT1 (see complete procedure below). Secondary antibodies used were Cy3- or Cy5-conjugated (Jackson Immunoresearch) or conjugated with Alexa-488 or Alexa-555 (Invitrogen). NMJs were visualised with Alexa-488 conjugated alpha-Bungarotoxin (1/2000), and F-actin with alexa-594 or Alexa-647 conjugated-Phalloidin (Invitrogen). Retinal vasculature was visualised with Alexa-488-conjugated GS-IB4 (Invitrogen), as described [83], including CaCl2 1 mM and MgCl2 1 mM in all incubating solutions. Image acquisition was performed with a Zeiss Axioplan equipped with Apotome.
For electron microscopy analysis, muscles were dissected from mice previously perfused in 4% PFA, and post-fixed in 2% PFA, 2.5% glutaraldehyde, 50 mM CaCl2 in 0.1 M cacodylate buffer (pH 7.4). Muscles were additionally postfixed with 1% OsO4, 0.1M cacodylate buffer (pH 7.4) for 2 h at 4°C and dehydrated in a graded series of ethanol, with a 2h incubation step with 2% uranyl acetate in 70% Ethanol at 4°C. Samples were further dehydrated and embedded in epon resin. Thin (70-nm) sections were stained with uranyl acetate and lead citrate and examined by transmission electron microscope (Zeiss EM 912). Images were acquired with a digital camera Gatan Bioscan 792, using the Digital Micrograph software.
Embryos were collected in PBS and fixed in 4% PFA. In situ hybridizations were performed with a MyoD RNA probe on whole mount E12.5 embryos, according to previously published procedures [95]. In order to assess the orientation of myoblasts, the CM muscle sheet was dissected and flat mounted, after completion of the ISH procedure, for high magnification imaging with a Zeiss Axioplan. For each muscle, three areas in stereotyped positions of the CM (3 positions in which the main chain direction made a 10°, 45°, and 70° angle with the DV axis, respectively) were imaged at 63X resolution. Scoring myoblast direction was done using AxioVision image software (Zeiss Imaging). For each picture, three to four chains were outlined. For each cell, an angle between the closest outlined chain and the nucleus direction was measured. Every cell for which the nucleus was visible was assigned such an angle. The distribution of angles was thus determined for each embryo side (two CM muscles per embryo), by defining angle ranges of 10°, and determining the percentage of cells showing an angle in the given angle range. This distribution was averaged between 3 wild types embryos sides (n = 3), and 5 Fat1LacZ/LacZ embryo sides (n = 5).
Tissue extracts were prepared in EBM buffer (20 mM Tris–HCl pH 7.5, 150 mM NaCl, 1% Triton, 5 mM EDTA, 5 mM EGTA, 10% glycerol) supplemented with protease inhibitors [95]. To enrich the lysates in membrane associated proteins, lysates were lectin-purified by incubation with Lectin-sepharose beads. For immunoblotting, 50 µg of protein extracts were separated by SDS–PAGE using 3–8% gradient gels (Invitrogen), blotted onto nitrocellulose membrane and detected with specific antibodies. Immunoblots were revealed by ECL (Amersham).
We first constructed a GST-FAT1 fusion protein, containing the C-terminal part of the intracellular domain of mouse FAT1 (from aa 4451 to 4588; an epitope entirely contained by exon 28, and comprising approximately one third of the cytoplasmic domain), in PGEX2 vector. Serum was collected from two rabbits immunized with the GST-FAT1 fusion protein (Rb 1465 and 1464). Antibodies were affinity purified from the two antisera, using the same GST-FAT1 (GST-Fat1-aa4451-4588) fusion protein, loaded on Affi-gel 15 support in poly-prep chromatography columns, following the manufacturer's instruction (Biorad).
In both mouse and human samples, total RNA was isolated using Trizol reagent (Gibco, BRL). RNA was resuspended in 100 µl DEPC-treated H2O and quantified by spectrophotometry; samples used for RT-PCRs had a 260/280 absorbance ratio greater than 1.8. cDNA was synthesized from 1 µg of total RNA using Superscript III (Invitrogen) or the First Strand cDNA Synthesis Kit (Fermentas RevertAid: K1622) and random oligonucleotides.
In mouse RNA samples, expression levels of Fat1, Creatine kinase B (CKB) or HPRT were determined by semi-quantitative and/or quantitative RT-PCRs using real-time sybrgreen PCR assay (life technologies), using the following primer sets. Fat1 Primer set exons 20–21 (product size: 511 or 525 bp), 5′ CCA CGC GGT TGT CAT GTA CG 3′ (exon 20-Fw), and 5′ TCC AGT AGG CGA GGG ATT GC 3′ (exon 21-rev). Fat1 Primer set exon 6–8 (product size: 545): 5′ AAG CCC CTT GAT GCA GAA CA 3′ (exon6-Fw); 5′ TCA GCG TTC CTC CCT TTG TC 3′ (exon8-rev). Fat1 primer set exons 24–25 (product size 142 bp) 5′ TGC TGT CTG TCA GTG TGA CTC AGG C 3′ (exon 24-Fw); 5′ GAG AGG CAT CCT CAC AGT GCT TCC C 3′ (exon 25-rev); Fat1 primer set exons 26–28 (product size varies according to splice variants expressed; 3 products are observed:268 bp, 304 bp; and 330 bp) 5′ CGC TTA GCT CCT TCC AGT CAG AGT CC 3′ (exon 26-Fw); 5′ GGG TGG GTG TAT GGA CTC GAA CTG G 3′ (exon 28-Rev); HPRT primer set: HPRT-fw: 5′ CAC AGG ACT AGA ACA CCT GC 3′ HPRT-rev: 5′ GCT GGT GAA AAG GAC CTC T 3′. Creatine kinase B-type (CKB); CKB-Fw: 5′ ACG ACC ACT TCC TCT TCG ATA A 3′; CKB-rev: 5′ TTT TCA GTG TCA GCA ACA GCT T 3′. For qPCR experiments, the HPRT gene was used as endogenous reference gene to normalize the data across all samples. For each gene examined, primers were chosen at the junction between two exons, to distinguish by their size the RT-PCR products from the genomic DNA PCR products. For Fat1 primer sets, sizes expected from genomic PCR amplicons, in case of genomic DNA contamination, have been indicated on Figures S4 and Figures S5).
Expression of the human FAT1 gene was monitored by a real time quantitative RT-PCR method using TaqMan gene expression assay reference number Hs00170627_m1 targeting the 5′ part of the FAT1 sequence (Applied biosystem), or using real-time sybrgreen PCR assay (Roche) (see primers below). The ubiquitous beta-glucuronidase (GUS), was used as endogenous reference gene to normalize the data across all samples. FAT1 primers were chosen at the exon2-3 junction: forward primer: 5′- CAT TAG AGA TGG CTC TGG CG-3′; reverse primer: 5′- ATG GGA GGT CGA TTC ACG-3′). (Fw GUS: 5′-CTC ATT TGG AAT TTT GCC GAT T-3′; Rev GUS: 5′- CCG AGT GAA GAT CCC CTT TTT A-3′). Primers used for other muscle genes: DHPR-Fw: 5′- CGC AAC TGG TGG GTT GCC AGC-3′; DHPR-Rev: 5′- GGC CCA TCC TCC AGC AAC GC -3′; MURF1-Fw: 5′- CTT GAC TGC CAA GCA ACT CA -3′; MURF1-Rev: 5′- CAA AGC CCT GCT CTG TCT TC -3′; DYSF-Fw: 5′- GAA GCC AAG GTC CCA CTC CGA -3′; DYSF-Rev: 5′- CAG GCA GCG GTG TGT AGG ACA -3′; Calp3-Fw: 5′- TCT CTT CAC CAT TGG CTT CGC -3′; Calp3-Rev: 5′- TGC TGC TTG TTC CCG TGC -3′; B2M-Fw: 5′- CTC TCT TTC TGG CCT GGA GG -3′; B2M-Rev: 5′- TGC TGG ATG ACG TGA GTA AAC C -3′; γSARC-Fw: 5′- CGA CCC GTT TCA AGA CCT TA-3′; γSARC-Rev: 5′- CCT CAA TTT TCC CAG CGT GA -3′. Similar results were obtained with two other normalizing genes (β-2-microglobulin (B2M) or the human acidic ribosomal phosphosprotein (PO)).
Each experiment was performed in triplicate and repeated at least three times against age matched unaffected foetuses used as controls (see Figure S6A). For quantitative RT-PCR experiments, relative quantities of RNA expression were calculated using the comparative cycle threshold (ΔΔCt) method [96] and were normalized with GUS RNA levels as endogenous reference gene. Briefly, the fold change of RNA expression levels was calculated by the equation 2−ΔΔCt, where Ct is the cycle threshold. The cycle threshold (Ct) is defined as the number of cycles required for the fluorescent signal to cross the threshold in qPCR. ΔCt was calculated by subtracting the Ct values of the endogenous control (GUS) from the Ct values of the RNA of interest (FAT1 or control muscle genes, such as DHPR, MURF1, DYSF, Calp3, γ-Sarcoglycan). ΔΔCt was then calculated by subtracting ΔCt of the sample used as control from the ΔCt of FSHD biopsies.
Human primary myoblasts unaffected by muscle disease were infected with lentivirus carrying either DUX4-fl or GFP as control for 24 hr. RNA was extracted with Qiagen RNeasy kit, DNAse'd with Ambion Turbo DNAse and reverse transcribed with Invitrogen SuperScript III according to manufacturers' instructions. Real time quantitative PCR was performed with the following primers: FAT1-f 5′ – GGA AAG CCT GTC TGA AGT GC - 3′; FAT1-r 5′ – TGT ATG TCC GGC AGA GGA AC -3′; RPL13a-f 5′ – AAC CTC CTC CTT TTC CAA GC - 3′; RPL13a-r 5′ – GCA GTA CCT GTT TAG CCA CGA - 3′. FAT1 values were normalized to the internal standard RPL13a and expressed as percent relative to control condition.
ChIP assays were performed on chromatin from fetal muscle biopsies (tissue samples obtained as described above) using the Magna A ChIP kit (Millipore/Upstate). For chromatin preparation, muscle samples (∼50 mg) were weighed, cut in small pieces, and cross-linked with 1.5% paraformaldehyde for 10 minutes, the cross-linking reaction being stopped by addition of glycin. Nuclei were extracted from the tissue samples by using a 2 ml dounce tissue grinder and the kit's cell lysis buffer. Chromatin was then sheared by sonication and quantified after DNA extraction. Immunoprecipitations were performed on 5 µg of Chromatin, with 3 µg of the following antibodies: anti-H3K4me3 (17–614, Millipore), anti-H3K27me3 (07–449, Millipore), and anti-H3K4me1 (ab8895, Abcam), following the ChIP kit instructions, using _proteinA-conjugated mareferredbeads. Immunoprecipitaded material was then washed, cross linking was reversed with proteinase K at 56° for 2 h, and DNA was extracted. The presence of individual regulatory regions in immunoprecipitated chromatin was analyzed by qPCR using sybr green (Invitrogen) on a Biorad CFX96 apparatus. Relative quantities of each chromatin bound fragment expression were calculated using the comparative cycle threshold (ΔΔCt) method again [96] and were normalized either relative to the amount of input DNA (in the same amount of chromatin before immunoprecipitation, quantified with the same PCR), or with levels of the promoter region of a normalizing genes GUSB.
Oligonucleotides for FAT1 Promoter are: FAT1_P1_Fw: 5′ CTT AAG TTT GCC CTG GTC GGA AGC C 3′; FAT1_P1_Rev: 5′ AAA GTC CTC GGC AGC TCC GTG ATC C 3′; Oligonucleotides for the 17/18 intronic enhancer were: FAT1_inton17_Fw: 5′ gga gtg ggg agg agg gaa gag tgg g 3′; FAT1_intron_Rev: 5 ctt ccc tct tgc tct tct tct agc c 3′. Primers for the normalizing GUSB promoter were: GUSB_E/P_Fw: 5′ AGA GGA TGT AGA CCA GGC AAA AGC C 3′; GUSB_E/P_Rev: 5′ TAG AGG ACA GGA CAT GAC ATC AGG C-3′. All sequences were selected based on the Encode ChIP tracks on the UCSC browser (available with the base genome Human Mar. 2006 (NCBI36/hg18)).
DNA was hybridized on a Nimblegen HD2.1 genomic array consisting of 135,000 probes targeting the 4q35.2 genomic region. Probes were designed with a spacing of 10 bp between consecutives probes in the exonic/intronic regions and 100 bp in the intergenic regions. Labeling of DNA and hybridization was based on the Nimblegen protocols. Arrays were scanned with the MS200 scanner (Roche) and the acquired paired files were analyzed using the CGHweb algorithms [97]. To visualize the deletion/duplication events, coordinates were formatted as a bed file and added in the custom tracks of the genome browser (genome.ucsc.edu). Patients that had CNVs in the intron 16–17 area are refered to with the number corresponding to their number in the patient summary table (Table S1).
A first PCR validation was performed using primers flanking the deleted area. This approach is expected to yield a PCR product of approximately 1500 bp from a control locus, and a smaller size product in the case of the deleted allele, the deletions being approximately 1 kb long. Primer sequences are: Del-Fw: 5′- CCT TCA CCT GCA GTA AG-3′; Del-Rev: 5′- CTA GGA TTC CTA AGA GC -3′. This approach led to validate the presence of both the 1500 band and the smaller band in the three independent patients (numbers 11, 12 and 25), plus patient 13 (sibling of 12), as carrying a deleted allele as well. Moreover, 20 unaffected controls were tested with this method and only yielded the control 1500 bp band, indicating the absence of deleted alleles.
Further validation of the deletion was performed by quantitative PCR, comparing the relative amount of PCR products scanning the deleted area, using as reference a PCR product outside the considered zone, a method used to quantify copy number variations [96], [98]. All DNA samples, whether from healthy controls or from FSHD patients, were normalized with the ADORA Reference PCR, against one healthy control DNA used as standard DNA (set to 100%). The reference primers are as follows: chr17p12 : ADORA2B-2F: 5′-GTC ACT CTT TTC CAG CCA GC-3′ ; ADORA2B-2R:5′-AAG TCT CGG TTC CGG TAA GC – 3′. The primers corresponding to the deleted area were as follows: qPCR-primer-1: 5′ – GCA ACA GAG GCC AAT GGA AA – 3′; qPCR-primer-2: 5′ – CTG AAA AGA TTT CAG GTT ACA CGC T – 3′; qPCR-primer-3: 5′ – TTC GGT AAG ATG GGA GCA GCC TTC C - 3′; qPCR-primer-4: 5′ – GGT CCT GAC AAG CTA ATC CTG AGG G - 3′; qPCR-primer-5: 5′ – GGA GTG TGG TGT GTT CTA GGT TAT GG – 3′; qPCR-primer-6: 5′ – AGC AGA CAA GAG CAC AAG GCA TTT C – 3′; qPCR-primer-7: 5′ – GGA ACA CAG CCA AAT CTA TAT GGG – 3′; qPCR-primer-8: 5′ – TCT TCC TCC TCA CAC TCC CTT TC – 3′; qPCR-primer-9: 5′ – CCT GGG CAA TGA GTG TAA CTC C – 3′; qPCR-primer-10: 5′ – CCA ACC TCC TCC CTA CTC CAC TT -3′; qPCR-primer-11: 5′ – CCA GTG GCA GCA GGT CTG ATT AAG C - 3′; qPCR-primer-12: 5′ – GGG AAA CGT AGA ATT CAA GAA GTC GC - 3′ (primers numbered as in Figure 10A and Figure S16A).
Among contraction-independent cases, a large proportion, referred to as FSHD2, harbours hypomethylated D4Z4 [63]–[64], while others do not and are expected to carry unrelated causal abnormalities. Hypomethylation was assessed for 8 patients as indicated in the patient summary table, by digesting genomic DNA, with BlnI, CpoI and Eco91I, and hybridizing the southern blot with the p13E-11 probe, as described [64]. Numbers indicated represent the percentage of methylated proximal D4Z4 unit at chromosome 4q35.
Results were expressed as the mean ± s.e.m. Statistically significant differences were assessed by unpaired t-Student test, or Mann-whitney test for non-Normally distributed data, X2 test or Fischer tests (for linkage studies), calculated with the StatEL add-in program to excel. The Kaplan-Meier plot was made with the StatEL add-in program to excel, and P value was calculated with the logrank test. * indicates P value<0.05; ** indicates P value<0.001.
|
10.1371/journal.pcbi.1003085 | Unsupervised Clustering of Subcellular Protein Expression Patterns in High-Throughput Microscopy Images Reveals Protein Complexes and Functional Relationships between Proteins | Protein subcellular localization has been systematically characterized in budding yeast using fluorescently tagged proteins. Based on the fluorescence microscopy images, subcellular localization of many proteins can be classified automatically using supervised machine learning approaches that have been trained to recognize predefined image classes based on statistical features. Here, we present an unsupervised analysis of protein expression patterns in a set of high-resolution, high-throughput microscope images. Our analysis is based on 7 biologically interpretable features which are evaluated on automatically identified cells, and whose cell-stage dependency is captured by a continuous model for cell growth. We show that it is possible to identify most previously identified localization patterns in a cluster analysis based on these features and that similarities between the inferred expression patterns contain more information about protein function than can be explained by a previous manual categorization of subcellular localization. Furthermore, the inferred cell-stage associated to each fluorescence measurement allows us to visualize large groups of proteins entering the bud at specific stages of bud growth. These correspond to proteins localized to organelles, revealing that the organelles must be entering the bud in a stereotypical order. We also identify and organize a smaller group of proteins that show subtle differences in the way they move around the bud during growth. Our results suggest that biologically interpretable features based on explicit models of cell morphology will yield unprecedented power for pattern discovery in high-resolution, high-throughput microscopy images.
| The location of a particular protein in the cell is one of the most important pieces of information that cell biologists use to understand its function. Fluorescent tags are a powerful way to determine the location of a protein in living cells. Nearly a decade ago, a collection of yeast strains was introduced, where in each strain a single protein was tagged with green fluorescent protein (GFP). Here, we show that by training a computer to accurately identify the buds of growing yeast cells, and then making simple fluorescence measurements in context of cell shape and cell stage, the computer could automatically discover most of the localization patterns (nucleus, cytoplasm, mitochondria, etc.) without any prior knowledge of what the patterns might be. Because we made the same, simple measurements for each yeast cell, we could compare and visualize the patterns of fluorescence for the entire collection of strains. This allowed us to identify large groups of proteins moving around the cell in a coordinated fashion, and to identify new, complex patterns that had previously been difficult to describe.
| High-content screening of fluorescently tagged proteins has been widely applied to systematically characterize subcellular localizations of proteins in a variety of settings [1]. Because they employ automated liquid handling and high-throughput microscopy, these experiments result in large numbers of digital images. Previous work has demonstrated that automated image analysis approaches based on machine-learning can classify these images into groups with shared subcellular localization patterns [2]. These approaches are typically ‘supervised’ in that they rely on predefined sets of example ‘training’ images for each pattern of localization to learn specific discriminative information that defines each class [3].
In contrast, unsupervised methods offer a more exploratory approach to high-throughput data analysis in which it is not necessary to predefine patterns of interest, and therefore can discover new patterns. This also enables the analysis of patterns that are very rarely observed, which typically are hard to capture in supervised analysis as a suitable training set for classification is difficult to construct [1]. Unsupervised analysis also has the advantage that it is unbiased by prior ‘expert’ knowledge, such as the arbitrary discretization of protein expression patterns into easily recognizable classes. For these reasons, unsupervised cluster analysis has become a vital tool of computational biology through its application to genome-wide mRNA expression measurements [4]–[7], and protein-protein interaction data [8]. It has also been applied in automated microscopy image analysis [9]–[13] where it has been shown to provide complementary capabilities to supervised approaches.
Here we apply unsupervised analysis to a set of high-resolution images of 4004 yeast strains, where each strain contains a different fluorescently tagged protein [14]. Because localization classes are not defined in advance, one difficulty is to identify a set of image features that reliably distinguish classes of protein expression [10]. Further, in order to allow identified statistical patterns to be directly related to our understanding of cell biology, we sought to define a small set of simple biologically interpretable measurements. This is in contrast to many automated image analysis approaches that use a large number of image features, which are typically used for object recognition in photographs [15], [16]. Although these features can be used to build powerful classifiers, the nature of the discriminative information does not need to be intelligible to allow class label recovery [3].
Recent work has demonstrated the power of incorporating cell-cycle stage into proteomics analysis (e.g., [12], [13], [17], [18]). Several studies have identified proteins whose abundance and localization change over the cell-cycle in mammalian cells. Furthermore, unsupervised analysis has been applied to identify novel, unexpected patterns. In general, these approaches have been applied to time lapse movies of mammalian cells, although it is also possible to acquire dynamic data from still images of mammalian cells [18].
One advantage of budding yeast as a model organism is that it shows stereotypical cell-cycle dependent morphological changes, which can be used to infer cell-stage based on cell morphology in still images of asynchronous cells. Previous work has demonstrated the feasibility of uncovering and analyzing yeast morphology using automated image analysis methods [19], [20]. Although the identification of cell boundaries in images has been shown to be unnecessary for subcellular localization classification [2], [21], [22], in order to extract dynamic protein expression profiles based on changes in cell morphology, in this work we sought to accurately identify individual cells. Here, we use an explicit model of yeast cell shape in order to (1) rapidly identify cells in high-resolution images, even when they occur in clumps, (2) obtain a probabilistic confidence measure for the identified cells and (3) define biologically interpretable measurements that describe protein expression in each cell over space and time.
We show that many previously defined subcellular localization patterns can be recognized in an unsupervised hierarchical cluster analysis. We find that protein complexes and small functional protein classes, which are not typically associated with their own subcellular localizations, cluster together in this analysis. Based on these observations, we show that the resolution of the hierarchical clustering is significantly higher than previous manual subcellular location assignments to discrete classes [14]. Further, we gain global insight into the cell stage dependence of protein localization; for example, we find a large cluster of nuclear proteins that seem to appear in the bud at a clearly defined time, which we believe corresponds to the inclusion of the nucleus in the daughter cell. Finally, we identify groups of proteins that show complex, dynamic patterns of localization that can not easily be predefined or described using simple localization classes; for example, many of the subunits of the exocyst complex are seen to localize to the bud periphery while the bud is small, but then move to the bud neck as the bud grows.
Starting with a collection of 4004 strains where each protein has been systematically tagged with green fluorescent protein (GFP) [14], a red-fluorescent protein (RFP) which appears everywhere in the cell was introduced into each strain using SGA [23]. These strains were then imaged in quadruplicate at high resolution to generate two-channel fluorescent images (see Methods). The RFP was introduced to facilitate automated analysis, as it provides both a signal for cell segmentation, as well as an internal control for methodological variation in fluorescence measurements.
Encouraged by the consistency and interpretability of our measurements relative to previous knowledge about yeast subcellular localization, we performed global unsupervised analysis of our time profiles of interpretable features for the mother and bud cells (Figure 6). To identify groups of proteins with similar patterns, we use agglomerative hierarchical clustering based on a maximum likelihood criterion [37] (see ‘Maximum likelihood agglomerative hierarchical clustering’ in Methods) because it does require the size (or number) of clusters to be specified, and we expect hierarchical relationships between functional classes, and a wide range in the number of proteins in each class. The hierarchical clustering results may be browsed online using the Java Treeview [38] applet at http://www.moseslab.csb.utoronto.ca/louis-f/unsupervised/.
Previous studies have demonstrated the feasibility of uncovering cell stage from images of unsynchronized cell populations, either from time lapse movies [17] or from still images [18]. We apply this approach to high-throughput still images of budding yeast. To do so, we devised a segmentation method to identify and separate the bud and mother cells, and uncover the cell stage based on measurement of the bud size. Our method depends critically on our estimates of bud size, and we show that the automatically estimated sizes were comparable to those obtained from manually identified cells. Several parts of the analysis may be improved. For example, since the bud-site selection is predetermined by the position of the preceding daughter cell [50], it could be used to help determine the correct mother-bud assignments. Similarly, a better model for the relation between daughter cell size and the cell cycle could be used to infer a more accurate estimate of cell stage.
We presented a cell identification pipeline that includes a confidence measure which summarizes the probability that an object identified in our images is actually a correctly identified cell. To do so, we characterized the deviation of real cells from an elliptical model using several quality measures whose distribution for real cells we inferred from ellipses that had been manually fit to cells by eye. Our confidence measure allows us to distinguish correctly identified cells from artifacts and misidentified objects, without specifying what the nature of artifacts might be (Suppl. Figure S2). We believe that this type of approach for measuring the confidence of automatically identified objects in image analysis will be generally useful, because artifacts tend to vary between microscope, experiments and computational methods, whereas cell shapes are expected to be much more consistent. In addition, this confidence measure is explicitly defined as a posterior probability of an identified object to be a properly identified cell. This allows us to weight probabilistically data points according to the posterior probability. For classes of cells where our model does not fit as well, such as very early non-ellipsoidal buds, we expect to downweight all the data points, but we can still include information from these data points in our analysis. This is in contrast to the situation where we used a hard threshold to exclude artifacts. In that case, certain classes of cells are preferentially excluded (Suppl Figure 1B), and the statistical significance of downstream analyses is reduced (Suppl. Tables S2, S3, S4, S5).
Typically, spatial patterns of protein expression are described by assigning labels [14] or functional annotations [51]. Such discrete classes are not sufficient to fully describe a protein's expression if it is present in quantitatively different localizations or abundances at different cell stages, or if a protein is simultaneously present in several locations with quantitatively different fractions [52]; because our approach assigns a quantitative expression profile to each protein, we can characterize protein expression at a finer scale than the resolution currently achieved by discrete classes. Approximating protein expression patterns as discrete classes has also led to challenges for computational analysis. For example, in previous work based on discrete classes [16], [21] many proteins are often filtered out because they have either been annotated as ‘ambiguous’ or are reported to be located in several localization classes.
Because we treat expression patterns quantitatively, our analysis identifies clusters of proteins that are significantly enriched in ‘ambiguous’ proteins and proteins that were manually annotated [14] as localized in multiple compartments. Furthermore, our analysis identified and organized a group of proteins that show complex patterns relating to the growth of the bud, that were not consistently annotated previously using discrete categorizations (Suppl. Table S6). To our knowledge no previous genome-scale analysis of still microscopy images has identified groups of proteins with subcellular localization patterns that change as a function of cell-stage, such as the MCM and exocyst complex subunits discussed above, although recent work on smaller collections of time-lapse images has demonstrated that functionally related proteins can be identified in unsupervised analysis of dynamic protein expression profiles [13].
One limitation of cluster analysis is that the members of each cluster identified are not always consistent between different parameter settings, or different clustering methods. Indeed, the remarkably specific groupings corresponding to specific regulatory mechanisms (such as the clustering of all 4 of MCM complex subunits and of all 3 DNA replication factor A complex subunits) were not always observed when we varied the distance metric or clustering method used (Suppl. Table S4).
Despite these limitations, our analyses consistently identified clusters that were enriched in functional groups of proteins (Ribosome, Proteasome, DNA-damage pathway, exocyst complex, etc.; see Suppl. Table S4, S5) that are not usually associated with their own subcellular compartments. Because we used hierarchical clustering of interpretable features, we could see that these functional groups of proteins showed patterns of localization similar to those localized in the same compartment, but in each case showed subtle differences in pattern that allowed them to be distinguished. These results suggest that high-resolution images could be used directly for functional discovery as has been reported for mammalian cells [12].
This work demonstrates that accurately identifying large numbers of cells for each protein allows quantitative characterization of spatial and temporal characteristics of protein expression patterns and permits direct interpretation of image-based measurements without requiring human inspection of large numbers of images to train classifiers. Our analysis gives new insight into the relationship between protein function and protein expression patterns inferred from high resolution microscope images.
Using yeast synthetic genetic array technology [23], a new GFP collection was generated from the existing collection [14]. In this new collection, a highly expressed RFP (a tdTomato [53] fluorescent protein from the constitutive RPL39 promoter), integrated at the HO locus, was introduced into the GFP collection to mark the cell in order to facilitate automated image analysis. Micrographs were acquired using a confocal microscope (Opera, PerkinElmer). Eight micrographs were imaged (at 1331×1017, 12 bit resolution) from each strain, 4 in the red channel and 4 in the green channel, yielding a dataset of 44 Gb of image data.
It was noted that the background noise had a mean and variance that was not uniform across the image. Therefore, we defined a background image that was subtracted from each image. This background image was obtained by averaging all the images. The background image intensity accounts at most for a third of the RFP signal expected in mother cells, except for several defective CCD pixels which systematically report the same value.
For each image, we modeled the background and foreground (cell) RFP intensity levels with Normal distributions. In order to account for punctuate noise, we used a Pseudo-2D hidden Markov model (P2DHMM) [54] to model the dependence of neighboring pixels. In order to recover the maximum likelihood parameters for the Normal distributions and state transition probabilities efficiently, we performed expectation maximization (EM) on both the image under the assumption that image rows are independent, and on the same image where columns are now assumed to be independent. Finally, we infer the probability for each pixel to belong to the foreground, as the average of the two probabilities that were calculated when we assumed rows and columns were independent.
Given an image for which we know the probability of each pixel to be from the background, we want to define a map of geometric distance to background for each foreground pixel. We estimate this quantity using an iterative motion on the image grid (which includes diagonals and knight moves), where transitions from a point deterministically select the neighbor through which the shortest path to background is expected. We then compute the expected path length under the assumption that pixels reached along paths have background/foreground state transitions described by a HMM with the parameters inferred from the segmentation. The transition probabilities for diagonal and knight moves are obtained by exponentiation of the transition matrix by the distance between the two points. Since it is enforced that transitions are only allowed from point of higher expected distance to lower ones, distances can be computed directly by dynamic programming, in linear time of the number of pixels in the image. The Edge Distance map () has several uses in our pipeline: to generate the clump contours, as a quality measure for identified objects and to evaluate the distance of a protein to the periphery.
We used robust regression for matching ellipsoidal shapes to the contour of the segmented area. An ellipse is characterized to be the set of points for which the algebraic error [55] is zero:(5)where is the coordinate of the ellipse centre and r is an additional parameter, proportional to the radius of a circle for a fixed matrix A.
The matrix A may make the set of points with zero algebraic error correspond to a hyperbole or a line, and a superfluous scale parameter is observed in this parameterization. We therefore constrain the form the of matrix A:(6)where is the angle corresponding to the orientation of the major axis, and is a parameter which determines the eccentricity of the ellipse. This choice of this matrix to ensure that for any value for the set of 5 parameters (in equation 5) generates an ellipse with minor to major axis length ratio larger than , as they are both determined by the eigen values of the matrix A and then scale with the parameter ‘r’ [55].
Contour pixels are first identified by finding foreground pixels which are pixels away from some background pixel (using the Edge Distance Map described above). Initial guesses for ellipses are generated by first fitting a circle to 3 randomly sampled contour points (that circle is unique). Initial guesses are rejected if the circle does not fit within the rectangle clamping the contour points, or if the center is a background pixel. The initial guess ellipse will be set to match width (diameter) and center of an accepted circle. A small eccentricity corresponding to and a random angle (drawn uniformly from 0 to ) is used to define its remaining parameters.
If the set of contour pixels matches a single ellipse, we could directly update the ellipse coordinates by minimizing the sum of the algebraic error of all contour pixels. However, if the set of contour pixels is best explained by several ellipses, the sum of algebraic errors is likely to have local minima that are not close to any of the true ellipse parameters. Therefore, we use robust regression [56] and minimize the objective function:(7)where C is the collection of coordinates of contour pixels and is the expected error, which is chosen to be 5, matching the thickness of the contour. This effectively weights down the importance of contour points with large deviations to the current ellipse, so that the many local minima can correspond to actual ellipses.
Upon convergence, we discard ellipses that are not bounded by the clamping rectangle, or that have a background pixel at the center. Since a large number of local minima are expected, we generate about 10 fold more sets of ellipse parameters than the number of expected ellipses (based on number of contour pixels) and select the ellipse with the best fit. Once we have identified the best ellipse, we remove all contour pixels that have an error smaller than , and find the next ellipse using the remaining contour pixels using the same procedure. Since some missed lone pixels may remain, we reject the ellipse and remove the corresponding pixels if the ellipse width is less than 3 pixels or if the number of removed contour pixels accounts for less than 10% of the amount expected from the ellipse parameters and known contour width. This process is repeated iteratively, until no more contour pixels can be removed. The running time of the segmentation is linear in the number of pixels in images, and the running time of cell-finding is linear with the number of randomly sampled circles for the initialization of geometric ellipse fit. On a single 2.83 GHz Intel core, 98 seconds were required to analyze a single 1331×1017 image, which on average contained 82 cells and 31 artifacts.
We want to precisely recover the cell shape, as we will be considering the size of the bud object as a cell-stage indicator, and the position of the bud neck as a point of interest for uncovering cell-stage dependent changes of protein localization. Because cells are not exactly ellipsoidal in our images, we sought to capture foreground pixels which partition the cell clumps into non-overlapping cell areas (which we refer to as ‘shapes’). In our images, cells are often separated by dim pixels, so we force boundaries to match these dim areas.
We first use the watershed [57] transform to identify regions of the clump that are entirely contained within single cells. For each pixel which brighter than any of its neighboring pixels, we find the set of pixels (catchment basin) which can be reached by a path of monotonically decreasing intensities [57]. Secondly, we assign each basin to a cell, by finding the ellipse closest to each maximum intensity pixel. The proximity of a point to an ellipse is evaluated using the algebraic error (Equation 5). This procedure ensures that if two neighboring basins are assigned to different cells, we are guaranteed that the boundary pixels are all dimmer than the nearby inner pixels found inside one of the two basins.
Each such basin is then assigned to the closest ellipse, such that the union of these regions forms the ‘shape’. The resulting shapes may be highly non-ellipsoidal (Figure 1iii); for example, if a cell has not be properly fitted by an ellipse, a portion of its area may appended to the area of a neighbor cell instead.
In addition to the mean RFP intensity in the object, we define three shape measurements based on geometrical properties of ellipses and circles. First, we compute the best fit of an ellipse to an arbitrary shape ‘’ by evaluating the following 6 statistics on the coordinates of pixels in the shape (eq. 8).(8)where is the coordinate for a pixel from the shape ‘’, and is the number of pixels in the shape (cell size). A function defined on which takes the value ‘D’ within the area of an arbitrary ellipse has 6 degrees of freedom as well:(9)We can derive that there is a closed form for the parameters of the above function that makes the corresponding statistics defined on a continuous space match the statistics from the pixel coordinate of any shape. For instance, the center of the fitted ellipse will correspond to the center of mass of the provided shape:(10)The major and minor axes length (‘’ and ‘’) are the square root of the two solutions to a quadratic equation:(11)Finally, the recovered density ‘’ is the ratio of number of pixels to fitted ellipse area. Since the coordinates are drawn from a bitmap, we observe that the measured densities typically are bounded above by 1, except for the smallest objects. Any shape whose density is above or equal to 1 is assigned to the artifact class, otherwise we use the following first quality measure:(12)The second quality measure is based on the relationship between the perimeter and the area of an ellipse. We compute the perimeter of the shape by counting the number of pixels that have 3 or more background pixels among their 8 neighboring pixels. The theoretical relationship between the perimeter length of an ellipse and its parameters has no simple form, but may be approximated using the Ramanujan first approximation [58]:(13)The log ratio for the number of contour pixel to Ramanujan ellipse perimeter length approximation is our second confidence measure .
A third quality measure captures the deviation of the shape to a circle, by reporting the log coefficient of variation of the sum of the distance to the ellipse center and the distance to the edge for each pixel in the area(eq 14). In a theoretical circle, there should not be any variance, since the two quantities are to sum up to be exactly the radius of the circle.(14)The last quality measure is the mean RFP intensity . We model each of the quality measures using a Normal distribution. We observe that the quality measure spread displays a non-trivial dependency on cell size. For this reason, we define 7 Normal distributions, for each of the 4 quality measures, which correspond to the distribution of quality measure for 7 bins of cell sizes. The quality measure vector is then modeled by the linear interpolation of a pair characterized random variables (eq. 15).(15)where and the are diagonal covariance matrices.
We used the automatically identified shapes that were mapped to the 4305 manually identified cells in order to infer the parameters of the normal distribution at the 7 sizes (7 means and 7 standard deviations). In order to define the posterior probability of cell, it remains to characterize the uniform distribution for the non-cell objects and the mixing parameter (eq. 1). The uniform distributions were chosen to correspond to the extremum in quality measure obtained from the complete collection of identified objects that have not been labeled as artifacts. Finally, we used soft expected maximization (soft-EM) [59] on the complete collection to infer the mixing parameter, which rapidly converged to 9.9% as all other parameters are already predefined.
In order to evaluate the accuracy of our cell identification method, we first compared the automatically identified ellipses to a set of 4305 ellipses that had been drawn around cells manually. We assigned each manually identified ellipse to the automatically identified ellipse with closest center. We found that for 94.2% of manually identified ellipses, there is an automatically identified one with center occurring within 10 pixels. In these cases, the average distance between the centers was 1.86 pixels (). The correlation between the areas of the automatically identified and manually identified matched ellipse pairs was 0.882.
We next compared the center and area of the automatically identified ‘shapes’ to the set of manually drawn ellipses. Here, 92.3% of the manually drawn ellipses have a corresponding recovered shape that has a center within 10 pixels of the manually drawn ellipse center. For these, the mean distance between the shape and the manually drawn ellipse center was 1.41 pixels (). The area of the ‘shapes’ have a correlation of 0.953 with area of the automatically identified ellipses, and 0.928 with the area of manually drawn ellipses. Hence, identifying the ‘shapes’ (the hybrid operation of assigning the watershed regions to their closest ellipse) produces cells that are on average closer both in location (1.41 vs. 1.86 pixels) and size (correlation 0.928 vs. 0.882) to the manually drawn ellipses than the automatically identified ellipses. We note that the ‘shape’ -based analysis led to a slight reduction in the fraction of cells identified (92.3% from 94.2%) but this was acceptable to us in the context of the improvement in cell size estimation (0.928 vs. 0.882 correlation) because we use the cell size as an indicator of cell stage.
In order to compare the accuracy of the simple cell-finding method described above with an established method for cell identification, we compared our results to Cell profiler [60]. For background correction, we used the polynomial fit to the ensemble of images, and subtracted the resulting amount from each image. We identified the primary objects under Otsu global threshold method, and used the ‘Shape’ method for defining boundaries between objects and to distinguish the clumped objects. We chose this method because the Cell profiler documentation suggests it as proper to recover round objects in clumps. Using the same method described above for our pipeline, we compared the cells identified by Cell profiler to the manually drawn ellipses. We found that 89.0% of the manually drawn ellipses have a corresponding identified cell within 10 pixels of their area centre. The mean distance in the paired centers was 2.23 () and the correlation in object sizes 0.876. Although these statistics are slightly lower than for our simple methods, it did perform significantly faster, identifying the cells in a typical image in seconds.
In addition, 139 artifacts were manually identified. We used this set to compute the false positive rate by pairing automatically identified cell areas to the manually identified cells and artifacts (Suppl. Figure S2). We also computed the false-positive rate as a function of cell probability threshold. For example, filtering all cells that have a cell probability below 0.8 reduces the false positive rate. This is in agreement with previously reported results using post processing [61]. Since we we found that the number of cells is critical for the robustness of the time profile estimates (Suppl. Figure S3) and that small buds have systematically lower cell probability estimates (Suppl. Figure S1B), we prefered not to choose a hard threshold. Indeed, we found that using a 0.8 cell probability threshold reduces the robustness of the time profiles (Suppl. Figure S3B). We also found that applying this threshold would discard of the small buds which were used to define the first four of our ten cell-stage time points.
We characterize the protein expression phenotype within each cell object using the absolute intensity of the GFP, as well as geometrical distances between proteins to identified points of interest. In both cases, we use the RFP signal to normalize the observations made for the GFP signal. The RFP intensity was found to be dependent on the object size, so we characterized the expected RFP, , and used to normalize the GFP signal by the fold difference to the expectation of the mean RFP intensity (eq. 2). We defined using three linear function segments which fits the mean level of RFP in the 1.4M automatically identified cells:(16)Some of the morphological distances require us to identify the coordinates of a point of interest; the cell center, protein mass-center and bud neck position are obtained by averaging the coordinates of the cell pixels, of GFP-tagged proteins and Mother-bud separation contour pixels, respectively. Assuming GFP intensities are proportional to protein amount, we derive the expected value for geometrical distances with respect to the position of a randomly selected protein. The position of cell center, protein mass center and bud neck are given by:(17)where is the sum of GFP intensities and ‘’ is the set of contour pixels which separates the bud from the mother cell. The other 2 distances have a slightly different form: first, the distance to the perimeter for any coordinate has been computed using Edge Map distance, so that:(18)Second, we derive the equation for the expected distance between proteins:(19)Once again we use the RFP marker to normalize these distances. In the case of distance between proteins, the distance is normalized by the expected distance between a protein and a RFP marker. For that case, the reported log ratio representing a morphological distance would be:(20)
First, we model cell stage as a function of the bud size. Under the assumption that the bud volume increases at a constant rate, we expect that time scales linearly with . Because we have a number of identified cells and distribution of object size that varies throughout the collection of 4004 yeast strains, a common basis is required to enable comparisons between the expression of different proteins. For each strain, time series are defined as expected feature values for objects observed at 10 equidistant cell stage keypoints . We use local regression (LOESS) to infer the mean and variance at each keypoint (eq. 21), where the ‘’ is Gaussian kernel function with bandwidth parameter equal to 1700. In addition, because we have developed a probabilistic cell confidence, which assigns to each identified cell a posterior probability of being a properly identified cell, we use the cell confidence to compute a weighted average, which is the expected profile conditioned on each identified object being drawn from the cell class:(21)where is feature value that is expected at the cell stage keypoint from feature values , which are measured for the identified object. are the quality measures for each shape and are cell sizes for bud objects.
Each protein profile is a vector of means and variances of observations. We use the Maximum likelihood clustering criterion [37] (eq. 22) in order to agglorameratively join pairs of protein profiles, proteins to cluster profiles, or pairs of cluster profiles:(22)where is the determinant of a covariance matrix. This criterion is the log likelihood ratio for two cluster of size to have their protein profiles modeled as two multivariate Normal distributions (with their corresponding parameters ), to a single multivariate Normal model explaining both expression groups.
Initial cluster profiles are build from individual protein profile, which corresponds to 12 concatenated time series of feature values. As such, initial covariance matrices are diagonal matrices whose values were estimated from the LOESS(see in eq. 21). New cluster profiles are characterized by a multivariate normal distribution whose parameters are obtained from merging two previous cluster profiles (eq. 23).(23)where ‘’ are cluster sizes and are normal distribution parameters for merged cluster profiles.
The 4004 proteins were grouped based on exact correspondence of subcellular location annotation, as defined by Huh et al. [14] (Suppl. Table S1). 22 classes correspond to unique subcellular locations. We merged member profiles into a class profile () using the operation defined above (eq. 23). The Bhattacharyya metric (eq. 24) was used to compare each class profile (Suppl. Table S1) and class profiles were clustered using euclidean distance (Suppl. Figure S5A).(24)
Protein subcellular location was characterized by Huh et al. [14] by assigning one or many annotations to each protein. We report about the enrichment of either separate annotations and/or exact localization set correspondence. For example, proteins that were only annotated to be nuclear obtained the label ‘pure nucleus’ and new labels, such as ‘nucleus AND cytoplasm’ were reported. The GO and PFAM annotation were obtained from Uniprot/SwissProt [51]. In the reported hierarchical clustering results (Figure 6), clusters were manually selected and the hypergeometric distribution was used to model the occurrence of annotations of proteins within them. Bonferroni correction was applied to the P-values (1990 hypotheses, accounting for 3.3 in log scale).
To assess the significance of the hierarchical clustering, we performed permuation tests. For each protein annotation, we find the cluster that yields the smallest P-value for annotation enrichment. We then assess the statistical significance of the sum of the smallest log P-values, ‘S’, by defining two background distributions for ‘S’. In the first, we preserved the structure of the tree, but chose random proteins to assign to each leaf. In the second, we preserved the structure of the tree, but randomly replaced the proteins with other proteins that had exactly the same set of annotations of subcellular localization. In other words, for this second ‘localization constrained’ permutation, we only allow proteins of identical characterization in subcellular location terms to be permuted, so that any enrichments of subcellular location (as displayed in Figure 6) will be preserved for any permutation. We found that the statistic ‘S’ was systematically higher in the 10000 permutations than for original hierarchical cluster. Therefore, we report the corresponding Z-score, but we note that background distribution for ‘S’ is not necessarily a normal distribution (Suppl. Table S3).
In order to evaluate the resolution of functional enrichments in the hierarchical clusters, we computed the significance for subsets of annotations. We show in supplementary table S3 that complexes characterized by GO annotation are found significantly enriched, and that the ribosomal and proteasomal proteins, which typically show high protein abundance, have a limited contribution in the sum. In addition, we applied the statistical tests on 14 subsets of GO annotations based on the number of annotated proteins. This analysis was also performed on 5 alternative hierarchical clustering results: This allows us to evaluate the robustness of the results to a change of clustering algorithm (Maximum likelihood clustering, Euclidean metric with complete linkage, Correlation metric with complete linkage), and the usage of the cell confidence (as a weight or using 0.8 as a filtering threshold) (Suppl. Table S2, S3, S4, S5).
Hierarchical clusters are available to be browsed online at: http://www.moseslab.csb.utoronto.ca/louis-f/unsupervised/. In addition, the source code for the cell identification and feature measurements, the set of 17 images in which 4305 ellipses corresponding to cells and 139 ellipses corresponding to artifacts were manually drawn, as well as a table of feature measurements for all 400 K mother-buds pairs are available.
|
10.1371/journal.pgen.1007889 | Integrating predicted transcriptome from multiple tissues improves association detection | Integration of genome-wide association studies (GWAS) and expression quantitative trait loci (eQTL) studies is needed to improve our understanding of the biological mechanisms underlying GWAS hits, and our ability to identify therapeutic targets. Gene-level association methods such as PrediXcan can prioritize candidate targets. However, limited eQTL sample sizes and absence of relevant developmental and disease context restrict our ability to detect associations. Here we propose an efficient statistical method (MultiXcan) that leverages the substantial sharing of eQTLs across tissues and contexts to improve our ability to identify potential target genes. MultiXcan integrates evidence across multiple panels using multivariate regression, which naturally takes into account the correlation structure. We apply our method to simulated and real traits from the UK Biobank and show that, in realistic settings, we can detect a larger set of significantly associated genes than using each panel separately. To improve applicability, we developed a summary result-based extension called S-MultiXcan, which we show yields highly concordant results with the individual level version when LD is well matched. Our multivariate model-based approach allowed us to use the individual level results as a gold standard to calibrate and develop a robust implementation of the summary-based extension. Results from our analysis as well as software and necessary resources to apply our method are publicly available.
| We develop a new method, MultiXcan, to test the mediating role of gene expression variation on complex traits, integrating information available across multiple tissue studies. We show this approach has higher power than traditional single-tissue methods. We extend this method to use only summary-statistics from public GWAS. We apply these methods to 222 complex traits available in the UK Biobank cohort, and 109 complex traits from public GWAS and discuss the findings.
| Recent technological advances allow interrogation of the genome to a high level of coverage and precision, enabling experimental studies that query the effect of genotype on both complex and molecular traits. Among these, GWAS have successfully associated genetic loci to human complex traits. GWAS meta-analyses with ever increasing sample sizes allow the detection of associated variants with smaller effect sizes [1–3]. However, understanding the mechanism underlying these associations remains a challenging problem.
Another approach is the study of expression quantitative trait loci (eQTLs), measuring association between genotype and gene expression. These studies provide a wealth of biological information but tend to have smaller sample sizes. A similar observation applies to QTL studies of other traits such methylation, metabolites, or protein levels.
The importance of gene expression regulation in complex traits [4–7] has motivated the development of methods to integrate eQTL studies and GWAS. To examine these mechanisms we developed PrediXcan [8], which tests the mediating role of gene expression variation in complex traits. Briefly, PrediXcan tests the hypothesis that genetic variants affect phenotypes through the regulation of gene expression traits. To do that, it correlates genetically predicted gene expression and the phenotype with the idea that causal genes are likely to show a significant association. Linear prediction models of expression using genetic variation in the vicinity of the gene are trained in reference transcriptome datasets such as Genotype-Tissue Expression project (GTEx) [9].
Due to sharing of eQTLs across multiple tissues, we have shown the benefits of an agnostic scanning across all available tissues [10]. Despite the increased multiple testing burden (for Bonferroni correction, the total number of gene-tissue pairs must be used when determining the threshold), we gain considerably in number of significant genes. However, given the substantial correlation between different tissues [9], Bonferroni correction can be too stringent increasing the false negative rate.
In order to aggregate evidence more efficiently, we present here a method termed MultiXcan, which tests the joint effects of gene expression variation from different tissues. Furthermore, we develop and implement a method that only needs summary statistics from a GWAS: Summary-MultiXcan (S-MultiXcan). We make our implementation publicly available to the research community in https://github.com/hakyimlab/MetaXcan. We apply this method to simulated and real data (222 traits from the UK Biobank study [11] and 109 public GWAS) to show the performance and proper calibration of p-values. We make all of the results publicly available at https://doi.org/10.5281/zenodo.1402225.
To integrate information across tissues, MultiXcan regresses the phenotype of interest on the predicted expression of a gene in multiple tissues as follows:
y = μ + t 1 g 1 + t 2 g 2 + ⋯ + t p g p + e (1)
where y is the n-dimensional phenotype vector, μ is an intercept term, ti is standardized predicted expression of the gene in tissue i, gi is its effect size, and e an error term with variance σ e 2; p is the number of available tissue models. We use an F-test to assess the joint significance of the regression.
Expression predictions across tissues can be highly correlated. We predicted expression for individuals from the UK Biobank cohort using models trained on 44 GTEx tissues (as presented in [10]), and found a median pair-wise correlation of rp50 = 0.56 (IQR = 0.69) between different tissue models in a given gene, across genes (see Methods for details). To avoid numerical issues caused by collinearity, we use principal components of the predicted expression data matrix as explanatory variables, and discard the axes of smallest variation (PCA regularization). Additional covariates can be added to the regression seamlessly. Fig 1-a displays an overview of the method; see further details in the Methods section. S1 Fig shows an example of the correlation between tissues of predicted expression of the gene SLC5A6.
We applied MultiXcan to 222 traits from the UK Biobank cohort. The traits were chosen based on several criteria, such as availability of well-established literature, binary traits having enough cases, or potential interest for a phenome-wide study (allergy, behavioral, metabolic and anthropometric phenotypes). We used Elastic Net prediction models trained on 44 tissues from GTEx, originally presented in [10].
We compared three approaches for assessing the significance of a gene jointly across all tissues: 1) running PrediXcan using the most relevant tissue; 2) running PrediXcan using all tissues, one tissue at a time; 3) running MultiXcan. Fig 1-b illustrates the results from each approach. We summarize a comparison between approaches 2) and 3) in Table 1. PrediXcan overcomes MultiXcan only in 21 traits, all of them with less than 50 significant associations across both methods. MultiXcan detects more associations in 103 traits.
Fig 2-a and 2-b show a comparison of detections for both MultiXcan and PrediXcan. See S1 Dataset for a summary of detections per trait, and S2 and S3 Datasets for the full list of significant MultiXcan and PrediXcan results respectively.
As an illustrative example, we examined more closely the results for self-reported high cholesterol phenotype (http://biobank.ctsu.ox.ac.uk/crystal/field.cgi?id=20002). We used 50,497 cases and 100,994 controls. After Bonferroni correction, MultiXcan was able to detect a larger number of significantly associated genes (251 detections) than PrediXcan using all tissues (196 detections) or only a single tissue (whole blood, 33 detections). 172 genes were detected by both PrediXcan and MultiXcan. Fig 2-c shows the QQ-plot for associations in these three approaches.
There are 79 genes associated to high cholesterol via MultiXcan and not PrediXcan. Among them, we find genes related to lipid metabolism (APOM [12], PAFAH1B2 [13]), glucose transport(SLC5A6 [14]), and vascular processes (NOTCH4 [15]). The well known gene SORT1 is detected by both MultiXcan and PrediXcan.
To evaluate MultiXcan’s performance in different known scenarios, we simulated traits as a function of different numbers of causal tissues for each gene: a single tissue, multiple tissues, all available tissues. We executed PrediXcan, MultiXcan without PCA regularization, and MultiXcan with PCA regularization. We show proper calibration under the null hypothesis of no association in S3 Fig, and robustness of the regularization approach in S6 Fig. See further details in S1 Supplementary Note.
As expected, when there is a known single causal tissue, PrediXcan with the known tissue yields more significant associations. However, when there are multiple causal tissues, MultiXcan yields more significant associations than the best single tissue PrediXcan results. In traits simulated from a single causal tissue, PrediXcan outperforms MultiXcan in 99.9% of the cases (AOV p-value < 10−16). MultiXcan performs best in scenarios with multiple causal tissues (84.4% of the times when a few tissues are causal, and 99.5% when all tissues are causal; AOV p-value < 10−16 in both cases).
One caveat is that the simulation does not cover cases when the prediction in the single tissue has low quality. In such an scenario, borrowing information from other tissues will still be beneficial.
To expand the applicability of our method to massive sample sizes and to studies where individual level data are not available, we extend our method to use summary results rather than individual-level data. We call this extension Summary-MultiXcan (S-MultiXcan).
We infer the joint estimates of effect sizes of predicted expression on phenotype (Eq 1) using the marginal estimates. We also compute the covariance matrix of the effect sizes and leverage the asymptotic multivariate normality of the estimates, to compute a statistic that is approximately χ p 2 (p number of tissues). The final expression is equivalent to the omnibus test mentioned in [16], which can be interpreted as a specific case of general weighted association analysis [17]. Fig 3-a illustrates our approach and the details can be found in the Methods section.
As with the individual level approach, the correlation between tissues leads to numerical problems (due to near singular covariance matrices that need to be inverted). We address this by using a pseudo inverse approach which, in a nutshell, uses singular value decomposition (SVD) of the covariance matrix to keep only the components of large variation. This is analogous to the PCA regularization used for the individual level approach. Thus we test for significance using χ k 2 with k the number of surviving components. See details in the Methods Section.
A robust implementation for calculating predicted expression correlation is critical to avoid unnecessary false positive results. In principle, it is possible to simply calculate the correlation between tissues using predicted expression in a reference set. However, we found that this approach can lead to large differences between the individual level data results (our gold standard) and the summary level ones when SNPs from the reference set are missing in the GWAS results. An example of this is shown in S8 Fig with the Type 1 Diabetes study from the Wellcome Trust Case-Control Consortium (WTCCC); association data is included in S9 Dataset. To avoid this problem, we calculate the covariance matrix between tissues using only the predictor SNPs that are common in both the GWAS summary and the reference LD set.
Fig 4 displays a few examples of the general agreement between the individual-level MultiXcan and S-MultiXcan. The summary-based version’s results tend to be slightly more conservative than MultiXcan, as illustrated in S2 Fig. As a general comparison to the individual-level method, we list a summary of S-MultiXcan’s application to the 222 UK Biobank traits on Table 2; we observe an adequate similarity between S-MultiXcan’s and MultiXcan’s summaries. The small loss in power arises from the imperfect match of LD between the UK cohort and the reference panel.
To reduce false positives due to LD misspecification when dealing with GWAS summary statistics, we discard any significant association result for a gene if the best single tissue result has p-value greater than 10−4 (“suspicious associations”). In other words, we keep significant associations if at least one single gene-tissue pair association is borderline significant or better (10−5 is the Bonferroni threshold for a typical tissue model). This is rather conservative since it is possible that evidence with modest significance from weakly correlated tissues can lead to very significant combined association when their effects get aggregated. For example among Bonferroni significant genes in the individual level analysis, a median of 8.3% across traits (IQR = 5.7%) have the most significant marginal (PrediXcan) p-value greater than 10−4. We list the number of such genes for each of the 222 UK Biobank traits in S8 Dataset.
We applied S-MultiXcan to 109 traits on publicly available GWAS, chosen with a similar criteria as UK Biobank’s traits. Like the individual level method, we observed S-MultiXcan to detect more associations than S-PrediXcan in most cases (average detection increase 10), as shown in Fig 3-b, after discarding suspicious associations. We also show the QQ-plots for a sample trait (Schizophrenia) on Fig 3-c and the total number of associations across all public GWAS traits in 3-d.
We display a summarized comparison between S-MultiXcan and S-PrediXcan in S1 Table, after discarding suspicious associations. The list of analyzed traits can be found in S4 and S5 Datasets contains a summary of significant associations for each trait and for each method. S6 Dataset lists the significant S-MultiXcan results for each trait. These results have been uploaded to https://doi.org/10.5281/zenodo.1402225.
Motivated by the widespread sharing of regulatory processes across tissues [9], we propose MultiXcan, a method that aggregates information by jointly fitting the phenotype on predicted expression across multiple tissues. In simulations and real data, we show that our approach can detect more associations. To expand the applicability of our approach, we derive the analytical expression to infer the association using summary results only, which we show is approximately equivalent to the omnibus test. An important benefit of our multivariate approach is that we can use the individual level data as gold standard to calibrate the type and degree of regularization needed to invert the near singular covariance matrices found in practice. The availability of a gold standard also allowed to identify the need for robust estimates of correlations between tissues.
We found high concordance, in general, between the individual level and summary version with the latter slightly more conservative. As any method relying on a reference panel, S-MultiXcan may be inaccurate when the study population has a different LD structure than the reference panel. We attempted to address this by flagging results where none of the marginal associations reached a somewhat arbitrary threshold of 10−4. This is far from perfect. To take full advantage of summary results and summary-based methods, reference sets that are the closest to the study population should be used. This also stresses the need to generate representative reference LD datasets for a wide variety of populations.
Via simulations, we show that MultiXcan is properly calibrated under the null hypothesis of no associations. This is reassuring, but it is possible that in real data there are hidden confounders that we did not capture in our simulations. For example, significant association results might arise due to LD contamination, i.e. when causal variants for the trait and expression are different but in LD with each other, inducing a spurious correlation between the predicted expression and the trait. This is a complex problem that we are currently working to address. In Barbeira et al [10], we sought to address the LD contamination issue by adding a colocalization filtering step where we discard associations with low colocalization probability, using COLOC [31] to keep only associations with Pcolocalized > 0.5. A similar strategy may be applied for MultiXcan by restricting the analysis to gene-tissue pairs with high colocalization probability in the marginal analysis.
In practice, we emphasize the need to further validate the significant associations with additional replication and experimental follow-up.
Importantly, we provide compelling examples where using multiple tissues rather than picking one considered to be relevant for the phenotype increases the list of candidate causal genes. In our simulations, we found that only when the single causal tissue is known and the regulatory mechanism is captured perfectly by predicted expression in that tissue, using PrediXcan with that tissue yields more significant associations than MultiXcan. This scenario is unlikely to occur in practice. Therefore, in general, we recommend jointly scanning of all tissues in addition to focusing on a few tissues selected based on prior knowledge.
We make our software publicly available on a GitHub repository: https://github.com/hakyimlab/MetaXcan. Prediction model weights and covariances for different tissues can be downloaded from http://predictdb.org/. A short working example can be found on the GitHub page; more extensive documentation can be found on the project’s https://github.com/hakyimlab/MetaXcan/wiki. The results of S-MultiXcan applied to the 44 human tissues and a broad set of phenotypes can be queried on http://gene2pheno.org. The data used in this paper is publicly available in https://doi.org/10.5281/zenodo.1402225.
This study uses de-identified genotype and phenotype data from public repositories including dbGaP, EGA, and UK Biobank. Our study has been determined to be non-human subject research by the University of Chicago’s IRB protocol number IRB16-0921.
MultiXcan consists of fitting a linear regression of the phenotype on predicted expression from multiple tissue models jointly:
y = ∑ j = 1 p t j g j + e = T g + e (2)
where y is a centered vector of phenotypes for n individuals, tj is an n-vector of standardized predicted gene expression for model j, gj is the effect size for the predicted gene expression j, e is an error term with variance σ e 2, and p is the number of tissues. Thus, T is a data matrix where each column j contains the values from tj, and g is the p-vector of effect sizes gj.
The high degree of eQTL sharing between different tissues induces a high correlation between predicted expression levels. In order to avoid collinearity issues and numerical instability, we decompose the predicted expression matrix into principal components and keep only the eigenvectors of non negligible variance. To select the number of components, we used a condition number threshold of λ max λ i < 30, where λi is an eigenvalue of the matrix Tt T. As a side effect, we observe moderate increases in significance levels because less informative components of tissue expression are discarded from the model. A range of values between 10 and 100 yielded similar results in the simulations described in S1 Supplementary Note as displayed in S6 Fig.
Lastly, we use an F-test to quantify the significance of the joint fit.
We use Bonferroni correction to determine the significance threshold. For MultiXcan, we use the total number of genes with a prediction model in at least one tissue, which yields a threshold approximately at 0.05/17500 ∼ 2.9 × 10−6. For PrediXcan across all tissues, we use the total number of gene-tissue pairs, which yields a threshold approximately at 0.05/200, 000 ∼ 2.5 × 10−7. Since the tested hypotheses are not independent, Bonferroni correction is overly conservative, as can be seen when counting the number of associations via FDR in S7 Fig.
We have demonstrated that S-PrediXcan can accurately infer PrediXcan results from GWAS Summary Statistics and LD information from a reference panel [10], with the added benefits of reduced computational and regulatory burden. Here we extend MultiXcan in a similar fashion.
Summary-MultiXcan (S-MultiXcan) infers the individual-level MultiXcan results, using univariate S-PrediXcan results and LD information from a reference panel. It consists of the following steps:
Prediction Models were obtained from http://predictdb.org/ resource. These models were trained using Elastic Net as implemented in R’s package glmnet [38], with a mixing parameter α = 0.5, on 44 tissue studies from GTEx’ release version 6p. The underlying GTEx study data was obtained from dbGaP with accesion number phs000424.v6.p1. Please see [10] for details. We implemented MultiXcan and S-MultiXcan using python scientific packages, working up from existing software in the MetaXcan package. S-PrediXcan, PrediXcan, MultiXcan and S-MultiXcan analysis were computed using the Center for Research Informatics’ high performance cluster at the University of Chicago. PrediXcan, S-PrediXcan, MultiXcan and S-MultiXcan results have been uploaded to the http://gene2pheno.org resources. The databases are open to the research community for arbitrary programmatic query.
|
10.1371/journal.pcbi.1003912 | Bistable Forespore Engulfment in Bacillus subtilis by a Zipper Mechanism in Absence of the Cell Wall | To survive starvation, the bacterium Bacillus subtilis forms durable spores. The initial step of sporulation is asymmetric cell division, leading to a large mother-cell and a small forespore compartment. After division is completed and the dividing septum is thinned, the mother cell engulfs the forespore in a slow process based on cell-wall degradation and synthesis. However, recently a new cell-wall independent mechanism was shown to significantly contribute, which can even lead to fast engulfment in 60 of the cases when the cell wall is completely removed. In this backup mechanism, strong ligand-receptor binding between mother-cell protein SpoIIIAH and forespore-protein SpoIIQ leads to zipper-like engulfment, but quantitative understanding is missing. In our work, we combined fluorescence image analysis and stochastic Langevin simulations of the fluctuating membrane to investigate the origin of fast bistable engulfment in absence of the cell wall. Our cell morphologies compare favorably with experimental time-lapse microscopy, with engulfment sensitive to the number of SpoIIQ-SpoIIIAH bonds in a threshold-like manner. By systematic exploration of model parameters, we predict regions of osmotic pressure and membrane-surface tension that produce successful engulfment. Indeed, decreasing the medium osmolarity in experiments prevents engulfment in line with our predictions. Forespore engulfment may thus not only be an ideal model system to study decision-making in single cells, but its biophysical principles are likely applicable to engulfment in other cell types, e.g. during phagocytosis in eukaryotes.
| When the bacterium B. subtilis runs out of food, it undergoes a fundamental development process by which it forms durable spores. Sporulation is initiated by asymmetric cell division after which the larger mother cell engulfs the smaller forespore, followed by spore maturation and release. This survival strategy is so robust that engulfment even proceeds when cells are deprived of their protective cell wall. Under these severe perturbations, 60 of the mother cells still engulf their forespores in only 10 of the normal engulfment time, while the remaining 40 of mother cells withdraw from engulfment. This all-or-none outcome of engulfment suggests decision-making, which was recently also identified in other types of engulfment, e.g. during phagocytosis when immune cells engulf and destroy pathogens. Here, we developed a biophysical model to explain fast bistable forespore engulfment in absence of the cell wall and energy sources. Our discovered principles may prove very general, thus predicting key ingredients of successful engulfment across all kingdoms of life.
| To survive starvation the Gram-positive bacterium Bacillus subtilis develops durable spores among other survival strategies [1]. During sporulation, bacteria go through a costly developmental process under limited energy resources. The initial morphological step of sporulation is asymmetric cell division, resulting in a large mother-cell and a small forespore compartment [2]. Subsequently, the dividing septum is largely degraded and the mother-cell membrane moves around the forespore. This membrane movement is similar to phagocytosis whereby immune cells clear our bodies from pathogens and other particles [3], [4]. Finally, the engulfed forespore matures into a spore and the mother cell lyzes for its release. The origin of the engulfment force has been a topic of current research [5]–[11]. Cell-wall degradation and new cell-wall deposition were shown to play a significant role in advancing the mother-cell membrane leading edge. Strikingly, when the cell wall is enzymatically removed engulfment still occurs, surprisingly taking only 1–2 min compared to 45 min with the cell wall (see Fig. 1, Movie S1) [8]. Furthermore, engulfment is successful in 60 of cells while the remaining 40 retract. This observation raises questions on the origin of bistability and decision-making in relatively simple systems under severe energy limitations.
In the absence of the cell wall, migration of the mother-cell membrane around the forespore depends on the two membrane proteins that bind each other with high affinity [12], constituting a backup mechanism under severe perturbations [8], [13]: SpoIIQ expressed in the forespore and SpoIIIAH expressed in the mother cell [8], [14], [15](see Fig. 1A–D). To facilitate engulfment a physical mechanism similar to a Brownian ratchet was proposed [8]. Specifically, thermal fluctuations move the leading membrane edge forward, thus establishing new SpoIIQ-SpoIIIAH bonds that prevent backward membrane movement. One striking feature, however, is that the membrane cup surrounding the forespore is very thin (Fig. 1A, top). This either indicates a fast nonequilibrium mechanism for engulfment or additional forces that produce high membrane curvatures around the cup's neck region. Even though modeling of similar processes such as membrane budding and phagocytosis helped us understand the role of physical constraints on engulfment [16]–[19], quantitative modeling of forespore engulfment as a fundamental development process is still missing.
Here, using image analysis, Langevin simulations and simple analytical approaches we show that fast forespore engulfment in the absence of the cell wall occurs below 1 min, consistent with out-of-equilibrium dynamics driven by strong SpoIIQ-SpoIIIAH binding. Furthermore, we find physical parameter regimes responsible for bistable engulfment, including the number of bonds necessary for threshold-like engulfment and suitable osmotic pressures. The former prediction matches previously published data, while we successfully tested the latter with time-lapse microscopy. Hence, our model makes testable predictions on the measurable physical parameters leading to fast, energy-efficient engulfment. Forespore engulfment in the absence of the cell wall is thus an ideal system to study phagocytosis-like processes and decision-making in single-cell organisms.
To better understand the process of engulfment in the absence of the cell wall, we analyzed the volume and surface area of sporulating cells treated with cell-wall removal enzyme (lysozyme) from previously published data [8]. For this purpose we used the semi-automated image-analysis software JFilament [20] (see Materials and Methods). Briefly, JFilament software allows assisted manipulation of active contours that comply with the bright linear structures of the images. Therefore, membranes that were fluorescently stained with FM 4–64 were easily tracked over time. Information about cell-membrane position in the medial focal plane was used to calculate volume and surface area assuming rotational symmetry around the axis connecting mother and forespore center of mass.
Upon cell-wall removal, a drastic volume loss of 35 in engulfing mother cells was observed (see Fig 2A). The onset of volume loss for each cell was set to 0 min, and surface area and engulfment measurements were aligned in time based on this time point. Similar analyses showed that no changes in forespore volume were observed (Fig. 2A). However, surface-area analysis during engulfment did reveal a minor reduction of the mother-cell surface by 5–15 (Fig. 2B), depending on the assumed shapes of the progressing membrane in the image analysis. To correlate the drastic volume loss with the onset of engulfment we created kymographs along the fluorescently labeled forespore membranes of the engulfing cells (Fig. 2C and D). Extracting from the kymographs the percentage of forespore engulfment over time revealed that engulfment in the absence of the cell wall occurs on a time scale of 1–2 min as opposed to 45 min with the cell wall [8], [10].
Calculating cross-correlations we found significant anticorrelations between volume and engulfment, and surface area and engulfment (Fig. 2D). No time delays were detected (the minimum of the cross-correlation function is at 0 min) at a sampling rate of 1.3 frame/min (for details see Materials and Methods). However, under the same experimental conditions no volume or surface-area losses were observed in cells lacking SpoIIQ or SpoIIIAH zipper-molecules after cell-wall removal (see Fig. S1). Therefore, this led us to conclude that the drastic volume loss may play a facilitating role in engulfment.
During fast engulfment the mother-cell membrane drastically changes shape. Based on fluorescence intensity measurements, a very thin and tight cup was proposed at the late stage of engulfment 20 min after cell-wall removal (Fig. 3A) [8]. However, this tight cup amounts to very high membrane curvatures. Therefore, we wondered how the thin cup with high membrane-bending energy may emerge from an initial broad cup, as observed in cells prior to cell-wall removal (Fig. 1A, top). To answer this question, we analyzed fluorescence images in two channels: fluorescent FM 4–64 (red) that uniformly binds to all membranes exposed to the medium, and SpoIIIJ-GFP (green), a labeled protein that is only expressed in the mother cell where it is uniformly recruited to the cell membrane [8].
In Fig. 3A the tight-cup model predicts four membrane folds at the mother-forespore boundary in the FM 4–64 channel and three membrane folds in the GFP channel. Likewise, the broad cup model predicts two membrane folds in the FM 4–64 channel and a single membrane in the GFP channel. To quantify the transition from a broad to the proposed thin cup we measured the average intensity of the mother-forespore boundary () and mother-cell membrane intensity far from the boundary (, see Fig. 3A). Ratio is then used to determine the number of membrane folds on the mother-forespore interface. Relative boundary analysis of FM 4–64 intensity revealed that during late-stage engulfment cup shape conforms in between a broad and tight cup with 3.5 membrane folds on the mother-forespore interface. Similarly, analysis of the noisy GFP channel (Fig. 3C, and Movie S2) suggests that a broad cup indeed undergoes a transition towards a thin cup through significant morphological changes at the onset of engulfment. To quantitatively explain the observed volume loss, membrane morphological changes, and fast bistable engulfment, we implemented a biophysical model of forespore engulfment using Langevin dynamics.
We hypothesized that thin cups form due to fast nonadiabatic engulfment away from equilibrium. If, however, it turns out that the engulfment dynamics are not fast enough under biophysical membrane constraints, then additional forces would need to be postulated to produce such high-curvature membrane features. To test our hypothesis we used Langevin dynamics that account for out-of equilibrium processes.
In our model the 3D mother-cell membrane is represented by a string of beads assuming rotational symmetry around the -axis, while the forespore membrane was modeled as a hard sphere (Fig. 4). Indeed, experiments show negligible deformation of the forespore during engulfment (Fig. 2A and B). Specifically, the Langevin dynamic equation of the bead at position is given by:
(1)where each bead at the position = (, ) represents a ribbon of width and length shown in Fig. 4A. The left-hand side of Eq. 1 depends on the drag coefficient [21], with is the effective medium viscosity (see Text S1). On the right-hand side of Eq. 1 we have contributions of membrane bending, stochastic thermal fluctuations, zipper-molecule binding, surface tension, and osmotic pressure. For each term we give a brief description, while the detailed model equations and analytical derivations can be found in Materials and Methods and Text S1, respectively.
The membrane-bending force () restores the curved membrane to the equilibrium flat configuration. The stochastic term () is used to simulate thermal fluctuations with an effective temperature representing a driving force for the leading-edge ratchet movement. The amplitude of the membrane thermal fluctuations is chosen to be 15 nm (see Fig. S2) as typically observed for lipid bilayers [22] and red-blood cells [23]. The zipper-molecule force () accounts for the high binding affinity between SpoIIQ and SpoIIIAH [15]. This zipper mechanism is a strong driving force of the membrane leading edge. For the zipper-protein surface density () we initially chose the maximum possible value that corresponds to a single molecule per 100 nm2 of membrane (based on 10 nm for the size of the protein [14], [15] and assumed dense packing). The surface-tension force () is characterized by linear () and nonlinear () surface tensions (see Materials and Methods) [16]. The force due to osmotic pressure () is characterized by the pressure difference () between inner and outer medium. While the surface-tension term causes membrane contraction, the pressure-difference term produces volume expansion. In thermal equilibrium these two forces balance each other. An example of an implemented simulation is shown in Fig. 4B.
To validate our simulations we considered a spherical mother cell prior to engulfment, allowing us to quantify membrane fluctuations in thermal equilibrium. Specifically, we used the Langevin equation without the zipper term on the right-hand side of Eq. 1. Obtained numerical results were compared with analytical results of thermal membrane fluctuations [24], [25] and fluctuation spectra [26] (Fig. S3). This validation showed that our 3D model of the mother cell has indeed appropriate membrane biophysical properties. For further details see Materials and Methods.
Using the model governed by Eq. 1 we numerically simulated forespore engulfment in real time (Fig. 5, Movie S3). At the beginning of the simulations mother cell and forespore have spherical shape and intersect at a single point of their perimeters as shown in Fig. 4A. To better understand contributions of linear surface tension () and pressure difference () on engulfment we varied them while keeping other parameters constant (for parameters see Text S1). We note that the explored surface tension ( 100 pN/µm) is smaller than the experimentally observed rupture tension ( 20 nN/µm) [27], [28] and that explored pressure differences ( 1000 Pa) are suitable for an osmotically balanced medium (see Materials and Methods).
In Fig. 5A we show simulation snapshots at s. Simulations that reached full engulfment earlier than 5 s were terminated and last snapshots were displayed. In Fig. 5B the white dashed line separates successful engulfment ( 575 Pa) from retraction ( 575 Pa). In the region of successful engulfment with 200 Pa and 30 pN/m we observed engulfment with thin, tight cups as experimentally observed (Fig. 5A and B). In the region between retraction and thin cups we also observed successful engulfment but with adiabatic broad cups resulting in almost spherical mother cells.
In the region of thin cups engulfment is fast, taking only 1–2 s which is even faster than 1–2 min from experiments (Fig. 2). We attributed this time discrepancy to the limiting factor of cell-wall removal in the experimental setup, since residues of the cell wall can prevent significant membrane fluctuations in our simulations, therefore delaying engulfment (Fig. S4). Indeed, movies of engulfment show elongated cells even minutes after addition of lysozyme [8]. For example, in Fig. 1A, top, at min (the onset of volume loss) the engulfing mother cell still has elongated shape even though this is min after lysozyme treatment. To explore possible simulated engulfment times we varied the kinematic parameter that represents the effective medium viscosity. This parameter sets the time scale of engulfment but does not influence the morphology of the engulfing cup (Fig. S5). We found that even for extremely high, experimentally observed , simulated engulfment times are still about an order of magnitude smaller than the experimentally observed engulfment times (Fig. S5).
To better understand physical changes of the simulated mother cell during engulfment, we plotted volume and surface area relative to their initial values (Fig. 5C and D). In the parameter region of successful engulfment we observed a mother-cell volume loss of 0.2 µm3, which corresponds to the forespore volume. However, the experimentally observed mother-cell volume loss (Fig. 2A) is about 2 times higher. Furthermore, simulated surface areas increase in this region, while experimentally observed surface areas decrease slightly (Fig. 2B). This discrepancy may be explained by effects not included in our simulations, e.g. cytosol leakage (see Discussion).
Experiments showed that engulfment depends critically on the number of expressed SpoIIQ zipper protein in a threshold-like fashion (Fig. 6) [8]; for SpoIIQ expression levels below a critical value mother-cell membrane retracts while for expression levels above the critical value the forespore is successfully engulfed (Fig. 6C). For wild-type cells with naturally occurring expression-level variation this presumably leads to the observed 60/40 bistable engulfment outcome [8].
To numerically determine the number of SpoIIQ proteins required for successful engulfment, we varied the protein-surface concentration () for model parameters = 50 pN/µm and = 500 Pa (Fig. 6A and B). Since in experiments the expression of SpoIIQ was reduced compared to wild-type cells, we assumed that the pool of SpoIIIAH molecules is large compared to SpoIIQ. Therefore, the limiting factor in engulfment is attributed to the surface concentration of SpoIIQ proteins. In Fig. 6A successful engulfment occurs for SpoIIQ surface densities above 3350 µm−2. Since forespore surface area is 2 µm2 (Fig. 2B), this produces a lower bound of 6700 SpoIIQ molecules in the forespore necessary for engulfment. To further explore the role of physical constraints on the number of critical zipper proteins that lead to successful engulfment, we scanned the parameter space for successful engulfment in Fig. 5B and found lower bounds on the number of molecules indispensable for engulfment (Fig. 6D). This lower bound can be as low as 120 SpoIIQ molecules for 30 pN/µm and 200 Pa and increases to 7200 SpoIIQ molecules for 550 Pa. Therefore, our simulations predict that under certain osmotic conditions and surface tensions, engulfment can occur with only 100 SpoIIQ molecules.
Our model predicts that high osmotic pressure differences across the mother-cell membrane lead to swelling, reduced membrane fluctuations and membrane retraction, therefore preventing the completion of engulfment (Fig. 5A and B). To assess the validity of these predictions, we followed the progression of engulfment in absence of the cell wall for different osmotic conditions by time-lapse fluorescence microscopy (see Materials and Methods). Briefly, we resuspended sporulating cells in SMM buffer, either with 0.5 M sucrose to support protoplast engulfment (Fig. 1A, Movie S1) [8] or without sucrose, leading to an increase in the osmotic pressure difference between the cytoplasm and the extracellular medium. We then stained cell membranes with FM 4–64, added lysozyme to remove the cell wall, and performed time-lapse microscopy as the cell wall was degraded (Fig. 7A). In the presence of 0.5 M sucrose, more than 50 of the cells completed engulfment as previously reported (Fig. 7B, Movie S1; [8]). However, in absence of sucrose the engulfing membrane retracted for almost all cells, and less than 2 of them completed engulfment (Fig. 7A and B, Movie S4). Furthermore, membrane retraction was accompanied by an increase in mother-cell volume and surface area of 30 and 5 , respectively (Fig. 7D and E). Hence, our data confirm the predictions of our biophysical model on the response to changes in medium osmolarity.
In this work we presented image analysis and modeling of forespore engulfment in the absence of the cell wall. Image analysis showed that engulfment occurs extremely fast ( 1–2 min compared to 45 min with the cell wall), accompanied by a drastic volume loss ( 35 ) of the mother cell. During engulfment the initial broad cup dynamically changes to a thin cup, forming high-curvature membrane folds at the intersection between mother cell and forespore. Using Langevin simulations we showed that a Brownian ratchet model reproduces fast, out-of-equilibrium engulfment. Additionally, we numerically determined regions of engulfment and retraction, and predicted the number of SpoIIQ molecules necessary for successful engulfment. Similar, out-of-equilibrium Brownian ratchet mechanisms were previously used to explain molecular-motor directional movement [29], Lysteria motility from actin comets [30], [31], filapodia protrusion [32], and unidirectional movement of other microscopic objects [33].
Our model makes a number of predictions. The phase diagram in Fig. 5B predicts engulfment success and mother-cell morphology for a wide range of surface tensions() and pressure differences (). For example, for a given SpoIIQ surface density, high surface tension restricts the engulfment region while high osmotic pressure prevents engulfment. To test this prediction we increased the osmotic pressure difference by lowering the osmolarity of the suspended buffer. Decreased osmolarity indeed caused mother-cell swelling and stopped engulfment in line with our predictions (Fig. 7). Another prediction, such as the need for an excess of mother-cell membrane, might be tested by controlling the production of FapR, the major lipid homeostasis regulator [34], [35].
There are a number of model limitations, which raise interesting issues. Based on our current model there is no stabilizing force for maintaining high curvatures (thin cups) once engulfment is accomplished. Therefore, the restoring constrain forces () and bending force () will flatten high curvatures within 1 s on the length scale of 1 µm (see Fig. S3C and Eq. 11). To estimate the force necessary to prevent membrane flattening we measured restoring forces in the neck region after engulfment is completed (Fig. S6). Typical radial forces were about 10 pN. Three factors can contribute to experimentally observed “snowman-like” shapes at late stages of engulfment (Fig. 1A, top): residues of the mother cell wall, potential outer/inner membrane binding, and lipid/protein sorting. These factors could be investigated experimentally. First, fluorescent cell-wall labeling [36], [37] could rule out possible remnants of cell wall after lysozyme treatment, which may preserve high membrane curvatures (see also Fig. S4). Second, thin cups could also form if an unknown cohesive factor binds to outer and inner cup membrane thus preventing their separation. Third, membrane lipids that localize to high negative curvatures [38]–[40] together with curvature sensing proteins such as SpoVM [41] could contribute to the formation of stable structures preventing membrane flattening. To better explore this stabilizing mechanism we simulated engulfment with membranes that have positive and negative intrinsic curvatures (Fig. S7). We found that successful engulfment proceeds in range (−120 20) µm−1. Interestingly, high negative curvatures prevent membrane flattening by forming tight and thin cups in the neck region typical for the snowman shape (Fig. S7, bottom, left). Therefore, the stability of tight cups after engulfment completion should be a topic of future experimental investigation, leading to better insights and advancement in modeling.
Our current model fell short in reproducing the exact volume loss and surface-area conservation. To numerically find parameter regimes that lead to thin cups and observed volume and surface-area changes, we constructed a new albeit less physically grounded model with explicit volume and surface-area constraints (see Fig. S8 and Text S1). Using this model we determined the parameter region in which experimentally observed volume and surface-area changes occur similar to Fig. 5. As a result, engulfment occurs when either volume or surface area is not conserved as previously proposed [42] (see Movie S3). Volume loss may contribute to engulfment by effectively decreasing the surface tension, therefore boosting excess membrane for forespore engulfment. However, by close inspection of sporulating cells, we observed leakage of the cytosol in one of the cells after membrane removal at the onset of volume loss (Fig. S9). This direct loss of cytosol was not explicitly included in our models. One plausible interpretation of volume loss can be attributed to the hypertonic solution of suspended buffer containing 0.5 M of sucrose and producing an osmotic pressure of 12 atm [8]. Although this pressure is comparable to osmotic pressures inside of bacteria to balance pressures across the membrane, water leakage from bacteria and/or partial lyses after cell-wall removal is expected [43], [44]. Therefore, future experiments may help address this issue.
An important future goal is the theoretical understanding of forespore engulfment in the presence of the cell wall, including membrane fission as the last stage of engulfment [45]. High turgor pressure and constraints from the cell wall and septum must provide difficult constraints for the engulfing mother cell, partially explaining the long engulfment time. In the presence of the cell wall, it has been proposed that peptidoglycan hydrolysis and new cell-wall deposition play major roles in the leading-edge membrane movement around the forespore [5]–[7], [9]–[11]. First, membrane proteins SpoIID, SpoIIM, and SpoIIP (DMP) form a complex and localize to the leading edge of the moving membrane. Since SpoIID and SpoIIP degrade peptidoglycans [5], [46] and play an important role for thinning the septum, it has been proposed that the DMP complex is a processive motor for membrane advancement. However, a mechanistic description of this motor is still missing. Second, new cell-wall deposition at the leading edge may provide an additional motor-like mechanism for membrane movement [10], [47]. A similar mechanism was proposed for cytokinesis of fission yeast Schizosaccharomyces pombe, where polymerizing septum fibrils contribute to inward septum ingression. Using Brownian ratchet modeling of this process it was estimated that a single -glucan fibril can exert polymerization force of 10 pN [48]. Together, future modeling will produce a better understanding of the complicated process of forespore engulfment and decipher the contributions from each mechanism towards the total force at the leading membrane edge.
In conclusion, our quantitative model of engulfment in absence of the cell wall provided first insights into mother-cell morphologies, such as cup shape, engulfment dynamics, and bistability. Due to simplicity from cell-wall removal, energy limitations, and absence of cytoskeletal cortex, forespore engulfment could present a minimal system for studying bistability and decision-making. Interestingly, bistability and commitment to engulf were previously identified in phagocytosis [18], [49], thus showing similarities to forespore engulfment. This similarity becomes enhanced, if we speculate that the SpoIIQ-SpoIIIAH backup mechanism may instead be the original core mechanism, which may have evolved before the more complex DMP-based mechanism. In fact, the C-terminal domain of SpoIIIAH is homologous to YscJ/FliF protein family forming multimeric rings in type-III secretion system and flagella motors [50], pointing towards an ancient mechanism. General biophysical principles may also apply to other types of engulfment including the penetration of the red blood cell by the malaria parasite [51].
We used the semi-automated active contour software JFilament [20] available as ImageJ plugin to extract the membrane position over time. All movies analyzed were previously published in [8]. The information about membrane positions obtained from medial focal plane is used to obtain 3D volume and surface area by assuming rotational symmetry around the axis connecting center of mass of mother cell and forespore. Kymographs as in Fig. 2C were created by collecting intensities along the forespore contours using JFilament. Subsequently, pixel angles were determined using pixel position relative to the mother-forespore frame as defined in inset of Fig. 2C. To test the image analysis method, which was used to estimate the number of membrane folds in the neck region of cup (Fig. 3), we measured fluorescence intensities along known single () and double () membranes (Fig. S10). This analysis produced = 2.3 0.6 as expected.
To compare two signals and that are given at discrete time points , we calculated the cross-correlation function:(2)
Here, and are the average signals, and is the total number of discrete time points.
A stochastic Langevin equation is used to simulate mother-cell membrane dynamics (see Eq. 1). Simulation time step and distance between neighboring beads were s and = 10 nm, respectively. When the distance between two neighboring beads exceeded the equilibrium distance by 25 the whole contour was rebeaded using a linear-interpolation method [20]. Additionally, membrane-excluding volume is also implemented; whenever the distance between two beads that are not nearest neighbors was less than a repelling radial force of 50 pN was applied to both beads. Analytical details of the model of Eq. 1 are explained in the following sections.
To validate our biophysical membrane model, we quantified thermal fluctuations from simulations and compared this with analytical results from [24]–[26] (see Fig. S3). Fourier mode with wave number is calculated as in [26]:(10)with the average membrane radius.
Fourier modes are collected during first 10 s of simulations. Autocorrrelation function () of each Fourier mode is calculated and fitted to exponential function () [56], where is the relaxation time for each Fourier mode that satisfies analytical expression [24], [25], [57]:(11)with the reduced membrane tension, the medium viscosity, and the bending stiffness.
To validate membrane shapes in thermal equilibrium, we collected 6000 simulated membrane contours and calculated the variance for each Fourier mode, also known as dimensionless spectrum. The analytical expression for dimensionless spectrum of planer membranes is given by [26]:(12)
Bacillus subtilis PY79 sporulation was induced by resuspension at 37°C similar to [58], except that the bacteria were grown in 25 LB prior to resuspension rather than in CH medium. Samples were taken two hours and forty-five minutes after resuspension, spun at 9000 rpm for 10 s, and resuspended either in 25 µl of SMM buffer (0.5 M sucrose, 20 mM maleic acid, 20 mM MgCl2, pH 6.5) or in the same buffer without sucrose. 10 µl of the resuspended culture were placed on a poly-L-lysine-treated coverslip and mixed with lysozyme and FM 4–64 (final concentrations 1 mg/ml and 5 µg/ml, respectively). Pictures were taken at room temperature, every 45 seconds for one hour, using an Applied Precision optical sectioning microscope equipped with a Photometrics CoolsnapHQ2 camera. Images were deconvolved and analyzed with SoftWoRx version 5.5 (Applied Precision) and ImageJ.
|
10.1371/journal.pgen.1003263 | Regulation of Drosophila Metamorphosis by Xenobiotic Response Regulators | Mammalian Nrf2-Keap1 and the homologous Drosophila CncC-dKeap1 protein complexes regulate both transcriptional responses to xenobiotic compounds as well as native cellular and developmental processes. The relationships between the functions of these proteins in xenobiotic responses and in development were unknown. We investigated the genes regulated by CncC and dKeap1 during development and the signal transduction pathways that modulate their functions. CncC and dKeap1 were enriched within the nuclei in many tissues, in contrast to the reported cytoplasmic localization of Keap1 and Nrf2 in cultured mammalian cells. CncC and dKeap1 occupied ecdysone-regulated early puffs on polytene chromosomes. Depletion of either CncC or dKeap1 in salivary glands selectively reduced early puff gene transcription. CncC and dKeap1 depletion in the prothoracic gland as well as cncCK6/K6 and dKeap1EY5/EY5 loss of function mutations in embryos reduced ecdysone-biosynthetic gene transcription. In contrast, dKeap1 depletion and the dKeap1EY5/EY5 loss of function mutation enhanced xenobiotic response gene transcription in larvae and embryos, respectively. Depletion of CncC or dKeap1 in the prothoracic gland delayed pupation by decreasing larval ecdysteroid levels. CncC depletion suppressed the premature pupation and developmental arrest caused by constitutive Ras signaling in the prothoracic gland; conversely, constitutive Ras signaling altered the loci occupied by CncC on polytene chromosomes and activated transcription of genes at these loci. The effects of CncC and dKeap1 on both ecdysone-biosynthetic and ecdysone-regulated gene transcription, and the roles of CncC in Ras signaling in the prothoracic gland, establish the functions of these proteins in the neuroendocrine axis that coordinates insect metamorphosis.
| Human Nrf2-Keap1 and the fruit fly CncC-dKeap1 protein complexes function both in response to foreign chemicals and in development. We found that CncC and dKeap1 control fruit fly development by regulating the production and actions of the principal hormone that controls the transformation of larvae into pupae. In hormone-responsive cells, CncC and dKeap1 bound to the genes that are activated by the hormone. When the amount of CncC or dKeap1 in these cells was reduced, the genes were not activated efficiently. When the amount of CncC or dKeap1 was reduced in the organ where the hormone is made, the genes whose products make the hormone were not activated efficiently. Because less hormone was made, it took longer for the larvae to turn into pupae, and the resulting pupae were bigger. Reduction of the amount of CncC intercepted previously identified signals for pupation. Nrf2 is required for the same signals to cause cancer in mice. The effects of CncC and dKeap1 both on genes that control hormone production and on genes that are switched on by the hormone in different organs indicate that they have multiple roles in the transformation of fruit fly larvae into pupae.
| Cellular responses to many xenobiotic compounds, including various toxins and pharmacological agents, are controlled by mammalian Nrf2 and Keap1, and by the homologous Drosophila CncC and dKeap1 proteins [1], [2], [3]. The Nrf2-Keap1 complex has multiple effects on carcinogenesis. Nrf2-deficient mice have increased susceptibility to chemical carcinogens, potentially because of defective activation of cytoprotective genes in response to carcinogen exposure [4]. Mutations in Nrf2 and Keap1 that are predicted to disrupt their interactions are found in many human cancers, suggesting that Nrf2 interactions with Keap1 counteract cancer progression [1], [5]. Conversely, the deletion of Nrf2 suppresses pancreatic and lung tumorigenesis in a mouse model with constitutively active K-RasG12D expression [6]. The mechanisms whereby Nrf2 promotes tumorigenesis in conjunction with K-RasG12D are not known. Nrf2 and Keap1 are investigated as potential targets for therapeutic interventions in cancer, neurodegenerative diseases and developmental disorders [1], [7].
Nrf2 (NF-E2-Related Factor 2) is a bZIP family transcription factor that can bind to genes whose transcription is induced by xenobiotic compounds [1]. Keap1 (Kelch-like ECH-Associated Protein 1) is a Kelch family protein that can interact with the N-terminal region of Nrf2, and inhibits the activation of many genes activated by Nrf2 [8]. Studies in cultured mammalian cells indicate that Keap1 is predominantly localized to the cytoplasm [9], where it promotes Nrf2 degradation and inhibits its accumulation in the nucleus [8], [10], [11], [12].
Studies of the Drosophila homologues of Nrf2 and Keap1 have provided insights into the functions of these protein families in adult flies. The Drosophila cap‘n’collar locus encodes CncC, which contains a bZIP domain homologous to that of Nrf2 and N-terminal DLG and ETGE motifs homologous to those that mediate Nrf2 interaction with Keap1 [13] (Figure 1A). Drosophila dKeap1 contains Kelch repeats homologous to those that mediate Keap1 interaction with Nrf2 as well as a sequence motif that is required for mammalian Keap1 export from the nucleus [3], [10]. Overexpression of CncC and depletion of dKeap1 in adult flies activates the transcription of many genes that protect cells from xenobiotic compounds, whereas dKeap1 overexpression represses their transcription, indicating that the functions of these protein families in the xenobiotic response are conserved between mammals and Drosophila [2], [3].
Several lines of evidence suggest that CncC and dKeap1 also affect cell proliferation and development. CncC overexpression and dKeap1 depletion inhibit intestinal stem cell proliferation, and counteract the proliferative effects of environmental stress in these cells [14]. Loss of function mutations in cncC and dKeap1 cause larval lethality [3], [15]. The genes regulated by CncC and dKeap1 during larval development had not been established. Elucidation of the relationship between CncC and dKeap1 functions in xenobiotic responses and in development is important to define how the transcription regulatory functions of CncC and dKeap1 are regulated in response to intrinsic and extrinsic stimuli.
In Drosophila and in other holometabolous insects, the onset of metamorphosis is triggered by an increase in the level of the endocrine hormone ecdysone [16], [17]. Ecdysone is synthesized in the prothoracic gland (PG) by a series of cytochrome P450 enzymes [18]. The expression of these ecdysone-biosynthetic genes and the timing of pupation are regulated by Ras signaling in response to prothoracicotropic hormone (PTTH) binding to the Torso receptor [19], [20]. Ecdysone facilitates the onset of metamorphosis by regulating transcription in many tissues, including the salivary glands where ecdysone-regulated transcription is manifest by puffs at specific polytene chromosome loci [21]. The transcription factors that bind to the ecdysone biosynthetic gene promoters and activate their transcription have remained unknown.
In the work presented here, we found that CncC and dKeap1 occupied the classical ecdysone-regulated puffs on polytene chromosomes. Depletion of CncC or of dKeap1 in salivary glands reduced ecdysone-regulated gene transcription. Depletion of CncC or of dKeap1 in the PG as well as cncC and dKeap1 loss of function mutations reduced ecdysone biosynthetic gene transcription in larvae and in embryos, respectively. The reduced ecdysteroid levels caused by CncC and by dKeap1 depletion in the PG delayed pupation and suppressed the premature pupation caused by constitutive Ras signaling. These observations establish roles for CncC and dKeap1 in transcriptional programs in different tissues that coordinate metamorphosis.
To investigate if the subcellular localization of CncC was regulated by dKeap1 in the manner that has been reported for mammalian Nrf2 and Keap1, we determined the distributions of CncC and dKeap1. Both CncC and dKeap1 immunoreactivity were predominantly nuclear in Drosophila salivary gland cells (Figure 1B, Figure S1A). Likewise, ectopic CncC and dKeap1 fused to fluorescent proteins were enriched within the nuclei of live salivary gland cells (Figure 1B, Figure S1A). CncC and dKeap1 were also present in the nuclei of prothoracic gland, imaginal disc and gut cells, though the proportions that were localized to the nucleus varied in different tissues (Figure 1C, Figure S1B). The intensity of anti-dKeap1 immunoreactivity was markedly reduced in dKeap1EY5/EY5 mutant larvae, and the bands corresponding to endogenous dKeap1 and CncC were not detected by immunoblotting of extracts from dKeap1EY5/EY5 and cncK6/K6 mutant larvae, demonstrating the specificity of these antibodies (Figure S1C, S1D). These observations establish that both endogenous as well as ectopically expressed CncC and dKeap1 were localized to the nuclei in many different tissues, in contrast to the predominantly cytoplasmic localization observed for Keap1 and Nrf2 in many cultured mammalian cell lines.
To establish if CncC and dKeap1 bound to specific chromatin loci, we visualized their occupancy on polytene chromosomes by immunostaining. Anti-CncC and anti-dKeap1 antibodies recognized overlapping sets of loci, including a majority of the classical ecdysone-regulated early puffs on polytene chromosomes (e.g. 2B, 74EF, 75B, 63F, and 25B) (Figure 1D). Anti-CncC antibodies also recognized several loci that were not detected by anti-dKeap1 antibodies (e.g. 22B and 97B) and vice versa (e.g. 50C and 94C). CncC and dKeap1 occupied many non-puff loci, and did not occupy all puffs, indicating that their occupancy was not controlled solely by chromatin decondensation. Ectopically expressed CncC and dKeap1 fusion proteins occupied loci that overlapped those occupied by endogenous CncC and dKeap1, though they also occupied additional loci (Figure 1D). Few other sequence-specific DNA binding proteins have been identified that bind to ecdysone-regulated puffs [22], [23], [24]. The overlapping sets of loci occupied by endogenous and ectopic CncC and dKeap1, as detected by several different antibodies, corroborate the specificity of CncC and dKeap1 binding at these loci.
To test if CncC and dKeap1 regulated transcription of the early puff genes that they occupied on polytene chromosomes, we investigated the effects of CncC as well as dKeap1 depletion in salivary glands on transcription of ecdysone-regulated genes. Expression of an shRNA that targets CncC [3] under the control of either the 71B-GAL4 or the Sgs3-GAL4 driver reduced the levels of almost all of the ecdysone-regulated early puff and glue gene transcripts examined (Figure 2A). In contrast, transcription of most of the late puff genes that were not prominently occupied by CncC or dKeap1 was not affected by CncC depletion (Figure 2A). 71B-GAL4 directs expression throughout salivary gland development and in imaginal discs [25]; Transcription directed by Sgs3-GAL4 is detected only in late 3rd instar salivary glands [22], establishing that the change in transcription of ecdysone-regulated genes was due to CncC depletion in salivary glands. Expression of a different shRNA that targets all Cnc isoforms also reduced the levels of all of the early puff and glue gene transcripts examined (Figure 2A). The cncC-RNAi transgene had no detectable effects on transcription in larvae that lacked a GAL4 driver (Figure S2A).
Expression of an shRNA that targets dKeap1 [3] under the control of the Sgs3-GAL4 driver also reduced the levels of almost all of the ecdysone-regulated early puff and glue gene transcripts examined, but had no effect on most of the late puff gene transcripts (Figure 2B). In contrast to the concordant effects of CncC and dKeap1 depletion on ecdysone-regulated early puff gene transcription, CncC versus dKeap1 depletion had opposite effects on transcription of the gstD1 and gstE1 xenobiotic response genes (Figure 2A, 2B) [3], [26].
To examine if CncC and dKeap1 depletion affected early puff gene transcription through indirect mechanisms, we measured the levels of ecdysone receptor subunit transcripts and ecdysteroids. CncC and dKeap1 depletion in the salivary glands had no effect on the levels of the ecdysone receptor (EcR) or ultraspiracle (usp) transcripts in the salivary glands (Figure 2A, 2B). CncC depletion in the salivary glands also had no effect on the level of 20-hydroxyecdysone (20E) in the larvae (Figure S2B). There was no detectable effect on the size or the morphology of the salivary glands, or on the time of pupation. CncC and dKeap1 therefore likely regulated transcription of the ecdysone-regulated genes directly by binding to these loci.
The effects of CncC and dKeap1 on ecdysone-regulated gene transcription in salivary glands, the arrested development of cncK6/K6 and dKeap1EY5/EY5 mutant larvae, and the presence of both CncC and dKeap1 in prothoracic gland nuclei prompted us to investigate their roles in ecdysone biosynthetic gene transcription. We investigated the effects of CncC and dKeap1 depletion in the prothoracic gland (PG) on ecdysone biosynthetic gene transcription. We measured the levels of the neverland (nvd), spookie (spok), phantom (phm), disembodied (dib), shadow (sad), and shade (shd) transcripts in the brain complexes of larvae that expressed the shRNA targeting CncC or dKeap1 in the PG. Expression of the shRNA targeting CncC under the control of either the 5015-GAL4 or the phm-GAL4 driver reduced the levels of all ecdysone biosynthetic gene transcripts that are expressed exclusively in the PG (Figure 3A, Figure S3A). 5015-GAL4 directs expression in the PG, the salivary glands and the lymph gland [27]; phm-GAL4 directs expression in the PG and at low levels in the wing and leg discs of 3rd instar larvae [28]. Expression of the shRNA targeting CncC also reduced Sad immunoreactivity in the PG (Figure 3C, Figure S3B). Expression of the shRNA targeting dKeap1 under the control of the phm-GAL4 driver reduced the levels of nvd, spok, phm, but not the levels of dib and sad in the brain complex (Figure 3A). CncC depletion in the PG therefore reduced transcription of all known ecdysone biosynthetic genes that are selectively expressed in the PG, and dKeap1 depletion reduced transcription of a subset of these genes.
To determine the specificity of the reduction in ecdysone biosynthetic gene transcription upon CncC or dKeap1 depletion in the PG, we examined transcription of shd, which is expressed throughout the brain, and start1, which is expressed predominantly in the PG [29]. The levels of shd and start1 transcripts in the brain complex were not reduced by CncC or dKeap1 depletion in the PG (Figure 3A, Figure S3A). Expression of the shRNAs targeting CncC or dKeap1 also did not alter the size, morphology or the number of nuclei in the PG (Figure 3C, Figure S3B and S3C). It is therefore unlikely that the effects of CncC or dKeap1 depletion on ecdysone biosynthetic gene transcription were caused by a disruption of PG development.
To examine if CncC or dKeap1 affected ecdysone biosynthetic gene transcription at a different stage of development, we examined the effects of the cncK6 and dKeap1EY5 loss of function mutations on transcription of these genes in late embryos. The levels of nvd, spok, dib, sad, and shd transcripts were lower in cncK6/K6 homozygous than in cncK6/+ heterozygous embryos (Figure 3B). Likewise, the levels of nvd, spok, phm, dib, and sad transcripts were lower in dKeap1EY5/EY5 homozygous than in dKeap1EY5/+ heterozygous embryos, whereas the level of shd transcripts was higher in the homozygous than in heterozygous embryos (Figure 3B). The moderate effects of the cncK6 and dKeap1EY5 loss of function mutations on ecdysone biosynthetic gene transcription and the consequent lack of complete developmental arrest during embryogenesis could be due to maternal deposition of CncC and dKeap1 mRNA or proteins in the egg. The effects of these mutations on the levels of ecdysone biosynthetic gene transcripts in embryos corroborate the effects of CncC and dKeap1 depletion on transcription of these genes in the PG. In contrast to the concordant effects of the cncK6 and dKeap1EY5 loss of function mutations on ecdysone biosynthetic gene transcription, these mutations had opposite effects on transcription of the gstD1 xenobiotic response gene (Figure 3B). We were not able to determine the effects of CncC or dKeap1 depletion on the level of gstD1 in the PG since gstD1 is expressed throughout the brain.
The cncK6 and dKeap1EY5 mutations could affect transcription of the ecdysone biosynthetic genes through several mechanisms, including direct binding to the promoters and indirect effects on other transcription factors. To test if CncC and dKeap1 bound to the ecdysone biosynthetic genes, we measured CncC and dKeap1 occupancy at their promoter regions in late embryos using ChIP analysis. CncC and dKeap1 occupancy were observed at the phm, shd, dib and sad genes at levels that were comparable to their occupancy at the dKeap1 and gstD1 genes (Figure 3D). Their occupancy was higher near the sad promoter compared to flanking regions (Figure S3D). No CncC occupancy above background and only low dKeap1 occupancy was observed at the Rp49, Actn3, and Gapdh1 housekeeping genes. CncC and dKeap1 are therefore likely to regulate ecdysone biosynthetic gene expression directly by binding to their promoter regions.
Defects in ecdysteroid biosynthesis in the PG can delay pupation and increase the size of the pupae [19]. We investigated if the reduction in ecdysone biosynthetic gene transcription caused by CncC or dKeap1 depletion affected the timing of pupation by altering larval ecdysteroid levels. Expression of the shRNA targeting CncC under the control of the phm-GAL4 or the 5015-GAL4 driver extended the average time between third instar molting and pupation by 40–125% (Figure 4A). Expression of a different shRNA targeting all Cnc isoforms under the control of the phm-GAL4 driver also delayed the time of pupation (Figure 4A). The cncC-RNAi transgene alone had no detectable effect. The mean size of the pupae formed by larvae that expressed the shRNA targeting CncC in the PG was larger than the mean size of the pupae formed by control larvae (Figure 4B), indicating that the delayed pupation was not a secondary consequence of a reduced rate of larval growth. Some larvae continued to grow and formed giant semi-pupae (Figure S4A).
To evaluate the role of ecdysteroid levels in the delayed pupation, we measured the level of 20E in the larvae. Expression of the shRNA targeting CncC in the PG delayed the rise in 20E after third instar molting (Figure 4C, Figure S4B). To establish if the reduced level of 20E was the cause of the delay in pupation, we added 20E to the food for the larvae that expressed the shRNA targeting CncC in the PG. Supplementation with 20E shortened the time between third instar molting and pupation in these larvae by almost 50%, restoring their time of pupation nearly to that of wild-type larvae (Figure 4A).
Expression of the shRNA targeting dKeap1 under the control of the phm-GAL4 driver extended the average length of the larval stage by 4 days (Figure 4D). The mean size of the pupae formed by larvae that expressed the shRNA targeting dKeap1 in the PG was larger than the mean size of the pupae formed by control larvae (Figure 4E). Taken together, these results establish that CncC and dKeap1 affected the time of pupation through their effects on ecdysone biosynthetic gene transcription and on the level of 20E.
We examined the functions of CncC in relation to the Ras signaling pathway, which controls the timing of pupation in response to prothoracicotropic hormone (PTTH) binding to the Torso receptor [19]. Constitutively active RasV12 expression in the PG causes early pupation and a smaller pupal size [20] (Figure 5A). Moreover, deletion of Nrf2 in mice suppresses the lung and pancreatic tumorigenesis caused by constitutively active K-RasG12D expression [6]. We determined the effect of CncC depletion in combination with RasV12 expression in the PG on the time of pupation and on pupal size. When the shRNA targeting CncC was co-expressed with RasV12 in the PG, the premature pupation was suppressed and the pupae were restored to nearly normal size (Figure 5A, 5B). CncC depletion in the PG not only suppressed premature pupation caused by RasV12 expression, but delayed pupation relative to wild type larvae, suggesting that CncC was required for both ectopic and endogenous Ras signaling.
We further examined if CncC depletion affected the consequences of constitutive Ras signaling for pupal development. Most of the animals that expressed RasV12 alone arrested at early pupal stages with no detectable eye pigmentation or wings (Figure 5C, Figure S5A). In contrast, co-expression of the shRNA targeting CncC with RasV12 enabled a majority of the pupae to develop to late stages, and some to eclose and produce adult flies (Figure 5C, Figure S5A). It is unlikely that CncC depletion affected RasV12 expression in the PG since CncC depletion did not alter the level of rasV12 transcripts in salivary glands (Figure 5E). The genetic interactions between CncC depletion and RasV12 expression suggest that CncC mediated the regulation of pupation by the Ras signaling pathway.
To determine if Ras signaling affected CncC binding to chromatin, we investigated if RasV12 expression in salivary glands affected endogenous CncC occupancy on polytene chromosomes. RasV12 expression increased both the number of loci occupied by CncC and the level of CncC occupancy at most loci, but did not affect the level of cncC transcripts in salivary glands (Figure 5D, Figure S5B). RasV12 expression reduced CncC binding at some loci (Figure 5D). Ras signaling therefore regulated both the efficiency and the specificity of CncC binding to chromatin.
To establish if Ras signaling and CncC affected gene transcription in concert, we examined the effects of ectopic RasV12 and CncC expression on transcription of genes at two of the loci where RasV12 expression affected CncC occupancy in salivary glands (Figure 5D, lower panels). Both RasV12 as well as CncC fusion protein expression activated transcription of these genes (Figure 5E). Conversely, CncC depletion by shRNA expression counteracted the activation of these genes by RasV12 expression. RasV12 expression had selective effects on the transcription of genes at these loci since the transcription of other CncC target genes, including gstD1 and gstE1, was not detectably affected by RasV12 expression (Figure 5E). These results suggest that Ras signaling regulated CncC transcriptional activity by altering its occupancy at selected target genes.
Visualization of the subcellular distributions of CncC and dKeap1 and their occupancy on polytene chromosomes revealed that both CncC and dKeap1 were predominantly nuclear and occupied specific chromatin loci. The nuclear localization of dKeap1 and its occupancy of specific chromatin loci indicated that it has functions distinct from those that have been previously attributed to mammalian Keap1. Analysis of the transcriptional and developmental consequences of tissue-specific depletion of CncC and dKeap1 as well as of mutations in cncC and dKeap1 established that these proteins control transcriptional regulons in different organs that coordinate the onset of metamorphosis. The direct roles of CncC and dKeap1 both in ecdysone biosynthetic gene transcription in the PG as well as in ecdysone-regulated gene transcription in salivary glands establish mechanistic links between these central processes in Drosophila metamorphosis.
Both CncC and dKeap1 depletion reduced transcription of ecdysone-regulated early puff genes. These loci were occupied by both CncC and dKeap1, suggesting that CncC and dKeap1 activated transcription of these genes in concert. Similarly, transcription of most ecdysone biosynthetic genes was reduced by both CncC and dKeap1 depletion as well as by the cncK6/K6 and dKeap1EY5/EY5 loss of function mutations in larvae and embryos, respectively. The ecdysone biosynthetic genes were also occupied by both CncC and dKeap1 in embryos, suggesting that CncC and dKeap1 activated their transcription in concert. In contrast, CncC and dKeap1 depletion as well as the cncK6/K6 and dKeap1EY5/EY5 mutations had opposite effects on transcription of the gstD1 and gstE1 xenobiotic response genes in salivary glands and in embryos, respectively. Similarly, opposite effects of CncC and dKeap1 on transcription of other xenobiotic response genes have been previously reported in adult Drosophila [2]. CncC and dKeap1 therefore regulated transcription of different classes of genes through distinct mechanisms. Whereas xenobiotic response genes are regulated by antagonistic effects of dKeap1 on transcription activation by CncC, ecdysone biosynthetic and response genes were activated by concerted chromatin binding by CncC and dKeap1. Chromatin binding by dKeap1 as well as its cooccupancy and cooperation with CncC have potential implications for Keap1 function and its effects on Nrf2 activity in mammalian cells. Keap1 can shuttle into the nucleus in some cells [10], [12], and could bind chromatin in association with Nrf2 or other interaction partners.
The effects of CncC and dKeap1 depletion on ecdysone biosynthetic gene transcription and on the timing of pupation indicate that CncC and dKeap1 are important components of the transcription regulatory circuit that controls ecdysone biosynthesis (Figure 6). Many parts of the neuro-endocrine signaling axis that induces ecdysone biosynthesis have been characterized [19], [20], [30], [31]. Previous studies had not identified the transcription factors that bind and regulate ecdysone biosynthetic genes. dSmad2 depletion in the PG reduces ecdysone biosynthetic gene transcription and inhibits pupation. dSmad2 depletion also reduces torso and InR transcription, and RasV12 or InR co-expression in combination with dSmad2 depletion restores both ecdysone-biosynthetic gene transcription as well as pupation [30]. It is therefore likely that dSmad2 affects ecdysone production indirectly by altering Torso or Insulin signaling. In contrast, CncC depletion suppressed the premature pupation caused by RasV12 expression in the PG, and RasV12 expression in salivary glands altered the loci occupied by CncC on polytene chromosomes. These results, together with CncC occupancy and regulation of ecdysone biosynthetic genes in embryos, suggest that CncC mediated the effects of Ras signaling in the PG on pupation by regulating ecdysone biosynthetic gene transcription.
CncC and dKeap1 also regulated transcription of the early ecdysone-inducible genes in the salivary gland. CncC and dKeap1 binding at ecdysone-inducible early puffs, and the absence of effects of CncC depletion on ecdysone receptor subunit or on late puff gene transcription indicate that CncC and dKeap1 regulated early puff gene transcription directly. The functions of CncC and dKeap1 in regulation of genes that control both ecdysteroid synthesis as well as the transcriptional responses to this hormone place CncC-dKeap1 complex at the nexus of a regulatory network that that coordinates the onset of insect metamorphosis (Figure 6).
The discovery that CncC and dKeap1 coordinate Drosophila metamorphosis has identified novel functions of Nrf2-Keap1 family proteins in normal cellular processes and development. The regulation of both metamorphosis and xenobiotic responses by CncC and dKeap1 suggests that these processes either share a common evolutionary ancestry, or that they are mechanistically or functionally interrelated. Most of the ecdysone biosynthetic genes encode cytochrome P450 class oxidoreductases [18]. P450 class oxidoreductases are also key mediators of the metabolic detoxification of many xenobiotic compounds [32].
The genes that were regulated by CncC and dKeap1 in the salivary and prothoracic glands during larval development and those that are regulated by CncC and dKeap1 in adult flies [2], [3] were mostly non-overlapping. Among the genes identified in this study, only nvd among the ecdysone biosynthetic genes and Sgs5 among the ecdysone-regulated genes were detected by microarray analysis of transcripts induced by CncC expression in adult flies [2]. Thus, the effects of CncC and dKeap1 on the transcription of most of the genes that controlled the onset of metamorphosis were restricted to specific tissues and stages of development.
The functions of CncC and dKeap1 in both hormonal regulation of development and in responses to toxic compounds and environmental stress could represent a mechanism that controls development in response to environmental conditions. Imaginal disc damage inhibits PTTH synthesis, resulting in reduced ecdysone synthesis and a delayed pupation [33]. Modulation of TOR signaling in the PG regulates ecdysone biosynthetic and ecdysone-regulated gene transcription and the timing of pupation [31]. Activation of TOR signaling in the PG suppressed the pupation delay caused by larval starvation, indicating that TOR signaling affected developmental timing in response to nutrient stress. Nutrient restriction and heat stress alter 20E and juvenile hormone levels in the ovaries, arresting oogenesis [34]. The interaction between CncC and dKeap1 could mediate responses to both external as well as endogenous signals that modulate developmental progression. Future studies of the effects of environmental stresses on the developmental functions of CncC and dKeap1 will test this hypothesis.
The premature pupation and developmental arrest caused by constitutively active RasV12 expression were suppressed by CncC depletion in the PG. Similarly, the lung and pancreatic tumorigenesis caused by constitutive K-RasG12D expression are suppressed by Nrf2 deletion in mice [6]. K-RasG12D expression can cause a two-fold increase in Nrf2 transcription, but the significance of this change in Nrf2 transcription for tumorigenesis has not been established. RasV12 expression in Drosophila did not alter the level of CncC transcription, but increased the overall level of CncC binding to chromatin, shifted the loci occupied by CncC on polytene chromosomes, and activated genes at those loci in concert with CncC. These results suggest that Ras signaling can regulate the functions of CncC/Nrf family proteins by altering their target gene specificities or transcriptional activities. The mechanisms whereby Ras regulated CncC occupancy remain to be determined, but are likely to include phosphorylation as the MAPK pathway has been proposed to regulate both Nrf2 and the C. elegans homologue of CncC [35], [36], [37].
The relationships between the roles of CncC and dKeap1 in Drosophila metamorphosis and the functions of their mammalian homologues in development remain to be elucidated. Two of the mammalian homologues of CncC, Nrf1 and Nrf2, appear to have partially overlapping functions during mouse development [38], [39], [40], [41]. Genome-wide analyses have identified many genes occupied by Nrf2 that have no known functions in the xenobiotic response [42], [43]. Although ecdysteroids are unique to invertebrates, steroid hormones have central roles in many aspects of mammalian physiology. Nrf2 can mediate the 1α,25-dihydroxyvitamin D3-induced differentiation of acute myeloid leukemia cells through multiple mechanisms, including VDR/RXRα transcription [44]. Further studies of the mechanisms of action of CncC/Nrf and dKeap1/Keap1 family proteins in different phyla are required to establish the evolutionary relationships among these proteins and their functions in development and disease.
Plasmids encoding CncC, CncB, and dKeap1 fused to intact fluorescent proteins and fluorescent protein fragments were constructed as described in supplemental materials and methods. Transgenic Drosophila lines carrying these expression constructs were generated by microinjection in the w1118 background. The transgenic lines carrying UAS-cncC-RNAi, UAS-dKeap1-RNAi and UAS-rasV12 transgenes were as described [3], [20]. The transgenic line carrying UAS-cnc-RNAi expressed an shRNA that targets all of the Cnc. Double transgenic lines were produced by crosses with Sgs3-GAL4, 71B-GAL4, 5015-GAL4, phm-GAL4 and tub-GAL4 driver lines [19], [22], [25], [27]. To minimize external sources of stress, all studies were conducted with larvae and embryos maintained at 25°C with the exception for larvae carrying the UAS-cnc-RNAi and UAS-dKeap1-RNAi transgenes, which were maintained at 29°C to improve the efficiencies of CncC and dKeap1 depletion. Homozygous and heterozygous embryos carrying the cncK6 and dKeap1EY5 alleles were identified by using the Dfd-YFP marker.
Anti-CncC and anti-dKeap1 antisera were raised against proteins encompassing residues 88–344 of CncC and residues 620–776 of dKeap1 fused to GST. The antigens were immobilized and used for affinity purification of the antibodies. Polytene chromosome spreads isolated from the salivary glands of early wandering 3rd instar larvae were prepared and immunolabeled as described in supplemental information. Salivary glands, brain complexes (including brain and prothoracic gland), and imaginal discs were isolated from early wandering 3rd instar larvae and were immunolabeled as described in supplemental information.
mRNA was isolated from the salivary glands and brain complexes of early wandering 3rd instar larvae as well as embryos, and was quantified by RT-qPCR. The relative transcript levels were calculated by assuming that they were proportional to 2−Cp, and were normalized by the levels of Rp49 transcripts. For ChIP analysis, chromatin was isolated from dechorionated embryos, sheared by sonication, and precipitated using the antisera indicated. The precipitated DNA was quantified by qPCR.
Newly molted 3rd instar larvae or newly hatched 1st instar larvae were collected and transferred into vials. The number of white prepupae (WPP) was scored every 12 hours or 24 hours. To determine the effect of 20E feeding on pupation, the larvae were grown on feeding plates topped with yeast paste containing 0.5 mg/ml 20E. 20E was extracted from larvae and white pre-pupae and was quantified using an enzyme immunoassay kit (Cayman Chemical). Detailed experimental procedures and references are included in Text S1, Table S1, Table S2.
|
10.1371/journal.pcbi.1003928 | Power Laws from Linear Neuronal Cable Theory: Power Spectral Densities of the Soma Potential, Soma Membrane Current and Single-Neuron Contribution to the EEG | Power laws, that is, power spectral densities (PSDs) exhibiting behavior for large frequencies f, have been observed both in microscopic (neural membrane potentials and currents) and macroscopic (electroencephalography; EEG) recordings. While complex network behavior has been suggested to be at the root of this phenomenon, we here demonstrate a possible origin of such power laws in the biophysical properties of single neurons described by the standard cable equation. Taking advantage of the analytical tractability of the so called ball and stick neuron model, we derive general expressions for the PSD transfer functions for a set of measures of neuronal activity: the soma membrane current, the current-dipole moment (corresponding to the single-neuron EEG contribution), and the soma membrane potential. These PSD transfer functions relate the PSDs of the respective measurements to the PSDs of the noisy input currents. With homogeneously distributed input currents across the neuronal membrane we find that all PSD transfer functions express asymptotic high-frequency power laws with power-law exponents analytically identified as for the soma membrane current, for the current-dipole moment, and for the soma membrane potential. Comparison with available data suggests that the apparent power laws observed in the high-frequency end of the PSD spectra may stem from uncorrelated current sources which are homogeneously distributed across the neural membranes and themselves exhibit pink () noise distributions. While the PSD noise spectra at low frequencies may be dominated by synaptic noise, our findings suggest that the high-frequency power laws may originate in noise from intrinsic ion channels. The significance of this finding goes beyond neuroscience as it demonstrates how power laws with a wide range of values for the power-law exponent α may arise from a simple, linear partial differential equation.
| The common observation of power laws in nature and society, that is, quantities or probabilities that follow distributions, has for long intrigued scientists. In the brain, power laws in the power spectral density (PSD) have been reported in electrophysiological recordings, both at the microscopic (single-neuron recordings) and macroscopic (EEG) levels. We here demonstrate a possible origin of such power laws in the basic biophysical properties of neurons, that is, in the standard cable-equation description of neuronal membranes. Taking advantage of the mathematical tractability of the so called ball and stick neuron model, we demonstrate analytically that high-frequency power laws in key experimental neural measures will arise naturally when the noise sources are evenly distributed across the neuronal membrane. Comparison with available data further suggests that the apparent high-frequency power laws observed in experiments may stem from uncorrelated current sources, presumably intrinsic ion channels, which are homogeneously distributed across the neural membranes and themselves exhibit pink () noise distributions. The significance of this finding goes beyond neuroscience as it demonstrates how power laws power-law exponents α may arise from a simple, linear physics equation.
| The apparent ubiquity of power laws in nature and society, i.e., that quantities or probability distributions satisfy the relationship(1)
where α is the power-law exponent, has for a long time intrigued scientists [1]. Power laws in the tails of distributions have been reported in a wide range of situations including such different phenomena as frequency of differently sized earth quakes, distribution of links on the World Wide Web, paper publication rates in physics, and allometric scaling in animals (see [1] and references therein). A key feature of power laws is that they are scale invariant over several orders of magnitude, i.e., that they do not give preference to a particular scale in space or time. There are several theories with such scale invariance as its fingerprint, among the most popular are fractal geometry [2] and the theory of self-organized critical states [3].
Conspicuous power laws have been seen also in the field of neuroscience [4], among the most prominent the observed power laws in the size distribution of neuronal ‘avalanches’ [5], [6] and in the high-frequency tails of power spectral densitites (PSDs) of electrical recordings of brain activity such as electroencephalography (EEG) [7], [8], electrocorticography (ECoG) [9]–[12], the local field potential (LFP) [13]–[16], and the soma membrane potential and currents of individual neurons [17]–[21]. To what extent these various power laws have the same origin, is currently not known [4], [6]. In any case, it is the latter type of power law, i.e., those observed in the PSDs of electrical recordings, which is the topic of the present paper.
Ever since Hans Berger recorded the first human electroencephalogram (EEG) in 1924 [22], its features have been under extensive study, especially since many of them are directly related to disease and to states of consciousness. In the last decades the frequency spectra of the EEG has, for example, attracted significant attention as the high-frequency part of the PSD in experiments with maximal frequencies typically in the range 30–100 Hz has often well fitted by a power laws with α typically in the range from 1 to 2.5 [7], [8]. Such apparent power laws have not only been seen in macroscopic neural recordings such as EEG, ECoG and LFP, they also appear at the microscopic level, i.e., in single-neuron recordings. PSDs of the subthreshold membrane potentials recorded in the somas of neurons often resemble a power law in their high-frequency ends ( 100–1000 Hz), typically with a larger exponent α ranging from 2 to 3 [17]–[21]. This particular power law seems to be very robust: it has been observed across species, brain regions and different experimental set-ups, such as cultured hippocampal layer V neurons [17], pyramidal layer IV–V neurons from rat neocortex in vitro [19], [20], and neocortical neurons from cat visual cortex in vivo [18], [21]. At present, the origin, or origins, of these macroscopic and microscopic power laws seen in PSDs of neural recordings are actively debated [4], [6].
Lack of sufficient statistical support have questioned the validity of identified power-law behaviors, and as a rule of thumb, it has been suggested that a candidate power law should exhibit an approximately linear relationship in a log-log plot over at least two orders of magnitude [1]. Further, a mechanistic explanation of how the power laws arise from the underlying dynamics should ideally be provided [1]. In the present paper we show through a combination of analytical and numerical investigations how power laws in the high-frequency tail of PSDs naturally can arise in neural systems from noise sources homogeneously distributed throughout neuronal membranes. We further show that the mechanism behind microscopic (soma potential, soma current) power laws will also lead to power laws in the single-neuron contribution (current-dipole moment) to the EEG. Moreover, we demonstrate that if all single-neuron contributions to the recorded EEG signal exhibit the same power law, the EEG signal will also exhibit this power law. We find that for different measurement modalities different power-law exponents naturally follow from the well-established, biophysical cable properties of the neuronal membranes: the soma potential will be more low-pass filtered than the corresponding current-dipole moment determining the single-neuron contribution to the EEG [23], [24], and as a consequence, the power-law exponent α will be larger for the soma potential than for the single-neuron contribution to the EEG [25] (see illustration in Fig. 1).
When comparing with experimental data, we further find that for the special case when uncorrelated and homogeneously distributed membrane-current sources themselves exhibit power laws in their PSD, the theory predicts power-law exponents α in accordance with experimental observations for the microscopic measures, i.e., the soma current and soma potential. The experimental situation is much less clear for the EEG signal where frequency spectra presently is limited upwards to 100 Hz. However, we note that under the assumption that such single-neuron sources dominate the high-frequency part of the EEG signal, the theoretical predictions are also compatible with the power-law-like behavior so far observed experimentally.
Both synaptic noise and intrinsic channel noise will in general contribute to the observed noise spectra, cf. Fig. 1. While our theory per se is indifferent to the detailed membrane mechanism providing the noisy current, our findings suggests that the dominant noise source underlying the observed high-frequency power laws seen in PSDs may be channel noise: prevalent theories for synaptic currents are difficult to reconcile with a power law in the high-frequency tail of power spectra, while potassium ion channels with such noise spectra indeed have been observed [26]. Note that this does not imply that channel noise in general dominates synaptic noise in electrophysiological power spectra: it only suggests that the high-frequency power-law part, which in the in vivo situation typically represents a tiny fraction of the overall noise power, is dominated by channel noise.
Through the pioneering work by Wilfred Rall half a century ago [27], [28] the ball and stick neuron model was established as a key model for the study of the signal processing properties of neurons. An important advantage is the model's analytical tractability, and this is exploited in the present study. We first demonstrate the relevance of this simplified model in the present context by numerical comparisons with results from a morphologically reconstructed multicompartmental pyramidal neuron model. Then we derive analytical power-law expressions for the various types of electrophysiological measurements. While a single current input onto a dendrite does not give rise to power laws, we here show that power laws naturally arise for the case with homogeneously distributed inputs across the dendrite and the soma [29], see Fig. 1. For this situation we show that the ball and stick neuron model acts as a power-law filter for high frequencies, i.e., the transfer function from the PSD of the input membrane currents, , to the PSD of the output (soma potential, soma current, or current-dipole moment setting up the EEG), , is described by a power law: . Notably the analytically derived power-law exponents α for these transfer functions are seen to be different for the different measurement modalities. The analytical expressions further reveal the dependence of the PSDs on single-neuron features such as the correlation of input currents, dendritic length and diameter, soma diameter and membrane impedance.
The theory presented here also contributes to -theory in general [30]: it illustrates that a basic physics equation, the cable equation, can act as a power-law filter for high frequencies when the underlying model has spatially distributed input. Furthermore, α may have any half-numbered value between 1/2 and 3, depending on the physical measure (some potential, soma current, single-neuron contribution to the EEG) under consideration, and the coherence of the input currents. Intuitively, the emergence of the power-law spectra can be understood as a result of a superposition of simple low-pass filters with a wide range of cutoff frequencies due to position-dependent dendritic filtering of the spatially extended neuron [23], . This is in accordance with the orginal idea of Schottky from 1926 [32] that the shot-noise observed in vacuum tubes by Johnson could be understood by the combined action of a continuous distribution of ‘exponential relaxation processes’ [33].
The paper is organized as follows: In the next section we derive analytical expressions for the soma potential, soma current and current-dipole moment for the ball and stick neuron for the case with noisy current inputs impinging on the soma ‘ball’ and homogeneously on the dendritic stick. While these derivations are cumbersome, the final results are transparent: power laws are observed for all measurement modalities in the high-frequency limit. In Results we first demonstrate by means of numerical simulations the qualitative similarity of the power-law behaviors between the ball and stick model and a biophysically detailed pyramidal neuron. We then go on to analytically identify the set of power-law exponents for the various measurement modalities both in the case of uncorrelated and correlated current inputs. While the derived power laws strictly speaking refer to the functional form of PSDs in the high-frequency limit (Eq. 1), the purported power laws in neural data have typically been observed for frequencies less than a few hundred hertz. Our model study implies that the true high-frequency limit is not achieved at these frequencies. However, in our ball and stick model, quasi-linear relationships can still be observed in the characteristic PSD log-log plots for the experimentally relevant frequency range. These apparent power laws typically have smaller power-law exponents than their respective asymptotic values. The numerical values of these exponents will depend on details in the neuron model, but the ball and stick model has a very limited parameter space: it is fully specified by four parameters, a dimensionless frequency, the dimensionless stick length, the ratio between the soma and infinite-stick conductances, and the ratio between the somatic and dendritic current density. This allows for a comprehensive investigations of the apparent power-law exponents in terms of the neuron parameters, which we pursue next. To facilitate comparison with experiments we round off the Results section exploring how PSDs, and in particular apparent power laws, depend on relevant biophysical parameters. In the Discussion we then compare our model findings with experiments and speculate on the biophysical origin of the membrane currents underlying the observed PSD power laws.
In the present study the idealized ball and stick neuron model will be treated analytically, while simulation results will be presented for a reconstructed layer V pyramidal neuron from cat visual cortex [34] (Fig. 2). Both the ball and stick model and the reconstructed layer V neuron model are purely passive, ensuring that linear theory can be used. The input currents are distributed throughout the neuron models with area density in the dendrite and in the soma. The input currents share statistics, i.e., they all have the same PSD, denoted , and a pairwise coherence . The coherence is zero for uncorrelated input and unity for perfectly correlated input.
For the ball and stick neuron, the cable equation is treated analytically in frequency space. We first provide a solution for a single current input at an arbitrary position, and then use this solution as basis for the case of input currents evenly distributed throughout the neuronal membrane. The resulting PSDs can be expressed as Riemann sums where the terms correspond to single-input contributions. In the continuum limit where the neuron is assumed to be densely bombarded by input currents, the Riemann sums become analytically solvable integrals. From these analytical solutions we can then extract the various transfer functions relating the output PSDs to the PSDs of the input current. Here the output modalities of interest are the net somatic current, the soma potential and the single-neuron contribution to the EEG, see Figs. 1 and 2.
Below we treat the ball and stick neuron analytically. For the pyramidal neuron (Fig. 2), the NEURON Simulation Environment [35] with the supplied Python interface [36] was used.
For a cylinder with a constant diameter d the cable equation is given by(2)
with the length constant and the time constant . , and denote the specific membrane resistance, the specific membrane capacitance and the inner resistivity, respectively, and have dimensions , and . Lower-case letters are used to describe the electrical properties per unit length of the cable: , and , with units , and . For convenience, the specific membrane conductance, , will also be used, see Table 1 for a list of symbols.
With dimensionless variables, and , the cable equation, Eq. 2, can be expressed(3)
Due to linearity, each frequency component of the input signal can be treated individually. For this, it is convenient to express the membrane potential in a complex (boldface notation) form,(4)
where is a complex number containing the amplitude and phase of the signal, and the dimensionless frequency is defined as . The complex potentials are related to the measurable potential through the Fourier components of the potential,(5)
where is the direct current (DC) potential. The cable equation can then be simplified to(6)
where , see [23], [31]. The general solution to Eq. 6 can be expressed as(7)
The expression for the axial current is given by(8)
and is applied at the boundaries to find the specific solutions for the ball and stick neuron. In complex notation and with dimensionless variables this can be expressed as(9)
where is the infinite-stick conductance. Similarly, the transmembrane current density (including both leak currents and capacitive currents) is given by(10)
with its complex counterpart,(11)
The ball and stick neuron [27] consists of a dendritic stick attached to a single-compartment soma, see Fig. 3A. Here we envision the stick to be a long and thin cylinder with diameter d and length l. The membrane area of the soma is set to be , corresponding to the surface area of a sphere with diameter , or equivalently, the side area of a cylindrical box with diameter and height .
The solution of the cable equation for a ball and stick neuron with a single input current at an arbitrary dendritic position is found by solving the cable equation separately for the neural compartment proximal to the input current and the neural compartment distal to the input current, These solutions are then connected through a common voltage boundary condition at the connection point. For the proximal part of the stick, Ohm's law in combination with the lumped soma admittance gives the boundary condition at the somatic site, and for the distal part of the stick, a sealed-end boundary is applied at the far end. In this configuration the boundary condition acts as the driving force of the system. The potential can, however, also be related to a corresponding input current through the input impedance, i.e., .
Above we derived transfer functions for the ball and stick neuron, connecting current input at an arbitrary position on the neuron to the various measurement modalities, i.e., the current-dipole moment (), the soma potential () and the soma current (). We will now derive expressions for the PSDs when the ball and stick neuron is bombarded with multiple inputs assuming that all input currents have the same PSD and a pairwise coherence [37]. The PSDs can then be divided into separate terms for uncorrelated () and fully correlated () input.
The PSD, , of the output can for the case of multiple current inputs be expressed as(39)
where is the PSD of the input currents, is their coherence and is the transfer function between the PSD of the input and the PSD of the output. The complex conjugate is denoted by the asterisk.
We now assume the first of the input currents to be positioned at the soma compartment, and the rest of the input to be spread homogeneously across the dendritic stick. The transfer function for the soma compartment, , is the same for all somatic inputs, for , while the input transfer function for the dendritic stick is position dependent, for . The PSD transfer function can then be expressed(40)
To allow for analytical extraction of power laws, we next convert the sums into integrals. By assuming uniform current-input density (per membrane area) in the dendritic stick (given by ), it follows that the axial density of current inputs is . In the continuum limit () we thus have(41)
where the last factor comes from the conversion to dimensionless lengths. The PSD transfer function, , in Eq. 40 can then be split into three parts,(42)
where(43)
is the PSD transfer function for uncorrelated input at the soma compartment,(44)
is the PSD transfer function for uncorrelated input distributed throughout the dendritic stick, and(45)
is the PSD transfer function for correlated input distributed both across the dendritic stick and onto the soma.
We have now derived (i) a general expressions for the PSD transfer function expressed by the general, single-input transfer functions and , and (ii) specific analytical expressions for the single-input transfer functions for the dipole moment, the soma potential and the soma current. We will next combine these results and analytically derive specific PSD transfer functions for the dipole moment, the soma potential and the soma current for distributed input.
For convenience we here summarize the results, now solely in terms of dimensionless variables (except for the amplitudes ), i.e., , , , and (see Table 2). The general expression for the PSD transfer functions reads:(79)
where represents the contributions from uncorrelated current inputs, represents the contributions from correlated inputs, and is the pairwise coherence function. The contributions from uncorrelated input currents are in turn given as sums over contributions from somatic and dendritic inputs , i.e.,(80)
The contribution to the PSD transfer functions for correlated input currents are given by(81)(82)(83)
with the squared norm of given by Eq. 49, and and defined by Eqs. 50 and 51, respectively.
The contributions from uncorrelated dendritic inputs are:(84)(85)(86)
In the special case with input to soma only, the PSD transfer functions are the same for uncorrelated (Eq. 43) and correlated input (Eq. 45), the only difference being the amplitudes,
( implies that the input is onto soma only.) The corresponding PSD transfer functions from uncorrelated somatic input thus become(87)(88)(89)
In an infinite, homogenous, isotropic Ohmic medium with conductivity , the extracellular potential recorded at a given position far away from a single-neuron current dipole is given by [25], [38].(90)
where designates the spatial position of the current dipole, is the magnitude of the current-dipole moment, and is the angle between the dipole moment vector and the position vector . An important feature is that all time dependence of the single-neuron contribution to the potential lies in so that factorizes as(91)
For the electrical potential recorded at an EEG electrode, the forward model in Eq. 90 is no longer applicable due to different electrical conductivities of neural tissue, dura matter, scull and scalp. Analytical expressions analogous to Eq. 90 can still be derived under certain circumstances such as with three-shell or four-shell concentric spherical head models (see Nunez and Srinivasan [38], Appendix G), but the key observation for the present argument is that the single-neuron contribution to the EEG will still factorize, i.e., where here is an unspecified function.
The compound EEG signal from a set of single-neuron current dipoles is now given by(92)
where the index runs over all single-neuron current dipoles. For each Fourier component (frequency) we now have(93)
For the special case where the different single-neuron current dipoles moments are uncorrelated we find that the power spectral density of the EEG is of the form [39](94)
(We have here introduced the notation ‘UC’, i.e. capitalized, to highlight the difference between the present assumption of uncorrelated single-neuron current dipoles and the separate assumption of uncorrelated membrane currents onto individual neurons in the above sections.) If the single-neuron current dipoles have the same power-law behavior in a particular frequency range, i.e., , it follows directly that the EEG signal will inherit this power-law behavior:(95)
where determines the PSD amplitude, but not the slope.
The inheritance of the single-neuron power-law behavior also applies to the case of correlated sources, provided that the pairwise coherences are frequency independent. By similar reasoning as above we then find(96)
Analogous expressions for the PSD for the EEG can also be derived when both correlated and uncorrelated single-neuron current dipoles contribute, but we do not pursue this here; see Lindén et al. [37] and Leski et al. [39] for more details.
The NEURON simulation environment [35] with the supplied Python interface [36] was used to simulate a layer-V pyramidal neuron from cat visual cortex [34]. The main motivation for pursuing this was to allow for a direct numerical comparison with results from the ball and stick neuron to probe similarities and differences, see Fig. 2. In addition, NEURON was also used on the ball and stick neuron model to verify consistency with the analytical results above. Both the layer-V pyramidal neuron and the ball and stick neuron had a purely passive membrane, with specific membrane resistance , specific axial resistivity m, and specific membrane capacitance F/m. Simulations were performed with a time resolution of 0.0625 ms, and resulting data used for analysis had a time resolution of 0.25 ms. All simulations were run for a time period of 1200 ms and the first 200 ms were removed from the subsequent analysis to avoid transient upstart effects in the simulations.
The digital cell reconstruction of the layer-V pyramidal neuron was downloaded from ModelDB (http://senselab.med.yale.edu/), and the axon compartments were removed. To ensure sufficient numerical precision compartmentalization was done so that no dendritic compartment was larger than 1/30th of the electrotonic length at 100 Hz (using the function lambda_f(100) in NEURON), which resulted in 3214 compartments. The soma was modeled as a single compartment.
The ball and stick neuron was modeled with a total of 201 segments, one segment was the iso-potential soma segment with length and diameter , and 200 segments belonged to the attached dendritic stick of length 1 mm and diameter .
Simulations were performed with the same white-noise current trace injected into each compartment separately. The white-noise input current was constructed as a sum of sinusoidal currents [24](97)
where represents a random phase for each frequency contribution. Due to linearity of the cable equation, the contributions of individual current inputs could be combined to compute the PSD of the soma potential, the soma current and the dipole moment resulting from current injection into all compartments. In correspondence with Eq. 39, the summation of the contributions from the input currents of different segments with membrane areas was done differently for uncorrelated and correlated input currents. The uncorrelated PSDs, , were computed according to(98)
while the correlated PSDs, , were computed according to(99)
Here, denotes the Fourier components of the signal (either soma potential, soma current or dipole moment due to input in one segment), the product gives the total number of input currents into one segment , and the density represents for dendritic input and for somatic input.
The total dipole moment was in the numerical computations assumed to equal the dipole moment in one direction only: the direction along the stick for the ball and stick model, and the direction along the apical dendrite for the pyramidal neuron model, both denoted as the -component, . For the pyramidal neuron this is an approximation as the dipole moment also will have components in the lateral directions. However, the prominent ‘open-field’ asymmetry of the pyramidal neuron in the vertical direction suggests that this is a reasonable approximation when predicting contributions to the EEG signal. The current-dipole moment is then given by(100)
where is the transmembrane current of compartment , and is the corresponding -position.
To establish the relevance of using the simple ball and stick neuron to investigate the biophysical origin of power laws, we compare in Fig. 2 the normalized power spectral densities (PSDs) of the transmembrane soma current (row 1), the current-dipole moment (row 2), and the soma potential (row 3) of this model (column 1) with the corresponding results for a biophysically detailed layer-V pyramidal neuron (column 2); the rightmost column gives a direct comparison of PSDs. Both neuron models have a purely passive membrane and receive spatially distributed current input. As described in the Models section, the PSD of the single-neuron contribution to the EEG will be proportional to the PSD of the neuronal current-dipole moment given the observation that the extracellular medium, dura matter, scull and scalp appear to be purely ohmic [24], [38]. We here stick to the term ‘current-dipole moment’ even if the term ‘single-neuron contribution to the EEG’ could equally be used.
A first striking observation is that unlike single-input PSDs (thin gray lines in Fig. 2), the PSDs resulting from numerous, homogeneously distributed input currents (thick lines) have a linear or quasi-linear appearance for high frequencies in these log-log plots, resembling power laws. This is seen both when the numerous current inputs are correlated (green thick lines) and uncorrelated (blue thick lines). We also observe that the decay in the PSD with increasing frequency is strongest for the soma potential, somewhat smaller for the current-dipole moment, and smallest for the soma current. This is reflected in the power-law exponents estimated at 1000 Hz from these PSDs, see legend in Fig. 2. Here we observe that is largest for the soma potential (bottom row) and smallest for the soma current (top row).
In the example in Fig. 2 we have assumed constant input current densities across the neurons, i.e., . For this special case, correlated current input will, at all times, change the membrane charge density equally across the neuron, and as a consequence the neuron will be iso-potential. In this case the axial current within the neuron will be zero, and likewise the net membrane current (with the capacitive current included) for any compartment, including the soma. As a consequence the current-dipole moment vanishes, and the model can effectively be collapsed to an equivalent single-compartment neuron. For the soma current and dipole moment we thus only show results for uncorrelated inputs in Fig. 2. However, correlated current input will still drive the soma potential (green curves in columns 1 and 2). Here we observe that the exponent is smaller for uncorrelated input than for correlated input both for the ball and stick neuron and for the pyramidal neuron.
The results above pertain to the situation with white-noise current inputs, i.e., flat-band PSDs. However, the results are easily generalized to the case with current inputs with other PSDs. Since our neuron models are passive and thus linear, the PSDs simply multiply. This is illustrated in column 3 of Fig. 2 which shows how our PSDs for uncorrelated input change with varying PSDs of the current input, . The blue curves correspond to white-noise input and are identical to the blue curves in column 2. The pink and brown curves illustrate the case of pink () and Brownian () input, respectively. Since the PSDs multiply, the power-law exponent of the input noise simply adds to the exponent . Thus, the pink and Brownian input increase the slope with and , respectively, compared to white-noise input.
Even though the dendritic structure of the reconstructed pyramidal neuron is very different from the ball and stick neuron in that it has both a highly branched structure and a varying diameter along its neural sections (tapering), both models seem to produce linear or quasi-linear high-frequency PSDs in the log-log representation. Also the power-law exponents are found to be fairly similar. This implies that the ball and stick neuron model captures salient power-law properties of the more biophysically detailed neuron model, and motivates our detailed analytical investigation of the power-law properties of the ball and stick neuron following next.
In the Models section above we derived analytical expressions for the PSD transfer functions of the soma current (), current-dipole moment () and soma potential () for the ball and stick neuron for spatially distributed input currents. The resulting transfer functions , summarized in Eqs. 79–89, were of the form(101)
where and represent the contributions from uncorrelated somatic and dendritic inputs, respectively, and represents the contribution from correlated inputs. is the pairwise coherence of the current inputs, all assumed to have the same PSDs ().
These mathematical expressions are quite cumbersome, but they are dramatically simplified in the high-frequency limit, , in which the dominant power can be found analytically by a series expansion of the mathematical expressions for the transfer functions in Eqs. 81-89.
The expressions for the PSD transfer functions contain terms which are both polynomial and superpolynomial (i.e., including exponentials/exponentially decaying functions) with respect to frequency. As these superpolynomial terms will dominate the polynomial terms in the high-frequency limit, it follows from Eq. 49 that for high frequencies the absolute square of the denominator can be approximated by(102)
where terms decaying exponentially to zero with increasing frequency have been set to zero. The frequency dependence is through and , see Eqs. 50 and 51. Note that since . In the high-frequency limit the PSD transfer functions Eqs. 81–89 become(103)(104)(105)(106)(107)(108)
where the amplitudes are found in Table 2. When the PSDs expressed in Eqs. 103-107 are expanded reciprocally for high frequencies, i.e., , we get(109)(110)(111)(112)(113)(114)
where is the dimensionless relative density, , and , with and denoting the somatic and dendritic diameter, respectively, and denoting the dendritic length constant. The expansions were done in Mathematica (version 7.0), and a list of parameters used throughout the present paper is given in Table 1 (along with the default numerical values used in the numerical investigations in later Results sections).
In Eqs. 109–114 terms which are exponentially decaying to zero for large have been approximated to zero. Note that Eq. 114 does not apply in the special case of no somatic input, , for which the series expansion gives(115)
The corresponding high frequency expansions of the PSD transfer functions for uncorrelated somatic input, , are not shown, as these expressions are identical to the corresponding transfer functions for correlated input into the soma only, (i.e., equal to Eqs. 110, 112 and 114 with ).
Eqs. 109–115 show that, due to position-dependent frequency filtering of the numerous inputs spread across the membrane (cf. Fig. 3B), all PSD transfer functions express asymptotic high-frequency power laws. Moreover, these genuine ‘infinite-frequency’ power-law exponents, denoted , span every half power from (for , Eq. 109) to (for , Eq. 115) for the different transfer functions. The results are summarized in Table 2.
To obtain the power-law exponents in the general case with contributions from both correlated and uncorrelated current inputs, we need to compare the different terms in the general expression for in Eq. 101. With different leading power-law exponents in their asymptotic expressions, the term with the lowest exponent will always dominate for sufficiently high frequencies. From Table 2 we see that for all three quantities of interest, i.e., , and , the lowest exponent always comes from contributions from uncorrelated inputs. Note that the correlated term in Eq. 101 also involves a frequency-dependent coherence term , but to the extent it modifies the PSD, it will likely add an additional low-pass filtering effect [39] and, if anything, increase the power-law exponent. If we assume that the coherence is constant with respect to frequency we identify the following asymptotic exponents (i.e., with ‘all’ types of possible input) for , and :
Note that these power-law exponents are unchanged as long as uncorrelated activity is distributed both onto the soma and the dendrite, but will increase to and if no uncorrelated input are present on the dendrite. Similarly, without input onto soma, the asymptotic value will change for the soma potential PSD: it becomes if uncorrelated input is uniformly distributed on the dendrite, and if the dendritic input is correlated.
Detailed inspection of the power-law slopes for the ball and stick model in Fig. 2 and comparison with the power-law exponents listed in Table 2 reveal that although the curves might look linearly decaying in the log-log plot for high frequencies, the expressed exponents are still deviating from their high-frequency values , even at 1000 Hz. As experimental power laws have been claimed for much lower frequencies than this, we now go on to investigate apparent PSD power laws for lower frequencies. For this it is convenient to define a low-frequency (lf) regime, an intermediate-frequency (if) regime and a high-frequency (hf) regime, as illustrated in Fig. 3C. The transition frequencies between the regimes are given by the frequencies at which is and of , respectively.
The log-log decay rates of the PSD transfer functions can be defined for any frequency by defining the slope as the negative log-log derivative of the PSD transfer functions,(116)
In Figs. 4, 5, and 6 we show color plots of for the soma current , current-dipole moment , and soma potential , respectively, both for cases with uncorrelated and correlated inputs. The depicted results are found by numerically evaluating Eq. 116 based on the expressions for listed in Eqs. 81–89. Note that since our model is linear, the log-log derivative is independent of the amplitude . Thus, with either completely correlated or completely uncorrelated input, the dimensionless parameters , , and span the whole parameter space of the model. The 2D color plots in Figs. 4–6 depict as function of and for three different values of the electronic length ( = 0.25, 1, and 4), i.e., spanning the situations from a very short dendritic stick () to a very long stick (). Electrotonic lengths greater than produced plots that were indistinguishable by eye from the plots for . The thin black contour line denotes the transition between the low- and intermediate-frequency regimes (), whereas the thick black contour line denotes the transition between the intermediate- and high-frequency regimes ().
The 2D color plots in Figs. 4–6 depicting the slopes of the PSDs of the transfer functions , give a comprehensive overview of the power-law properties of the ball and stick model as they are given in terms of the three key dimensionless parameters , , and . To get an additional view of how the model predictions depend on biophysical model parameters, we plot in Figs. 7 and 8 PSDs, denoted , for a range of model parameters for the soma current, current-dipole moment and soma potential when the neuron receives homogeneous white-noise current input across the dendrite and/or the soma. We focus on biophysical parameters that may vary significantly from neuron to neuron: the dendritic stick length , the specific membrane resistance , the dendritic stick diameter , and the soma diameter . The specific membrane resistance may not only vary between neurons, but also between different network states for the same neuron [40], [41].
To predict PSDs of the various measurements, and not just PSDs of the transfer functions , we also need to specify numerical values for the current-input densities and (and not only the ratio ), as well as the magnitude of the PSDs of the current inputs. These choices will only affect the magnitudes of the predicted PSDs, not the power-law slopes. As the numerical values of the high-frequency slopes predicted by the present work suggest that channel noise from intrinsic membrane conductances rather than synaptic noise dominates the observed apparent high-frequency power laws in experiments (see Discussion), we gear our choice of parameters towards intrinsic channel noise. We first assume the input densities and (when they are non-zero) to be 2 m, in agreement with measurements of the density of the large conductance calcium-dependent potassium (BK) channel [42]. Next we assume the magnitude of PSD of the white-noise current input to be = const = 1 fA2/Hz. This choice for gives magnitudes of predicted PDSs of the soma potential, assuming uncorrelated current inputs, in rough agreement with what was observed in the in vitro neural culture study of [17], i.e., about 10−3–10−2 mV2/Hz for low frequencies. Note that the shape of the PSDs, and thus estimated power-law exponents, are independent of the choice of current-noise amplitude.
Figs. 7 and 8 show PSDs for uncorrelated and correlated input currents, respectively. A first observation is that the predicted PSD magnitudes are typically orders of magnitude larger for correlated inputs, than for uncorrelated inputs. With the present choice of parameters, the cases with correlated inputs predict PSDs for the soma potential and soma current much larger than what is seen in in vitro experiments [17],[19],[20]. A second observation is that variations in the dendritic stick length (first column in Figs. 7–8) and membrane resistance (second column) typically have little effect on the PSDs at high frequencies, but may significantly affect the cut-off frequencies, i.e., the frequency where the PSD kinks downwards. This may be somewhat counterintuitive, especially that the PSDs for the current-dipole moment are independent of stick length as one could think that a longer stick gives a larger dipole moment. For the ball and stick neuron, however, this is not so: input currents injected far away from both boundaries (ends) of a long stick will not contribute to any net dipole moment, as the input current will return symmetrically on both sides of the injection point and thus form a quadrupole moment. This symmetry is broken near the ends of the stick: for uncorrelated input a local dipole is created at each endpoint; for correlated input the dendrite will be iso-potential near the distal end of the stick, while a local dipole will arise at the somatic end if . Note though that this is expected to be different for neurons with realistic dendritic morphology, since the dendritic cables typically are quite asymmetric due to branching and tapering.
The effects of varying the dendritic stick diameter and soma diameter are quite different (cf., two rightmost columns in Figs. 7–8). Here both the magnitudes and the slopes of the high-frequency parts are seen to be significantly affected. On the other hand, the cut-off frequency is seen to be little affected when varying the soma diameter , in particular for the current-dipole () and soma potential () PSDs. (Note that for the case with homogeneous correlated input, (row 4 in Fig. 8), the ball and stick model is effectively reduced to a single-compartment neuron for which the PSD is independent of and .)
In Figs. 4–6 regions in the log-log slope plots were observed to have positive double derivatives, i.e., concave curvature. The effect was particularly prevalent for the soma potential transfer function in the case of short dendritic sticks () with dominant current input to the soma. This feature is also seen in the corresponding ‘soma-input’ curves (bottom rows of Figs. 7–8), also for non-compact sticks, i.e., for the default value = 1 mm ( = 1).
In the present work we have taken advantage of the analytical tractability of the ball and stick neuron model [27] to obtain general expressions for the power spectral density (PSD) transfer functions for a set of measures of neural activity: the somatic membrane current, the current-dipole moment (corresponding to the single-neuron EEG contribution), and the soma potential. With homogeneously distributed input currents both onto the dendritic stick and with the same, or another current density, onto the soma we find that all three PSD transfer functions, relating the PSDs of the measurements to the PSDs of the noisy inputs currents, express asymptotic high-frequency power laws. The corresponding power-law exponents are analytically identified as for the somatic membrane current, for the current-dipole moment, and for the soma potential. These power-law exponents are found for arbitrary combinations of uncorrelated and correlated noisy input current (as long as both the dendrites and the soma receive some uncorrelated input currents).
The significance of this finding goes beyond neuroscience as it demonstrates how power laws with a wide range of values for the power-law exponent may arise from a simple, linear physics equation [30]. We find here that the cable equation describing the electrical properties of membranes, transfers white-noise current input into ‘colored’ -noise where may have any half-numbered value within the interval from to 3 for the different measurement modalities. Intuitively, the physical underpinning of these novel power laws is the superposition of numerous low-pass filtered contributions with different cut-off frequencies (i.e., different time constants) [32], [33] due to the different spatial positions of the various current inputs along the neuron. (Note, however, that power laws with integer coeffients (1 and 2) also are obtained with purely somatic input; cf. Table 2.) As our model system is linear, the results directly generalize to any colored input noise, i.e., transferring spectra of input currents to output spectra.
Our ball and stick model expressions for the PSDs cover all frequencies, not just the high frequencies where the power-law behavior is seen. When comparing with results from neural recordings, one could thus envision to compare model results with experimental results across the entire frequency spectrum. However, the experimental spectra will generally be superpositions of contributions from numerous sources, both from synapses [41] and from ion channels [17]. These various types of input currents will in general have different PSDs, i.e., different . A full-spectra comparison with our theory is thus not possible without specific assumptions about the types and weights of the various noise contributions, information which is presently not available from experiments. However, the presence of power-law behavior at high frequencies implies that a single noise process (or several noise processes with identical power-law exponents) dominates the others in this frequency range.
In the following we first discuss apparent power laws observed in the soma potential and soma current in vitro [17], [19], [20]. Next, we discuss apparent power laws seen in vivo, both in the soma potential [18], [21], [43] and, briefly, in the EEG [7]. Here synaptic noise is expected to provide almost all of the noise variance, but our results suggest that the power law at the high-energy tail of the spectrum nevertheless may be due to ion-channel noise.
Power laws have also been reported in recordings of extracellular potentials inside (local field potential; LFP) and at the surface of cortex (electrocorticography; ECoG). However, the reported power-law exponents vary a lot, with 's between 1 and 3 for LFPs [13]–[16] and between 2 and 4 for ECoG signals [9]–[12], [50]. From a modeling perspective the single-neuron contribution to putative power-law exponents for these signals is more difficult as, unlike the EEG signal, the single-neuron contributions are not determined only by the current-dipole moment: dominant contributions to these signals will in general also come from neurons close to the electrode (typically on the order of hundred or a few hundred micrometers [37]), so close that the far-field dipole approximation relating the current-dipole moment directly to the contributed extracellular potential [25] is not applicable [37].
A point to note, however, is that it may very well be that power laws observed in the LFP or ECoG are dominated by other current sources than the power laws observed in the EEG spectra: As observed in [37], [39] (see also [51]) the LFP recorded in a cortical column receiving correlated synaptic inputs can be very strong, and it is thus at least in principle conceivable that power laws in the LFP may stem from synaptic inputs from neurons surrounding the electrode, whereas the EEG signal, which picks up contributions from a much larger cortical area, may be dominated by uncorrelated noise from ion channels. Further, the soma potential and soma current of each single neuron may also still be dominated by uncorrelated channel noise, even if the LFP is dominated by correlated synaptic activity. This is because correlated synaptic inputs onto a population of neurons add up constructively in the LFP, whereas the uncorrelated inputs do not [37], [39]. For single-neuron measures such as the soma potential and soma current there will be no such population effects, and the uncorrelated inputs may more easily dominate the power spectra.
As a final comment it is interesting to note that in the only reported study we are aware of for the frequency range 300-3000 Hz, the PSD of the LFP exhibited a power law with a fitted exponent of = 1.1 [15]. This is very close to what would be predicted if the LFP was dominated by the soma current from uncorrelated (pink) noise sources: In Table 2 we see that the ‘infinite-frequency’ power-law exponent for the transfer function from dendritic current inputs to soma current is . With a pink () PSD of the input noise current, the ‘infinite-frequency’ prediction for the soma current exponent will thus be 1.5. This is already fairly close to the experimental observation of 1.1. Further, from Fig. 4 it follows that the apparent power-law coefficient for the transfer-function power law may be somewhat smaller than 0.5 in the frequency range of interest, suggesting that the agreement between experiments and model predictions assuming uncorrelated noise may be even better. If so, it may be that the LFP power spectra are dominated by synaptic inputs for frequencies below a few hundred hertz (with rapidly decaying LFP contributions with increasing frequency, i.e., higher power-law exponents in accordance with [13], [14], [16]), while uncorrelated inputs, and thus power laws with smaller exponents, dominate at higher frequencies.
In the present analysis we have modeled the membranes of somas and dendrites as simple passive linear (RC) circuit elements. This implies a strictly linear response to the current inputs, allowing for the present frequency-resolved (Fourier) analysis. However, the same kind of analysis can be done for active dendritic membrane conductances, at least close to the resting potential of the neuron: In the so called quasi-active membrane models, the active conductances are linearized and modeled by a combination of resistors, capacitors and inductors [52], [53]. These extra circuit elements will change the PSD. For example, the inductor typically introduces a resonance in the system. In Koch [53] the impedance for this ‘quasi-active’ membrane was however found to coincide with the impedance for a purely passive membrane for frequencies above 200 Hz, implying that the predicted high frequency power laws will be about the same. This is in accordance with experimental results from neocortical slices, where blocking of sodium channels were shown mainly to affect the soma potential PSD for frequencies below 2 Hz [19]. Nevertheless, the investigation of the role of active conductance on PSDs is a topic deserving further investigations.
Here we modeled the noise-generating membrane mechanism as a simple current, i.e., , making the system fully linear. As a (non-linear) alternative, these noise currents could have been modeled as conductance-based currents, i.e., where is the conductance, and is the channel reversal potential. In the case of potassium channels, will typically be around -80 mV. However, when exploring the situation when the membrane potential is not too close to the channel reversal potential, we observed in simulations the same high-frequency power-law behavior for conductance-based and current-based noise-current models (results not shown). That these two models give the same power law can be understood as follows: In the conductance-based case the channel current has two terms, i.e., . The conductance is here dependent on the incoming spike trains, but not on the membrane potential. The first term involves a product of and , while the second term has the same mathematical form as the current-based noise model. Since the potential membrane potential always will be low-pass filtered compared to the input, the linear term is expected to dominate the product for high frequencies. If so, it follows that the linear term will determine the power-law behavior, and that the power-law behavior will be the same as for the current-based model.
A key conclusion from the present work is that the power-law predictions from our models are in close agreement with experimental findings for the soma potential and the soma current provided the transmembrane current sources are assumed to be (i) homogeneously distributed throughout the whole neuron, (ii) uncorrelated, and (iii) have a pink () noise distribution. It should be stressed that we do not argue against synaptic noise being a major component underlying neural noise spectra; the importance of synaptic inputs in setting the noise level has been clearly demonstrated, for example by the large difference in membrane potential fluctuation between in vivo and in vitro preparations [41], [43]. We rather suggest that the power-law behavior seen at the high-frequency end of these noise spectra may be dominated by intrinsic channel noise, not synaptic noise.
We also speculate that potassium channels with inherent noisy current with PSDs following a distribution in the relevant frequency range, underlie the observed high-frequency power laws, and the slow voltage- and calcium-activated BK channel, reported to have a very large channel conductance [47], is suggested as a main contributor [17]. If future experiments indeed confirm that the BK channel is a dominant source of membrane noise, this may have direct implication of the understanding several pathologies. Not only has the BK channel been implicated as a source of increased neural excitability [54] and epilepsy [55], but also disorders such as schizophrenia [56], autism and mental retardation [57] have been linked to the BK channel through a decrease in its expression [58].
|
10.1371/journal.pgen.1003330 | Duplication and Retention Biases of Essential and Non-Essential Genes Revealed by Systematic Knockdown Analyses | When a duplicate gene has no apparent loss-of-function phenotype, it is commonly considered that the phenotype has been masked as a result of functional redundancy with the remaining paralog. This is supported by indirect evidence showing that multi-copy genes show loss-of-function phenotypes less often than single-copy genes and by direct tests of phenotype masking using select gene sets. Here we take a systematic genome-wide RNA interference approach to assess phenotype masking in paralog pairs in the Caenorhabditis elegans genome. Remarkably, in contrast to expectations, we find that phenotype masking makes only a minor contribution to the low knockdown phenotype rate for duplicate genes. Instead, we find that non-essential genes are highly over-represented among duplicates, leading to a low observed loss-of-function phenotype rate. We further find that duplicate pairs derived from essential and non-essential genes have contrasting evolutionary dynamics: whereas non-essential genes are both more often successfully duplicated (fixed) and lost, essential genes are less often duplicated but upon successful duplication are maintained over longer periods. We expect the fundamental evolutionary duplication dynamics presented here to be broadly applicable.
| Duplicate genes occur in all organisms. It has been found that mutations in duplicate genes cause defects much less often than when single copy genes are mutated. It is widely believed that this is due to functional redundancy—that is, the two genes can carry out similar functions so that the non-mutated duplicate gene can cover for or “mask” the phenotype of the mutation in the first duplicate. To determine whether this hypothesis is true, it is necessary to test systematically whether defects indeed occur in the organism when both duplicate genes are inhibited. We have for the first time carried out such an analysis in a multicellular organism, the nematode Caenorhabditis elegans. In contrast to expectations, we observed that when both copies of duplicate genes are inhibited deleterious effects are very rare. We show that this is because duplicate genes are much more often non-essential compared to genes where there is only a single copy. Non-essential genes are also lost from the genome much more often than essential genes. However, when essential genes are duplicated, they remain present in the genome over longer periods. Our results give a framework to explain the evolutionary dynamics of duplications in the genome.
| Duplication of genes is an important source of evolutionary novelty [1], [2]. Duplicate genes may also provide stability to an individual organism, by buffering the effect of harmful mutations [3]–[9], although it is unlikely that this explains why a duplication is initially favoured [10]. Immediately after duplication two new paralogs are probably similar in both sequence and expression. As a consequence, it is hypothesized that the effects of mutations in one paralog can be masked by the other: although the first paralog has a mutation that would normally (in the absence of masking) reduce fitness, the second paralog compensates for the mutation, so that the reduction in fitness is less than expected. This was proposed by Haldane, who hypothesised that paralogous genes could undergo mutations without disadvantage to the organism [4]. This phenomenon has been variously termed masking, functional redundancy, compensation, or phenotype buffering; we will refer to it as masking. Masking is proposed to occur because of overlap in the biochemical and physiological functions of the paralogs, which allows the second paralog to carry out the functions of the first (Figure 1).
In Caenorhabditis elegans, 17.7% of single-copy genes have been observed to have an ‘essential’ function, defined as a phenotypic defect easily observable upon knockdown under laboratory growth conditions [11]. Compared to single-copy genes, paralogous genes in yeast (Saccharomyces cerevisiae), worm (C. elegans), fly (Drosophila melanogaster) and mouse (Mus musculus) are significantly less likely to have a loss-of-function phenotype [11]–[16]. The low loss-of-function phenotype rates have been interpreted as evidence for functional redundancy, leading to masking of phenotypes. An alternative proposal is that duplicate genes may be biased to have originated from non-essential ancestors and that this may contribute to the lower loss of function phenotype rate of duplicate genes [17]. Phenotype masking however, remains the prevailing theory to explain why genes with paralogs more rarely have obvious loss of function phenotypes, because it is supported by relatively high observed masking rates in tests where selected samples of yeast and worm duplicate pairs have been simultaneously inhibited (∼12–55%) [18]–[22]. However, this question is still open because the incidence of masking has not yet been investigated genome-wide.
Here we report the first unbiased study of masking of duplicate gene-pairs lacking any other close homolog in a multicellular eukaryote, C. elegans. We observe phenotypic masking in only 6% (50/790) of duplicate gene-pairs, far less often than observed in studies of selected gene sets. Strikingly, there is an age-related bias in masking rates with younger paralog pairs (which duplicated after the C. elegans-C. briggsae speciation) displaying masking 4.9 times less frequently than older pairs (which arose before this speciation). We demonstrate that this rate difference is due to a large over-representation of non-essential gene pairs among younger duplicates. When considering only duplicates for which the double knockdown has a phenotype, masking rates are highest for the youngest duplicates, as expected. Our findings support a model whereby non-essential genes are both more likely to be successfully duplicated (duplicated and subsequently fixed in the population) and to be lost in the long term. However, when fixed, essential duplicates are more likely to be maintained in the long term. Overall, these evolutionary dynamics lead to a low observed loss of function phenotype rate upon knockdown of duplicate genes either singly or in pairs because they are frequently non-essential. The results indicate that phenotype masking should not be the default explanation as to why genes that have a paralog do not exhibit a discernable phenotype on single gene knockdown; it is more likely that they were derived from non-essential genes, this being especially true if they are recently duplicated.
To measure the incidence of masking among paralogous genes in an unbiased way and on a genome-wide scale, we carried out single and double gene RNA interference (RNAi) knockdown experiments for 790 C. elegans paralog pairs (see Methods). As RNAi is a sequence-based process, a single RNAi probe will knock down both members of a pair of paralogs that have nearly identical sequences, preventing assessment of single-gene knockdown phenotypes. Therefore, a paralog pair was only included in the set of 790 pairs if they had diverged sufficiently so that a different RNAi probe could uniquely target each gene (see Methods). For each pair, the two genes are each other's closest homolog within C. elegans and lack any closely related paralog, although pairs may belong to a larger C. elegans gene family (see Methods).
To test for masking between two genes, we used the standard procedure of comparing the phenotype of each single-gene inhibition to that of the double [23]. If w is fitness and s1 and s2 are the reductions in fitness associated with inhibiting genes 1 and 2, then, in the absence of masking, the fitness of the single and double loss of function individuals is expected to be wi = 1−si, w2 = 1−s2, and w1,2 = (1−s1)(1−s2), respectively. Fitness w1,2 lower than expected is interpreted as evidence of masking [18],[19],[24]. In some cases, both single-gene and double-gene inhibitions have no observable, or very little, reduction in fitness, (w1≈w2≈w1,2≈1), presumably because the genes are of relatively low importance to the organism in the conditions studied. Typically, genes or gene pairs where an obvious defect is observed upon knockdown (wi<1) are classified as ‘essential’ and those where no obvious phenotypic defect is observed upon knockdown (wi≈1) ‘non-essential.’ This definition of ‘essential’ genes includes those that may not have a lethal knockdown phenotype, and ‘non-essential’ genes might display a loss of function phenotype under other assay conditions or only require a very low level of gene activity to maintain fitness. In addition, classification as non-essential does not mean that the gene is evolutionarily dispensable.
Single and double RNAi knockdown experiments were conducted in duplicate using the RNAi hypersensitive strain eri-1(mg366);lin-15B(n744) [25]–[27]. P0s were scored for fertility and lethality of F1 embryos; P0s and F1s were additionally scored for a host of other post-embryonic phenotypes, and all observed phenotypes were confirmed by rigorous analysis of additional replicates (see Methods).
Of the genes having a single-gene knockdown phenotype in any of four previous RNAi screens [11], (n = 198 genes), our screen detected a single-gene knockdown phenotype in 90% of cases (Table S6). This level of concordance is similar to that observed for replicate genome-wide RNAi screens in C. elegans [29]. We further observed that each of the individual genes were effectively inhibited using the double RNAi feeding protocol: a phenotype was observed for 99% of double knockdowns where either of the single-gene knockdowns showed a phenotype (n = 175).
As described above, we considered a paralog pair to exhibit masking if the double knockdown displayed a more severe phenotype than expected under a multiplicative model of interaction when compared to the two single-gene knockdowns, i.e. w1,2<(1−s1)(1−s2) (see Methods). This includes both full and partial masking, where one member of a paralog pair either fully or partially compensates when the other member is knocked down.
We observed phenotype masking for just 6.3% (50 of 790) of paralog pairs. Surprisingly, we found that phenotype masking was very rare for genes showing no phenotypic defect upon single knockdown (5.1%, n = 1382). Instead, duplicate genes with single knockdown phenotypes much more often showed masking (15.2%, n = 198). Overall, 30% of genes displaying masking showed a single knockdown phenotype compared to 17.7% of single copy genes and 12.5% of duplicate genes.
It is expected that masking would be more common in younger duplicates, since they generally are more similar to each other in sequence and expression [8], [13], [16], [18], [24], [31]. To investigate this we used phylogenetic analysis to identify duplicate pairs which arose from a duplication that occurred in (i) the C. elegans lineage after the speciation separating C. elegans from C. briggsae ∼30 Mya [32]; (ii) the ancestor of Caenorhabditis species; (iii) the ancestor of Bilateria; or (iv) the ancestor of eukaryotes (see Methods). We identified 178 duplicate pairs where the duplication occurred in the C. elegans lineage after speciation from C. briggsae, and 533 pairs that arose before this speciation (see Methods). We will refer to the 178 C. elegans-lineage pairs as ‘younger’ pairs, and to the 533 pairs that arose before the C. elegans-C. briggsae speciation as ‘older’ pairs.
Despite the expectation that masking would be most common for younger paralog pairs, we found that just 1.7% of the younger pairs (3/178 pairs) exhibited masking and 1.7% of genes in this set (6/356) exhibited a fully masked phenotype. The single-gene knockdown phenotype rate for the 356 genes in the 178 younger duplicate pairs is 1.4%, far lower than the rate of 17.7% for single-copy C. elegans genes (455 of 2566 genes, X2-test: P<10−14). This 16.3% difference cannot be due to phenotypic masking since full masking is very rare among younger duplicate genes (1.7%).
Since a pair of duplicates will diverge over time, we would predict a lower rate of masking amongst older duplicate pairs than for younger pairs. However, surprisingly we find that overall (full or partial) and full masking rates are much higher for the 533 older pairs than the 178 younger pairs (4.9-fold and 3.4-fold, respectively Figure 2A and Figure S2).
We consider a paralog pair to exhibit masking if the double knockdown displays a more severe phenotype than expected compared to the single-gene knockdown phenotypes. If a gene-pair was relatively unimportant (i.e. non-essential) under the conditions studied, then there would be no obvious phenotypic defect upon single or double knockdown and masking would not be observed. Therefore, a possible reason why masking is observed less frequently for younger than older duplicate pairs could be that a greater fraction of the younger pairs are non-essential.
If we assume that the younger duplicates have not gained or lost essential functions since the duplication events that generated them, then the extant C. elegans genome should be a good surrogate for the gene pool from which the duplicates arose. If so, we would predict that the fraction of younger paralog pairs that are ‘essential’ pairs (for which the double knockdown has an obvious phenotypic defect) should be approximately equal to the fraction of all C. elegans genes that have a single-gene knockdown phenotype. In striking contrast to this prediction, the double knockdown phenotype (essentiality) rate for the 178 younger pairs is only 4.5%, compared to 13.4% for single-gene knockdowns across the C. elegans genome (1917 of 14327, X2-test: P<10−3; Figure 2B). On the other hand, the essentiality rate for the 533 older duplicate pairs is 27.6%, significantly higher than the single-gene knockdown rate for the whole C. elegans gene set (X2-test: P<10−15; Figure 2B). The finding that non-essential genes are over-represented among the younger paralog pairs relative to the whole C. elegans gene set can explain why the observed rate of masking is low among younger paralogs: they tend to be non-essential, so display no evident phenotype upon single or double knockdown.
Why are younger duplicate pairs more often non-essential compared to the whole C. elegans gene set (4.5% vs. 13.4%)? The young duplicate genes do not appear to be biased for particular functional classes that could explain this difference (Table S1). We also considered the possibility of masking by more distant paralogs. However, the essentiality rate for duplicate pairs with no detectable other paralog is still lower than the knockdown phenotype rate for single copy genes (Figure S1). An alternative explanation is that non-essential genes may be more likely to successfully duplicate (i.e. duplicate and subsequently become fixed in the population) compared to essential genes, as hypothesised by He and Zhang [17]. They showed that single-copy S. cerevisiae genes whose orthologs had duplicated in another yeast species were more often non-essential than those whose orthologs remained single-copy [17]. Bias favouring successful duplication of non-essential genes could explain why knockdown of duplicate pairs rarely show loss of function phenotypes. Different mechanisms could contribution to such a bias. For example, genes that are not dose sensitive on knockdown may be more prone to duplication because changes in dose are of lesser phenotypic impact.
To explore a possible duplication bias, we compared the knockdown phenotype rate of 960 C. elegans single-copy genes whose orthologs have remained single-copy in two other nematode species (C. briggsae and C. remanei) to that of 269 single-copy C. elegans genes whose orthologs have duplicated in at least one of these nematode species (see Methods). We found that the single-copy C. elegans genes whose orthologs have duplicated have a significantly lower knockdown phenotype rate than those whose orthologs have remained single-copy (19.3% vs. 30.2%, X2-test: P = 0.0006). This agrees with a similar trend previously observed for a small C. elegans dataset [17]. Therefore, non-essential genes in Caenorhabditis duplicate more often than essential genes, which can explain why C. elegans paralog pairs are so often non-essential.
It is often the case that genes with an essential phenotype are more likely to have orthologs in distant species than do genes lacking any strong knockdown or knockout phenotype. Does the same hold for gene duplicates whose double knockdowns are essential or non-essential? That the duplicates with a phenotype tend to be evolutionarily more ancient (Figure 2B) would suggest that they would be more likely to have orthologs in distant species. To analyse this, and to ensure that the result is not biased by different rates of evolution, we considered a recently assembled worm-human ortholog set [33].
This set was assembled using four different orthology calling tools (InParanoid, OrthoMCL, HomoloGene and Ensembl Compara). We consider a set of worm genes with evidence for orthology in humans through any of these methods (a liberal list of 7663 genes) and a set found by all of these methods (a conservative list of 3386 genes). For each list we considered whether each member of a duplicate pair was identified as having an ortholog in humans or not. We find that duplicate genes whose double knockdown has no evident phenotype are less likely to have an ortholog in humans than duplicate genes with a knockdown phenotype (from the liberal list, 58% of non-essential genes have a human ortholog versus 84% of those with a phenotype, chi squared test, P<<0.0001; from the conservative list, 23% of non-essential genes have a human ortholog versus 48% of those with a phenotype, chi squared test, P<<0.0001). As duplicate genes without knockdown phenotype evolve faster than those with a phenotype (Figure S5), the finding of fewer genes with knockdown phenotype having an ortholog may simply reflect a higher rate of sequence evolution and hence weakened homology searching. To address this problem, we performed a logistic regression in which we predict presence or absence of orthologs in humans as a function of the knockdown phenotype and the rate of protein evolution derived from the C. elegans-C. briggsae comparison. This revealed that, while rate of protein evolution is a predictor of presence/absence of a human ortholog (liberal set: P = 2×10−6; conservative set: P = 4×10−6), duplicate genes with a double knockdown phenotype are more likely to have an ortholog in humans controlling for the rate of evolution (liberal set: P = 1×10−5; conservative set: P = 1×10−7). We conclude that duplicates genes with an underlying phenotype are more likely to be phylogenetically preserved. This result comes with the caveat that we presume the rate of evolution of a gene in the intra-worm comparison is a fair reflection of its rate of evolution in other lineages.
It is expected that genes are most likely to exhibit masking immediately after duplication and then to show a lower rate of masking with increasing age, as they diverge in sequence and expression. This view is supported by previous studies in yeast, C. elegans, fly and mouse where it was observed that the single-gene knockdown phenotype rate for duplicated genes increases with protein divergence between the two members of a pair (Kapair) [8], [13]–[16], [18], [24], [31] (Figure 3A; logistic regression: P<10−11). Measurement of masking rates would be made difficult by the preponderance of non-essential genes and indeed we did not find a significant correlation between Kapair and the full masking rate (Figure 3A; logistic regression: P = 0.7). To avoid this difficulty, we restricted analysis to essential duplicate pairs, where phenotypes are readily observed. This analysis showed a significant negative correlation between the rate of phenotype masking and Kapair (logistic regression: P = 0.002, Figure 3B), supporting the hypothesis that duplicate pairs with greater sequence similarity are more likely to exhibit masking. We also find a prevalence of masked phenotypes amongst the youngest duplicates (those that arose in the C. elegans lineage since divergence from C. briggsae, or in the Caenorhabditis ancestor; Figure 2C and Figure S2). Therefore, the youngest and most sequence similar duplicates are most likely to exhibit masking.
We were interested to test whether phenotypic masking was evolutionarily conserved. We could not compare our data to that of yeast, because only two pairs are orthologous to a yeast duplicate pair screened in yeast [24]. To assess the level of conservation of masking in a closer relative, we identified 31 duplicate pairs that arose prior to the C. elegans-C. briggsae speciation and tested whether the C. briggsae ortholog pairs showed masking (see Methods). We observed phenotype masking for 19 of the 31 C. briggsae duplicate pairs (61.3%), indicating significant retention of masking between duplicates over the estimated ∼30 million years [32] since the C. elegans-C. briggsae speciation.
Lynch and Conery [34] observed that young duplicate pairs tend to evolve fast at the protein level in C. elegans, mouse, human and fly and inferred that “early in their history, many gene duplicates experience a phase of relaxed selection or even accelerated evolution at replacement sites” [34]. A possible explanation for the rapid protein evolution of young duplicate pairs is that they are usually similar enough in sequence for masking to occur, and since masking compensates for mutations in either member of a duplicate pair, this may allow them to accumulate substitutions relatively rapidly [35]. The bias for successful duplication of non-essential genes suggests an alternative possibility: that this duplication bias is also a bias for successful duplication of intrinsically fast-evolving genes. This could be the case if non-essential genes evolve faster than essential genes (as some previous studies suggest [36], [37]). Indeed, when we estimated the evolutionary rate of each duplicate pair by calculating the mean protein divergence between orthologous members of the pair in C. elegans and C. briggsae (KaCeCb), we find that non-essential duplicate pairs have a higher rate of protein sequence evolution than essential pairs (mean KaCeCb 0.120 vs. 0.092, Wilcoxon test: P<10−4).
Expression level is strongly negatively correlated with the rate of protein sequence evolution in many species [37], . We find that non-essential duplicate pairs have lower expression levels than essential pairs (average of 8.3-fold lower; log2 means 8.60 vs. 11.66; Wilcoxon test: P<10−15), suggesting that the higher rate of protein sequence evolution of non-essential pairs could be related to their lower expression level. In support of this, expression level is a good predictor of the rate of protein evolution (KaCeCb) in an ANCOVA model (using Ln(KaCeCb) as the response variable: P<0.0001; Figures S3 and S4). Essentiality/non-essentiality of duplicate pairs in the ANCOVA is not a significant predictor indicating that it is expression level rather than dispensability per se that is the important variable (Figures S3 and S4). We also find for singleton genes the difference in evolutionary rate between those with and without a phenotype on knockdown is related to differences in expression level rather than essentiality per se (Figures S3 and S4).
Given these results, we propose that the relatively fast protein sequence evolution of young duplicates [34] is partly due to a bias towards successful duplication of lowly-expressed, non-essential genes, which, given their expression level, tend to evolve fast. Consistent with this, more recent duplicate pairs that arose in the Caenorhabditis ancestor have lower expression levels than duplicates that arose in the Bilaterian or Eukaryotic ancestors (average of 10.3-fold lower; log2 means 8.70 vs. 12.06; Wilcoxon test: P<10−15). Therefore, fast evolution of young/nonessential duplicates is not prima facie evidence that duplicates are under weak purifying selection owing to masking (as classically presumed), as young duplicates are biased towards lowly expressed non-essential genes with intrinsically high rates of evolution and, for non-essential genes, there is little or no possibility of phenotype masking. We can, however, use our data to examine this hypothesis more directly.
If duplication enabled phenotype masking and so permitted fast evolution we would expect singleton genes with an underlying phenotype to evolve slower than duplicates with an underlying phenotype. Against these expectations, for genes with a phenotype, the evolutionary rate is the same for singletons and duplicated genes (dN for singletons with knockdown phenotype = 0.087+/−0.094; dN for duplicate genes with a double knockdown phenotype = 0.092+/−0.076, t-test P = 0.56). Controlling for expression level does not alter this conclusion (P = 0.43; Figure S4). Similarly, if we compare duplicates genes with a double knockdown phenotype that show evidence of masking with those with a double knockdown phenotype but no evidence of masking we find in the ANCOVA, controlling for expression level, that presence/absence of masking is not a predictor of the rate of evolution (P = 0.24) (see Supplementary Result 1.1 in Text S1). Likewise singletons with a phenotype evolve no slower than duplicates with masking when controlling for expression level (P = 0.36) (see Supplementary Result 1.2 in Text S1). Incidentally, we also find that singleton genes without phenotype evolve at the same rate as duplicates genes without double knockdown phenotype (singleton genes without phenotype, dN = 0.13+/−0.1 (sd), duplicate genes without phenotype, dN = 0.12+/−0.9, t-test, P = 0.16). In sum, where there exists the possibility of phenotype masking (i.e. when the double knockdown has a phenotype), we see no evidence that the duplicated genes evolve any faster than expected of genes of similar dispensability/expression level and find no evidence that masking promotes rapid sequence evolution.
Through systematic double knockdown analyses, we showed that non-essential genes in C. elegans are more likely to be successfully duplicated than essential genes. A similar bias is supported by the finding of a paucity of orthologs of murine essential genes in segregating CNVs in humans [39] and the observation of lower than expected numbers of genes associated with lethal phenotypes that have copy number variants in flies [40]. The mechanism for this bias might be mutational, selectionist, or both. In a mutational model, non-essential genes could be more prone to duplication, but once duplicated no more prone to fixation than essential duplicates. Under a selectionist model, a non-essential gene could be equally prone to duplication, but the duplicate could be more likely to be fixed in the population.
Mutation bias could arise if chromosomal regions vary in their propensity for duplication, and regions with a higher density of non-essential genes have higher duplication rates. This is plausible as duplications are commonly caused by non-homologous recombination events [41], which in turn are more likely in chromosomal regions with high homologous recombination rates [42]. C. elegans chromosome arms have high recombination rates, are rich in duplicate genes, and are poor in essential genes [11], [43]–[45]. We hypothesise that the location of non-essential genes in chromosomal arms where the recombination rate is high might contribute to their higher propensity for duplication. Indeed, we find that 60% of younger duplicate pairs lie on the arms, compared to 30% of older pairs (Fisher test: P = 10−8), suggesting that most new duplicates arise on the arms, regions rich in non-essential genes.
The selection bias hypothesis is also plausible. In yeast, many essential genes show dosage sensitivity because they belong to protein complexes [46], [47]. Duplications of essential genes may therefore often be deleterious and purged by selection, giving rise to a net selection bias for duplications of non-essential genes. The finding that segregating CNVs in humans are depleted for orthologs of murine essential genes was interpreted in this manner [39].
As well as the bias towards duplication of non-essential genes, over the longer term we also see a retention bias for essential duplicates: essential duplicate pairs are enriched among older duplicate pairs compared to younger pairs (27.6% vs. 4.5%; Figure 2B). It is well described that in the majority of instances one of a pair of duplicates will be lost [34]. It is plausible that this death/retention process is biased, such that in the long term essential genes are more likely to persist [37]. Our data suggest that those genes that are easily duplicated (i.e. non-essential genes) are also more easily lost. The loss of non-essential duplicates could occur by gene loss of one of the two members (e.g. deletion, pseudogenization). Alternatively, it could be that the gene is retained but no longer recognizable as having a paralog because sequence divergence is so great. If this were the case, we would expect that essential duplicate pairs would be more slowly evolving than non-essential pairs, and as noted above, we find some evidence for this (Figure S3). However, we also find that, as noted above, the presence/absence of orthologs in humans cannot be accounted for simply in terms of differential rate of evolution; although this is a significant predictor, the presence/absence of a phenotype on knockdown also contributes significantly.
A further mechanism for loss of non-essential duplicates over time could be re-duplication of one of the members of a paralogous gene-pair. As members of a pair are defined here as each other's closest homologs, re-duplication of a non-essential duplicate gene would result in simultaneous loss of an old and creation of a young non-essential duplicate pair in our dataset. Since non-essential genes are more likely than essential genes to undergo successful duplication, they may be also more likely to undergo re-duplications.
Another possible mechanism for loss of non-essential duplicates over time could be gain of new essential functions by non-essential duplicates (e.g. by neofunctionalization), although experiments in yeast did not find evidence for this phenomenon [18]. Therefore, we consider that the retention bias for essential duplicate pairs is probably due to both the slower rate of divergence of essential duplicates and preferential re-duplication of non-essential duplicates.
Genome-wide, we observed masking for 6% (50/790) of C. elegans duplicate pairs, roughly half that observed in the previous C. elegans study (11%), which was based on a smaller sample of gene-pairs (n = 143 [22]). This difference is probably due to a bias towards older gene-pairs in their sample compared to our genome-wide sample (Figure S6). Masking is more common among older duplicate pairs, which will have increased the observed masking rate. A masking rate of 6% for C. elegans paralog pairs appears to be at odds with the much higher rate of 30% observed in yeast [24]. However, our estimate for the masking rate for ‘essential’ genes, where we can confidently detect loss of function phenotypes, is 29%, very similar to the yeast estimate. Nonetheless, we note that this resemblance should be taken with the caveat of methodological differences (e.g. the yeast study used gene deletions whereas ours used RNAi knockdowns).
In conclusion, we have shown that phenotype masking makes a minor contribution to the low knockdown phenotype rate of duplicate genes. The primary reason that the knockdown phenotype rate is low is because the rate of gain and loss (or reduplication) of duplicates derived from non-essential genes is much higher than for essential genes, so that the majority of duplicate pairs are young and have arisen from non-essential precursors. While the rates of masking may differ among organisms due to the influence of varying duplication rates affecting the abundance of young non-essential duplicates, we expect the fundamental duplication dynamics presented here to be broadly applicable. In support of this, recent studies in mouse have shown that younger genes are less likely to be essential than older genes, and that there is an age dependent increase in the proportion of duplicate genes that are essential [12], [15], [16]. We conclude that phenotype masking should not be the default explanation as to why genes that have a paralog do not exhibit a discernable phenotype on single gene knockdown. It is simply more likely that they were derived from non-essential genes in the first place.
An all-against-all protein-sequence WU-BLAST search [48] was carried out using the longest isoform of each protein-coding gene in C. elegans (19735 peptides from WormBase release WS140; https://www.wormbase.org). 2690 duplicate pairs (paralog pairs) were defined as reciprocal-best matches (Table S2), requiring BLASTP matches to have an e-value less than 10−9 and the HSP (high-scoring pair) alignments to span a minimum of 60% of each protein. Single-copy genes were defined as proteins without a BLASTP match of e-value <0.01.
C. elegans RNAi bacterial reagents were obtained from Fraser et al, 2000 [28] and Kamath et al, 2003 [11]. For genes where no RNAi reagent was available, clones from the library of Rual et al, 2004 [30] were used. Of the 2690 pairs, 1183 pairs existed for which each gene was uniquely targeted by an RNAi reagent with no expected non-target RNAi. Unique reagents are defined in WormBase as having one primary target (gene has at least 95% nucleotide identity over 100 bp) and no predicted secondary targets (gene has at least 80% nucleotide identity over 200 bp and is not a primary target). We sequenced both clones for the 1183 pairs of RNAi reagents and found that both were correct for 932 pairs; these pairs were used for screening (Table S2).
To identify duplicate pairs without a close third paralog, we generated a measure of duplicate isolation and applied it as a filter. The ‘duplicate isolation value’ measures the protein sequence similarity between the duplicate pair relative to their similarity to the next closest BLASTP hit that they have in common (considering BLASTP hits with e-values <0.01). For comparison of relative protein-protein similarity, the negative log10 of BLASTP e-values was used as previously described [49]. Duplicate isolation was calculated as: negative log10 of the maximum e-value of the BLAST matches between the protein sequences of the duplicate pair and their closest shared hit, divided by the negative log10 of the e-value for the protein-sequence BLAST match between the genes of the duplicate pair (Table S2). The maximum of the e-values to the closest third paralog was used, as it should best represent the match to the third paralog from sequence shared between members of the duplicate pair. Isolation values range from 0 to 1, with 1 signifying a common best match that is equally as strong as the match between the genes of the duplicate pair, and 0 signifying that the duplicates have no match in common (i.e. they belong to a gene family with just two members).
To identify a subset of duplicate pairs that lack a close third paralog (but may have a distant third paralog), we filtered our set of 932 duplicate pairs using a cutoff of ≤0.83 for the duplicate isolation value. An isolation value of 0.83 would correspond, for example, to a duplicate pair with a protein-sequence BLAST match to each other of e-value 10−100 (or e.g. 10−15) and a best common BLASTP match to a third paralog of (maximum) e-value 10−83 (or e.g. 3.5×10−13). Filtering removes 142 duplicate pairs from the screened set of 932 pairs, leaving 790 duplicate pairs that lack a close third paralog, which we used for our analysis (Table S2). This threshold retains all 50 duplicate pairs that exhibited phenotype masking, indicating that the paralog pairs showing masking probably lack a third paralog that is close enough to provide masking activity.
RNAi bacteria were grown at 37°C in 96-well format in LB containing 50 µg/ml ampicillin for 6–8 hours. Cultures were concentrated 2-fold by centrifugation and removal of half of the medium before resuspension of the bacterial pellet. Aliquots of bacterial cultures targeting single genes of each paralogous pair were mixed in a 1∶1 ratio. Approximately 100 µl of each individual culture or the mixed culture was spotted onto a well of a 6-well plate containing NGM agar including 25 µg/ml carbenicillin, 1 mM IPTG and 50 µg/ml Nystatin and left to dry and induce for 36 hours.
Single and double RNAi knockdown experiments were conducted in duplicate using the RNAi hypersensitive strain eri-1(mg366);lin-15B(n744) [25], [26]. 5–10 L1 eri-1;lin-15B larvae were aliquoted per well in 40 µl drops using a WellMate liquid handling device (Matrix) from a solution of M9 buffer with 0.01% Triton X-100. Plates were incubated at 15°C for 6 days, when controls had been laying eggs for ∼24 hrs. P0s were then scored for a host of post-embryonic phenotypes (see below for F1s) before being removed by aspiration. Approximately 42 hours later, P0 fertility (Ste and Lbd) and lethality of F1 embryos (Emb) was scored. P0 mothers were scored as sterile (Ste) or low brood (Lbd) where wells contained fewer than 10 or 30 F1 progeny, respectively. Embryonic lethality (Emb) was assigned where at least 10% of the brood failed to hatch. When controls had reached mid-larval (∼66 hours) and late-larval/young adult developmental stages (∼90 hours), the F1s were scored for the following post-embryonic phenotypes: Unc (uncoordinated), Prz (paralyzed), Dpy (dumpy), Bmd (body morphology defect), Sck (sick), Bli (blister), Mlt (molting defect), Him (high incidence of males; F1s only), Pvl (protruding vulva), Muv (multivulva), Lon (long), Sma (small), Gro (growth defect), Egl (egg laying defect; P0s only), Stp (sterile progeny; F1s only), Adl (adult lethal), Ooc (oocytes laid; P0s only), Rup (ruptured), and Lvl (larval lethal). Phenotypes were assigned when at least one of the replicates had a penetrance of ≥10% in the F1 population or ≥50% for the P0 mothers. Phenotype data are given in Table S3.
Effectiveness of the double RNAi feeding was monitored by comparing single- and double knockdown phenotypes; in 99% of cases (n = 175) a phenotype was observed in the double feeding well when either of the single-gene knockdowns showed a phenotype. Additionally, in 92% of these cases, the double knockdown phenotype was as least as strong as that of either single knockdown indicating that the double feeding procedure was effective. Duplicate pairs showing potential phenotype masking were defined as those where double RNAi knockdown of the pair showed a stronger phenotype than either of the single-gene knockdown experiments. These candidates were retested for reproducibility; 60 confirmed pairs were subjected to a final round of quantitative testing as described below.
For 3–5 replicates, quantitative tests of brood size in the P0 generation, embryonic lethality in the F1 generation, post-embryonic lethality in the F1 generation and abnormal morphology defects in the F1 generation were carried out for single and double RNAi experiments from the progeny produced by a single P0 mother in the first 48 hours as an adult. In addition, in cases where candidates showed quantifiable phenotypes in the P0 generation, quantitative scoring of post-embryonic phenotypes was carried out for 20–30 P0s. Qualitative scoring of 20–30 P0s was also carried out, indicating the severity (e.g. severe Dpy vs. mild Dpy) or developmental stage of the phenotype (e.g. Lvl L1 vs. Lvl L4).
All quantitative phenotypes were statistically analysed to determine if the double RNAi experiment was more severe or merely mulitiplicative compared to the corresponding single RNAi experiments. Phenotype masking was defined as a genetic interaction where the double RNAi phenotype of the paralogous pair was greater than the product of each of the single-gene RNAi phenotypes, using a method adapted from Baugh et al, 2005 [50] as follows. Quantitative assessment of brood size, embryonic lethality, post-embryonic lethality and abnormal morphology defects were expressed as a percentage of normal development, through normalization to the same measures of 111 control animals or their progeny. The normalized phenotype is used as an estimate of the fitness of the knockdown (w). For each quantified phenotype, the null hypothesis was that the normalized phenotype of the double RNAi experiment (w1,2) is equal to the product of the normalized phenotypes of each of the single RNAi experiments (w1 and w2). Phenotype masking was inferred when w1,2 was significantly lower than the expected value of w1×w2 (Mann-Whitney-U test: P<0.05). Phenotype masking of qualitative phenotypes was inferred when either the developmental stage of the observed phenotype was earlier (e.g. Lvl L1 vs. Lvl L3), or the class of phenotype observed was more severe (e.g. Lvl L3 vs. Gro L3) in the double RNAi experiment compared to both single RNAi experiments. Following the detailed quantitative and qualitative scoring, 50 duplicate pairs were identified as showing phenotype masking (Table S4).
We compared our RNAi phenotype data to that from genome-wide RNAi-by-feeding screens [11], [28], [29], supplemented by data from Rual et al, 2004 [30] where a gene lacked an RNAi reagent in the above three screens. These screens scored the same range of phenotypes as in our study. Only reagents with one primary target and no predicted secondary targets were considered (coverage for 14327 protein-coding genes). Genes targeted by an RNAi reagent that was annotated as having a loss-of-function phenotype in at least one study were assigned as having a knockdown phenotype. Of the genes having a single-gene knockdown phenotype using the combined data from Fraser-Kamath-Simmer-Rual (FKSR) screens (n = 198 genes), our screen also detected a single-gene knockdown phenotype in 90% of cases (Table S6). This level of concordance is similar to that observed for replicate genome-wide RNAi screens in C. elegans [29].
Because duplicates by nature have related sequences, we investigated the possibility that some single-gene knockdown phenotypes observed for duplicate genes that showed masking were due to RNAi off-targets (i.e. unintended knockdown of the other gene member of the pair) that were not predicted in WormBase. Among the set of duplicates that showed masking, we found that 100% of single genes assigned an RNAi knockdown phenotype also showed a phenotype in the genetic mutant (n = 19 genes, based on allele data available in WormBase). This indicates that unpredicted RNAi off-targets in the other member of a paralog pair are unlikely to have confounded estimates of phenotype masking.
We classified each of the paralogous gene-pairs as ‘essential’ if it showed an obvious phenotypic defect upon double knockdown, or ‘non-essential’ otherwise. For this purpose, a duplicate pair was taken to have a phenotypic defect upon double knockdown if: (i) at least one gene of the pair had a (non-wildtype) phenotype based on the Fraser-Kamath-Simmer-Rual screens, or (ii) the pair showed phenotype masking in our data.
In Figure S1, as well as the gene-pairs where an RNAi probe uniquely targets each gene, we also included double knockdown phenotypes inferred for duplicate pairs that are so similar in sequence that a single RNAi reagent targets both members of the pair (i.e. two primary targets and no predicted secondary targets; Table S2).
Of the 50 C. elegans duplicate pairs showing phenotype masking, 41 duplicate pairs were identified where the duplication giving rise to the gene-pairs occurred in the C. elegans-C. briggsae ancestor, resulting in two extant C. elegans-C. briggsae ortholog pairs (see Identification of orthologs below). For 31 C. briggsae gene-pairs, we were able to generate dsRNA to uniquely target each C. briggsae gene by RNAi (i.e. one primary target and no predicted secondary targets). Primers to amplify C. briggsae genomic fragments contained 5′ T7 polymerase promoter sequences (5′ TAATACGACTCACTATAGG 3′) to allow in vitro transcription from PCR products as described by Zipperlen et al, 2001 [51]. RNAi experiments were conducted with the C. briggsae wild-type strain (AF16). Each single dsRNA or a mixture of the two dsRNAs targeting the duplicate pair was injected into 8–10 young adult hermaphrodites at a final concentration of 1–2 µg/µl. Worms were grown at 15°C on NGM plates (2.2% agar to prevent burrowing) seeded with OP50. Injected C. briggsae single P0 mothers were transferred to a fresh well after 24 hours, transferred again after 48 hours and finally removed at 72 hours. Brood size and F1 progeny laid on these plates were scored beginning 24 hours after P0 transfer or final removal. Qualitative and quantitative phenotypes were scored in the same manner as described above with the exception of P0 post-embryonic scoring, given that RNAi was initiated in young adults. Quantitative data was normalized to the same measures of the P0 brood and F1 progeny of 57 C. briggsae worms injected with loading buffer. Following detailed quantitative and qualitative scoring, 19 duplicate pairs were identified as showing phenotype masking (Table S5).
To estimate the dates of duplication that gave rise to duplicate pairs of C. elegans genes, we analysed data from the TreeFam database of animal gene families [52]. Where two genes of the pair belonged to the same TreeFam family, the duplication date was taken to be the taxonomic level of the common ancestor node for the two genes in TreeFam's phylogenetic tree for that family. Duplications were inferred to have occurred either in the C. elegans lineage, in the common ancestor of Caenorhabditis species, or in the common ancestor of Bilaterian species. The age estimate was considered confident if the same date was estimated from at least two of the three most recent TreeFam releases, or if there was strong support for the estimated date from the most recent TreeFam release (6). For ‘strong’ support in TreeFam 6, we required that the bootstrap for the common ancestor node was ≥70%; and that all internal nodes on the lineages back from the two genes to their common ancestor node had bootstraps of ≥70%, or were speciation nodes at which no genes had been lost. Where two genes of a duplicate pair belonged to different families, we investigated whether both families in TreeFam release 6 contained genes from human, Drosophila, and a Saccharomyces or Arabidopsis outgroup. If they did, the duplication must have occurred either in the ancestor of all eukaryotes, or in a pre-eukaryotic ancestor (e.g. the eukaryote-prokaryote common ancestor). We refer to the age of such pairs as ‘Eukaryota’. Using the above approach, we could make confident estimates for the ages of 92% of all duplicate pairs identified in C. elegans (n = 2690; Table S2).
C. elegans-C. briggsae one-to-one orthologs were inferred from Treefam releases 4, 5 and 6, where a duplication had occurred in the C. elegans-C. briggsae ancestor, resulting in two extant C. elegans-C. briggsae ortholog pairs. We only retained orthologs inferred from at least 2 releases or inferred from release 6 with an orthology bootstrap of ≥70% [53], [54]. Orthologs of C. elegans single-copy genes in the nematodes C. briggsae and C. remanei were inferred if the orthology bootstrap was ≥70% in TreeFam release 6. Orthologs between C. elegans and Saccharomyces cerevisiae (yeast) were identified using TreeFam, and the Ensembl-Compara database [55], and found to agree in all cases examined. Protein divergence between the two members of each C. elegans duplicate pair (Kapair, Table S2), and between C. elegans-C. briggsae orthologs (KaCeCb), was measured using Li's 1993 protocol (correcting for multiple hits using Kimura's 2-parameter model) [56], [57].
Gene Ontology (GO) analysis was carried out using files obtained from the GO Consortium website http://www.geneontology.org/ (downloaded January 2012). Enrichment was assessed using Ontologizer 2.0 [58]; P-values were corrected for multiple hypothesis testing by Bonferroni correction.
|
10.1371/journal.ppat.1006421 | Experimental study of tuberculosis: From animal models to complex cell systems and organoids | Tuberculosis (TB) is a devastating disease to mankind that has killed more people than any other infectious disease. Despite many efforts and successes from the scientific and health communities, the prospect of TB elimination remains distant. On the one hand, sustainable public health programs with affordable and broad implementation of anti-TB measures are needed. On the other hand, achieving TB elimination requires critical advances in three areas: vaccination, diagnosis, and treatment. It is also well accepted that succeeding in advancing these areas requires a deeper knowledge of host—pathogen interactions during infection, and for that, better experimental models are needed. Here, we review the potential and limitations of different experimental approaches used in TB research, focusing on animal and human-based cell culture models. We highlight the most recent advances in developing in vitro 3D models and introduce the potential of lung organoids as a new tool to study Mycobacterium tuberculosis infection.
| Tuberculosis (TB) is the number 1 killer in the world due to a bacterial infection. The study of this disease through clinical and epidemiological data and through the use of different experimental models has provided important knowledge on the role of the immune response generated during infection. This is critical for the development of novel vaccines and therapeutic strategies. However, in spite of the advances made, it is well accepted that better models are needed to study TB. This review discusses the different models used to study TB, highlighting the advantages and disadvantages of the available animal and cellular models and introducing recently developed state-of-the-art approaches based on human-based cell culture systems. These new advances are integrated in a road map for future study of TB, converging for the potential of lung organoids in TB research.
| Tuberculosis (TB) kills over 1.8 million people every year and thus remains the leading cause of death by an infectious agent [1]. Additionally, TB afflicts over 10.4 million new individuals per year and is estimated to exist in a latent form in nearly 2 billion people worldwide [1]. In addition to the human toll, TB imposes a significant economic burden, corresponding to 0.52% of the global gross national product, with a cost of over 500 million euros per year in the European Union alone [2]. Tackling TB is therefore a matter of urgency, as reflected in the current WHO End TB Strategy, which targets a 90% reduction in the incidence of TB to less than 100 cases per million people by 2035 [3]. Achieving this target requires a much quicker decline in TB incidence, from the current annual reduction of 2% to a 20% decrease per year [4, 5]. For this, 3 areas in TB research are generally accepted as critical: development of novel vaccines, improved diagnostic tools, and better treatment options [5, 6]. Succeeding in advancing these areas requires fresh approaches and ways of thinking, notably the development of better experimental models to study TB [7]. In this review, we discuss the different experimental approaches used in TB research, from in vivo models to human-based cell culture ones (Table 1). We also propose a road map of the available experimental approaches to study TB and of alternatives that are envisaged in a near future (Fig 1). We highlight the most recent advances in developing in vitro 3D models and introduce the potential of lung organoids as a new tool to study host—pathogen interactions during Mycobacterium tuberculosis infection. The development of such models requires a deep understanding of the disease pathogenesis and of the immune players, which are not the focus of this review and have been extensively reviewed elsewhere [8–10].
Several animal models are used in TB research (Fig 2), ranging from zebrafish to nonhuman primates (NHPs) [11, 12]. Mice are preferred model animals for a number of practical reasons, such as availability of immunological-based tools for mice, the existence of genetically modified mouse strains, and the small size and cost-effectiveness of maintaining mice in the laboratory [13–15]. Whereas many important aspects of the immune system are indeed conserved, there are also important differences that hamper the use of the mouse model of infection in our understanding of TB pathogenesis. The mouse is not a natural host for M. tuberculosis, and lung cavitation, a key characteristic for the disease transmission in humans [16], is not observed for the 2 most-used mouse strains (Balb/c and BL6) [13]. Necrotizing responses to M. tuberculosis occur in other mouse strains [17], indicating the impact of genetic variability on the outcome of infection. A recent study illustrates this fact by demonstrating that the susceptibility to TB infection and the efficacy of Bacillus Calmette-Guerin (BCG) vaccination varied greatly when genetically different mouse strains were used [18]. It is thus not surprising that, depending on the mouse strain used, different studies report different data. Furthermore, variability in the reported results is enhanced by different experimental end points used [19, 20]. The route and dose of M. tuberculosis administration and the mouse microbiome are also thought to contribute to variable findings.
Since currently used mouse models fail to fully reflect human immunity to TB, several studies were performed using humanized mice. Humanized mice can be generated through the reconstitution of immunocompromised mice with human hematopoietic cells of different origins [21]. Infection of humanized mice with M. tuberculosis reproduced important hallmark features of human TB disease pathology, such as the formation of organized granulomatous lesions, caseous necrosis, and bronchial obstruction [22, 23]. However, abnormal T-cell responses and an impaired bacterial control were also observed [23]. In line with this, humanized mice generated by engraftment of human leukocyte antigen (HLA)-restricted cells showed partial function of innate and adaptive immune systems, culminating in antigen-specific T-cell responses to mycobacterial infection but also in lack of protection [24]. Other approaches consist in infecting transgenic mice expressing human-specific molecules such as, for example, the human cluster of differentiation group 1 CD1, which allows for the study of a humanized immune system using the mouse model of infection [25]. In all, humanized mice are a good tool to study TB, being particularly relevant for the study of HIV/TB, as recently shown [26]. However, this model requires further improvement to reach its full potential for TB research.
To address some of these limitations, other animal models have been used. For example, guinea pigs and rabbits may be considered better models to study the humanlike granuloma formation, a hallmark of M. tuberculosis infection in the lung [14, 27, 28], although they still fail to display other characteristics of the human disease. Additionally, they are much more difficult to maintain in the lab and a lot less immunology tools are available for these 2 species, which greatly limits their use. Infection of zebrafish (Danio rerio) embryos with the natural fish pathogen M. marinum is also used as a model for the study of granuloma formation [29–31]. Several similarities were found in the cellular and molecular events presiding M. marinum and M. tuberculosis infections [32–34], despite the many differences between these 2 diseases. Research on zebrafish embryos benefits from the similarities between M. marinum and M. tuberculosis, i.e., from the optical transparency of the embryos, which facilitates the use of advanced imaging techniques, and from the easy genetic manipulation of zebrafish, which allows for deep mechanistic molecular studies. Because zebrafish embryos lack a fully developed immune system, the study of later stages of infection requires the use of adult fish, thus abrogating the advantages of using embryos. Furthermore, the physiological differences between zebrafish and humans are enormous, which inevitably imposes some limitations to the use of this model. As for the other animal models, specific facilities for housing zebrafish are required. NHPs are so far considered as the best animal model for TB research [35, 36], as the disease pathogenesis parallels that observed in humans [37]. NHPs present lung cavitation [38]; show a spectrum of disease overlapping that of humans, namely, with the establishment of latent TB infection [38]; display a susceptibility to TB in the presence of comorbidities such as HIV and anti—tumor necrosis factor (TNF) treatment similar to that reported in humans [39, 40]; and present a transcriptomic signature of disease comparable to the human one [41]. However, the ethical, practical, and economic problems that are inherent to NHP research [36, 42], exacerbated when the animals are made to develop a potentially fatal infection, hinder the generalized use of this animal model, which in fact accounts for only 1% of the papers published in TB (Fig 2). In conclusion, important advances in our understanding of TB have been made through the use of different animal models. However, in addition to each model’s specific limitations, all animal-model research into human diseases is ultimately restricted by the need to translate findings across species. This calls for the wider use of human-based models to complement and reduce the use of experimental in vivo research.
Owing to the central role of the macrophage as host and effector cell during M. tuberculosis infection [43, 44], many studies have been centered in macrophage cell cultures. In terms of human-based systems, monocyte-derived macrophages are the most widely used culture. Among these is the human monocytic leukemia cell line, THP-1, which is easy to culture, yielding a nearly unlimited amount of cells for experimental purposes. THP-1 cells are typically differentiated to macrophages through the stimulation with phorbol 12-myristate 13-acetate (PMA) for 3 days, although different protocols are found in the literature [45, 46], which may contribute to some variable findings. Macrophages can alternatively be freshly derived by extracting and culturing human peripheral blood mononuclear cells (PBMCs) in the presence of differentiating factors, namely, granulocyte-macrophage colony stimulating factor (GM-CSF) or macrophage colony stimulating factor (M-CSF) [47], or of human serum [48]. In these cases, the macrophages are of primary origin, but because of the in vitro differentiation process, their properties are most likely different from tissue-resident cells. Although alveolar macrophages would be ideally used, access to these cells is a costly procedure that requires lengthy ethical approvals, which limits their use. In vivo, M. tuberculosis is found in foamy macrophages. These cells result from pathogen-induced dysregulation of host lipid synthesis and sequestration and play a key role in both sustaining persistent bacteria and contributing to the tissue pathology [49]. Therefore, in vitro differentiation of foamy macrophages is an excellent tool for the study of macrophage-pathogen interactions. A protocol to convert cultured macrophages (THP-1 or primary) into foamy cells has been developed by incubating these cells under hypoxia [50]. Other alternatives for the differentiation of foamy cells include the exposure of cell cultures to palmitic acid, oleic acid, or lipoproteins [51] or to surfactant lipids [52].
Given the importance of working with primary, unmanipulated cells, many studies have been performed using freshly isolated human PBMCs [53]. Human PBMCs are easily accessible, cost-effective, and readily infected with M. tuberculosis, responding to the infection with the production of relevant immune mediators such as TNF and other interleukins as well as chemokines [53]. Furthermore, the PBMC response captures interactions between different immune cell types, such as monocytes, T cells, and B cells, which are in fact interacting during natural immune responses. However, these cells still differ from the tissue-resident ones and when used in in vitro cultures lack the environmental stimuli that ultimately shape cellular responses to infection. In addition to the standard monolayer cultures, PBMCs have been used to develop in vitro models of human mycobacterial granulomas. In 1 study [54], a sequential recruitment of human monocytes and lymphocytes towards mycobacterial antigen-coated artificial beads or live mycobacteria was observed. This recruitment culminated with the formation of a cellular structure reminiscent of natural mycobacterial granulomas in terms of morphology and cell differentiation [54]. This or similar/improved models have been used in several studies [55–57]. A different approach based on the culture of human PBMCs in a collagen matrix with a low dose of M. tuberculosis was used to develop an in vitro model of human TB granuloma with dormant bacteria [58]. This model recapitulated important characteristics of the mycobacterial granuloma, such as the aggregation of lymphocytes surrounding infected macrophages, the formation of multinucleated giant cells, the presence of secreted cytokines and chemokines in the culture supernatants, and the reactivation of M. tuberculosis upon immune suppression caused by TNF blockade [58]. These models offer the possibility of studying the infection by M. tuberculosis in a more physiological environment, resembling the structure of the infected human tissue. They constitute valuable approaches for the study of cell—cell interactions, cell differentiation, and bacterial control.
To further reflect the complex environment and structure of the human lung, a growing body of studies are resorting to the use of new technologies in the tissue-engineering field to advance human-based TB research models into the 3D era (Fig 3) [59]. Tissue bilayer systems consisting of epithelial and endothelial cell layers were initially developed to study the early events of alveolar infection [60, 61]. More recently, through the use of these systems, microfold (M) cells were shown to play a critical role in translocating M. tuberculosis to initiate lung infection [62]. A study combining lung-derived epithelial cells and fibroblasts with peripheral blood primary macrophages reported the establishment of a lung tissue model that upon infection led to the clustering of macrophages reminiscent of early TB granuloma formation [63]. Similarly, another report showed the implementation of an in vitro human 3D lung tissue model to study M. tuberculosis infection that allowed the analysis of human granuloma formation and resembled some features of TB [64].
A novel bioengineering approach utilized bioelectrospray technology to generate microspheres of M. tuberculosis–infected human PBMCs in a 3D extracellular matrix [59, 65]. This model takes advantage of the high throughput potential of the bioelectrospray system and allows the interrogation of host—pathogen interactions in 3D in the context of an extracellular matrix [59, 66]. When combined with a microfluidic system to enable pharmacokinetic modeling, this model also showed great potential to monitor the efficacy of new antibiotic regimens or anti–M. tuberculosis drugs [65]. Although these experimental systems facilitate the discovery of the interactions between mycobacteria and host cells in a more physiological environment, they still bear some limitations, namely, the lack of vasculature and absence of other immune cells (e.g., neutrophils) that play a role in the multifaceted response in TB infection. Also, not all the models include epithelial and stromal cells, which are known to play important roles during infection [67, 68]. Finally, the spatial organization of the lung is mostly lost, and so is the role of the anatomical constraints during infection. The advances made in the development of all these models will certainly contribute to moving the field forward into novel strategies that overcome current limitations. In this context, other 3D and tissue-chip models are being explored.
Organoids are in vitro 3D cell cultures generated from embryonic stem cells (ESCs), induced pluripotent stem cells (iPSCs), or adult stem cells (aSCs) that functionally and structurally mimic the organ they model [69, 70]. This technology is emerging as a promising tool to study organ development and disease “in a dish” [69, 70]. The potential of organoids to study infectious processes has been increasingly demonstrated in many original papers and recently reviewed by Mills and Estes [71], with most examples coming from human gastric [72], brain [73, 74], and gut [75, 76] organoids. So far, lung organoids have not been explored as a model to study infection.
Human lung organoids have been generated through different technologies [77]. The most advanced studies involve the differentiation of human embryonic stem cells into endoderm cells, anterior foregut endoderm cells, lung progenitor cells, and, finally, various types of airway epithelial cells. If this procedure is performed in a 3D structure, a human lung organoid is formed, as initially described by Dye et al. [78] and Konishi et al. [79]. These relatively immature organoids may be transplanted into mice to complete their differentiation in an in vivo environment, into adultlike airways [80].
Despite some limitations, lung organoids recapitulate important features of the lung, such as heterogeneous cell composition, spatial organization, and retention of a stem cell population harboring the capacity for both self-renewal and differentiation [70]. There is increasing evidence that human lung organoids may be used to investigate the cellular and molecular pathways implicated in lung development and lung diseases as well as screening platforms for drugs directed at respiratory diseases [77]. At the disease level, the application of lung organoids to cancer development, cystic fibrosis, and infection is envisaged although still is unexplored in TB research. The obvious advantage of lung organoids over 2D and 3D cultures relies on their spatial organization and heterogeneity of the cellular components. As compared to the animal model, infection of lung organoids allows the inclusion of very early time points, which are difficult to follow in in vivo infections, whilst at the same time overcoming species differences and reducing the use of animals in research. Thus, as the lung organoid technology stands, human-derived lung organoids could be explored to study the early events of infection, namely, the initial interactions of M. tuberculosis with the lung epithelium [67]. Of the aforementioned experimental models, both 2D and 3D cultures based on PBMCs may also be explored to investigate the early immune events during infection, although to a lesser complexity than organoids.
Although there are indeed exciting perspectives for the use of lung organoids as a model for TB research, some important challenges remain before they can be more systematically used as experimental models. Chief among these is the introduction of immune cells in the structure of lung organoids. Only then will lung organoids cover the complexity of immune response and of the stromal-immune cells’ cross talk upon in vitro infection. Also, the introduction of the vasculature would be an important improvement to create a more dynamic model in which the microenvironment of an airway could be experimentally controlled. This dynamic lung organoid would be an interesting model for drug screening. In this context, microfluidic cell culture devices called “organs on a chip” have also generated airway epithelium from human adult airway cells grown on an air—liquid interface platform [81, 82]. Another important step forward would be the development of lung organoids from iPSCs instead of ESCs, as this will offer the possibility of including in the disease modelling individual variability, either genetic or caused by extrinsic conditions. In the context of TB research, this would allow for the study of host–M. tuberculosis–microenvironment interactions at an individual level by infecting lung organoids generated from individuals with HIV or diabetes versus controls or from smokers versus nonsmokers. This would be of utmost importance as the molecular mechanisms underlying the impact of comorbidities and life habits on the course of infection remain incompletely understood. Additionally, comorbidities are very difficult to incorporate in the other complex experimental system—the animal. Generation of personalized lung organoids would also open new avenues for the study of individual responses to therapies and thus for the implementation of personalized medicine.
TB remains a devastating disease to mankind and a huge challenge for the scientific community. From many epidemiological studies, it is clear that the progression of the disease is highly related to the host immune status, and as such, a deep understanding of the immune response to M. tuberculosis is critical for the development of novel preventive and therapeutic strategies. However, the lack of experimental systems that parallel the complexity of the human disease remains a major gap hindering the in-depth study of the immune response in TB. Critical species differences mean that the widely used animal models only partly recapitulate the human disease. NHP models are the most representative ones but bear high operational and maintenance costs. Traditional human cellular systems overcome the interspecies translation problem but are limited by their low level of complexity and the abnormal characteristics of cell lines. Recent development of human-based tissue models is promising real alternatives for the experimental study of human TB. State-of-the art in vitro models have now incorporated several important characteristics of “real-life” tissues, namely, the presence of different cell types and of the extracellular matrix. Technological advances coupled to these models allow for the experimental manipulation of different variables, which is critical in studies of host—pathogen interactions or in drug-screening processes. A key next step will be to introduce in these models the anatomical constraint associated with the lung tissue. Albeit at very early days, lung organoids hold a great promise here. The road from lung organoids to complete lungs “in a dish” is still a long one, but creating a lung structure composed of different stromal cells and coupled with a competent immune system would unquestionably provide a major leap forward in TB research. Being able to use as starting points cells from different individuals (TB patients or latently infected people with different genetic backgrounds and comorbidities) would constitute a revolutionary way of studying TB. This would open many new avenues to investigate long-standing questions and put us in a privileged position to effectively tackle TB.
In sum, recent advances in tissue engineering and future steps in this area will certainly play an important role in the development of new tools for the study of infectious diseases. Such tools hold the potential to replace some animal experiments and overall lead to a reduction of the number of animals used in TB research. Most importantly, these tools will allow for a series of key questions to be answered in a more precise way by including individual variability at the single-cell and tissue levels.
|
10.1371/journal.pcbi.0030188 | A Statistical Framework for Modeling HLA-Dependent T Cell Response Data | The identification of T cell epitopes and their HLA (human leukocyte antigen) restrictions is important for applications such as the design of cellular vaccines for HIV. Traditional methods for such identification are costly and time-consuming. Recently, a more expeditious laboratory technique using ELISpot assays has been developed that allows for rapid screening of specific responses. However, this assay does not directly provide information concerning the HLA restriction of a response, a critical piece of information for vaccine design. Thus, we introduce, apply, and validate a statistical model for identifying HLA-restricted epitopes from ELISpot data. By looking at patterns across a broad range of donors, in conjunction with our statistical model, we can determine (probabilistically) which of the HLA alleles are likely to be responsible for the observed reactivities. Additionally, we can provide a good estimate of the number of false positives generated by our analysis (i.e., the false discovery rate). This model allows us to learn about new HLA-restricted epitopes from ELISpot data in an efficient, cost-effective, and high-throughput manner. We applied our approach to data from donors infected with HIV and identified many potential new HLA restrictions. Among 134 such predictions, six were confirmed in the lab and the remainder could not be ruled as invalid. These results shed light on the extent of HLA class I promiscuity, which has significant implications for the understanding of HLA class I antigen presentation and vaccine development.
| At the core of the human adaptive immune response is the train-to-kill mechanism in which specialized immune cells are sensitized to recognize small peptides from foreign pathogens (e.g., HIV virus). Following this sensitization, these cells are then activated to kill other cells that display this same peptide (and that are infected by this same pathogen). However, for sensitization and killing to occur, the pathogen peptide must be “paired up” with one of the infected person's other specialized immune molecules—an HLA (human leukocyte antigen) molecule. The way in which pathogen peptides interact with these HLA molecules defines if and how an immune response will be generated, which has implications for vaccine design where one may artificially introduce select peptides to pre-train the immune system. Furthermore, there is a huge repertoire of such HLA molecules, with almost no two people having the same set. We introduce a statistical approach for identifying which HLA molecules interact with which pathogen peptides, given a particular kind of laboratory data. Our approach takes as input, data that tells us only which pathogen peptides generate a response, but not which HLA molecules support the response. Our statistical approach fills in this missing information.
| The human adaptive immune response is composed of two core elements: antibody-mediated response (sometimes called humoral response), and T cell–mediated response (sometimes called cellular response). Research on HIV vaccines initially focused on the antibody-mediated response but more recently has included the cellular response [1,2], which is the focus of our application.
At the core of the cellular response is the ability of certain antigen-presenting cells to digest viral proteins into smaller peptides, and then to present these peptides at the surface of the cell. Presentation of a peptide depends on the peptide first forming a complex with an HLA (human leukocyte antigen) molecule. If a peptide is presented, it can then be recognized by (naive) T cells, allowing activation of these T cells so that they may subsequently recognize and attack virally infected cells displaying the same complex. Any peptide that is able to generate such an immune response in the context of a given HLA allele is called an epitope, and, in particular, an epitope restricted by that allele. Only certain HLA alleles can form a complex with any given peptide, and hence the compatibility of these two elements is essential for the adaptive immune response just described.
Several types of T cells exist, each playing its own, though interdependent, role. In ongoing HIV vaccine research, the elicitation of a CD8+ T cell response has shown promise. Since CD8+ T cells recognize only HLA class I bound epitopes, our data, and hence our paper, focus on epitopes recognized in the context of these particular molecules, although the statistical framework is not tailored or limited to this domain and could be immediately applied to HLA class II epitopes, for example. Humans have up to six HLA class I alleles arising from the A, B, and C loci. Currently, there are hundreds of possible alleles at each of these loci, with more being discovered every year [3].
A crucial task in HIV vaccine development is the identification of epitopes and the alleles that restrict them, since it is thought that a good vaccine will comprise a robust set of epitopes [4–6]. By robust, we mean a set which broadly covers regions that are essential for viral fitness in a given human population characterized by a particular distribution of HLA alleles. Also, note that beyond vaccine design, epitope identification may have important applications such as predicting infectious disease susceptibility and transplantation success.
Traditional methods for identifying epitopes involve time-consuming, technically demanding, and expensive culturing of T cells. Recently, a more expeditious laboratory technique using ELISpot assays has been developed [7]. Unfortunately, the ELISpot assay gives only information about which individual donors generated an immune response to a particular peptide, but does not provide any information about which of a donor's HLA alleles are restricting this reaction; it is this HLA specificity that is crucial and in which we are most interested. However, by leveraging information contained in ELISpot reactivity across a large set of donors with known HLA types, in conjunction with the statistical model presented in this paper, we can determine (probabilistically) which HLA alleles are likely to be responsible for the observed reactivities. Thus we are able to learn about new HLA-restricted epitopes in an efficient, cost-effective, and high-throughput manner.
A related, though distinct problem from our problem of epitope identification is that of epitope prediction (e.g., [8–11]), in which new epitopes are predicted in silico, on the basis of amino acid sequence and other information, but not on the basis of assays that directly measure binding energies or other measures such as the ELISpot assay. The work presented here focuses strictly on the identification of restricting (i.e., epitope presenting) HLA class I alleles from ELISpot data, although newly identified epitopes can aid the task of epitope prediction by providing more known examples to learn from.
Our statistical model takes as input, measured CD8+ T cell ELISpot reactivities from a set of donors with known HLA class I alleles, for a number of epitopes, and deduces which of the donor's individual HLA alleles are likely to be responsible for the observed reactivities. That is, our model deduces which epitopes are restricted by which HLA alleles. Additionally, we can provide a good estimate of the number of false positive epitope hypotheses returned by our analysis (i.e., the false discovery rate (FDR) [12,13]) so that we have a sense of how many new epitope hypotheses to pursue (if any).
We assume that a given epitope is or is not restricted by a given HLA allele. If an epitope is restricted by a particular HLA allele, it is still likely that a donor with the restricting HLA allele will not react to the epitope. Such false negatives arise from factors including immunodominance, (immunodominance can be thought of as biology's “waste not want not”—that is, the immune system focuses its efforts in a few areas that work well, to the exclusion of others), T cell repertoire, lack of previous peptide exposure (e.g., exposure arising from infection or vaccination), suboptimality of the epitopes (i.e., if a peptide that optimally binds a particular HLA is of length nine amino acids, a peptide of length ten which contains the nine-mer may sometimes bind, but not as efficiently), and experimental noise. Furthermore, an epitope reaction may be falsely associated with some HLA alleles in ELIspot data due to linkage disequilibrium of a nonrestricting and a restricting HLA allele. (For example, if a restricting HLA allele is in linkage disequilibrium with a nonrestricting HLA allele, then the nonrestricting allele will very often be present in a donor with the restricting allele, and so the ELISpot data for this allele will also correlate with positive reactivities—though only as a result of the linkage.) Thus, the task of recovering HLA-restricted epitopes from ELISpot data is not straightforward. As a brief example, if one examines the HIV ELISpot dataset used in this paper and considers any HLA-epitope pair that has any observed reactivity to consist of an HLA-restricted epitope, then one incurs a false positive reactivity rate of roughly 70%. One could then imagine a next logical step of setting a threshold for what minimum fraction of donors must react and so on, soon finding oneself with a rather ad hoc model for which there would be no principled way to set the parameters nor to determine statistical significance. The task of identifying restricting HLA alleles from ELISpot data is in fact nontrivial and well-suited to statistical modeling. Next, we formally outline our statistical model.
For a set of J epitopes (more precisely, each peptide under examination may contain one, or several, epitope(s), but for simplicity of presentation, we refer to the peptides as epitopes) and K donors, we have a set of measured binary ELISpot reactivities (actual laboratory assays provide real values which are thought by the laboratory scientists to convey mostly binary information [14]), which are used as input to our model. We are also given the six HLA class I alleles for each donor.
Let hi = 1 denote that a donor has HLA allele i, and hi = 0 denote that the donor does not have that allele. Let yj be the observed, binary reactivity for epitope j in a donor (as measured by the ELISpot assay). An important assumption in our model is the following: whether an epitope is restricted by a particular HLA allele is independent of whether that epitope is also restricted by any other HLA allele. This assumption is commonly referred to as an assumption of causal independence [15]. From this assumption, it follows that the probability of not observing a reaction to a particular epitope, in a given donor, is the probability that none of that donor's HLA alleles cause a reaction. Because of the independence assumption, this is simply the product over the probability of each HLA (that the donor has) not causing a reaction. Formally, if epitope j is restricted by HLA i, then we let qij be the probability that we observe a reaction in a donor with HLA i and no other HLAs restricted by epitope j. Also, let lj be the probability that a reaction is observed to epitope j when a donor has none of the restricting HLAs for epitope j—a so-called leak term (corresponding to unrepresented causes such as reactivity due to HLA E molecules). Given settings for these parameters, qij and lj, our model stipulates that the probability that a donor does not react to epitope j, p(yj = 0 | {qij},lj), or does react, p(yj = 1 | {qij},lj), is given by
Such a model is sometimes referred to as a noisy-OR model. It can be viewed as a probabilistic version of the common (deterministic) logical OR, and has been shown to be useful in a number of settings [16]. The model can be represented in graphical form as shown in Figure 1. Here, nodes represent the variables {hi} and {yj} and an arc is drawn from hi to yj if epitope j is restricted by HLA i (i.e., if qij > 0). The characteristics of how the probability of an observed reaction changes with an increasing number of restricting alleles depends on the values of {qij}. For example, if qij ≡ qj ≈ 1, then for a given donor with M restricting alleles, each additional restricting allele beyond one allele would do little to increase the probability of a reaction to epitope j (as with a deterministic logical OR). Alternatively, if qij ≡ qj ≈ 0, then according to the Taylor series expansion
, the probability of reactivity to epitope j would increase roughly linearly with M. The likelihood of the ELISpot data under this model is simply the product of likelihood terms for the reaction in each patient k, to each peptide (given the HLA types for each patient) :
Given the model just described, and experimental ELISpot and donor HLA data, we wish to infer which epitopes are restricted by which HLA alleles. That is, we wish to know which qij should be included in the model (which arcs should appear in the graphical model). This is a problem of model selection. Note that this problem breaks down into J separate problems, one for each epitope under consideration, since under our model, qij and qi′j′ are independent from one another when j ≠ j′.
To tackle this problem of inferring HLA-restricted epitopes from our data and model, one might consider simply learning a maximum likelihood value for all possible qij simultaneously, and concluding that those for which qij > 0 are those which support the hypothesis of an HLA-restricted epitope. However, in practice, with finite and noisy-data, almost all qij > 0, and this approach would lead to a huge number of false epitope hypotheses. Instead, we need a more robust way of deciding which qij to include in the model. There are a variety of standard approaches to this problem, most centered on some form of model selection score, such as the Akaike Information Criterion (AIC) [17], the BIC (Bayesian Information Criterion) [18], or the MDL (Minimum Description Length) [19]—all of which are forms of penalized likelihood scores. These scores are but three commonly used model selection scores, and many variations of these exist as well. However, all of these scores have an intuitive interpretation of balancing the fit of the data to the model, with model complexity (controlling the model complexity so that overfitting does not occur). The fit of the data to the model is usually assessed by the maximum likelihood of the data under the model in question, while the model complexity is usually controlled by penalizing for the number of free parameters in the model—hence the term penalized likelihood. For example, the AIC of a model, M, is given by
, where
is the maximum likelihood, and Q is the number of independently adjusted parameters in the model.
Given a model selection score, one then chooses a search procedure to select qij (arcs) for inclusion in the model. The ideal way to do so is to try every subset of arcs and choose the subset which gives the highest model score (for example). However, with n possible arcs per epitope there are 2n subsets, and this approach is not feasible for most problems. Thus, in practice, it is common for some form of greedy, stepwise procedure to be used, such as greedily adding arcs to the model, or greedily adding/deleting arcs, terminating when the model score can no longer be increased. Then the final model built in the greedy sequence of models is taken as the model to be used and/or interpreted. Commonly, the search is started with the empty model (no arcs). In synthetic experiments with our model, we found that a greedy add/delete procedure, starting from the empty set, worked well (see Results for details), and thus we use such a procedure to identify specific HLA alleles restricting given epitopes.
It may at first seem counterintuitive that deleting an arc could increase the score when in a previous step adding that same arc had increased the score. However, when one considers that different variables can explain the same data to differing degrees, then it becomes clear how this can arise. Suppose one arc most explains some part of the data, followed next by, say, two other arcs, each of which explains that part of the data less well than the first arc, but which together explain the data better than the first arc by itself. In this case, after addition of the first arc, followed by addition of the next two arcs, the first arc would become redundant in light of the other arcs, and so removing it can increase the model selection score (it will not improve the likelihood, but will incur a parameter penalty). In practice, for our problem and data, the delete operator was used only occasionally.
Different model selection scores used in a given search procedure lead to different recovered models. In particular, AIC is known to be generally less conservative (allowing more arcs) as compared with, say, BIC and MDL. Note that if one were to use an add-only procedure (where deletion of an arc is not allowed) for noisy-OR based models, then the AIC, BIC, MDL, and the Likelihood Ratio Test (LRT) [20] would each add arcs in the same greedy order, though with each score stopping at a different point in the search (except for BIC and MDL which are equivalent). So the fundamental difference between these scores is not so much which arc to add next, but when to stop adding arcs.
Rather than dogmatically choosing one score with which to find restricting HLA alleles, we develop a novel approach in which we use a parameterized family of model scores. Then, for any chosen model score parameter setting, we are able to estimate the FDR of the resulting model (that is, we are able to estimate the proportion of recovered qij which are not truly HLA-restricted epitopes). Then we choose a model score parameter setting which produces an FDR that we find reasonable for our purposes (i.e., one producing an FDR that gives us enough epitope hypotheses to pursue, but not too many false leads). This approach to model selection confers two advantages over the more traditional approach described: (1) we do not depend in a fundamental way on the choice of a single model selection score, and (2) regardless of which model selection score we use (within the parameterized family), we are able to estimate the FDR of our selected arcs, providing us with a good sense of what (interpretable) features the model has actually recovered, rather than, say, far less interpretable measures of quality such as the maximum likelihood of the data under the recovered model compared with that under some baseline model.
We call the parameterized family of model scores XIC, (to denote that it encompasses various Information Criterion such as AIC and BIC). The XIC for model M is parameterized by f and is given by
where
is the maximum likelihood of model M (M represents, for example, a model consisting of a particular subset of {qij}), Q is the number of independently adjusted parameters in the model, and f parameterizes the family of scores represented by XIC. When f = 1, the XIC behaves identically to the (negative) AIC during search, because it is directly proportional to it. When f = ½ log N, where N is the sample size of the data, then the XIC is identical to the BIC. When f = 0, the XIC is the maximum likelihood. Thus by varying f, the XIC spans a range of model selection scores, from very liberal ones for low values of f, to increasingly conservative ones for higher values of f.
Leaving aside the issue of estimating the FDR for the moment, our model selection procedure is the following:
1. Select a value for the XIC parameter, f = f*.
2. Start with the empty set of arcs under consideration (that is, no qij are in the initial model), but include all of the leak terms, lj. Compute the XIC of this “leak-only” base model, M0.
3. For every qij under consideration, compute the XIC of the model which is the same as M0 but also includes qij. If none of these models has a higher XIC than M0, stop the search. Otherwise, add the qij whose corresponding XIC was largest, and call the resulting model, M1.
4. Repeat the previous step, except using M1 in place of M0, and also allowing arc deletions: for all qij in M1, compute the XIC of the model which is the same as M1, except that it does not contain qij. Among all the possible arc additions and deletions, choose the operation which most increases the XIC, and call the resulting model M2.
5. If possible, continue greedily adding/deleting arcs, stopping when the XIC can no longer be increased.
Then we use the last model in the sequence as our final model from which to infer HLA-restricted epitopes. That is, for all qij included in the final model, we will call the hypothesis that epitope j is restricted by HLA i, true. The smaller f* is, the more qij will be included in the final model. Next we show how to estimate the number of qij recovered using this procedure that we expect to be spurious (i.e., arising from chance alone, rather than from true HLA restrictions).
For any specified value of the model selection parameter, f, we want to know how many qij in the recovered model are likely to be true (rather than spuriously generated). That is, we want some sort of statistical significance measure for the epitope hypotheses we have generated. We compute such a measure using a method that we have recently developed [21]. Next we provide some background to this area of research, followed by presentation of our approach.
When inferring whether a single hypothesis is true or not, statisticians have traditionally relied on the p-value, which controls the number of false positives (type I errors). However, when testing hundreds or thousands of hypotheses simultaneously, the p-value needs to be corrected to help avoid making conclusions based on chance alone (known as the problem of multiple hypothesis testing). A widely used, though conservative correction, is the Bonferroni correction, which controls the Family Wise Error Rate (FWER). The FWER is a compound measure of error, defined as the probability of seeing at least one false positive among all hypotheses tested. In light of the conservative nature of methods which control the FWER, the statistics community now places great emphasis on estimating and controlling a different compound measure of error, the false discovery rate (FDR) [12,13].
In a typical computation of FDR, we are given a set of hypotheses where each hypothesis, i, is assigned a score, si (traditionally, a test statistic, or the p-value resulting from such a test statistic). The FDR is computed as a function of a threshold, t, on these scores, FDR = FDR(t). For threshold t, all hypotheses with si ≥ t are said to be significant (assuming, without loss of generality, that the higher a score, the more we believe a hypothesis). The FDR at threshold t is then given by
where S(t) is the number of hypotheses deemed significant at threshold t and F(t) is the number of those hypotheses which are false, and where expectation is taken with respect to datasets of the same sample size as the observed data drawn from the true joint distribution of the variables. When the number of hypotheses is large, as is usually the case, one can take the expectation of the numerator and denominator separately:
Furthermore, it is often sufficient to use the observed S(t) as an approximation for E[S(t)]. Thus, the computation of FDR(t) boils down to the computation of E[F(t)]. One approximation for this quantity which can be reasonable is E[F(t)] ≅ E0[F(t)], where E0 denotes expectation with respect to the null distribution (the distribution of scores obtained when no hypotheses are truly significant), and it is this approach that we take. (For traditional applications of FDR, Storey and Tibshirani offer a clever method to compute E[F(t)] which is less conservative than using E[F(t)] ≅ E0[F(t)] [13]. However, this approach is not appropriate in the present context.)
Applying this approach to estimating the number of true qij recovered by our model selection procedure (i.e., the number of true HLA-restricted epitopes found by our model), we generalize S(·) and F(·) to be functions of f, the XIC parameter in Equation 4. In particular, S(f) is the number of qij found by our model selection procedure when the XIC is used with parameter setting f and F(f) is the number of those qij which do not truly correspond to HLA-restricted epitopes (i.e., false positives). As in the standard FDR approach, we use the approximation E(S(f)) ≅ Q(D,f), where Q(D,f) is the number of qij found by applying our model selection procedure with XIC parameter f to the observed data D (in our application, D ≡ {yj}). In addition, we estimate E0(F(f)) to be N(Dr,f) averaged over multiple datasets Dr, r = 1,…,R, drawn from a null distribution. That is, we estimate the FDR of our HLA-restricted epitopes using the following:
The addition of 1 to the numerator smoothes the estimate of E0[F(f)] so as to take into account the number of random permutations performed. Without this smoothing, if one performed too few random permutations such that ∑fQ(Dr,f) = 0 due to sampling error, then the estimate of E0[F(f)] and hence FDR (f) would also be 0. We prefer our more conservative estimate, especially as the bias it induces diminishes as the number of permutations increases.
We sample Dr from a null distribution for each epitope by permuting the ELISpot data for that epitope, but leaving the HLA types of the donors intact. This permutation guarantees that any qij recovered from the model selection procedure on this data are only spuriously recovered. Also note that although the parameters qij are independent for different epitopes, j, and thus the model selection procedure, can operate independently on each epitope, for the purposes of estimating the FDR, we pool all of the epitopes together, so that the approximations we make in computing the FDR are more reasonable.
As shown in the Results section, by way of synthetic experiments, we find that these approximations for estimating the FDR work quite well in practice. There is, however, one concern about the use of the null distribution described, for which we refer the reader to [21], but which, to our knowledge, does not affect our use of this methodology in this paper.
By construction, the emphasis of our FDR approach is on the accuracy of the estimate of the number of false positives, and does not examine the number of false negatives. Whereas this emphasis may seem undesirable, it is common for experimenters to be more interested in how many hypothesized interactions are real, rather than how many were missed, because experimenters will in most cases be using resources to pursue the positive hypotheses, not the negative ones. A similar line of reasoning is mentioned in [13,22].
The problem of finding a meaningful ranking of the individual HLA-restricted epitope hypotheses does not immediately fall out of the FDR framework. However, we can naturally construct a ranking algorithm for the epitope hypotheses by using a Likelihood Ratio statistic. Let M denote the model that we learn with our model selection procedure (regardless of the value of f used). Then we rank our hypotheses using a likelihood ratio statistic, vij, which is the log of the ratio of the likelihood of the final model, to that of the final model without the qij we are evaluating. Specifically, our ranking algorithm is:
For each qij included in M, do the following: construct a model, M′ij defined to be model M, but without qij, and then compute the likelihood ratio:
Assign a rank to each qij equal to the rank of vij in the set {vij}.
This ranking assesses each qij based on how much it contributes to the likelihood of the data in the model, M, in the context of all qij recovered from the model selection procedure. (The likelihood ratio, vij, viewed from a Bayesian perspective, is a quantity proportional to a BIC approximation to the Bayes factor [18], which, under the assumption of a uniform prior over arc sets, amounts to the posterior probability of qij being included in the model, given the remaining arcs in the model.)
For our experiments, we used two types of datasets: laboratory-generated HIV ELISpot data, as well as synthetic data based on our model and this real data. The HIV ELISpot data is from a set of previously optimally defined CTL epitopes derived from HIV [14], which were generally optimized for length so as to be recognized at the lowest antigen concentration in the context of a specific restricting HLA class I allele. Note that these “optimal” peptides may not be optimal for other HLA class I alleles which could also restrict them—for example, other alleles could restrict epitopes that are embedded within the longer peptide sequence tested. There were 140 epitopes and 102 donors with a total of 70 unique HLA-I alleles (although HLA alleles are ideally described by a four-digit number; in many cases, this was not available, and as such, we truncated all HLA-I alleles to two digits, with the exception of the HLA-B15 family alleles, which always had the full four digits available since these “subtypes” may present vastly different sets of epitopes [23,24]. The number of unique HLA alleles reported is the number obtained after this compression.). First we use synthetic experiments to show that (1) the FDR estimate we have described is reasonably accurate, and (2) the model selection procedure can recover a good proportion of ground-truth HLA-restricted epitopes from data. Finally, we apply our algorithm to the real data.
Note that to compute the XIC score for our models, we need to find the maximum likelihood solution for noisy-OR nodes. Fortunately, this is a convex optimization problem [25] and therefore local minima are not a problem.
The synthetic model used to generate data was our epitope model, as described earlier, fitted to the real HIV ELISpot data by using our model selection procedure. We used an XIC setting for f that resulted in an estimated FDR ≅ 0.3 (f = 2.9). This produced 165 qij in our synthetic model. Additionally, we retained the learned maximum likelihood values for these qij (and the leaks, lj), so as to be able to generate data from the model. To generate synthetic data from this fitted model, donor HLA data was left as it appeared in the real data, and then Equations 1 and 2 were used to compute the probability that a particular donor would react to a particular epitope, pj, conditioned on the learned values of {qij} and {lj}. Then samples for each donor, sj, were drawn from a uniform distribution on (0,1] and the reactivities, yj, were set to yj = ( sj ≤ pj). Three synthetic datasets (each consisting of 102 donors and 140 epitopes) were generated in this manner, all from the same synthetic, generative model.
Plots of actual versus expected FDR for the three datasets are shown in Figure 2A. Estimates of FDR are quite accurate at the lower end, which is the region of interest for our problem and also most other problems of interest (where not too many spurious hypotheses are included). That the FDR becomes increasingly conservative (i.e., it peels away from the idealized line) can likely be explained by the approximation we make in generating a null distribution. Further discussion of this issue, and a suggested resolution, can be found in [21]. For the XIC parameter, f, we used the range [1.97,3.46], with f = 3.46 producing actual and estimated FDRs around 0.02, and f = 1.97 producing actual FDRs around 0.67 and estimated FDRs around 0.95. Note that BIC corresponds to the cluster of points that have estimated FDRs around 0.7 (f = 2.3). AIC (f = 1) corresponds to something even less conservative than anything shown (even higher FDRs).
Not only do we want to know that our FDR estimate is accurate, but we also want to know that our model selection procedure is a reasonable one. We therefore examine how many ground truth qij were recovered, and at what cost in false negatives. This information is displayed in Figure 2B. Note that because we created a synthetic model with what were presumed to be 30% spurious qij, many of these qij are likely quite small (signifying weak associations), and therefore would be more difficult to recover in synthetic experiments using data generated from this model. Such difficulties are also likely to arise with real data in real applications. The points in Figure 2B that have about 50 false positives correspond to an estimated/actual FDR of around 0.3. The points which have about 150 false positives are those corresponding to XIC = BIC (for which FDR ≅ 0.7). Overall, the tradeoff between the number of false positives and false negatives is very reasonable.
Using the real HIV data, we found 134 HLA-restrictions at FDR ≅ 0.2 among the possible 140 × 70 possible HLA restrictions. To validate our predictions on the real HIV dataset, we performed in vitro assays that specifically measured particular HLA restrictions [26]. Ideally, all 134 pairs should have been evaluated, but this was too expensive and work-intensive. Consequently, six pairs for which the HLA-peptide association is biologically interesting (i.e., unlikely based on current understanding of peptide–HLA binding) were evaluated. All six relationships were confirmed [26]. Prior to this study (partially reported in [26]), it was thought that HLA class I epitopes were restricted mainly by a single HLA allele, or if by more than one allele, then only a few that were structurally highly related and commonly fell into the same HLA supertype [27] (supertypes group together HLA alleles with similar amino acid binding motifs). However, our analysis suggests that a single epitope is frequently restricted by numerous HLA alleles. Additionally, when viewed through the traditional lens of supertypes, we found restrictions across supertypes. For example, IYQEPFKNLK was previously known to be restricted by A11, and we found that it is also restricted by A24 (confirmed experimentally), where A11 and A24 belong to two different supertypes. Table 1 shows a summary of the number of previously known HIV epitopes restricted by one HLA allele, and up to four HLA alleles (none were known to be restricted by more than four alleles) [14]. After adding our newly statistically identified HLA-restricted epitopes, these numbers change dramatically, as shown in Table 2. These tables suggest that HLA class I epitopes are far more “promiscuous” than originally thought, a notion that has significant implications for the understanding of HLA class I antigen presentation and vaccine development. We refer the reader to [26] for a more detailed account of the biological findings. (Note that there are a few differences between the results reported in [26] and the current presentation of results. In [26], the previously known HLA-restrictions were “fixed” to be present in the model before model selection was used to search for new HLA restrictions. We thought it would be of interest to see the results when this a priori information was not used. Additionally, the number of HIV “optimal” epitopes tested was reported as 162 in [26], whereas we report 140—this is due to the fact that epitope–HLA pairs were counted in the former, while here we count only unique epitopes—of which some were repeated across HLA restrictions. The raw data are, however, identical.)
Table S1A lists all epitopes identified by our statistical analysis, sorted by rank from most to least important, along with their learned qij values, and noting which epitopes were previously known, which were confirmed, and what other HLA alleles were previously known to restrict each epitope. Of the 134 identified epitopes we identified, 46 were previously known (eight of our top ten ranked epitopes were known).
We have introduced, implemented, and examined use of a statistical approach for identifying epitope-restricting HLA alleles from ELISpot data. This approach provides a high-throughput, efficient, and cost-effective method for the screening of novel HLA-restricted epitopes. Additionally, our methodology introduces a new approach to the model selection problem, wherein a parameterized family of model selection scores can be explored, by estimating the FDR resulting from the use of each score, and choosing one which suits the needs of the user. In other words, we are able to customize the tradeoff between high discovery rates, and false leads, rather than relying on a single model selection criterion.
Several improvements to the model are possible. (1) Some donors tend to have a higher overall reaction level, thus it may be fruitful to include a latent variable which models this donor-specific bias. (2) A confounding factor in our analysis is the existence of false negatives due to a failed chemical reaction in the ELISpot assay. One could add an observation component to model this type of experimental noise. (3) We stated that the ELISpot data are real-valued, but thought to be informative at a mostly binary level. However, it might be possible to extract more information by using the actual real-valued measurements.
Lastly, by applying our methodology to real HIV data, we have helped to shed light on the extent to which HLA class I epitopes are promiscuous. This has significant implications for the understanding of HLA class I antigen presentation and vaccine development.
|
10.1371/journal.pcbi.1003507 | Investigation of Inflammation and Tissue Patterning in the Gut Using a Spatially Explicit General-Purpose Model of Enteric Tissue (SEGMEnT) | The mucosa of the intestinal tract represents a finely tuned system where tissue structure strongly influences, and is turn influenced by, its function as both an absorptive surface and a defensive barrier. Mucosal architecture and histology plays a key role in the diagnosis, characterization and pathophysiology of a host of gastrointestinal diseases. Inflammation is a significant factor in the pathogenesis in many gastrointestinal diseases, and is perhaps the most clinically significant control factor governing the maintenance of the mucosal architecture by morphogenic pathways. We propose that appropriate characterization of the role of inflammation as a controller of enteric mucosal tissue patterning requires understanding the underlying cellular and molecular dynamics that determine the epithelial crypt-villus architecture across a range of conditions from health to disease. Towards this end we have developed the Spatially Explicit General-purpose Model of Enteric Tissue (SEGMEnT) to dynamically represent existing knowledge of the behavior of enteric epithelial tissue as influenced by inflammation with the ability to generate a variety of pathophysiological processes within a common platform and from a common knowledge base. In addition to reproducing healthy ileal mucosal dynamics as well as a series of morphogen knock-out/inhibition experiments, SEGMEnT provides insight into a range of clinically relevant cellular-molecular mechanisms, such as a putative role for Phosphotase and tensin homolog/phosphoinositide 3-kinase (PTEN/PI3K) as a key point of crosstalk between inflammation and morphogenesis, the protective role of enterocyte sloughing in enteric ischemia-reperfusion and chronic low level inflammation as a driver for colonic metaplasia. These results suggest that SEGMEnT can serve as an integrating platform for the study of inflammation in gastrointestinal disease.
| Mucosal histology plays a key role in the diagnosis, characterization and propagation of a host of gastrointestinal diseases, and the development of computational models capable of producing spatial architecture comparable to histology will enhance the evaluation of hypotheses for those diseases. Inflammation is a significant factor in the pathogenesis of a series of gastrointestinal diseases, and affects the maintenance of the mucosal architecture by morphogenic pathways. We have developed the Spatially Explicit General-purpose Model of Enteric Tissue (SEGMEnT) to investigate the behavior of enteric epithelial tissue as influenced by inflammation. SEGMEnT integrates cellular and molecular pathways governing morphogenesis and inflammation to generate a variety of pathophysiological processes from a common platform and knowledge base. Beyond reproducing healthy and disease ileal mucosal dynamics, SEGMEnT provides insight into a range of clinically relevant cellular-molecular mechanisms, including a novel putative role for Phosphotase and tensin homolog/phosphoinositide 3-kinase (PTEN/PI3K) as a key point of crosstalk between enteric inflammation and morphogenesis, the protective role of enterocyte sloughing in enteric ischemia-reperfusion and the mechanism of chronic low level inflammation as a driver for colonic metaplasia. These results suggest that SEGMEnT can serve as an integrating platform for the study of inflammation in gastrointestinal disease.
| The gut epithelium faces unique challenges in striking a balance between its receptive role in the absorption of nutrients and fostering synergistic interactions with commensal microbes versus retaining sufficient defensive barrier function to prevent microbial invasion and heal tissue injury effectively within this complex environment. The clinical relevance of the structure-function relationship of the gut mucosa is readily evident in the fact that histological characterization of intestinal mucosal architecture is a mainstay in the diagnosis of intestinal disease [1]. We propose that a broad spectrum of intestinal disease can be unified by a view that the mucosal tissue architecture, as maintained by morphogenesis pathways, is subject to a series of control modules that effectively balance the complex interplay of multiple functional objectives when in a state of health, but can become disturbed to generate pathological conditions (Figure 1). Of these control modules, inflammatory pathways are among the most clinically significant, playing an important pathophysiological role in a host of intestinal diseases ranging from environmental enteropathy [2], necrotizing enterocolitis [3], inflammatory bowel disease [4], gut-derived sepsis [5] and cancer [6]. While the specific manifestation of gut inflammation may be different in each of these situations, the same general set of processes are involved across this spectrum of diseases, and are correspondingly associated with specific histological changes of the gut mucosa. Given the complexity of the processes and interactions present, dynamic computational modeling can be a useful tool for instantiating conceptual models (hypotheses) of the structure-function relationship in the gut and its role in the pathogenesis of gastrointestinal disease. Towards this end, we have developed an agent-based computational model to simulate the cellular and molecular interactions that maintain and modify the enteric mucosal architecture, the Spatially Explicit General-purpose Model of Enteric Tissue (SEGMEnT). SEGMEnT models the spatial dynamics of the crypt-villus tissue architecture as generated by the behavior of gut epithelial cells as they undergo replication, migration and differentiation, with the novel incorporation of the effect of inflammation on those morphogenic processes (Figure 1).
SEGMEnT provides a platform for instantiating conceptual models (hypotheses) representing baseline, healthy systems, which can then be used to explore, through differential perturbations, different disease trajectories [7]–[9]. Previous work in spatially-explicit modeling of the gut epithelium is broadly split into two categories: 1) models depicting de novo morphogenesis to examine the generative processes governing crypt-villus growth and patterning [10]–[13], or 2) models with a fixed crypt architecture that considers corresponding issues such as morphogen spatial distribution [14], [15], cell type distribution [16]–[18], and homeostasis [17], [18]. We adopt a distinct but complementary approach that combines aspects of both of these approaches, where we assume the existing tissue architecture has already been produced and the morphogenic pathways maintain the subsequent homeostatic state. However, we allow for a mutable morphologic topology, specifically via alterations in the crypt-villus configuration and tissue architectural features (e.g., the location of the crypt-villus junction, relative and absolute sizes of the crypt and villus). This capability provides metrics for comparing SEGMEnT's output with various histological phenotypes representing different states of health and disease in the gut. Furthermore, we incorporated inflammation as an additional controller of morphogenesis, thereby providing the means to represent many different pathophysiological processes and outcomes.
SEGMEnT is a cell-level ABM: an object-oriented, discrete event, rule-based computational modeling method consisting of populations of computational entities (agents) that follow programmed rules governing their behavior with respect to the environment and interactions with other agents [19]–[24]. SEGMEnT represents the enteric mucosal surface with a 3-dimensional topology that has fixed central axes for the crypts and villi, with square grids over-laid onto an array of rectangular prisms to represent the epithelial surfaces of the crypts and villi (Figure 2). The primary cell types represented in SEGMEnT are gut epithelial cells (GECs), including their subtypes of stem cells, differentiating and mature enterocytes, and two main lineages of inflammatory cells, neutrophils and macrophages/monocytes. SEGMEnT integrates multiple functional modules including an intra-cellular morphogenic signaling pathway, an intra-cellular inflammatory signaling pathway, cell state transitions for proliferation, differentiation and movement, and spatial diffusion of morphogens to define gut epithelial cell behavior for the homeostatic maintenance of the spatial architecture of the enteric mucosa (Figure 3). Figure 3 depicts cellular processes (Letter 3A) represented, those processes linked to GEC type (Letter 3B), their spatial location in terms of the crypt-villus architecture (Letter 3C), and the spatial distribution of expected gradients of different morphogens (Wingless-related integration site (Wnt), Bone Morphogenetic Protein (BMP), Sonic Hedgehog Homolog (Hh), β-catenin, and Protein Kinase B (Akt) are shown in Letter 3D). For a comprehensive depiction and list of the molecules in the signaling networks included in SEGMEnT see Table S1 and Figure S1. More detailed information regarding network connection strengths and signaling molecule functions can be found in Tables S2 and S3. For an overview of the signaling pathways involved see Figure S1; implementation details of these pathways are presented in the Materials and Methods and Supplementary Materials Text S1. All epithelial and inflammatory cells share the same inflammatory signaling pathway structure, though with differential behaviors defined through altered signaling rates. Abstracted blood vessels are also incorporated as points of arrival for circulating inflammatory cells. SEGMEnT implements an expandable architecture that will allow the subsequent addition of functional modules, and the ability to represent additional enteric features (Figure 1B).
SEGMEnT aims to replicate the dynamically stable morphology and cellular populations of the healthy ileum, while matching the spatial distribution of molecular signaling gradients reported in the literature. SEGMEnT also reproduces the effects of inhibition of various signaling pathways by matching resultant phenotypic alterations of the crypt-villus architecture as reported in the literature. SEGMEnT is used to investigate two general types of potentially pathogenic conditions: 1) those resulting from acute insults/perturbations, namely local tissue injury resulting in a wound and the response to enteric ischemia-reperfusion, and 2) chronic conditions that result in a persistent alteration of the homeostatic set-point of the crypt-villus architecture. For this latter case we use the example of colonic metaplasia of the ileal pouch following restorative surgery for ulcerative colitis, thought to be induced by chronic, low-level inflammation due to fecal stasis [25]–[27]. In the course of model development it became apparent that the recognized links between inflammation and tissue patterning in the gut were not sufficient to generate the metaplasia phenotype. This insufficiency in the current state of knowledge led us to hypothesize a link between inflammation and the induction of apoptosis through the Phosphotase and tensin homolog/phosphoinositide 3-kinase (PTEN/PI3K) pathway, a relationship present in other tissues [28] but not previously suggested in the gut. We demonstrate that SEGMEnT, with this putative mechanism, is then able to generate colonic metaplasia of the ileum as seen after surgery for ulcerative colitis. SEGMEnT is the dynamic knowledge representation of a minimally sufficient set of components, mechanisms and interactions for the maintenance of enteric mucosal architecture (morphogenesis) and the effect of inflammation. The eventual goal for SEGMEnT is to serve as a community resource by acting as a “virtual sandbox” that allows researchers to instantiate and investigate their own knowledge, and try out novel and innovative hypothesis for a range of intestinal diseases.
Simulations with SEGMEnT fall into three categories: 1) those representing the baseline healthy state of the gut mucosa, 2) those reproducing previously published single-molecule/gene knockout/inhibition experiments, and 3) those reproducing published experiments investigating specific pathophysiological conditions. For details of the underlying biology in SEGMEnT and the development process see the Materials and Methods and the Supplementary Materials Text S1.
Integral to the use of computational models as a means of integrating mechanistic knowledge is the ability of those models to replicate baseline, healthy behavior [29], [30]. Therefore, the first set of simulations calibrated SEGMEnT to generate and sustain dynamically stable configurations corresponding to baseline enteric tissue behavior. While SEGMEnT does not generate the spatial architecture de novo, it is able to produce the appropriate cell distributions and spatial morphogen gradients leading to homeostasis. The homeostatic crypt-villus configuration in the ileum consists of a crypt depth approximately 1/4 the villus height (approximately 1 mm) [31]. Spatial gradients exist for morphogens such as Wnt, which is highest at the base of the crypt and zero at and above the crypt-villus junction [32]–[37]. The BMP molecule exists everywhere in an isotropic distribution [38], while the gradient of BMP activity is initiated at the crypt-villus junction progressively maximizes towards the tip of the villus [38]–[40]. Other morphogens included in SEGMEnT, Ephrin-B ligand/EphB receptor, Hh, and β-catenin, also exhibit spatial gradients [32]. Furthermore, the cellular components of the epithelium undergo a complete renewal every five days.
Figure 4 shows the output from a simulation for healthy baseline ileal tissue at homeostatic equilibrium. Stable cell populations in the crypt and villus are seen over time (Figure 4A) with complete renewal of the epithelial tissue every 5 days of simulated time. A SEGMEnT screenshot at homeostasis is seen in Figure 4B. At the crypt-villus junction there is a region of undifferentiated gut epithelial cells (GECs) with no Wnt. Since activated Wnt signaling is identified and localized by the accumulation of β-catenin, the absence of Wnt in this zone denotes a region after the activation of β-catenin destruction complex and before all the β-catenin has been destroyed, at which point differentiation is triggered and occurs. This pattern of β-catenin activity is consistent with patterns seen in published histological data seen in Figures 3a and 3c in Ref [33].
Figure S2 shows SEGMEnT's output in terms of both extracellular BMP and binding of BMP with its epithelial cell receptor. These simulation results can be compared with histology images from Reference [38] that demonstrate free extracellular BMP is evenly distributed throughout the tissue (Figure 1d from Ref [38]), whereas the binding of BMP with its cellular receptor, and hence its activity, is maximal in the enterocytes at the villus tip with decreasing binding down towards the crypt-villus junction (Figure 1f from Ref [38]). Output from SEGMEnT visualizing the distribution of free and bound BMP compare well with these images, where the extracellular BMP molecule concentration is roughly uniform distribution (Figure S2, Panel A, shown in blue, compare Figure 1d from Ref [38]). Simulated BMP activity/binding gradient matching experimental data can be seen in Figure S2B, with brown gradients representing bound BMP (compared to brown stained enterocytes seen in Figure 1f from Ref [38]), with corresponding areas of low BMP activity in the crypts (green brackets), initiation of BMP activity at the crypt-villus junction (yellow brackets) and high BMP activity at the tip of the villi (red brackets).
Selective pathway component knockout and/or inhibition studies are a mainstay of molecular biology[41]. The results of these types of experiments are often used to infer the mechanistic role of the targeted pathway component. From the standpoint of modeling and simulation, this approach can be viewed as model component sensitivity testing, and form a ready set of real-world experimental reference sets against which to evaluate the plausibility of a dynamic computational model. In order to provide baseline validation of SEGMEnT's implementation of enteric morphogenesis pathways, a series of simulation experiments were performed for three knockout/inhibition conditions, Wnt inhibition, Hh inhibition, and PTEN inhibition, and compared to previously published experimental data concerning the effects of these interventions. Note that we use the broad definition of “validity” as given in the recognized dictionary of modeling terms [42], namely in terms of establishing the believability or trust in a particular model by reproducing sets of identified behaviors in the real world referent.
The Wnt pathway is a fundamental pathway to gut morphology, and the consequences of its inhibition are well documented [32], [34]–[36]. When healthy ileal tissue is exposed to a strong Wnt inhibitor, the system exhibits a loss of proliferating cells, followed by a loss of crypts, followed by total organ failure and death [32], [34]–[36]. A Wnt inhibitor is applied to the system after it evolves to a steady state, and SEGMEnT produces results that match experimental observations (Figure 5). Figure 5A shows the average population of differentiated and undifferentiated cells (which can be considered a proxy for crypt depth and villus height respectively) vs. time, with the initial loss of crypt cells followed subsequently by loss of villus cells and complete population collapse. Figure 5B–D displays three frames from a SEGMEnT simulation showing the temporal change in the spatial organization of the crypt-villus architecture from baseline homeostasis (Figure 5B) to shortly after administration of the Wnt inhibitor, which leads to a decrease in the crypt population (Figure 5C). The loss of proliferating cells percolates to the villus populations (Figure 5D), preceding the death of the entire villus. These behaviors correspond to the outcomes reported in the literature concerning the effect of Wnt inhibition in producing first crypt-villus atrophy and then death [32], [34]–[36].
Computational models provide detail into the progression of events; here, one can observe that gut epithelial cells quickly differentiate as the β-catenin destruction complex is activated. The increased Hh signal from the newly differentiated cells leads to the production of the Wnt inhibitor, secreted frizzled-related protein 1 (SFRP1), which quickly eliminates all Wnt activity. At this point, the crypt no longer exists, there are few proliferating cells, and the villi have lost significant cellularity; within two days, the last of the differentiated villus cells will have undergone apoptosis and the tissue is dead. Unfortunately, at this time, quantitative experimental data characterizing the dynamic properties of the respective cell populations and their respective molecular signatures is not available, but qualitatively, the output from SEGMEnT matches that seen in experiment in terms of crypt-villus atrophy and subsequent death [32], [34]–[36].
Hh inhibition, while less studied, represents the diametrically opposed case to the Wnt pathway [43], [44]. Experimentally, the inhibition of Hh signaling in the ileum results in increased Wnt activity, crypt hyperplasia, and crypt fission [45]. When an Hh inhibitor is applied to the model at homeostasis, the system's SFRP1 is quickly depleted, allowing the Wnt source to operate unopposed. As Wnt activity increases, the villus shrinks slightly and the crypt displays significant growth (Figure 5E). Once the crypt has reached a stable equilibrium with regards to the new Wnt gradient (unaffected by Hh), the villus grows back to a normal size, leaving us with a normal villus and a deeper than normal crypt (Figure 5H). When simulating Hh inhibition, SEGMEnT generates increased Wnt activity and crypt hyperplasia as reported in the literature [43], [44]. Crypt fission is currently beyond the scope of this work.
While experimental inhibition of PTEN does not change epithelial architecture (in the long term) from the normal state in the absence of inflammation [46], some observations suggest the putative down-regulation of Hh by PTEN [39]. Additional studies suggest that Hh is down regulated by inflammation [37]; therefore we posit that it is the up-regulated PTEN that is responsible for the down-regulated Hh, with subsequent effect on GEC apoptosis.
Simulation of PTEN inhibition demonstrates a “null” consequence of SEGMEnT's PTEN representation on baseline morphogenesis; we hypothesize that its role would only become significant in the perturbation case involving inflammation. Figure 6 displays output from SEGMEnT when PTEN has been strongly inhibited. SEGMEnT simulations of unperturbed tissue show Hh and PTEN sharing the responsibilities of apoptotic regulation and villus maintenance. When Hh activity is inhibited, PTEN takes over; when PTEN in inhibited, the Hh pathway assumes these responsibilities. The simulation experiments demonstrate this transient effect on the crypt GEC population and which subsequently recovers with compensation via Hh. The resulting crypt villus architecture at the end of 24 hrs appears relatively normal, consistent with experimental findings [46]. It is notable that the compensatory dynamics seen in the simulation experiments occur in a period not captured by the sampling interval in Ref [46]. While the end result of a SEGMEnT simulation of PTEN inhibition closely matches experiment, further experiments are necessary to elucidate the progression of events from PTEN inhibition to the establishment of a new equilibrium, which happens to be similar to the PTEN active equilibrium.
All organisms have evolved in a heterogeneous and potentially dangerous environment and thus have to be able to deal with injury in order to maintain their bodily integrity. The inflammatory response is a highly conserved system that represents an organism's initial active response to these threats, providing initial containment and control and prompting the subsequent healing process [47], [48]. This function physiologically manifests in response to local tissue damage level as successful healing, and the expectation is that a model of normal tissue should be able to accomplish successful wound healing up to a certain level of injury. There is extensive literature concerning the computational modeling of wound healing, primarily in terms of epithelial cell dynamics and migration [49]–[51], though application to intestinal injury and healing is limited to necrotizing enterocolitis. However, even though they are not specific to the gut, the models of epithelial injury and recovery do share many of the same pathways concerning inflammation with SEGMEnT.
SEGMEnT simulations of localized epithelial injury demonstrate the ability of the simulated tissue to heal itself after the infliction of a localized wound to the mucosa. A tissue wound/injury is represented by activating sufficient toll-like receptor (TLR4) signaling to cause all cells on the villi to die simultaneously by necrosis (Figure 7A Arrow 1). In the time period directly subsequent to villus death the crypt grows rapidly; this is due to the sudden loss of Hh signaling as most of the differentiated cells on the villus have died. The death of the villi cells reduces the Wnt inhibition of the surviving crypt GECs, resulting in a growth spike in the crypt population (Figure 7A, Arrow 2). As these undifferentiated GEC migrate back up the crypt they lead to the reconstitution of the villus (Figure 7A, Arrow 3). The inflammatory response to the tissue damage clears the necrotic cells (see Figure 7B–D below) and allows restoration of the regulatory functions of the morphogen pathways, with subsequent regrowth of the villus back to the homeostatic state (7A, Arrow 4). Figure 7, Panels B–D display three screenshots of output from the SEGMEnT during this process: Panel B depicts the period immediately following the injury, with necrotic cells seen in place before they are cleared. In Figure 7 Panel C inflammatory cells begin to infiltrate the tissue, both from the activation of local inactive monocytes and the arrival of active neutrophils and macrophages from local blood vessels. These inflammatory cells follow gradients of interferon-γ (IFN-γ) and TLR4 ligand, respectively. The inflammatory cells clear out the necrotic debris via phagocytosis, halting further damage signaling from the necrotic debris. Once sufficient debris has been cleared, the villus begins to repair itself at a regular rate. Figure 7 Panel D depicts 24 hours post-injury, by which time necrotic debris is completely cleared, allowing for the epithelium to quickly return to homeostasis.
Intestinal ischemia/reperfusion (I/R) is a significant clinical pathophysiological entity. When an individual is under extreme stress, e.g. severe injury, blood loss or shock, the body's response is to divert blood flow away from the enteric circulation towards the more immediately critical organs such as the heart and brain, sacrificing intestinal ischemia for immediate survival. Additionally, certain types of major thoracic or abdominal surgery require a temporary occlusion of intestinal blood flow, leading to a period of enteric ischemia. The reperfusion phase involves a rapid influx of blood and circulating immune/inflammatory cells into an endothelial surface that has been primed by the ischemic period, which results in an acute activation of the inflammatory response [52]. This leads to a dramatic change in the enteric mucosal tissue, with induction of tissue inflammation, cellular death and turnover [53], and also affecting resident gut microbes [54]. Reperfusion also leads to sloughing/shedding of ischemic cells into the intestinal lumen). It should be noted that I/R-induced enterocyte sloughing represents a relatively unique means of eliminating ischemic and impending necrotic cells: as opposed to clearance of necrotic cells via enzymatic degradation and phagocytosis as in most tissues, I/R induced enterocyte sloughing takes advantage of the spatial fact that there is a place (i.e. the intestinal lumen) where these potentially necrotic cells can be eliminated. It has been suggested by other researchers that this mechanism for removing ischemic enterocytes, thereby reducing the forward feedback loop propagating inflammation, represents a protective mechanism against I/R of the small intestine [55], [56]. While the specific mechanisms for I/R-induced sloughing are still under investigation, one proposed hypothesis involves the accumulation of Intestinal Fatty Acid Binding Protein (I-FABP) in ischemic cells and its role in promoting sloughing at the onset of reperfusion [55]. We implemented this mechanism into SEGMEnT and then tested the conceptual basis of the hypothesis that sloughing was a protective mechanism by simulating the contra-factual condition by artificially reducing the sloughing rate of the GECs and measuring its effect on the degree of crypt/villus injury (Section 2.4.2).
Metaplasia is the reversible transformation of one type of tissue architecture into one resembling another type of tissue. It is distinct from neoplasia or dysplasia insomuch that the cells themselves do not exhibit dysfunctional growth but rather alter their differentiation path towards a different terminal cellular phenotype. Metaplasia is reflective of an alteration of the tissue environment that subsequently favors the new cellular phenotype, and is therefore often seen in chronic disease states. One example of this is seen in the small intestine following definitive surgery for ulcerative colitis resulting in the creation of an “ileal pouch:” after removal of the colon and rectum, the rectal vault is reconstructed by creating a looped pouch of the terminal ileum. This ileal pouch then serves as a neo-reservoir for the stool in an attempt to more closely approximate normal bowel habits for these patients. However, this process now changes the tissue environment for the ileal mucosa in the pouch. Even though the composition of the intestinal contents entering the pouch may not be considerably different than before, now the previously transitive luminal contents are subject to stasis, which results in an alteration of the environment for the ileal mucosa and can result in colonic metaplasia of the pouch epithelial tissue [25]–[27]. While not a true conversion to colonic tissue, the metaplastic epithelial architecture exhibits defined changes that more closely resemble colonic tissue: a change in the crypt-villus relationship where the crypts deepen and the villi become shortened and an increase in the relative population of goblet cells (mucous producing cells) [25]–[27].
Chronic, low-level inflammation has been associated with colonic metaplasia, and has been implicated as a mechanism driving the alterations seen in the mucosal architecture [27]. However, previously identified connections between inflammatory signaling and the morphogenesis pathway [57]–[62] did not produce the appropriate tissue architecture alterations, i.e. increasing crypt depth and shortened villi, consistent with the metaplasia phenotype. To address this issue we identified the key role of apoptosis (programmed cell death) in the generation of the metaplasia architecture: apoptosis plays a crucial role in morphogenesis by regulating life span of the GECs, subsequently affecting the height of the villus. However, existing knowledge links inflammation primarily to either anti-apoptotic (i.e., NFκB) or necrotic behavior (i.e. Receptor Interacting Protein Kinase, or RIP), neither of which would generate or is associated with a colonic metaplasia phenotype in the ileal pouch. Therefore, there existed a gap between the recognized role of inflammation and the actual processes needed to generate the target phenotype. A search of the literature identified that one proposed link between inflammation and the induction of apoptosis is through the Phosphotase and tensin homolog/phosphoinositide 3-kinase (PTEN/PI3K) pathway [28]. Based on this report, we hypothesized a putative link between gut epithelial inflammation and its effect on enteric mucosal tissue patterning: GEC apoptosis induced via the PTEN/PI3K pathway (Figure S1). The incorporation of this hypothetical mechanism not only increases the rate of GEC apoptosis, thus shortening the villus, but also inhibits Hh production. Inhibition of Hh production leads to reduced inhibition on the Wnt pathway and increases the size of the proliferative compartment in the crypt, thereby generating the essential crypt-villus architectural features characteristic of colonic metaplasia.
Simulations were performed with SEGMEnT to demonstrate the plausibility of a hypothesis previously published in the literature that prolonged low-level inflammation, acting as a persistent perturbation to the signaling network, would lead to a change in the morphology of the epithelial layer [27]. The simulation experiments involved implementing a continuous low-grade stimulation of TLR4s on SEGMEnT's GECs to represent a chronic low-level inflammatory milieu, mimicking the effects of luminal stasis and bacterial overgrowth in an ileal pouch. The effect of this condition on the crypt/villus architecture was evaluated in terms of alterations of the crypt/villus ratio as well as absolute changes in both crypt and villus dimensions. Figure 9A displays crypt and villus GEC populations when the system is exposed to chronic low-level TLR4 signaling (an abstraction of fecal stasis). This up-regulation leads to an increased rate of apoptosis, shortening the villus, as well as an inhibition of the Hh pathway, which leads to an increase in the size of the proliferative compartment. Figure 9B displays output from SEGMEnT when simulating conditions leading to colonic metaplasia. Crypt hyperplasia and villus atrophy are clearly evident (compare with normal homeostatic condition in Figure 9C, and as seen in Figure 5C), along with a villus to crypt height ratio that matches the alterations seen in colonic metaplasia [27], suggesting the plausibility of this mechanism as the driver for colonic metaplasia.
SEGMEnT dynamically represents and integrates existing knowledge concerning homeostasis and inflammation in the ileum and provides a computational platform to augment the exploration of the cellular/molecular processes involved in intestinal wound repair, ischemia/reperfusion injury, and colonic metaplasia/pouchitis. SEGMEnT successfully replicates the dynamically stable morphology and cellular populations of the healthy ileum, while qualitatively matching the spatial distribution of molecular signaling gradients consisting with the existing qualitative histological criteria utilized and reported in the literature [32]–[38], [40]. SEGMEnT also reproduces the effects of inhibition of various signaling pathways by successfully representing the resulting phenotypes in terms of alterations of the crypt-villus architecture [32]–[38], [40]. With the integration of inflammation as a control input to the morphogenic signaling pathway, SEGMEnT demonstrates the ability to withstand certain acute perturbations (local tissue injury, ischemia/reperfusion) as would be expected from normal intestinal tissue, as well as reproduce a specific chronic pathological state associated with inflammatory stimulation (colonic metaplasia). Furthermore, simulations reinforce the protective role of enterocyte sloughing in enteric ischemia-reperfusion and demonstrate the plausibility of PTEN as a cross-talk nexus between inflammation and epithelial patterning in the continuum effect of inflammation on the genesis of colonic metaplasia. Our hypothesis of the role of PTEN in the generation of metaplasia is an example of the type of exploratory modeling that could be performed using computational tools such as SEGMEnT. Further exploration of this concept could follow several investigatory paths. From a cell culture/cell biology standpoint, enteroid preparations could be evaluated under various types of inflammatory cytokines/simulation for levels of PTEN expression, corresponding downstream mediator expression and rates of apoptosis. From an in vivo perspective, a potential animal model of metaplasia and pouchitis should demonstrate an up-regulation of PTEN corresponding to greater histological changes, and the application of a local PTEN inhibitor should reverse or block the generation of metaplasia. Finally, clinical investigations could take a similar cohort to Ref [26] (two groups of patients with an ileal pouch – one that obtained the pouch as a result of ulcerative colitis and one that received the pouch as a result of familial adenomatous polyposis) and perform gene expression analysis on tissue collected from the ileal pouch.
SEGMEnT follows in a line of prior spatially explicit computational models investigating the dynamics of cell populations and signaling within the intestinal crypts [14]–[18]. The Meineke model represents an early attempt to characterize crypt cell population distributions in a spatial context, using an abstracted fixed cylindrical topology and an off-lattice, Voronoi tessellation to represent continuous cell movement, but did not include any representation of cell signaling [16]. Murray et al investigated cell transit between colonic crypts arising from various perturbations to a simulated Wnt pathway [14]. The Van Leeuwen model examines how the Wnt pathway and differences in its signaling progression might affect epithelial cell proliferation [15]. Buske et al investigated the dynamics of cell-fate determination under different conditions and how maintenance of cellular organization could proceed without the need to invoke stem cell behavior [17]. Wong et al utilized a cellular Potts model to investigate the role of EphB/Ephrin-B on cell migration velocities and trajectories within the crypt [18]. Pin, et al use a 3-dimensional Monte Carlo model with a fixed spiral architecture to simulate the dynamics of intestinal epithelial migration and their work includes simulations of crypt cell population regeneration following injury approximating radiation-induced damage, though the response to injury in this model is limited to the morphogenic processes associated with crypt repopulation and does not include inflammatory factors [63]. There have also been several projects involving computational modeling of intestinal inflammation and injury, focusing on intestinal host-microbe effects [64]–[66], the effects of sepsis on the gut [67], or on a specific disease process, namely necrotizing enterocolitis [68]–[71], or inflammatory bowel disease [72]. While none of these prior models have attempted to examine the specific tissue architectural consequences of inflammation, nearly all of them incorporate the same general set of inflammatory pathways seen in SEGMEnT.
SEGMEnT incorporates and integrates the features seen in the spatial gut models concerning the Wnt/Notch pathways with the inflammatory components and processes seen in the gut inflammation models. As such, it is able to extend the scope of its representation to investigate how inter-pathway specific interactions can affect other pathways in the overall signaling network. This capability allowed the instantiation of a novel hypothesis concerning a link between inflammation and morphogenesis through the PTEN/PI3K pathway. Additionally, prior spatial gut models all utilized a fixed crypt topology (i.e. height/depth) upon which their cellular components moved, and therefore are not able to examine metaplasia, ischemia/reperfusion or injury and repair, all conditions that result in some distortion of the baseline crypt-villus architecture. SEGMEnT expands on these prior approaches by allowing for dynamic alteration of these critical histological dimensions; in this fashion it crosses over into some of the capabilities represented in the other arm of spatial crypt-villus modeling that is more focused on looking at factors that actually generate the tissue topology [10]–[13]. The ability to modify the dimensions of the crypt-villus complex is particularly important in being able to represent the differential tissue configurations seen in the Wnt and Hh knockouts and colonic metaplasia.
SEGMEnT also allows the representation of multiple crypt-villus complexes and their associated external tissue structures, simulating significantly larger sections of intestinal tissue. The benefits of this have already been demonstrated in the simulations of healing local tissue injury, where a damaged villus is represented as being adjacent to the uninjured tissue involved in affecting it repair. The current version of SEGMEnT can represent up to 1.4 mm2 area of tissue on a single processor with 2 GB of memory. Capturing the healing of surgical anastomosis or the “patchiness” of inflammatory bowel exacerbations or environmental enteropathy will require the representation of substantially larger surface area of the gut; as such, current work is being done to implement SEGMEnT HPC to investigate the scaling issues associated with simulating physiologically/anatomically relevant areas and volumes of enteric tissue. SEGMEnT HPC is designed to represent arbitrarily large and customizable epithelial surfaces.
SEGMEnT is a work in progress, representing an instantiation of a minimally sufficient set of components and interactions to achieve a face valid representation of the mechanisms involved in maintaining the enteric mucosal architecture (morphogenesis) and its response to an initial set of perturbations (inflammation). The modular control organization of SEGMEnT (Figure 1) has been designed with expansion in mind; in Figure 1B we have positioned some of the potential avenues for development as additional control modules within this framework. Table S4 lists a set of recognized current limitations that have all been targeted for future development via the iterative refinement development process described by Hunt et al [7], [73], [74], in which additional detail is added only when new reference system features are desired, with subsequent confirmation that prior calibration/validation metrics continue to be met. This approach emphasizes the utilization of parsimonious yet functionally useful levels of abstraction and detail, providing a check to the tendency to incorporate as much detail as possible. Our eventual goal is for SEGMEnT, in its multiple forms, can serve as a community resource by acting as a “virtual sandbox” that can be used by different researchers to instantiate and investigate their own knowledge, try out novel and innovative hypotheses, and, hopefully, communicate their insights and findings to the wider community. With this in mind, work is progressing on integrating the SEGMEnT development process with our Computational Modeling Assistant (CMA) [75], a cyberworkspace that is intended to augment the ability of bioresearchers to construct and modify dynamic computational models and run and analyze simulation experiments. By linking SEGMEnT with the CMA we hope to accelerate the Scientific Cycle for the study of intestinal diseases, and aid the gastrointestinal research community in being able to overcome the Translational Dilemma [76], [77].
SEGMEnT is implemented using the agent-based modeling software, Repast Simphony 2.0 developed at Argonne National Laboratory [78]. Construction and calibration of SEGMEnT involved assembling agent rule sets based on a review of the literature, implementation of those rules and agents within the model topology, and adjusting the components such that it could generate a dynamically stable ileal crypt-villus architecture of plausible dimensions, defined as the ratio of crypt-to-villus as equal to 1/4 [31]. It should be noted that SEGMEnT is not designed to generate the mucosal tissue architecture de novo, i.e. it is not a model of embryogenesis or organogenesis, nor is it intended to recreate the physiomechanical forces involved in determining why the architecture has the shape it does. Rather, SEGMEnT starts with the assumption that these processes have already occurred to produce an existing tissue architecture that is subsequently maintained by the behavior of the gut epithelial cells (GECs). Chemokine/mediator dynamics were tuned to create biologically realistic morphogen gradients and maintain constant cellular migration velocities. Cellular proliferation rates were adjusted so that biologically plausible populations of differentiated and undifferentiated cells are maintained when the system is in homeostasis. The list of parameters present at the end of the calibration phase can be seen in Table S1.
SEGMEnT is a knowledge-based model incorporating the inter-relationships between the various cellular and molecular entities present in the enteric tissue. Parameters for these relationships were extracted from the literature, when possible; fitting emphasized the relational qualities between processes. The multi-scale nature of the SEGMENT is apparent in the hierarchical ordering of events and components. Molecular events (occurring on the order of 10−8 m) are aggregated by cells (spanning 10−6 m), which interact to form tissue (10−3 m2). This is also true with respect to temporal scales, where the time scales on which SEGMEnT operates span several orders of magnitude. Signaling events (i.e., the binding of a ligand to a receptor) occur on a time scale from approximately 10−3 s to 1 s, while it can take ∼103 s for the consequences of that event (the secretion of a protein) to manifest. An individual cell performs physical actions (i.e., migration, division) approximately every 103 s, and ultimately survives for ∼104 s. SEGMEnT then makes the implicit assumption in the time it takes for a cell to perform a physical action, any signaling events that would occur in the region of space which the cell occupies will have already occurred; thus, simple time-delayed rules give similar results to a differential equation based model, but at a fraction of the computational cost. Explicit details regarding the underlying biology and modeling methodology can be found in the Supplementary Materials Text S1.
|
10.1371/journal.pntd.0003252 | Crovirin, a Snake Venom Cysteine-Rich Secretory Protein (CRISP) with Promising Activity against Trypanosomes and Leishmania | The neglected human diseases caused by trypanosomatids are currently treated with toxic therapy with limited efficacy. In search for novel anti-trypanosomatid agents, we showed previously that the Crotalus viridis viridis (Cvv) snake venom was active against infective forms of Trypanosoma cruzi. Here, we describe the purification of crovirin, a cysteine-rich secretory protein (CRISP) from Cvv venom with promising activity against trypanosomes and Leishmania.
Crude venom extract was loaded onto a reverse phase analytical (C8) column using a high performance liquid chromatographer. A linear gradient of water/acetonitrile with 0.1% trifluoroacetic acid was used. The peak containing the isolated protein (confirmed by SDS-PAGE and mass spectrometry) was collected and its protein content was measured. T. cruzi trypomastigotes and amastigotes, L. amazonensis promastigotes and amastigotes and T. brucei rhodesiense procyclic and bloodstream trypomastigotes were challenged with crovirin, whose toxicity was tested against LLC-MK2 cells, peritoneal macrophages and isolated murine extensor digitorum longus muscle. We purified a single protein from Cvv venom corresponding, according to Nano-LC MS/MS sequencing, to a CRISP of 24,893.64 Da, henceforth referred to as crovirin. Human infective trypanosomatid forms, including intracellular amastigotes, were sensitive to crovirin, with low IC50 or LD50 values (1.10–2.38 µg/ml). A considerably higher concentration (20 µg/ml) of crovirin was required to elicit only limited toxicity on mammalian cells.
This is the first report of CRISP anti-protozoal activity, and suggests that other members of this family might have potential as drugs or drug leads for the development of novel agents against trypanosomatid-borne neglected diseases.
| The pathogenic trypanosomatid parasites of the genera Leishmania and Trypanosoma infect over 20 million people worldwide, with an annual incidence of ∼3 million new infections. An additional 400 million people are at risk of infection by exposure to parasite-infected insects which act as disease vectors. Trypanosomatid-borne diseases predominant in poorer nation and are considered neglected, having failed to attract the attention of the pharmaceutical industry. However, novel therapy is sorely needed for Trypanosoma and Leishmania infections, currently treated with ‘dated’ drugs that are often difficult to administer in resource-limiting conditions, have high toxicity and are by no means always successful, partly due to the emergence of drug resistance. The last few decades have witnessed a growing interest in examining the potential of bioactive toxins and poisons as drugs or drug leads, as well as for diagnostic applications. In this context, we isolated and purified crovirin, a protein from the Crotalus viridis viridis (Cvv) snake venom capable to inhibiting and/or lysing infective forms of trypanosomatid parasites, at concentrations that are not toxic to host cells. This feature makes crovirin a promising candidate protein for the development of novel therapy against neglected diseases caused by trypanosomatid pathogens.
| The pathogenic trypanosomatids from the genera Leishmania and Trypanosoma infect over 20 million people worldwide, with an annual incidence of ∼3 million new infections in at least 88 countries. An additional 400 million people are at risk of infection by exposure to insect vectors harboring parasites [1]–[3]. Leishmania and trypanosome infections predominate in poorer nations, and are considered neglected diseases that have “fallen below the radar of modern drug discovery” [4].
Leishmania parasites cause five different disease forms – cutaneous (CL), mucocutaneous (MCL), diffuse cutaneous leishmaniasis (DCL), post-kala-azar dermal leishmaniasis (PKDL) and visceral leishmaniasis (VL, also known as ‘black fever’ or ‘kala-azar’ in India) [5]. VL is the most severe and debilitating form of leishmaniasis, and can be fatal if left untreated. First-line treatment for leishmaniasis is based on pentavalent antimonials such as meglumine antimoniate (Glucantime) and sodium stibogluconate (Pentostan). Amphotericin B and pentamidine are used as second-line drugs in patients resistant to first-line therapy [1], [6]. Recently, miltefosine has been used in India as part of combination therapy regimens to treat VL, and the largest increase in miltefosine activity was seen in combination with amphotericin B [7], [8].
There are two forms of HAT (also known as sleeping sickness), caused by two subspecies of T. brucei parasites (T. b. gambiense or T. b. rhodesiense). Both HAT forms culminate in parasite invasion of the central nervous system, with gradual nervous system damage if untreated. The currently used anti-HAT drugs - melarsoprol, eflornithine, pentamidine, and suramin - are highly toxic and have lost efficacy in several regions. Also, treatment is difficult to administer in resource-limiting conditions, and often unsuccessful [9], [10].
Chagas' disease, caused by T. cruzi, affects the cardiovascular, gastrointestinal, and nervous systems of human hosts and has become, in recent decades, a worldwide public health problem due to travelers and migratory flow [2], [11]. Chagas' disease chemotherapy is based on the use of nifurtimox and benznidazole, two very toxic nitroheterocyclic compounds with modest efficacy (especially against late stage chronic disease), and ‘plagued’ by the emergence of drug resistance [12].
Given the high toxicity and limited efficacy of current treatments for leishmaniasis, Chagas' disease and HAT, the development of novel chemotherapeutics against these neglected diseases is essential. Animal venoms and poisons are natural libraries of bioactive compounds with potential to yield novel drugs or drug leads for pharmacotherapeutics [13]. In particular, snake venoms have proven to be interesting sources of potential novel agents against neglected diseases, including Chagas' disease [14]– and leishmaniasis [18]–[23].
Cysteine-rich secretory proteins (CRISPs) are single chain bioactive polypeptides with molecular masses of ∼20–30 kDa found in snake venom, reptilian venom ducts [24]–[26] and also in the salivary glands, pancreatic tissues, reproductive tracts [27]–[31]. In mammals, CRISPs are also expressed at low levels in non-reproductive tissues and organs, including skeletal muscle, spleen and thymus [32]. CRISPs belong to the CAP (Crisp, antigen 5, and pathogenesis-related) superfamily of proteins [33].
CRISP amino acid sequences have high degree of sequence identity and similarity, and include a highly conserved pattern of 16 cysteine residues which form 8 disulfide bonds [34]. Ten of these cysteine residues form an integral part of a well-conserved cysteine-rich domain at the C-terminus, although CRISP N-terminal sequences are overall more conserved than other regions of these proteins [33]–[35]. Snake venom CRISPs belong to the CRISP-3 subfamily [36], one of four subgroups of CRISPs, according to amino acid sequence homology. Most biological targets of snake venom CRISPs described to date are ion channels [37]–[43], although the functions and the molecular targets of most snake venom CRISPs remain to be determined. Some snake venom CRISPs had their biological activities tested on crickets and cockroaches [35]. Snake venom CRISPs have been shown to block the activity of L-type Ca2+ and/or K+-channels and also of cyclic nucleotide-gated (CNG) ion channels, thereby preventing the contraction of smooth muscle cells [26], [37], [40]–[43]. The CRISPs catrin, piscivorin and ophanin, from the snake Crotalus atrox, caused moderate blockage of L-type calcium channels, partially inhibiting the contraction of smooth fibers from mouse caudal arteries [26]. The Philodryas patagoniensis (green snake) CRISP patagonin was capable of generating myotoxicity when injected into the gastrocnemius muscle, but did not induce edema formation, haemorrhage or inhibition on platelet aggregation [44]. Despite their myotoxicity, there are no reports of CRISP protein lethality to mice, in concentrations of up to 4.5 mg/kg [35], [45], and patagonin did not induce systemic alterations in mice, or histological changes in tissues from the cerebellum, brain, heart, liver and spleen [44].
In a previous publication, we showed that crude venom from the rattlesnake Crotalus viridis viridis had anti-parasitic activity against all forms of T. cruzi, and could be a valuable source of molecules for the development of new drugs against Chagas' disease [46]. In search for the molecular source of the anti-parasitic activity found in Cvv crude venom, we purified a Cvv CRISP that will be henceforth referred to as ‘crovirin’. Here, we describe the purification, biochemical characterization and biological activity of crovirin against pathogenic trypanosomatids parasites and mammalian cells, showing that crovirin is active against infective developmental forms of trypanosomes and Leishmania, at doses that elicit no or minimal toxic effects on human cells.
Crude venom from the rattlesnake Crotalus viridis viridis (Cvv) and adjuvants such as parasites growth media, were purchased from Sigma–Aldrich Chemical Co (St. Louis, MO, USA). Benznidazole (Bz) (Laboratório Farmacêutico do Estado de Pernambuco [LAFEPE], Brazil), diminazene aceturate (Berenil [Ber], Hoechst Veterinãr GmbH, München, Germany), and Amphotericin B (Amp-B) (Sigma) were used as a reference drugs for Chagas disease, sleeping sickness and leishmaniasis treatment, respectively. The material and reagents used in SDS-PAGE were from Bio-Rad Laboratories, Inc. Molecular weigh markers LMW were from Fermentas Life Sciences. Mass spectrometry grade Trypsin Gold was from Promega. All other reagents and chemicals were from Merck (Darmstadt, Germany), Tedia Company and Eurofarma Laboratórios SA.
Lyophilized Cvv venom (10 mg) was dissolved in 1 ml of 20 mM Tris–HCl, 150 mM NaCl, pH 8.8 and centrifuged at 5,000 g for 2 min. The supernatant was applied onto a reverse phase analytical C8 column (5 µm, 250×4.6 mm) (Kromasil, Sweeden), previously equilibrated with the same buffer. Venom proteins were separated by reverse phase HPLC (Shimadzu, Japan). Fractions (0.7 ml/tube) were collected at a 1 ml/h flowrate. A linear gradient of water/acetonitrile containing 0.1% trifluoroacetic acid (TFA) was used. The elution profile was monitored by absorption at 280 nm, and the molecular homogeneity of the relevant fractions was verified by SDS-PAGE. Fractions containing protein peaks were dried in a Speed-Vac (Savant, Thermo Scientific, USA) and resuspended in distilled water prior to protein quantification by the Bradford method. Molecular mass determination was performed by MALDI-TOF and by electrospray ionization (ESI) mass spectrometry using a Voyager-DE Pro and a QTrap 2000 (both from Applied Biosystems), respectively.
Protein bands were excised from Coomassie Brilliant Blue-stained SDS-PAGE gels and cut into smaller pieces, which were destained with 25 mM NH4HCO3 in 50% acetonitrile for 12 h. The pieces obtained from the non-reducing gels were reduced in a solution of 10 mM dithiothreitol and 25 mM NH4HCO3 for 1 h at 56°C, and then alkylated in a solution of 55 mM iodoacetamide and 25 mM NH4HCO3, for 45 min in the dark. The solution was removed, the gel pieces were washed with 25 mM NH4HCO3 in 50% acetonitrile, and then dehydrated in 100% acetonitrile. Finally, all pieces from reducing and non-reducing gels were air-dried, rehydrated in a solution of 25 mM NH4HCO3 containing 100 ng of trypsin, and digested overnight at 37°C. Tryptic peptides were then recovered in 10 µl of 0.1% TFA in 50% acetonitrile.
The peptides extracted from gel pieces were loaded into a Waters Nano Acquity system (Waters, MA, USA) and desalted on-line using a Waters Symmetry C18 180 µm×20 mm, 5 µm trap column. The typical sample injection volume was 7.5 µl, and liquid chromatography (LC) was performed by using a BEH 130 C18 100 µm×100 mm, 1.7 µm column (Waters, MA, USA) and eluting (0.5 µl/min) with a linear gradient of 10–40% acetonitrile, containing 0.1% formic acid. Electrospray tandem mass spectra were performed in a Q-Tof quadrupole/orthogonal acceleration time-of-flight spectrometer (Waters, Milford, MA) linked to a nano ACQUITY system (Waters) capillary chromatograph. The ESI voltage was set at 3300 V, the source temperature was 80°C and the cone voltage was 30 V. The instrument control and data acquisition were conducted by a MassLynx data system (Version 4.1, Waters), and experiments were performed by scanning from a mass-to-charge ratio (m/z) of 400–2000 using a scan time of 1 s, applied during the whole chromatographic process. The mass spectra corresponding to each signal from the total ion current (TIC) chromatogram were averaged, allowing for accurate molecular mass measurements. The exact mass was determined automatically using Q-Tof's LockSpray (Waters, MA, USA). Data-dependent MS/MS acquisitions were performed on precursors with charge states of 2, 3 or 4 over a range of 50–2000 m/z, and under a 2 m/z window. A maximum of three ions were selected for MS/MS from a single MS survey. Collision-induced dissociation (CID) MS/MS spectra were obtained using argon as the collision gas at a pressure of 40 psi, and the collision voltage varied between 18 and 90 V, depending on the mass and charge of the precursor. The scan rate was 1 scan/s. All data were processed using the ProteinLynx Global server (version 2.5, Waters). The processing automatically lock mass calibrated the m/z scale of both the MS and the MS/MS data utilizing a lock spray reference ion. The MS/MS data were also charge-state deconvoluted and deisotoped with the maximum entropy algorithm MaxEnt 3 (Waters, MA, USA).
Proteins corresponding to the tryptic peptides from peak 3 were identified by correlation of tandem mass spectra and the NCBInr database of proteins (Version 050623), using the MASCOT software (Matrix Science, version 2.1). Settings allowed for one missed cleavage per peptide, and an initial mass tolerance of 0.2 Da was used in all searches. Cysteines were assumed to be carbamidomethylated, and a variable modification of methionine (oxidation) was allowed. Identification was considered positive when at least two peptides matched the protein sequence with a mass accuracy of less than 0.2 Da.
T. cruzi tissue culture trypomastigotes (CL-Brener clone) were obtained from the supernatants of 5 to 6-day-old infected LLC-MK2 cells maintained in RPMI-1640 medium (Sigma) supplemented with 2% FCS for 5–6 days at 37°C in a humidified 5% CO2. Theses trypomastigotes were also used to obtain intracellular amastigotes in macrophage cultures.
The MHOM/BR/75/Josefa strain of L. amazonensis, isolated from a patient with DCL by C. A. Cuba-Cuba (Universidade de Brasilia, Brazil), was used in the present study. Amastigote forms were maintained by hamster footpad inoculation, while promastigotes were cultured axenically in Warren's medium [47] supplemented with 10% fetal bovine serum (FBS) at 25°C. Infective promastigotes were used to obtain intracellular amastigotes in macrophage cultures, as described previously [48]. Bloodstream form (BSF) T. brucei rhodesiense (strain IL1852) were cultivated in HMI-9 medium (Invitrogen) supplemented with 10% inactivated FBS (Biosera-South America) and 10% of serum plus supplement (SAFC Bioscience, USA), at 37°C in a humidified 5% CO2 incubator [49]. Procyclic-form (PCF) T. brucei rhodesiense (strain 457) were grown in SDM-79 medium (LGC Biotecnologia) supplemented with 10% heat-inactivated FBS, at 28°C [50].
In this study, we used 5-week-old female CF1 mice as sources of peritoneal macrophages and of muscle sample for ex vivo assays (described below). All animal experimentation protocols received the approval by the Commission to Evaluate the Use of Research Animals (CAUAP, from the Carlos Chagas Filho Biophysics Institute - IBCCF), and by the Ethics Committee for Animal Experimentation (Health Sciences Center, Federal University of Rio de Janeiro – UFRJ) (Protocol no. IBCCF 096/097/106), in agreement with Brazilian federal law (11.794/2008, Decreto n° 6.899/2009). We followed institutional guidelines on animal manipulation, adhering to the “Principles of Laboratory Animal Care” (National Society for Medical Research, USA) and the “Guide for the Care and Use of Laboratory Animals” (National Academy of Sciences, USA).
Crovirin was purified as described above and stored at −20°C, in 3.6 mg/ml stock solutions prepared in PBS (pH 7.2). All experiments were carried out in triplicates. Stock solutions of Bz (14 mg/ml) and Amp-B (10 mg/ml) were prepared in dimethyl sulfoxide (DMSO), and the final concentration of the solvent never exceeded 0.5%, which is not toxic for parasites and mammalian cells. Ber stock solution (0.188 mg/ml) was prepared in pyrogen-free water.
Axenically grown parasite forms were treated with crovirin for up to 72 h in the same culture conditions used for growth (described above). The following crovirin concentrations were used to treat axenic forms: 1.2–4.8 µg/ml (L. amazonensis promastigotes) and 0.6–4.8 µg/ml crovirin (T. brucei rhodesiense BSF and PCF). IC50 values were calculated based on daily counting of formalin-fixed parasites using a hemocytometer. Positive controls were run in parallel with 4.7 µg/ml Amp-B [51] and 39.8 ng/ml Ber [52], respectively.
T. cruzi tissue culture trypomastigotes were treated with crovirin (0.45–4.8 µg/ml) at a density of 1×106 cells/ml, for 24 h at 37°C (in RPMI media containing 10% FCS). LD50 (50% trypomastigote lysis) values were determined based on direct counting of formalin-fixed parasites using a hemocytometer. Bz was used as reference drug, in a 3.39 µg/ml concentration [53].
To evaluate the effects of crovirin on T. cruzi and L. amazonensis intracellular amastigotes, peritoneal macrophages from CF1 mice were harvested by washing with RPMI medium (Sigma), and plated in 24-well tissue culture chamber slides, allowing them to adhere to the slides for 24 h at 37°C in 5% CO2. Adherent macrophages were infected with tissue culture T. cruzi trypomastigotes (at 37°C) or L. amazonensis metacyclic promastigotes (at 35°C) at a macrophage-to-parasite ratio of 1∶10, for 2 h. After this period, non-internalized parasites were removed by washing, cultures were incubated for 24 h in RPMI with 10% FCS, and fresh medium with crovirin (0.45–3.6 µg/ml for T. cruzi, and 0.6–9.6 µg/ml for L. amazonensis) was added daily for 72 h. At different time-points (24, 48 and 72 h) cultures were fixed with 4% paraformaldehyde in PBS (pH 7.2) and stained with Giemsa for 15 min. The percentage of infected cells and the number of parasites per 100 cells were determined by light microscopy examination. Positive controls of T. cruzi and L. amazonensis amastigotes infected cells were run in parallel with cultures treated with 0.73 µg/ml Bz [53] and 0.07 µg/ml Amp-B [54], respectively.
LLC-MK2 cells were maintained in RPMI medium supplemented with 10% FCS. Prior to treatment with crovirin, cells were seeded in 24-well plates containing glass coverslips and incubated in RPMI medium supplemented with 10% FCS for 24 h at 37°C. Cells were then treated with 4.8, 10 and 20 µg/ml crovirin at 37°C for 72 h. LC50 values (concentrations that reduces by 50% the cellular viability) for crovirin were calculated from daily counts of the number of viable cells, using trypan blue as an exclusion dye. At least 500 cells were examined per well, on a Zeiss Axiovert light microscope (Oberkochen, Germany).
In addition, mouse peritoneal macrophages were seeded on 96-well plates, incubated in RPMI medium with 10% FCS for 24 h at 37°C and treated with 4.8, 10 and 20 µg/ml crovirin at 37°C, for 72 h. After this period, cells were washed with PBS (pH 7.2), and the wells were filled with RPMI medium without phenol red containing 10 mM glucose and 20 µl of a solution of 2 mg/ml MTS (3-(4,5-dimethylthiazol-2-yl)-5-(3-carboxymethoxyphenyl)-2-(4-sulfophenyl)-2H-tetrazolium salt) and 0.92 mg/ml PMS (phenazine methosulfate), prepared according to the manufacturer's instructions (Promega, Madison, WI, USA). Following 3 h of incubation at 37°C, formation of a soluble formazan product by viable cells was measured using a plate reader, by absorbance at 490 nm. All cytotoxicity experiments were carried out in triplicates.
The myotoxicity of crovirin was studied ex vivo using a muscle creatin kinase (CK) activity assay [55]. The analysis consisted of monitoring the rate of CK release from isolated mouse extensor digitorum longus (EDL) muscle bathed in a solution containing crovirin (10 µg/ml). Adult male and female Swiss mice (25.0±5.0 g) were anesthetized with ethyl ether and killed by cervical dislocation. EDL muscles were collected, freed from fat and tendons, dried and weighed. Muscle samples were then homogenized in 2 ml saline/0.1% albumin and their CK content was determined using a commercial diagnostic kit (Bioclin, Brazil). Four EDL muscles were mounted vertically on a cylindrical chamber and superfused continuously with Ringer's solution equilibrated with 95% O2/5% CO2. At 30 to 90-min intervals, the perfusing solution was collected and replaced with fresh solution. The collected EDL samples were used for the measurement of CK activity as described above. Muscles were weighed at the end of the experiment (2 h later). Enzyme activity is reported as international units corrected for muscle mass.
Mean value comparisons between control and treated groups were performed using the Kruskal-Wallis test in the BioEstat 2.0 program for Windows. Differences with p≤0.05 were considered statistically significant.
In a previous study, we showed that the Cvv venom had anti-parasitic activity against T. cruzi [46]. Preliminary analysis of Cvv venom fractions by reverse-phase chromatography (not shown) indicated that the activity eluted with fractions containing peak 3 of the chromatographic profile (Fig. 1A). Thus, we analyzed the main chromatographic fraction corresponding to peak 3 by SDS-PAGE and MALDI-TOF mass spectrometry (Fig. 1B–C). SDS-PAGE analysis of peak 3 showed a single polypeptide, with a relative molecular mass of 24 kDa (Fig. 1B) and 28 kDa (data not shown), under reducing and non-reducing conditions, respectively. We will refer to this protein henceforth as crovirin. MALDI-TOF analysis of the intact protein showed a molecular mass of 24,893.64 Da (Fig. 1C). The peaks of 12,424.36 and 12,477.62 Da in the MS profile correspond to doubly-charged (z = 2) cationic forms of the protein. The amino acid sequence of tryptic crovirin peptides (produced by Nano LC-MS/MS mass spectrometry analysis) is nearly identical to a partial sequence of a Cvv CRISP (GenBank gi:190195319) (Fig. 2). The MS/MS-derived sequences are also nearly identical to those of a CRISP protein from Calloselasma rhodostoma (GenBank gi:190195317) and have high degree of sequence similarity to several other snake venom CRISPs, including ablomin (Fig. 2). The MS/MS spectrum of the fragmented peptide ions was matched by MASCOT displayed a coverage of 48% of identical peptides, with a p≥355 indicating extensive homology to the CRISP from C. rhodostoma. The MS results strongly suggested that a CRISP from Cvv snake venom had been purified, and corresponded to crovirin.
First of all, we investigated crovirin citotoxicity over mammalian host cells before proceeding with our analysis of the anti-parasitic activity of this venom protein.
LLC-MK2 cells were treated with crovirin for 72 h and examined for viability using a trypan blue exclusion assay (Fig. 3A). None of the tested crovirin concentrations (4.8, 10 or 20 µg/ml) were capable of inducing significant loss of cell viability, even after 72 h of treatment. In addition, we tested the activity of crovirin against murine peritoneal macrophages to investigate its cytotoxicity towards primary host cells. Treated cells were examined using an MTS assay, and no significant toxicity (p≤0.05) was observed in any treatment conditions (Fig. 3B). Creatine kinase (CK) activity was measured before and two hours after extensor digitorum longus (EDL) muscle exposure to 10 µg/ml crovirin. We did not observed significant CK release from treated EDL muscles compared to control (saline) after 2 hours of incubation with crovirin, indicating that this protein did not generated appreciable myotoxicity at the concentration tested.
After establishing that crovirin had only minimal cytotoxic effects towards mammalian cells at concentrations of up to 20 µg/ml, we tested the anti-parasitic activity of purified crovirin against relevant developmental forms of three different species of pathogenic trypanosomatid parasites, namely L. amazonensis, T. cruzi and T. brucei rhodesiense.
We tested crovirin activity against the two infective T. cruzi forms, trypomastigotes and amastigotes. Trypomastigote forms do not multiply and do not remain viable after several days in culture media at 37°C. Therefore, the effect of crovirin towards T. cruzi trypomastigotes was evaluated as the ability of the protein to lyse cells after 24 h of treatment (Fig. 4A). The calculated LD50 of crovirin for trypomastigotes was 1.10±0.13 µg/ml (Table 1). This concentration displayed the second higher selectivity index (18.2) (Table 1) among all crovirin treatments.
The treatment with 3.39 µg/ml Bz exhibited a 65.8% of parasites lysis at same conditions. T. cruzi amastigotes multiply in the intracellular environment. Crovirin inhibited the growth of amastigotes inside peritoneal macrophages in a dose-dependent manner (Fig. 4B), with an IC50 of 1.84±0.53 µg/ml when cells were treated with crovirin for 72 h (Table 1). Crovirin presented a discret superior trypanocidal activity against the intracellular forms as compared with Bz (Fig. 4B).
Crovirin activity was also tested against infective promastigote and amastigote forms of L. amazonensis, one of the species responsible for CL. None of the crovirin concentrations tested inhibited significantly the proliferation of L. amazonensis promastigotes in axenic media, unlike Amp-B treatment, which resulted in a reduction of a little over 80% in the number of parasites after 72 h of treatment. In contrast, crovirin inhibited the proliferation of intracellular amastigotes of L. amazonensis in a concentration-dependent manner (Fig. 4C–D). The effect of crovirin on amastigote proliferation was evident as early as 24 h after the start of treatment, and the IC50 for crovirin after 72 h of treatment was 1.21±0.89 µg/ml (Table 1). After 48 h incubation, the IC50 of 1.05 µg/ml also resulted in the highest selectivity index (19.1), being less toxic treatment to mammalian host cells. However, no tested concentration of crovirin had superior leishmanicidal activity against amastigotes forms as compared with Amp-B (Fig. 4D).
Both developmental forms of T. brucei rhodesiense tested here (PCF and BSF) were sensitive to crovirin treatment. A different profile of growth inhibition in the presence of crovirin was observed for PCF (Fig. 4E) and BSF (Fig. 4F) parasites, with IC50 values of 1.13±0.31 and 2.06±0.12 µg/ml, respectively, after 72 h of treatment. The 39.8 ng/ml Ber treatment resulted in a remarkable growth inhibition of both BCF and PCF than crovirin treatment (Fig. 4E–F).
There is an urgent need for the development of novel compounds for the treatment of trypanosomatid-borne diseases, currently treated with ‘dated’ chemotherapeutic agents with high toxicity and limited efficacy, partly due to the emergence of drug resistance. Animal venoms and toxins, including snake venoms, can provide compounds directly useful as drugs, or with potential as drug leads for the synthesis of novel therapeutic agents [22]. Previously, our group showed that Cvv crude venom displayed anti-parasitic activity against different T. cruzi developmental forms [46]. We have now extended this research with the purification of crovirin, a CRISP from Cvv venom with promising activity against key infective stages of the life cycle of T. cruzi, T. brucei rhodesiense and L. amazonensis. Furthermore, we show that crovirin has low toxicity towards host cells and mouse muscle, in agreement with the low or absent toxicity reported for most CRISPs proteins [35], [44]–[45].
CRISPs proteins are often given names that refer to the organism from which they were isolated. The first CRISP described in reptiles was isolated from the skin secretion of the lizard Heloderma horridum, and was named helodermin [56]. Examples of proteins isolated from snake venoms are patagonin, isolated from Philodryas patagonensis [44], latisemin, isolated from sea snake Laticauda semifasciata, tigrin isolated from Rhabdophis tigrinus tigrinus [41], and ablomin, isolated from Gloydius blomhoffi [41]. CRISPs sequences have also been identified in transcriptome analysis of venom glands [57]–[58] or are deposited at databanks but were not purified or studied. A partial CRISP sequence from Crotalus viridis viridis (GenBank gi:190195319) likely corresponding to central and C-terminal regions of crovirin was identified by transcriptome analysis of venom gland tissue. However, this is the first report on the purification and study of crovirin.
One of the most important findings of the present study was the activity of crovirin against the intracellular proliferation of trypanosomatids. Amastigotes are key developmental forms during the development and maintenance of infections by Leishmania and T. cruzi, representing the replicative intracellular stages of these protozoan parasites. Substantial inhibition of both T. cruzi and L. amazonensis intracellular amastigote proliferation was observed at crovirin concentrations significantly lower than those required to cause damage to host cells, including mouse EDL muscles. These results are particularly important because the currently available drugs to treat leishmaniasis and Chagas' disease are known to have lower anti-amastigote activity [1], [6].
The effects of crovirin over both the procyclic and the bloodstream form of T. brucei rhodesiense are also encouraging, suggesting that crovirin might be useful in the development of new anti-HAT chemotherapeutics. In conclusion, our results demonstrate that crovirin has promising trypanocidal and leishmanicidal effects, and represents a potential avenue for drug development against leishmaniasis, Chagas' disease and HAT, since its anti-parasitic effects are matched by low toxicity to host cells and muscles. Further studies are now required to extend our knowledge on the potential use of crovirin as an alternative compound to improve the effectiveness of treatment of trypanosomatid-borne neglected diseases.
|
10.1371/journal.pcbi.1003130 | Hierarchical Modeling for Rare Event Detection and Cell Subset Alignment across Flow Cytometry Samples | Flow cytometry is the prototypical assay for multi-parameter single cell analysis, and is essential in vaccine and biomarker research for the enumeration of antigen-specific lymphocytes that are often found in extremely low frequencies (0.1% or less). Standard analysis of flow cytometry data relies on visual identification of cell subsets by experts, a process that is subjective and often difficult to reproduce. An alternative and more objective approach is the use of statistical models to identify cell subsets of interest in an automated fashion. Two specific challenges for automated analysis are to detect extremely low frequency event subsets without biasing the estimate by pre-processing enrichment, and the ability to align cell subsets across multiple data samples for comparative analysis. In this manuscript, we develop hierarchical modeling extensions to the Dirichlet Process Gaussian Mixture Model (DPGMM) approach we have previously described for cell subset identification, and show that the hierarchical DPGMM (HDPGMM) naturally generates an aligned data model that captures both commonalities and variations across multiple samples. HDPGMM also increases the sensitivity to extremely low frequency events by sharing information across multiple samples analyzed simultaneously. We validate the accuracy and reproducibility of HDPGMM estimates of antigen-specific T cells on clinically relevant reference peripheral blood mononuclear cell (PBMC) samples with known frequencies of antigen-specific T cells. These cell samples take advantage of retrovirally TCR-transduced T cells spiked into autologous PBMC samples to give a defined number of antigen-specific T cells detectable by HLA-peptide multimer binding. We provide open source software that can take advantage of both multiple processors and GPU-acceleration to perform the numerically-demanding computations. We show that hierarchical modeling is a useful probabilistic approach that can provide a consistent labeling of cell subsets and increase the sensitivity of rare event detection in the context of quantifying antigen-specific immune responses.
| The use of flow cytometry to count antigen-specific T cells is essential for vaccine development, monitoring of immune-based therapies and immune biomarker discovery. Analysis of such data is challenging because antigen-specific cells are often present in frequencies of less than 1 in 1,000 peripheral blood mononuclear cells (PBMC). Standard analysis of flow cytometry data relies on visual identification of cell subsets by experts, a process that is subjective and often difficult to reproduce. Consequently, there is intense interest in automated approaches for cell subset identification. One popular class of such automated approaches is the use of statistical mixture models. We propose a hierarchical extension of statistical mixture models that has two advantages over standard mixture models. First, it increases the ability to detect extremely rare event clusters that are present in multiple samples. Second, it enables direct comparison of cell subsets by aligning clusters across multiple samples in a natural way arising from the hierarchical formulation. We demonstrate the algorithm on clinically relevant reference PBMC samples with known frequencies of CD8 T cells engineered to express T cell receptors specific for the cancer-testis antigen (NY-ESO-1) and compare its performance with other popular automated analysis approaches.
| Flow cytometry is the prototypical assay for multi-parameter single cell analysis, and is essential in vaccine development, monitoring of T cell-based immune therapies and the search for immune biomarkers. In many clinical research applications, the cell subsets of interest are antigen specific T lymphocytes that are often found in extremely low frequencies (0.1% or less). These antigen-specific T cells can be detected using HLA-peptide multimers or by their expression of effector proteins upon specific antigen stimulation in intracellular staining (ICS) assays. Current methods of flow cytometry analysis rely on visual gating of cell events to identify and quantify cell subsets of interest. However, the choice of sequence for the dot plots (gating strategy) and where to draw the gating boundaries is highly dependent on assay protocols and operator experience and may not be easily harmonized, as illustrated in recent international proficiency panels [1], [2].
There has therefore been increasing interest in the use of objective, automated methods for cell subset identification [3]. One approach that we and others have promoted is the use of statistical models to estimate the data distribution [4]–[6], followed by a mapping of summaries of the statistical distribution to cell subsets of biological interest. This model-based approach tends to be more numerically intensive than other ad hoc approaches to data clustering, but as we have previously demonstrated, this can be overcome by exploiting the cheap massively parallel capabilities of modern graphical processing units (GPUs). Importantly, the model-based approach has the advantage of using a declarative probabilistic framework that can be extended using well-established and understood mechanisms to improve discriminative power. In particular, hierarchical models that incorporate information from both the individual and group levels when fitted to flow cytometry data samples can increase both interpretability and sensitivity. These hierarchical models increase interpretability by aligning clusters in a way that enables direct comparison of cell subsets across data samples, and increase sensitivity for detecting very low frequency cell subsets by sharing information across multiple samples. Hierarchical models thus improve the ability of model-based approaches to detect low frequency event subsets, and enable the comparative analysis that is essential to any downstream analysis of multiple data samples.
We briefly describe three alternative software packages for automated analysis to contrast the approach of HDPGMM. FLOCK 2.0 (FLOw cytometry Clustering without K) [7] is widely used because it is a resource provided by IMMPORT (Immunology Database and Analysis Portal), a repository of data generated by investigators funded through the NIAID/DAIT. Similar to DPGMM and HDPGMM, FLOCK is able to estimate the optimal number of data partitions from the data. However, FLOCK uses an adaptive multi-dimensional mesh to estimate local density followed by hierarchical merging of adjacent regions based on density differentials rather than a mixture model, and does not appear to either provide a statistical model (e.g. for goodness-of-fit calculations) or methods for alignment of cell subsets across different samples. In contrast, flowClust [6] and FLAME (FLow analysis with Automated Multivariate Estimation) [5] both use a statistical mixture model approach for density estimation and clustering. Both packages are likely to be widely used, since flowClust is provided as a library in R/BioConductor, and FLAME is part of GenePattern. Apart from the choice of base distribution (T distribution for flowClust and skewed distributions for FLAME), the main differences with DPGMM are the use of optimization (Expectation-Maximization) rather than simulation (MCMC) to estimate the density, the need for the user to specify the number of partitions and differences in the type of transform applied in data pre-processing. FlowClust does not provide any method to align cell subsets across samples, while FLAME provides a heuristic algorithm to do so as described in their original publication [5]. Unlike HDPGMM, none of the three algorithms use a hierarchical approach to model group and individual specific effects.
With this in mind, the developments reported here concern the implementation of a hierarchical Gaussian mixture model based on a Dirichlet process prior, and extensions of the basic model to identify and quantify rare cell subsets in flow cytometry data. Simulated data is first used to demonstrate the advantages of hierarchical models over conventional clustering approaches. This is followed by validation of the model on experimental samples, using retrovirally TCR-transduced T cells that are spiked into autologous peripheral blood mononuclear cell (PBMC) samples to give a defined number of antigen-specific T cells [8]. Finally, the reproducibility and accuracy of this approach for rare cell quantification is compared to that of standard DPGMM and manual analysis performed by a group of ten flow cytometry users, and compared with the results from FLOCK, FLAME and flowClust.
The basic concept in model-based approaches is to consider events in a flow cytometry data set as being random samples drawn from a multi-dimensional probability distribution. The objective of analysis is then to define the probability distribution model and evaluate inferences over the model parameters based on fit to the specific data set. Statistical mixture models are a standard approach for the construction of the underlying distribution, using the sum of many simpler probability distributions (e.g. multivariate Gaussian, Student-t or skewed distributions) to approximate arbitrary multi-dimensional distributions. For biological interpretation, fitted models are then used for clustering, i.e. using statistical properties of individual events to assign them to biological cell subsets. For example, with statistical mixture models, this can be done by grouping events with the highest probability of coming from a specific mixture component together, or merging of multiple components using specified criteria such as having a common mode in the estimated distribution over markers [9], [10].
Of course, the number of distinguishable cell subsets and Gaussian components necessary to fit the model satisfactorily is not known in advance. To avoid having to specify the number of mixture components needed in the model, we use a Dirichlet process prior in which the number of components necessary is directly estimated from the data [11]. Computationally, the use of Dirichlet process priors is more efficient than fitting multiple models with different numbers of components and testing with some penalized likelihood (e.g. Akaike or Bayesian information criteria) to choose the best model, as only a single model fit is performed. Since we use multivariate Gaussian distributions as components, the overall approach is described as a Dirichlet process Gaussian mixture model (DPGMM). DPGMM are extremely flexible models that can fit flow data from flow cytometry experiments using different antibody-fluorochrome labels (e.g. 4-color HLA-peptide multimer and 11-color intracellular staining (ICS) panels), and a natural evolution of the fixed Gaussian mixture models we originally proposed [4]. Finally, while the model uses Gaussian components, cell subsets are identified with merged components using the consensus modal clustering strategy described in Methods. As a result, cell subsets can have arbitrarily complex distributions and are not restricted to symmetric Gaussian clusters.
Clustering methods applied to data samples independently face two major limitations. The first is that cluster labels are not aligned across data samples, posing a problem for comparing subsets across multiple samples which is usually the purpose of the original experiment. The second is that there are limits to the ability of clustering models to identify very rare event clusters due to masking by abundant event clusters [12]. In particular, this makes it difficult to identify clusters matching antigen-specific HLA-peptide multimer labeled or polyfunctional T cells in ICS assays that may be biologically meaningful at frequencies of 0.1% or lower. We show in this paper that both issues are successfully addressed by the use of hierarchical Dirichlet process Gaussian mixture models (HDPGMM).
Hierarchical, or multi-level models, represent individual events in flow cytometry data as being organized into successively higher units. For example, individual events belong to a sample, and a sample may belong to a collection of similar samples. The critical idea is that cell subset phenotypes that are common across data samples can be used to inform and hence better characterize events in individual samples. For example, one hierarchical Dirichlet process model formulation partitions components into those common across data samples and those unique to a specific sample [13], [14] – this provides a different notion of sharing that is useful for identifying fixed and variable components across heterogeneous data samples but lacks a straightforward alignment of all clusters necessary for multi-sample comparison.
Instead, we model information sharing by placing all data samples under a common prior, such that the mean and covariance in any of the individual sample Gaussian components are shared across all samples, but the weight (proportion) of the component in each sample is unique. As described by Teh et al (2006) [15], this can be achieved by using a set of random measures , one for each data sample, where is distributed according to a sample-specific Dirichlet process . The sample-specific DPs are then linked by a common discrete prior defined by another . This hierarchical model leaves the cluster locations and shapes constant across datasets, and hence aligns the clusters in that the location of the normal components is common to all data samples.
As depicted in the summary schematic of the HDPGMM model shown in Figure 1, there are basically 6 parameters that control the sensitivity. The parameter controls the spread of the (standardized) cluster means and controls how informative our prior is about the shape of the covariances. The default for these parameters is vague and it is our opinion that and should not be tuned since it is unlikely that a user is knowledgeable about these constraints. The next set of parameters and are hyper-parameters for the Gamma distribution on which controls the overall number of clusters. Small values of will encourage fewer clusters and large values of will encourage more clusters. The mean and variance of the Gamma distribution are and respectively, and the default is set such that both mean and variance are 1. As an example of how we can tune this, if we set , the variance will be fixed, and the mean will vary as – in that case we can encourage larger values of and more clusters by choosing small values of . The final set of parameters and are hyper-parameters for the Gamma distribution on which specifies how similar the weights for each sample are to the other samples' distribution – when is small, the amount of information shared is small (weights for each batch can be very different from the overall distribution); when is large, the weights for each batch are likely to be similar to the base distribution. Tuning of via and is analogous to tuning via and .
In the context of flow cytometry, a data sample typically consists of an by data matrix from a single FCS file, where there are events and features reporting scatter and fluorescent intensities. The HDPGMM is a model that fits a collection of such data samples, and makes the assumption that the same cell subsets are present in every sample with frequencies that vary from sample to sample. The model does not make any further assumptions about whether the samples in a collection come from the same or different subjects, experimental conditions, treatment groups etc. Different flow cytometry technologies generate data sets that mainly vary in the maximum number of features that can be observed rather than in the standardized locations of cell subsets or their covariances, and hence and do not need tuning. With more features, it is likely that more cell subsets can be distinguished, and it would be reasonable to tune and to encourage larger values of . The values of and do not depend on the flow cytometry technology, but rather on how similar or different samples are from each other, and can be tuned accordingly. The number of mixture components that are needed for a good model fit is also likely to increase, and we present a diagnostic for model goodness-of-fit that can be used to guide choice of the lower bound for the number of components used in the results and discussion.
The hierarchical DP mixture model allows information sharing over data sets. In the hierarchical model, each flow cytometry data sample can be thought of as a representative of the collection of data samples being simultaneously analyzed. The individual data samples then provide information on the properties of the collection, and this information, in turn, provides information on any particular data sample. In this way, an HDPGMM fitted to a single data sample “borrows strength” from all other samples in the collection being analyzed. In other words, if a rare cell subtype is found in more than one of the samples, we share this information across the samples in the collection to detect the subtype even though the frequency in a particular data sample may be vanishingly small. HDPGMM thus increases sensitivity for clustering cell subsets that are of extremely low frequency in one sample but common to many samples or present in high frequency in one or more samples. In principle, there is no lower limit to the size of a cluster that can be detected in a particular sample. In practice, vanishingly small clusters (e.g. 3–5 events out of 100,000) require expert interpretation to distinguish background from signal, but it is not uncommon for biologically significant antigen-specific cells to be present at such frequencies.
We illustrate the ability of hierarchical modeling to simultaneously overcome the problem of masking of rare event clusters and provide an alignment of cell subsets over multiple data samples. Four simulated data sets were created, each with up to 4 bivariate normal clusters in 4 quadrants. Clusters in each quadrant may have different means or covariance matrices, or be absent entirely; see Figure 2. We compared four different approaches to clustering the data – independent fitting of DPGMM to each data sample, using a reference data set, using pooled data, and using hierarchical modeling.
To evaluate the utility of HDPGMM for identifying rare event clusters in real data, we used reference cell samples containing a predefined number of T cells with known TCR specificity for the NY-ESO-1 cancer-testis antigen. TCR-transduced cells were added to autologous PBMC samples at final concentrations of 0%, 0.013125%, 0.02625%, 0.0525%, 0.105% and 0.21% [8]. There is also a small background contribution by antigen-specific T cells that are already present in the unspiked sample, which is estimated to be 0.0154% using the mean frequency from manual gating by 10 flow practitioners. A total of 50,000 events was then collected from each sample for analysis. At the highest spike frequency, we would therefore expect to detect a maximum of 0.2254%, or 113 antigen-specific T cell events out of 50,000 total events. This is a challenging clustering problem as the frequency of expected multimer-positive events is extremely low, but ideal for validation since the expected number of T cells that bind with high-affinity to the HLA-peptide multimer is known.
DPGMM and HDPGMM models were separately fitted to these six data samples using the FSC, SSC, CD45, CD3 and HLA-multimer channels (5 dimensional), using a truncated Dirichlet process with 128 mixture components, 20,000 burn-in steps and 2,000 identified iterations to calculate the posterior distribution as described in Methods. The trace plots of log-likelihood shown in Figure 4 provides evidence for model convergence, and the distribution of mixture component proportions in Figure 5 provides evidence for model goodness of fit. After consensus modal clustering, the multimer positive clusters were defined using the gating scheme shown in the left panel of Figure 3, but applied to event clusters found by HDPGMM rather than individual events. Since the clustering is done in the full set of markers rather than in two-dimensional slices, events that look close together in a particular projection but are further apart when all dimensions are considered will not belong to the same cluster. The frequency of multimer-positive events as a percentage of all 50,000 events was then calculated. We also ran trials of HDPGMM to evaluate the lower bound needed to find the antigen specific clusters in all samples; 3 out of 4 runs were successful with 32 components, and all runs were successful when 40 or more components were used.
A side-by-side comparison of manually gated, DPGMM and HDPGMM classifications is shown in Figure 3. All 3 approaches are comparable in terms of being able to identify and quantify the antigen-specific cluster of events. Across all runs, DPGMM consistently finds occasional outlier events that are likely to be false positives (e.g. the CD3 negative to low events in the DPGMM fits shown in rows 1 and 4). HDPGMM does not appear to suffer from the same false positive detection, and is also more sensitive for the samples with the lower spiked-in frequencies than DPGMM. However, the most striking advantage of HDPGMM over DPGMM is the interpretability of the hierarchical modeling – cell subsets are consistently labeled across data samples, allowing direct comparison of any cell subset of interest, not just of the multimer positive events.
Figure 6 shows the results from the application of FLOCK, FLAME and flowClust on the same data set. FLOCK only detects the antigen-specific cell subset at the highest spiked-in concentration with a moderate number of probable false positive events that are CD3-negative. As indicated by the color coding of events, FLOCK does not provide any alignment of cell subsets across samples. Using the default settings, FLAME failed to identify any antigen-specific cell subsets. Cell subsets found were aligned but there were alignment artifacts when the event partitioning was different across samples (arrowed example). Using a 64 component mixture, flowClust only detects antigen-specific clusters at the highest spiked-in concentration, and does not provide any alignment of cell subsets. Unlike FLOCK and Dirichlet process based models, the number of components for FLAME and flowClust is not estimated from the data. Hence, in practice, one would have to fit a variety of models with different numbers of components and subsequently perform model selection when using FLAME or flowClust.
In Figure S1, we compare HDPGMM, FLAME and flowClust models with 48 components fitted to the same data set. HDPGMM completed in 3 hours and 30 minutes (20,000 burn-in and 2,000 identified iterations), FLAME took 4 days 12 hours and 28 min, and flowClust completed in 25 minutes (1,000 iterations). With 48 components, HDPGMM found antigen-specific clusters in all samples. FLAME found the clusters when the spiked in concentration was greater or equal to 0.02625%, but cluster alignment failed with the error “missing value where TRUE/FALSE needed”. In contrast, flowClust did not detect any antigen-specific clusters. Both HDPGMM and FLAME clusters included a fair number of CD3-negative events, in agreement with the goodness-of-fit analysis shown in Figure 5 that 48 components is inadequate for modeling rare event clusters in this data set. We tried to run FLAME with 128 components but this was not practical since the program did not terminate after more than 10 days. It took 26 hours for flowClust to run 1,000 iterations with 128 components, and 4 out of 6 samples gave “NA” indicating missing data for all cluster centroids. The wide variation in run-times seen with flowClust (25 minutes to 26 hours) probably reflects early termination with fewer than 1,000 iterations due to tolerance thresholds being met in the 48 component case. We suspect that the missing data might be caused by the Expectation-Maximization algorithm failing when there are zero-event components, but cannot confirm this since the program terminated with no error messages.
Finally, to evaluate the robustness of the DPGMM and HDPGMM frequency estimates, the fitting was repeated 10 times for each algorithm using different random number seeds, and compared to manual gating results from 10 users. Manual gating was performed by operators who were instructed to gate using the same sequence of 2D plots (common gating strategy), but were free to set gate boundaries within any given plot. The results are shown in Figure 7. With respect to linear regression, all three methods perform comparably well with respect to correlation coefficient, but manual gating has slightly less deviation from a straight line fit than HDPGMM which in turn is better than DPGMM. From the figure, it can also be seen that HDPGMM is more accurate than manual gating in that the absolute deviation of the median of the estimates from the “true” concentration is lower than that for manual gating at every concentration. Since the “true” value is taken to be background estimated from 10 manual estimates in the autologous PBMC only sample added to the known spiked-in frequency, accuracy is not evaluated for autologous sample alone. In Figure 8, we show that the algorithm is robust to changes in the hyper-parameters across a 9-fold range.
We have shown that HDPGMM improves on fitting individual samples with DPGMM in two ways - 1) it aligns clusters, making direct comparison of cluster counts across samples possible, and 2) by sharing information across samples, it can identify biologically relevant cell subsets present at frequencies in the 0.01–0.1% range, since “real” cell subsets would naturally be expected to be present in multiple data samples. The hierarchical model is also preferable to using a reference data sample or pooling the data from all samples, since individual sample characteristics are lost with these alternative strategies.
Unlike HDPGMM, other approaches for automated flow cytometry analysis treat data in the same way as DPGMM, that is, fitting a model to independent samples separately, then using a heuristic or algorithm to match up clusters in one data set with another. However, since the model fitting is performed independently, the way that events are partitioned in individual data sets into clusters may be different even across samples that are otherwise very similar, resulting in poor alignment as seen in the FLAME analysis. We are not aware of any other automated flow cytometry analysis software that directly models contributions from individual and grouped samples to align cell subsets, and believe that the HDPGMM approach fills a useful niche in multi-sample comparisons, especially for the quantification of rare event clusters.
One limitation of the HDPGMM model is that all the data to be fitted need to be simultaneously available. This is not an issue for most studies, but may be limiting for longitudinal studies that collect samples serially over an extended period where interim analyses need to be performed. Even in these cases, it may be useful to batch process cell samples in stages using a hierarchical model, then perform post-processing to align cell subsets over different stages. Because of information sharing, cell subsets that are consistent across data samples will be extremely robust features in the posterior distribution. Hence, it is likely that features across batches will be more consistent and easier to align for HDPGMM-fitted batch samples than if every sample was independently fitted.
As described in the text, HDPGMM achieves alignment by assuming that the cluster locations and shapes are constant across datasets, and only their proportions vary from sample to sample. This is similar to the standard practice of using a gating template common to all samples for manual analysis. However, the HDPGMM approach has several advantages over the use of a common gating template. Because the locations and shapes of the clusters are inferred from the data and not imposed top-down by an expert, there is less risk of a subjective bias and failure to detect novel cell subsets. Since classification of events is done by assignment to the maximum probability cluster, cell subsets are not demarcated by arbitrary (typically polygonal) boundaries. In addition, it is simple to tune for higher sensitivity or specificity depending on experimental context by setting the probability necessary for an event to be included in a cluster; events that fall below this threshold are considered to be indeterminate. However, clusters that are doubly rare in the sense of being found in only a small proportion of the samples, and which also constitute a tiny fraction of the total events in any given sample, risk being masked by other more common and high abundance clusters. In many cases, this limitation can be addressed by the inclusion of appropriate positive controls in the samples. Where such positive controls are not available, a post-processing step to scan for “anomalous” events that are found in extremely low probability regions of the posterior distribution at higher frequencies than predicted, may be effective for identifying these doubly rare events.
Technically, our implementation of the HDPGMM integrates several innovations necessary to make such hierarchical models a practical tool for flow cytometry analysis, including the use of a Metropolis-within-Gibbs step for sampling, an identification strategy to maintain consistent component labels across iterations that allows us to calculate the posterior distribution from multiple MCMC iterations, and a consensus modal map to merge components in such a way that non-Gaussian cell subsets are aligned across multiple data sets. To ensure scalability, we have implemented Message Passing Interface (MPI) and Compute Unified Device Architecture (CUDA) optimized code that can take advantage of multiple CPUs and GPUs from a cluster of machines to fit a single HDPGMM model to multiple data sets.
We provide software for HDPGMM fitting to flow cytometry data sets, together with pre-specified robust default parameters and hyper-parameters that make practical usage simple. In our experience, we have never needed to adjust these parameters for data sets ranging from 3-color to 11-color flow cytometry data sets. The only parameters we individually set are the number of burn-ins, the number of iterations to collect for the posterior distribution, and the maximal number of components for the truncated DP algorithm. These parameters are tuned mainly for computational efficiency since conservative defaults that would be expected to be effective in all common use cases can be given, with the trade-off being longer run times. In addition, the use of prior information to set the starting values for component means and covariances (e.g. from fits to previously collected similar data) would reduce the number of iterations necessary to achieve convergence.
The fitting of HDPGMM is computationally demanding but can be accelerated with cheap commodity graphics cards as previously described [16]. For example, running an MCMC sampler for 20,000 burn-in and 2,000 identified iterations to fit a 128-component HDPGMM to the six multimer data sets shown in Figure 3 took less than 6 hours on a Linux workstation using a single NVidia GTX 580 card costing under USD 500. The algorithm has runtime complexity of , and benchmark experiments shown in Figure 9 confirm that the performance is linear in the number of events and samples and quadratic in the number of markers. Open source code for fitting DPGMM and HDPGMM models to flow cytometry data is available from http://code.google.com/p/py-fcm/. The code is written in the Python programming language, and will run on regular CPUs, but is optimized for massively parallel computing using the CUDA interface (a suitable Nvidia GPU is required for CUDA). Flow cytometry data samples, source code and a sample script to fit a HDPGMM model to the data are provided in Supplementary Materials.
In summary, we describe and provide code for a hierarchical modeling extension to statistical mixture models that improves on the robustness, sensitivity and interpretability of model-based approaches for automated flow cytometry analysis. We demonstrate the consistency of frequency of HDPGMM estimates on reference data samples spiked with extremely low frequencies of antigen-specific cells, a scenario directly relevant to many clinical research studies in vaccine development, immune monitoring and immune biomarker discovery where the frequency of rare antigen-specific T cells is of interest.
We give posterior computational details only for HDPGMM since details for our implementation of DPGMM have been previously published [16]. First, let and so that . Furthermore, let and . These along with equations (3) and (4) give a complete specification of the model. Metropolis within Gibbs is performed by updating each parameter with a draw from its conditional distribution in turn and when the conditional distribution is intractable, use a Metropolise Hastings update instead. We give the specifics of the sampling in the remainder of this section.
The generation of the standard samples with a defined number of antigen-specific CD8 T cells spiked into autologous PBMC for use in HLA-peptide multimer has been described [8]. Briefly, Phytohemagglutinin (PHA; ) and IL-2 (20 U/ml) stimulated HLA-A*0201 positive PBMC were retrovirally transduced with an HLA-A*0201 restricted specific TCR construct after the CD4 T cells were depleted using Dynabeads (Invitrogen). After 5 days, the transduced cells were harvested and purified using APC-conjugated NY-ESO-1 specific HLA multimer and magnetic cell sorting. Purified cells were clonally expanded, harvested and spiked at the desired percentage of NY-ESO-1 specific TCR expressing CD8 T cells into autologous PBMC. These samples were stained with monoclonal antibodies specific for CD45 (pan leukocyte) CD3 (T-lymphocytes) and HLA-A*0201 NY-ESO-1 157–165 multimers to identify spiked T cells. For details, please refer to reference [8]).
Sample preparation conditions were set so that results (i.e. generated FCS files) would be as comparable as possible: Cell staining was performed simultaneously by the same operator, using the same batches of staining reagents, and data acquisition was subsequently done in a single experiment using the same cytometer settings (voltages, compensations) for all samples. The data were generated using a FACSCalibur and CellQuest Pro 6.0, with values ranging from 0 to 1023. No further transformations were performed on the data but standardization to have zero mean and unit standard deviation was performed before fitting the mixture model so all markers would have equal contributions. The standardization was reversed before plotting - i.e. all plots are based on the original 0 to 1023 scale. For gating estimates, frequency estimates from 10 flow cytometry operators using the same gating strategy were collected.
|
10.1371/journal.pmed.1002246 | Subjective and objective cognitive function among older adults with a history of traumatic brain injury: A population-based cohort study | Traumatic brain injury (TBI) is extremely common across the lifespan and is an established risk factor for dementia. The cognitive profile of the large and growing population of older adults with prior TBI who do not have a diagnosis of dementia, however, has not been well described. Our aim was to describe the cognitive profile associated with prior TBI exposure among community-dwelling older adults without dementia—an understudied but potentially vulnerable population.
In this population-based cohort study, we studied 984 community-dwelling older adults (age 51 y and older and their spouses) without dementia who had been randomly selected from respondents to the 2014 wave of the Health and Retirement Study to participate in a comprehensive TBI survey and who either reported no prior TBI (n = 737) or prior symptomatic TBI resulting in treatment in a hospital (n = 247). Mean time since first TBI was 38 ± 19 y. Outcomes assessed included measures of global cognitive function, verbal episodic memory, semantic fluency, and calculation as well as a measure of subjective memory (“How would you rate your memory at the present time?”). We compared outcomes between the two TBI groups using regression models adjusting for demographics, medical comorbidities, and depression. Sensitivity analyses were performed stratified by TBI severity (no TBI, TBI without loss of consciousness [LOC], and TBI with LOC). Respondents with TBI were younger (mean age 64 ± 10 y versus 68 ± 11 y), were less likely to be female, and had higher prevalence of medical comorbidities and depression than respondents without TBI. Respondents with TBI did not perform significantly differently from respondents without TBI on any measure of objective cognitive function in either raw or adjusted models (fully adjusted: global cognitive function score 15.4 versus 15.2, p = 0.68; verbal episodic memory score 4.4 versus 4.3, p = 0.79; semantic fluency score 15.7 versus 14.0, p = 0.21; calculation impairment 22% versus 26%, risk ratio [RR] [95% CI] = 0.86 [0.67–1.11], p = 0.24). Sensitivity analyses stratified by TBI severity produced similar results. TBI was associated with significantly increased risk for subjective memory impairment in models adjusted for demographics and medical comorbidities (29% versus 24%; RR [95% CI]: 1.26 [1.02–1.57], p = 0.036). After further adjustment for active depression, however, risk for subjective memory impairment was no longer significant (RR [95% CI]: 1.18 [0.95–1.47], p = 0.13). Sensitivity analyses revealed that risk of subjective memory impairment was increased only among respondents with TBI with LOC and not among those with TBI without LOC. Furthermore, the risk of subjective memory impairment was significantly greater among those with TBI with LOC versus those without TBI even after adjustment for depression (RR [95% CI]: partially adjusted, 1.38 [1.09–1.74], p = 0.008; fully adjusted, 1.28 [1.01–1.61], p = 0.039).
In this population-based study of community-dwelling older adults without dementia, those with prior TBI with LOC were more likely to report subjective memory impairment compared to those without TBI even after adjustment for demographics, medical comorbidities, and active depression. Lack of greater objective cognitive impairment among those with versus without TBI may be due to poor sensitivity of the cognitive battery or survival bias, or may suggest that post-TBI cognitive impairment primarily affects executive function and processing speed, which were not rigorously assessed in this study. Our findings show that among community-dwelling non-demented older adults, history of TBI is common but may not preferentially impact cognitive domains of episodic memory, attention, working memory, verbal semantic fluency, or calculation.
| Prior studies have found that a history of traumatic brain injury (TBI) is associated with an increased risk of developing dementia later in life. It is unclear, however, whether a history of TBI has any measurable impact on cognitive function for the large and growing population of older adults without dementia who are living in the community.
Our aim was to find out if a history of TBI is associated with measurable cognitive impairment among older adults without dementia who are living in communities across the United States.
We used data collected in a large nationally representative survey study of older adults (age 51 y and older and their spouses) called the Health and Retirement Study.
We studied 247 older adults with a history of TBI and 737 older adults with no history of TBI. We compared them on measures of subjective memory impairment (self-rated memory problems) and objective cognitive function (neuropsychological tests of attention, working memory, short-term memory, calculation, and verbal category fluency).
We found no differences on any measures of objective cognitive function between older adults with and without a history of TBI. We did find that older adults with a history of TBI that involved loss of consciousness (LOC), but not those with a history of TBI without LOC, were 28% more likely to report subjective memory impairment compared to those with no history of TBI—even after accounting for differences in demographics, medical conditions, and depression.
Among community-dwelling older adults across the United States, a history of TBI is not associated with any measurable impairment in the cognitive domains of attention, working memory, short-term memory, calculation, and verbal category fluency—cognitive domains that are often affected very early in Alzheimer disease. This study did not, however, rigorously evaluate the cognitive domains of processing speed and executive function—cognitive domains that have been shown to be impaired among older military veterans with a history of TBI.
Further research is needed to determine the underlying cause of the higher rate of subjective memory impairment among older adults with TBI with LOC.
| An estimated 2.5 million people in the US seek hospital-based care for traumatic brain injury (TBI) annually [1]. An additional 3.2 to 5.3 million are estimated to be living with TBI-related disability [2]. Lifetime prevalence of TBI is up to 40% among civilian adults [3] and is likely higher among military veterans [4]. TBI may cause immediate cognitive deficits of varying degree and duration across a variety of cognitive domains [5]. TBI is also a risk factor for dementia [6–13], suggesting that in certain vulnerable individuals TBI may contribute to progressive cognitive decline. The majority of TBI-exposed older adults, however, do not develop dementia [10,12]. Risk factors for post-TBI cognitive decline remain to be elucidated [14]. Thus, the clinical importance of a prior TBI exposure to the average community-dwelling non-demented older adult is unclear. In order to determine whether prior TBI is associated with significant cognitive deficits in this population, we need high-quality population-based studies pairing validated TBI exposure measures and detailed cognitive outcome assessments.
Most studies that have assessed detailed long-term cognitive outcomes among individuals with prior TBI have focused on clinical or convenience samples [15–17]. While these studies have greatly advanced our understanding of the long-term cognitive consequences of TBI in these specific populations, they may not be generalizable to the broader US population of community-dwelling older adults. One of the major challenges of studying the long-term consequences of prior, often remote, TBI on a population level is that it is necessary to rely upon self-report of TBI exposure, as it is usually not feasible to confirm diagnoses that may have occurred decades earlier without accessible documentation. A number of prior cohort studies of TBI and dementia have assessed lifetime TBI exposure using very brief self-report screens [18–20] that are known to miss more than 35% of people who would have endorsed prior TBI with a more comprehensive screen [21,22]. Of the few studies that have applied comprehensive TBI screens on a population level, a couple have identified a high prevalence of subjective cognitive impairment [23,24] among individuals with prior TBI. To our knowledge, however, there have been no prior population-based studies of the long-term consequences of TBI that have assessed detailed measures of objective cognitive function.
Our aim in this study was to address the knowledge gaps that have limited our understanding of the long-term cognitive consequences of prior, often extremely remote, TBI on the average community-dwelling older adult without dementia. Specifically, we sought to determine whether a history of TBI (ascertained via a comprehensive TBI screen) is an independent risk factor for not only subjective but also objective cognitive impairment in a population-based community-dwelling sample of older adults without dementia.
This was a cross-sectional cohort study using publicly available secondary data from the Health and Retirement Study (HRS). The study is reported as per STROBE guidelines (S1 STROBE Checklist). All HRS respondents provided oral consent for the data used in this analysis. This study was deemed exempt by the University of California San Francisco Human Research Committee due to the use of publicly available de-identified data.
Data are from the HRS, a nationally representative longitudinal survey study of community-dwelling older adults (defined as adults age 51 y and older and their spouses) that began in 1992 and has continued with repeat surveys every 2 y. The HRS uses national area probability sampling of US households with supplemental oversampling of black individuals, Hispanic individuals, and Florida state residents. De-identified data are made publicly available within 1–2 y of survey completion. Detailed information about the sampling procedures, study design, instruments, and data access are available online (http://hrsonline.isr.umich.edu). For the present study, we used data from the TBI module survey that was administered to a random sub-sample of the 2014 survey respondents, as follows. The 2014 survey collected “core interview” data on 18,748 respondents. Of these, 17,698 (94%) respondents participated in the survey independently, while 1,050 (6%) respondents were unable to answer for themselves and underwent interviews by proxy respondents. Upon completion of the core interview, the 17,698 non-proxy respondents were offered participation in additional non-core surveys (“modules”). Of these, 16,642 (94%) agreed to participate in additional modules and were randomly assigned to one of 11 different survey modules. Of these 16,642 non-proxy respondents, 1,489 were randomly assigned to the TBI module (see Fig 1). Compared to the rest of the 2014 core interview non-proxy respondents, those who completed the TBI module were slightly more educated (mean [standard deviation] years education 13.6 [8.0] versus 13.2 [6.7], p < 0.01) but did not differ significantly on age, race, or ethnicity.
The TBI module consisted of a modified version of the Ohio State University TBI Identification Method (OSU TBI-ID). Because of its demonstrated reliability and predictive validity [25–27], the OSU TBI-ID is considered the gold-standard instrument for self-report of lifetime exposure to TBI and is recommended by the National Institute of Neurological Disorders and Stroke for use by clinical researchers studying TBI [28].
The modified OSU TBI-ID used in the HRS began with the statement “I am going to ask you about injuries to your head or neck that you may have had any time in your life.” The survey continued with six questions about injuries to the head or neck that may have been sustained in high-risk situations such as vehicle accidents and falling (e.g., “In your lifetime, have you ever injured your head or neck in a car accident…?”). If a respondent screened positive on one of these initial questions, then the survey continued with more detailed questions about the timing, associated symptoms, and need for medical attention of each reported injury up to a maximum of six injuries. The initial screening questions of the OSU TBI-ID are extremely sensitive but poorly specific as they detect injuries not only to the head but also to the neck, many of which would not fulfill clinical criteria for a TBI, which requires a traumatic impact to the head followed by neurological symptoms such as confusion, amnesia, or loss of consciousness (LOC) [29]. Furthermore, self-report of lifetime TBI, particularly early-life TBI, has been proven unreliable for TBIs that did not result in hospitalization [30]. Thus, in the present study, “TBI” was conservatively defined as any prior injury to the head or neck that (1) required treatment in a hospital (“Were you hospitalized or treated in an emergency room?”) and (2) resulted in LOC (“Were you knocked out or did you lose consciousness?”) or peri-traumatic amnesia/feeling dazed (“Were you dazed, or did you have a gap in your memory?”) or both. “No TBI” was defined as no prior head or neck injury of any kind. In order to reduce the likelihood of misclassification of TBI exposure, “ambiguous injury” was defined as any prior head or neck injury that did not meet the above criteria for TBI; respondents with ambiguous injury were excluded from subsequent analyses. For the purpose of sensitivity analyses, TBI severity (with versus without LOC) was also classified. A more precise measure of mild TBI could not be obtained because duration of LOC was not coded in number of minutes. For this reason, we chose to code severity according to LOC status rather than specific LOC duration. Although verification of TBI exposure via medical record review was not feasible in this study, our highly conservative approach to TBI exposure definition using a sensitive gold-standard instrument and exclusion of ambiguous cases makes misclassification unlikely.
Of the 1,489 respondents to the TBI module, those reporting a prior diagnosis by a physician of Alzheimer disease (AD, n = 6) or non-AD dementia (n = 20) were excluded. Of the remaining 1,463 non-demented respondents to the TBI module, 479 (33%) were classified as having ambiguous injury and were excluded. The remaining 984 respondents comprised the final study cohort and consisted of 247 respondents with TBI and 737 respondents with no TBI.
We included the following covariates that may modify or confound the association between TBI exposure and cognition: age, sex, ethnicity, race, education, self-report of physician-diagnosed medical comorbidities (hypertension, diabetes, cancer, lung disease, heart disease, stroke, smoking, and arthritis), and a measure of depression. Depression was measured via the eight-item Center for Epidemiologic Studies Depression Scale (CES-D 8), which asks respondents to report whether they experienced eight specific depressive symptoms over the past week (score 0–8). Based on prior studies, depression was defined as a score of 3 or higher [32].
All analyses were conducted using Stata version 13 [33]. Demographics, medical comorbidities, and depression were compared between TBI and no-TBI respondents using t-tests for continuous variables or chi-square tests for categorical variables. Objective cognitive outcomes with scores ranging from 0 to 10 or higher (global score, verbal episodic memory, and semantic fluency) were compared between TBI and no-TBI respondents using linear regression. Ordinal categorical outcomes (calculation and subjective memory) were binarized as described above and compared between TBI and no-TBI respondents using Poisson regression with a robust variance estimator as recommended by Cummings to estimate adjusted risk ratios (RRs) for binary outcomes in Stata [34]. In order to elucidate the complex relationship between TBI, demographics, comorbidities, depression, and cognitive function, regression models were first adjusted for demographics and medical comorbidities that significantly differed between groups and were then additionally adjusted for depression (CES-D 8 score ≥ 3). To evaluate the role of TBI severity, sensitivity analyses were conducted stratified by LOC status (no TBI, TBI without LOC, TBI with LOC) and differences between TBI severity groups were assessed via tests of trend and interaction. Significance was set at p < 0.05.
The analysis plan was initially determined via in-person meetings of the authors in advance of obtaining the data. Once the data were obtained and cleaned, the analysis plan was modified to optimize the feasibility and scientific value of the analyses within the limitations of the available secondary data (such as missing outcome data and lack of optimally detailed coding of mild TBI exposure). Following peer review, additional minor modifications were made, including the use of an alternative HRS smoking variable with less missing data and the addition of tests of trend and interaction to the sensitivity analyses as described above.
Respondents with TBI were younger; were less likely to be female; reported higher rates of lung disease, heart disease, and arthritis; and had more depression compared to respondents without TBI (Table 1). There were minimal missing covariate data. Each individual covariate had 0% to less than 1% missing data, with the exception of the smoking covariate, which had 1.5% missing data. Overall, 2.5% of the sample was missing one or more of the Table 1 covariates, and only 0.6% of the sample was missing one or more of the covariates included in the fully adjusted model (99.4% complete caseness rate).
The majority of respondents who reported TBI endorsed only a single prior TBI, and most TBIs involved loss of consciousness (Table 2). While average time since first TBI ranged from 0 to 98 y, the majority of TBIs occurred more than four decades ago (median 41 y). There were minimal (<3%) missing data in the outcomes assessed, except for the outcome of semantic fluency (82% missing), which was systematically missing as described above.
Measures of objective cognitive function—global cognition, verbal episodic memory, semantic fluency, and calculation—were not significantly different between respondents with and without prior TBI in both raw and adjusted analyses (Table 3). These negative findings were unchanged in sensitivity analyses stratified by TBI severity (Table 4).
Respondents with prior TBI were significantly more likely to endorse subjective memory impairment in both unadjusted models and models adjusted for demographics and medical comorbidities (35% versus 27%; RR [95% CI]: raw, 1.28 [1.04–1.57], p = 0.021; partially adjusted, 1.26 [1.02–1.57], p = 0.036; Table 3). Following further adjustment for depression, however, risk for subjective memory impairment was no longer significant (RR [95% CI]: fully adjusted, 1.18 [0.95–1.47], p = 0.13; Table 3).
In sensitivity analyses stratified by TBI severity, risk for subjective memory impairment was significantly elevated among respondents with TBI with LOC but not among respondents with TBI without LOC in all raw and adjusted models (Table 4). Adjustment for depression reduced, but did not attenuate, the risk of subjective memory impairment among those with TBI with LOC (RR [95% CI]: partially adjusted, 1.38 [1.09–1.74], p = 0.008; fully adjusted, 1.28 [1.01–1.61], p = 0.039; Table 4). While differences in the risk of subjective memory impairment between the TBI with LOC group and the TBI without LOC group did not reach statistical significance (test of TBI with LOC = TBI without LOC: raw p = 0.13, partially adjusted p = 0.14, fully adjusted p = 0.22), there was a statistically significant trend for greater risk of subjective memory impairment across TBI severity groups (test of trend across TBI severity groups [no TBI, TBI without LOC, TBI with LOC]: raw p = 0.006, partially adjusted p = 0.013, fully adjusted p = 0.065).
This study was powered (at alpha = 0.05; power = 0.8) to detect differences (effect sizes) as small as 0.85 points in global cognition, 0.39 points in verbal episodic memory, 3.2 points in semantic fluency, 9.3% in prevalence of calculation impairment, and 9.1% in prevalence of subjective memory impairment between TBI groups.
In this population-based study of community-dwelling older adults without dementia, prior TBI with LOC—but not prior TBI without LOC—was associated with a 38% increased risk for subjective memory impairment that was partially but not entirely mediated by comorbid depression. Furthermore, prior TBI of any severity was not associated with significant objective cognitive impairment in the domains of episodic memory, attention, working memory, verbal semantic fluency, or calculation.
Our finding that history of TBI is associated with subjective cognitive impairment is consistent with the results of two prior population-based survey studies [23,24]. One of these studies reported that 22% of respondents with prior TBI reported subjective difficulty with memory or thinking [24]. This study did not include a comparison to respondents without TBI, an assessment of this outcome according to TBI severity, or an assessment of objective cognitive function. Additionally, the survey non-response rate in this study was quite high (63%) [24], limiting the generalizability of the findings. The other survey study, which assessed multiple self-reported outcomes in a population-based sample of adults of all ages with and without TBI, reported mixed results [23]. On one measure, risk for subjective cognitive complaints was elevated among those with TBI with LOC (adjusted prevalence ratio [95% CI]: 1.44 [1.08–1.91]) but not among those with TBI without LOC (adjusted prevalence ratio [95% CI]: 1.15 [0.77–1.71]) compared to those without TBI—which is similar to our own findings. On another measure of post-concussive symptoms, however, there was increased risk for self-reported poor memory and poor concentration among all respondents with prior TBI, even those without LOC [23]. This study did not, however, adjust for medical comorbidities or depression and also did not assess measures of objective cognitive function. Our study expands upon these prior findings by demonstrating that a history of TBI with LOC is a risk factor for subjective cognitive impairment among community-dwelling older adults without dementia but that this risk is partially mediated by comorbid depression. This more nuanced understanding of post-TBI subjective cognitive impairment is important because depression may be a long-term consequence of prior TBI [16,24,35] and because depression is a treatable condition, raising hope that this negative outcome may be partially modifiable. Furthermore, our study is the first population-based study to our knowledge to demonstrate that the subjective cognitive impairment reported by older adults with a history of TBI may not be subserved by measurable impairments in objective cognitive function.
The lack of objective cognitive impairment in individuals with prior TBI found in our study may be due to a variety of factors. TBI-associated cognitive impairment among older adults may primarily impact domains not rigorously covered by the HRS cognitive battery—namely, executive function and processing speed. For example, prior studies of older military veterans have identified TBI-associated impairments only on rigorous measures of processing speed (Trail Making Test Part A [36] and Wechsler Adult Intelligence Scale–Revised Digit Symbol [37]) and executive function (Trail Making Test Part B [36] and NIH-EXAMINER fluency factor and cognitive control factor scores [38])—neither of which are part of the HRS cognitive battery—and not on detailed measures of attention, working memory, episodic memory, or language [16,17]. Alternatively, it is possible that the HRS cognitive battery was not sufficiently sensitive to detect mild impairments in the domains tested. For example, subjective cognitive impairment has been associated with the development of objective cognitive impairment decades later [39]. Thus, it is possible that measurable cognitive deficits will develop in these respondents over the next several years as they are followed longitudinally. The lack of objective cognitive impairment in individuals with prior TBI may also be due to the younger age of the respondents with TBI compared to those without TBI, though our models that adjusted for age make this interpretation less likely. Survival bias may have influenced our results if this community-dwelling non-demented sample of older adults represents a resilient segment of the US population. Lastly, the lack of evidence for objective cognitive impairment among older adults with a history of TBI may suggest that, in the general population (non-clinical, non-convenience sample), a history of TBI is not a significant risk factor for cognitive impairment or dementia. Indeed, some prior epidemiological studies have failed to identify an association between TBI and dementia [40–45]. Thus, our finding may provide support for the hypotheses that prior studies that have identified an elevated risk of dementia following TBI may have been limited by recall bias due to self-reported TBI [8,46–49], reverse causation [50], misdiagnosis of post-concussive syndrome as dementia [51], or inclusion of only clinical populations (not population-based).
Of note, the cognitive domains assessed in this study mainly include domains that are typically affected very early in typical AD: episodic memory, semantic fluency, and calculation. In fact, these domains may even be impacted in the preclinical phase of AD [52,53]. Performance on tests of episodic memory, in particular, has been found to correlate with amyloid pathology among cognitively normal older adults [54]. Thus, our finding that non-demented older adults with prior TBI do not have greater impairment in episodic memory compared to their non-TBI-exposed counterparts provides support for the hypothesis that TBI-related dementia and cognitive impairment, when present, are likely not of the AD type. Indeed, as mentioned above, prior cohort studies of older military veterans have identified impairments in executive function and processing speed among veterans with a history of fairly remote TBI compared to veterans without TBI, but have failed to identify differences on tests of attention, working memory, episodic memory, or language [16,17]. Prior cohort studies comparing cognitive profiles of patients with dementia and prior TBI to patients with dementia without prior TBI have similarly identified a non-AD pattern of deficits [18,19]. And while TBI is a fairly well-established risk factor for all-cause dementia, whether TBI is a risk factor for AD in particular remains unclear [55], with many epidemiological studies reporting increased risk for AD after TBI [6,7,9,11,12,47,56] and others reporting no increased risk [20,40] (see [55] for an excellent, well-balanced review of this topic). In light of growing evidence that the purely clinical (non-biomarker/neuropathological) diagnosis of AD is frequently inaccurate [57], it is notable that nearly all prior studies reporting an association between TBI exposure and AD-type dementia have relied upon clinical or ICD-9-code-based diagnosis of AD [6,7,9,11,12,47,56], while the few studies that have included neuropathological confirmation of AD diagnosis have failed to find an association between AD neuropathology and history of TBI [19,20].
Strengths of this study include the use of population-based, nationally representative data, a detailed cognitive assessment of both subjective and objective cognitive function, a gold-standard comprehensive TBI screen, a conservative approach to TBI classification that makes misclassification unlikely, and a careful assessment of the role of comorbid depression. This study was adequately powered to detect clinically relevant effect sizes in all outcomes assessed except for the outcome of semantic fluency. Additional strengths include the focus on community-dwelling older adults without dementia—an understudied but potentially vulnerable population.
Limitations of this study include the lack of detailed measures of executive function and processing speed, the potential for recall bias on all self-reported measures, the inability to determine onset of depression relative to TBI, and the slight bias towards higher education among respondents to the TBI module compared to the rest of the non-proxy HRS respondents, suggesting that respondents in this study may have slightly greater cognitive reserve than the general older adult population in the US [58]. Assessment of multiple outcomes without adjustment for multiple comparisons raises the possibility of a false-positive association between TBI and subjective cognitive impairment. Additional limitations include substantial systematic missing data in the outcome of semantic fluency, as this outcome was given only to first-time HRS respondents over age 65 y. The semantic fluency outcome may therefore be more prone to survival bias than the other outcomes and may also have inadequate power to detect a smaller, but potentially clinically relevant, effect size. It is reassuring, however, that the results in the domain of semantic fluency are consistent with the results across all of the other objective cognitive domains assessed. Lastly, as reflected by our high proportion of cases with TBI with LOC (72% of all TBI cases), our approach to TBI classification likely led to underrepresentation of mild TBI, thus limiting generalizability, particularly of our positive findings, to those with mild TBI.
In conclusion, this study provides, to our knowledge, the first population-based evidence that while community-dwelling non-demented older adults with a history of TBI may have increased subjective cognitive impairment compared to their counterparts without TBI, they may not have measurable impairments in the cognitive domains of episodic memory, attention, working memory, verbal semantic fluency, or calculation. While this study may provide some degree of reassurance to both older adult patients with prior TBI and their providers, further population-based studies are needed that include more comprehensive multi-domain cognitive batteries as well as measures of cognitive and depressive trajectories over time. The high prevalence of TBI of sufficient severity to require medical attention (17%) identified in this population-based study highlights the critical importance of ongoing TBI prevention efforts [59]. The high prevalence of depression identified among older adults with prior TBI (24%), in combination with the role that depression may play in increasing risk for subjective cognitive impairment, highlights an opportunity for more aggressive management of depression in this population.
|
10.1371/journal.pgen.1002728 | Target Gene Analysis by Microarrays and Chromatin Immunoprecipitation Identifies HEY Proteins as Highly Redundant bHLH Repressors | HEY bHLH transcription factors have been shown to regulate multiple key steps in cardiovascular development. They can be induced by activated NOTCH receptors, but other upstream stimuli mediated by TGFß and BMP receptors may elicit a similar response. While the basic and helix-loop-helix domains exhibit strong similarity, large parts of the proteins are still unique and may serve divergent functions. The striking overlap of cardiac defects in HEY2 and combined HEY1/HEYL knockout mice suggested that all three HEY genes fulfill overlapping function in target cells. We therefore sought to identify target genes for HEY proteins by microarray expression and ChIPseq analyses in HEK293 cells, cardiomyocytes, and murine hearts. HEY proteins were found to modulate expression of their target gene to a rather limited extent, but with striking functional interchangeability between HEY factors. Chromatin immunoprecipitation revealed a much greater number of potential binding sites that again largely overlap between HEY factors. Binding sites are clustered in the proximal promoter region especially of transcriptional regulators or developmental control genes. Multiple lines of evidence suggest that HEY proteins primarily act as direct transcriptional repressors, while gene activation seems to be due to secondary or indirect effects. Mutagenesis of putative DNA binding residues supports the notion of direct DNA binding. While class B E-box sequences (CACGYG) clearly represent preferred target sequences, there must be additional and more loosely defined modes of DNA binding since many of the target promoters that are efficiently bound by HEY proteins do not contain an E-box motif. These data clearly establish the three HEY bHLH factors as highly redundant transcriptional repressors in vitro and in vivo, which explains the combinatorial action observed in different tissues with overlapping expression.
| NOTCH signaling is a central developmental pathway that influences a multitude of cell fate decisions and differentiation steps as well as later tissue homeostasis and regeneration. The three HEY genes encode basic helix-loop-helix transcription factors that are critical effectors to convey signaling by NOTCH receptors and similar signaling systems. This is underscored by the multitude of developmental defects observed in HEY single- and double-mutant mice. The mode of action of HEY proteins remained largely unexplored, however. By gene expression analysis and chromatin immunoprecipitation we have now identified a large set of HEY target genes. While only 500–2,000 mRNAs are regulated by HEY1 or HEY2, there are around 10,000 binding sites in the genome. HEY proteins act as transcriptional repressors that bind close to transcriptional start sites in all cases tested. In contrast, gene activation seems to be mediated by indirect/secondary mechanisms. The extent of regulation is rather limited, implicating HEY genes in modulating rather than switching on or off target gene expression. All our in vitro and in vivo data point to a high degree of redundancy between the three HEY genes, suggesting that tissue specific patterns and expression levels determine the final outcome of HEY induced cellular responses.
| NOTCH signaling is a key regulatory pathway for cardiovascular development and homeostasis [1]. Its receptors mainly act through transcriptional activation of target genes by a complex of the NOTCH intracellular domain, released by gamma-secretase, the transcription factor CBF1 (RBP-Jk) and the Mastermind co-activator proteins (Maml1-3). Without NOTCH binding CBF1 has a repressive function and associates with additional co-repressor proteins. Upon activation different and in part cell type specific target genes are induced, the most prominent ones encoding members of the HEY and HES family of bHLH repressor proteins. There are three HEY genes (HEY1, HEY2 and HEYL) and several HES genes, with HES1 being the closest relative. All are related to the Drosophila hairy and Enhancer-of-split genes, which are well known transcriptional repressor proteins. HEY and HES proteins have a similar domain architecture with a DNA binding and dimer-forming bHLH (basic helix-loop-helix) region, an Orange domain that may also participate in dimerization and conserved C-terminal WRPW (HES) or YRPW (HEY) motifs. The WRPW peptide mediates interactions with groucho-type co-repressor proteins, but YRPW interaction partners for HEY proteins are still unknown.
Mouse knockout studies have revealed a striking overlap in phenotypes between NOTCH and HES or HEY mutants suggesting that these bHLH factors convey a significant fraction of the known biological responses [1], [2]. Loss of HEY2 or HEY1/HEYL leads to identical cardiac phenotypes with ventricular septum defects (VSD) and valve anomalies that appear to be due to an impaired EMT process of endocardial cells in the atrioventricular canal [3]. Since HEY1 and HEYL single knockouts do not show evidence of cardiac developmental defects these genes are obviously less critical in this process. Nevertheless, the overlap in endocardial expression and the overlap in phenotypes clearly argue for a combined and partially redundant action of all three HEY genes. Interestingly, there is also a cooperation of HEY1 and HEYL in skeletal muscle since double knockout mice lack quiescent satellite cells, which are essential for regeneration [4]. When HEY1 and HEY2 are deleted together a much earlier embryonic vascular defect is observed with failure of angiogenic remodeling and a lack of arterial differentiation [5], [6]. Additional critical sites for HEY functions are the inner ear [7], brain [8] and bone [9]. The apparent redundancy of HEY factors in many sites and their high degree of sequence identity in the bHLH region suggest that they may be functionally interchangeable, but there are also claims for a completely different function of HEYL e.g. in neurogenesis [10]. Evidence for this scenario is limited, however.
Despite extensive analyses of mouse phenotypes surprisingly little is known about the direct targets of HEY or HES genes [2]. Microarray analyses of overexpressing cells or tissues from knockout animals have provided evidence for HEY dependent changes in gene expression in several settings [11]–[14]. There is very little overlap between target lists, however, and evidence for direct regulation of these genes by HEY proteins is largely lacking.
To better characterize the network of genes that mediate NOTCH-HEY signaling effects in target cells we generated HEK293 cells that express HEY1, HEY2 or HEYL in a highly regulatable fashion. These cells were used to search for HEY-dependent changes in transcript levels by microarray analysis and to identify direct binding sites of all three HEY factors in target genes. We could define putative binding motifs and validate DNA targets by promoter analysis. Analysis of cardiac tissue from knockout mice validates a number of these genes as direct in vivo targets.
To identify target genes of HEY factors we employed HEK293 cells with tightly regulated HEY transgene expression. HEK293 cells were chosen since they express endogenous HEY genes at significant levels ensuring that these cells are capable of responding appropriately. According to transcript profiling HEY1 ranks as number 525 of all expressed genes, while HEY2 and HEYL are expressed at lower levels with a rank of around 10.000 [15]. To generate a system with tunable HEY gene expression, cells were first transfected with pWHE134 [16], encoding a reverse tetracycline transactivator (rtTA) plus a tetracycline-dependent repressor (tetR-KRAB) driven by a CMV promoter (293tet cells). Flag-HA-(FTH)-tagged HEY1 and Flag-HEY2 sequences under the control of a tetracycline-responsive promoter were subsequently introduced via lentiviral vectors (for details see Materials and Methods and Figure S1A–S1D). For some of the experiments Flag-Strep-(FS)-tagged constructs were employed with similar results using transposon-mediated insertion (FS-HEY1, FS-HEY2, and FS-HEYL). The N-terminal epitope tags do not affect localization or transcriptional activity of the proteins in vitro. Western Blot analysis and quantitative RT-PCR of individual clones confirmed tightly regulated doxycycline (Dox)-dependent expression (Figure S1E). Experiments were either conducted with 30–50 ng/ml Dox for low level expression (e.g. raising endogenous HEY1 mRNA level by 2–10 fold) or with 1–2 µg/ml Dox for stronger overexpression.
HEY regulated genes were identified after strong induction for 48 h (293tet-FTH-Hey1) or 72 h (293tet-Flag-Hey2), respectively, using Affymetrix microarrays. Under more stringent cut-off values only a small number of genes appeared regulated. With relaxed criteria of ≥1.3× similar numbers of up- and down-regulated genes were identified (Figure 1A and Table S1). The number of target genes and the extent of regulation were greater for HEY2, which may result from differences in protein levels due to longer induction, differential potency of the bHLH protein or cell-intrinsic mechanisms. Comparison of gene lists identified a strong overlap, but in all cases the span of regulation is small.
Validation of microarray results was done by quantitative real-time RT-PCR (qRT-PCR) on a subset of genes (Figure 1B). Repression could generally be confirmed and the extent of regulation tended to be higher, in the range of 2–6-fold. For upregulated genes validation was also successful in most cases, but the extent of regulation was more limited. Importantly, values obtained for HEY1 and HEY2 were more similar now, likely due to the same 72 h induction period. Especially for repressed genes the longer induction time may lead to larger changes since the half-life of target mRNAs becomes less of a problem. Expression of the endogenous HEY1 as well as HEY2 and the related HES1 gene was repressed, pointing to a negative feedback loop for these factors.
These experiments were repeated for HEK293 cells with a regulated expression of Flag-Strep-tagged HEYL and all the genes tested exhibited very similar direction and extent of regulation. Thus, the three HEY factors appear functionally interchangeable, at least in HEK293 cells.
Since HEK293 cells express endogenous HEY1 this may already lead to a repression at baseline. We therefore tested expression of target genes in a HEY1 knockdown situation. Using stably expressed shRNA we managed to reduce HEY1 RNA expression by 75%. Even this rather limited reduction had a clear impact on target genes expression (Figure 1B). Several HEY-repressed genes were up-regulated up to 3.4-fold (BMP2), while at least some of the genes induced upon HEY overexpression tended to be repressed by HEY1 knock-down, further confirming the validity of our target genes.
Gene ontology analysis of regulated genes identified a striking overrepresentation of genes related to transcriptional control as well as development and differentiation (Figure 1C). The prevalence of transcriptional control genes suggests that HEY genes are positioned higher up in the hierarchy of signaling cascades and modulate other transcription factors rather than effector genes. The terms identified for morphogenetic processes include limb/skeletal development, neurogenesis, organogenesis of branching organs (kidney, lung), cardiac and vascular development, which agrees with the dynamic spatio-temporal HEY expression patterns in embryos and implicates HEY genes in a broad spectrum of developmental decisions. For HEY1, apoptosis-related genes are over-represented especially among upregulated genes. However, most of the strongly enriched GO terms are preferentially found for down-regulated genes, indicating that they form a more focused group. These also tend to exhibit higher ratios of expression changes, which is in agreement with the primarily repressive nature expected for HEY bHLH factors.
The mode of HEY action has been questioned by publications implicating indirect mechanisms of transcriptional control despite the presence of a classical bHLH domain [summarized in 2]. Especially the lack of E-box target sequences in some of the few known HEY-repressed genes has cast doubt on direct DNA binding as the mode of action. We therefore tested four strong target genes (HEY1, KLF10, BMP2 and FOXC1) in HEK293 cells for direct HEY binding by chromatin immunoprecipitation (ChIP). 293tet-FTH-Hey1 cells were induced at low level to avoid oversaturation and HEY-bound DNA was captured using a Flag-antibody. In each case sequences from the proximal promoter region (within 2 kb of the transcriptional start site (TSS)) were efficiently enriched (10- to 60-fold) in Dox-induced cells (see also Figure 2D). Controls with non-induced cells or immunoprecipitations using unspecific IgG antibodies were both negative, demonstrating high specificity. Other conserved sequences further upstream (−1.4 to −6.5 kb from TSS) or intronic regions were not enriched. Experiments with HEY2 and HEYL (not shown) generated essentially the same results, indicating that HEY proteins bind to the proximal promoter regions of target genes in a similar, if not identical fashion.
Further support for direct DNA binding came from experiments with a subtle HEY1 mutant, where conservative point mutations were introduced at three sites in the basic domain, which alter presumptive DNA contacting amino acid residues (R50K, R54K, R62K; HEY1-RK3) (Figure 2). The mutant protein was expressed at similar levels upon Dox induction, it exhibits nuclear localization and it efficiently dimerizes with wild-type HEY1 (not shown). Expression analysis of the target genes HEY1, KLF10, BMP2 and FOXC1 revealed that only wild-type HEY1, but not HEY1-RK3 is capable of repression (Figure 2C). In ChIP analysis HEY1-RK3 did not bind to the corresponding target promoters (Figure 2D). Thus, the basic domain and its presumptive DNA contacting side chains are essential for the transcriptional activity of HEY1.
Since HEY proteins can directly bind to promoters of target genes we sought to identify the complete repertoire of potential HEY regulated genes through next-generation sequencing of ChIP-enriched DNA fragments (ChIPseq). Non-induced cells were used as a reference. A total of 13–14 million reads were generated for HEY1 and HEY2 and around 90% of these could be mapped back to the human genome (Tables S2, S3, S4). In both cases approximately 10,000 high confidence binding sites could be identified (criteria being a p-value of <10−5 and a peak height of ≥10). To validate candidate genes of HEY1 and HEY2, we tested peaks from 23 genes with different height (11 to 380) individually by quantitative PCR (Figure S2). Each binding site could be validated and the same DNA regions were also found to be targets of HEY2 (not shown).
HEY1 and HEY2 exhibit a remarkable similarity of binding profiles and in most cases peaks of ChIP-enrichment are superimposable (Figure 3A). When binding sites are ranked, 59% of the top 1000 sites are shared between HEY1 and HEY2. A further 37% of these sites are still among the top 5000 binding sites of the other factor, respectively (Figure 3B). Thus, only a small minority of binding sites (≈4%) may be divergent between HEY1 and HEY2 and upon manual inspection most of these are small peaks or the divergence is only of technical nature. The strong similarity of binding is also evident from the heat map generated for all HEY peaks, where very similar distributions of peaks are evident and potential differences seem to be limited to low-scoring sites (Figure 3C).
Binding sites are preferentially located in the proximal promoter region of genes or within exon 1: 55–62% of all peaks are within 500 bp of transcriptional start sites (TSS) and 66–76% fall within +/−2 kb (Figure 4A). When the strictly intragenic peaks were counted more than one third each is located in exon 1 or in intron 1, respectively (Figure 4B). This suggests that HEY proteins likely act directly on promoter associated protein complexes and not through long range enhancer or silencer type mechanisms. The vast majority of binding sites (>90%) are located within CpG islands. This is especially true for peaks within +/−2 kb of the TSS (98%) and to a lesser extent for more distal peaks (75%).
HEY binding sites are located preferentially at active or poised promoters exhibiting H3K4me3 histone marks. In HEK293 cells approximately 20.000 promoters are characterized by the presence of Pol II [15] and the histone mark H3K4me3 [17]. Around one third of these sites is also bound by HEY1 or HEY2, representing around 70% of all Hey peaks (Figure 4C). In contrast, there is no evidence at all for HEY binding at silent promoters that lack Pol II/H3K4me3 marks. HEY bound active promoters have somewhat reduced average H3K4me3 values, which may correspond to the repressive capacity of HEY proteins. Gene Ontology analysis of the top 1000 peaks revealed that the promoters bound by HEY proteins are strongly biased towards transcriptional control and embryonic development genes (Figure 4D). This corresponds well to the data obtained from the microarray analyses described above.
The striking similarity of HEY1 and HEY2 binding patterns posed the question whether HEYL has the same preferences. This is clearly the case for genes used for ChIPseq validation, listed in Figure S2 (data not shown). Preliminary analysis of ChIPseq data from induced 293tet-FS-HEYL cells revealed that the vast majority of HEY1/2 bound sequences are again bound by HEYL (95% of Hey1/2 peaks, Table S5). We also identified a large number of additional binding sites that tend to be barely present and/or not significant in the analysis of HEY1 and HEY2 (Table S5). HeyL was expressed at somewhat higher levels compared to Hey1/2, which may contribute to the detection of additional, previously not significant peaks. On the other hand, we have little evidence to major changes in Hey binding when cells were induced at lower or higher levels or even transiently transfected. Therefore the basis for the increase in binding sites for HeyL will have to be clarified in future studies before final conclusions can be drawn. Nevertheless, all three HEY proteins appear to bind to the same core of genomic sites with very similar preferences.
To identify potential DNA binding motifs for HEY1 or HEY2 we searched the top 300 target sites (+/−100 bp of peak location) using bioinformatic tools. Sequence motifs that are overrepresented tend to be highly GC-rich since the average GC content in HEY peak regions is around 85%. To reduce the influence of this bias we carefully selected control regions from a set of promoters that are not bound by HEY factors, but display very similar GC profiles. Using the motifRG package we identified two motifs that resemble E-box sequences (Figure 5A). Through binding site selection we had previously identified a class B E-box motif (CACGTG/CACGCG) as a preferred HEY binding site [18], which turned out to be one of the two sequences in our list. One other sequence (GCGCGC) reached a similar score, but its relevance remains unclear.
Electrophoretic mobility shift assays with purified HEY1 protein expressed in E.coli showed strong E-box binding (CACGTG) and efficient competition by the unlabeled oligonucleotide (Figure 5B). The related CACGCG and CGCGCG sequences were much poorer competitors and their own binding to HEY1 could easily be competed by an excess of the prototypic class B site. Nevertheless, only a fraction of Hey peaks contain the CACGYG E-box sequence, suggesting that in vivo binding may employ an even more relaxed consensus or depend on additional interacting proteins.
Promoter analysis with luciferase reporter assays validated HEY-dependent repression in vitro. In transient co-transfections several promoters like HEY1, JAG1, BMPR1A and NGN3 were efficiently repressed by cotransfection of HEY1 (Figure 6). Transfection of an activator construct encoding a fusion of the HEY1 bHLH-Orange sequences with the VP16 activation domain (VP16-HEY1) in turn induced luciferase expression from the same reporter. The HEY1-RK3 protein with its impaired DNA binding capacity was incapable of efficient repression or activation. Each promoter contains at least one sequence motif that could serve as a HEY binding sequence. In the case of JAG1 a targeted mutation of the putative E-box motif (gggCACGCGtca to gggCAtca) fully abrogated responsiveness to HEY1 or VP16-HEY1. This again demonstrates that HEY proteins directly bind DNA through E-box motifs and mediate repression of their target genes.
Comparison of target lists for gene regulation and DNA binding further supports the concept of HEY proteins as direct repressors. Especially the genes with stronger repression on mRNA level frequently had ChIP peaks close by and peak height was much higher (median 27 and 22) (Figure 7). Importantly, genes that were induced upon HEY expression did not have significant associated ChIP peaks (median peak height 0). This underscores the notion that repression of transcription appears to be a direct effect of HEY proteins on the corresponding promoters, while gene induction rather tends to be a secondary and indirect phenomenon.
Hey genes are important for development of several organ systems including the heart as reported in numerous knockout studies. We therefore aimed to validate our HEY target genes in the mouse heart. To confirm the presence of HEY binding sites at corresponding genomic locations we repeated our ChIP experiments in HL-1 cells, a murine cardiomyocyte cell line, which maintains cardiac morphology and biochemical and electrophysiological properties in cell culture [19]. We were able to confirm 16 out of 18 HEY binding sites (Figure 8A), indicating that the majority of HEY binding sites detected in HEK293 cells are also present in murine cardiomyocytes.
Hey2 and Hey1/L knockout mice exhibit membranous VSDs and valve defects and overlapping expression in the critical endocardial cells of the AV canal [3], [20]. Hey2 knockout hearts in addition show evidence of cardiomyopathy in the ventricles, which corresponds well to the fact that ventricular cardiomyocytes express only Hey2, but not Hey1 or HeyL. We therefore tested dissected ventricles of Hey2−/− embryos at E14.5 for deregulation of Hey target genes. A series of genes tested exhibited a clear and highly significant up-regulation in knockout embryos by quantitative real-time RT-PCR (Figure 8B). In contrast, Hey1 and HeyL are not expressed in the ventricles and in the knockout situation there is only limited deregulation of the same set of genes where only induction of Sema6d reaches statistical significance. To extend these findings we also tested ventricles from animals with a global Hey1 over-expression [9]. In this case, most of the genes up-regulated in Hey2−/− mice were down-regulated (Figure 8B) with the lower amplitude likely being due to endogenous Hey2 already being present. This clearly documents that Hey repression of target genes is functional in the mouse in vivo with an induction of these genes in the knockout situation.
The strong phenotypes of Hey knockout mice raised the question of potential target genes that may mediate the effects observed in various cell types. To gain such insight we have characterized the genome-wide regulatory potential of HEY proteins by performing microarray gene expression analysis combined with ChIPseq to identify all potential HEY binding sites.
HEY genes have been described as repressors of a small number of individually tested target genes that had been identified serendipitously by various means [summarized in 2]. To search for additional HEY regulated genes we chose HEK293 cells as these are easy to manipulate and they express endogenous HEY genes suggesting that they can react to altered HEY protein levels in a physiologically relevant manner. We obtained very similar patterns of expression changes for HEY1 and HEY2, both on microarrays as well as in confirmatory real-time RT-PCR. Even the more divergent family member HEYL led to concordant regulation of the target genes tested. Surprisingly, the level of regulation was rather limited in all cases. HEY1 itself is the strongest down-regulated gene (3–6-fold), indicative of an important negative feedback loop. For HEY2 and HEYL the repression was also seen, but less pronounced. This negative feedback loop may be similar to the ones described for Hes1 and Hes7 that are important in somitogenesis and neural stem cell biology [21]. The generally modest expression changes suggest that HEY genes rather modulate existing gene transcription instead of completely switching expression states. On the other hand, preexisting HEY1 mRNA and protein in HEK293 cells may already induce a level of repression that can be further enhanced to a limited extent only and this is supported by our experiments with Hey1 shRNA, where an induction of several target genes could be seen. The study by Xin et al. [13] likewise reported a small number of HEY2 regulated genes, where only three structural genes showed regulation in the range of 5–9-fold, which is in line with our data.
Gene regulation on mRNA level could be due to direct or indirect effects of HEY proteins on target promoters. This distinction became more relevant as HEY proteins led to induction and repression of comparable numbers of transcripts. ChIP analyses are an excellent tool to generate additional evidence for a direct mode of action. In these experiments we relied on a rather limited overexpression of HEY genes in order to still mimic a physiological situation. Nevertheless, we identified a very large number of around 10,000 target sites in HEK293 cells with almost identical profiles for HEY1 and HEY2. Differences are mostly restricted to less enriched target sites. This translates to a Pearson's correlation of r = 0.75 between HEY1 and HEY2, which is close to the value of r = 0.83 obtained for biological replicas in other studies [22], indicating that HEY1 and HEY2 regulate the same targets. For HEYL an even larger number of peaks was identified. While the reason for the increased number of binding sites is still unclear, the vast majority of HEY1 and HEY2 peaks were again seen in our HEYL dataset, supporting the idea of strongly overlapping functions.
There is a striking discrepancy, however, between the large number of ChIP peaks and the much smaller number of genes regulated by HEY proteins. The vast majority of binding sites observed may thus not contribute to gene regulation, or else endogenous HEY proteins may have already exhausted the regulatory potential at some of these sites. On the other hand, an overabundance of bound DNA sequences has been observed for other transcription factors before, like the bHLH factors MYC or MYOD that yielded comparable results [23], [24]. Given the probably limited protein concentration it even appears questionable if all sites will be occupied simultaneously in any given cell and rather points to a high turnover rate. For c-MYC, another E-box binding protein, a two-step model of initial binding to open chromatin followed by more relevant sequence specific binding has been put forward [25]. Functionally active binding sites may also emerge only through additional modifications or concomitant binding of additional factors to form fully functional complexes. A possible scenario to explain HEY functions might therefore include a general preference of HEY factors for genes with an open chromatin configuration, where the actual transcriptional change then depends on circumstances like cell type and differentiation status. It remains to be established if and how HEY functions can be described by such models.
The mode of regulation by HEY proteins appears to be rather uniform. The vast majority of binding sites were found in close proximity to transcriptional start sites. This rather implicates HEY proteins in direct interactions with the basal transcriptional machinery or local chromatin at promoters as opposed to long range enhancer type mechanisms. The majority of HEY-repressed genes appears to be direct targets since they contain strong HEY binding sites within the promoter or 5′ UTR regions. On the other hand, genes activated by HEY proteins are likely regulated in an indirect manner: more than half of them do not contain relevant peaks at all and peak height was generally rather small, suggestive of HEY expression leading to a reduction in other critical transcriptional activators for those genes.
Direct repression of target promoters could also be verified in vitro by luciferase reporter assays. As reported in earlier studies by us and others, HEY1/2/L can repress target promoters up to ten-fold [14], [26]. Our current experiments provide important additional evidence for a HEY function as direct transcriptional repressor: Firstly, the bHLH-Orange domain can be turned into an activator of transcription when fused to the strong VP16 activation domain. Furthermore, changing only three arginine residues that presumably contact DNA into lysine completely abolished DNA binding and transcriptional response of this mutant. This clearly establishes HEY proteins as direct DNA binding transcriptional repressors, while gene activation by HEY proteins appears to be indirect as the promoters are largely devoid of HEY target sites.
A putative DNA binding motif for HEY proteins of tggCACGYGcca has previously been defined by in vitro oligonucleotide selection [18]. However, in most studies the core consensus E-box site CACGYG was either not present in the small number of putative target promoters analyzed previously, or deletion of related E-box sites did not alter expression of luciferase reporter constructs [summarized in 2], leading to proposals of indirect HEY functions. Here we could show that deletion of an E-box site in the JAG1 promoter abolishes HEY regulation in luciferase assays. Related findings have recently been published for the IDE promoter [27]. This shows that at least for some HEY target genes the E-box motif is required for Hey regulation. De novo motif discovery in our ChIPseq data set also led to the identification of an E-box motif of CACGYG as one of the two top candidates. Finally, a search of all known DNA binding motifs likewise recovered myc-type E-box sequences as being highly enriched (not shown). While these data clearly demonstrated that E-box sequences can be bound by HEY proteins in vivo, this does not fully explain the genomic binding patterns observed since many of the bound regions do not contain such motifs. HEY proteins may either use less stringent criteria for DNA binding in vivo or they might also bind in a sequential manner that initially does not fully rely on sequence specificity as suggested by Perna et al. [25]. Another possibility would be the need for additional cooperating factors that bind in the vicinity or form ternary complexes to ultimately affect gene expression, but this will depend on the characterization of novel HEY binding partners. The observed co-occurrence of binding sites for factors like SP1, E2F, AP2, NRF and EGR is expected at promoter-proximal regions, but may also hint at potential interactions of HEY proteins with some of these factors.
The rather limited extent of gene repression by HEY proteins is also reflected in the chromatin signature of the corresponding promoters. There is a striking overlap of HEY bound sequences with the presence of polymerase II (Pol II) and the active chromatin mark H3K4me3, which are preferentially found at active and poised promoters. This again argues in favor of a modulatory role of HEY proteins with just limited alterations in gene expression. In human ES cells Pol II and H3K4me3 marks have been identified at silent genes as well, however, and it has been suggested that the critical step lies in transcription elongation. Interestingly many developmental regulators fall within this group of genes [28], [29], to which a significant fraction of HEY target genes belongs as well. The location of a large number of HEY binding sites just downstream of the transcriptional start site would ideally position HEY to influence the pausing vs. elongation switch of Pol II.
Previous studies have suggested redundancies between HEY1, HEY2 and HEYL that manifest in distinct phenotypes in single and combined KO mice due to partly overlapping expression profiles [3], [5], [6]. The striking overlap in gene regulation and the highly related patterns of ChIPseq peaks indicates that all three HEY proteins indeed elicit very similar responses in a given cell type. This is consistent with the idea that the expression patterns of HEY factors largely define the outcome of knockout studies, whereby no individual, intrinsic functional properties, but overall and cumulative Hey expression levels would be critical. On the other hand, the substantial divergence in the poorly conserved C-terminal half of the proteins is suggestive of a significant potential for paralog-specific functions that may yet have to be uncovered. The identification of either fully shared or paralog-specific protein interaction partners of HEY factors may help to shed light on this important issue.
HEK293 cells were cultured in DMEM medium (PAN Biotech, Aidenbach, Germany) containing 10% FCS, 50 U Penicillin and 50 µg/ml Streptomycin. 293tet cells were generated by transfection with PvuII linearized pWHE134 plasmid [16] using polyethylenimine (3 µg DNA, 6 µl PEI per 6-well plate for 8 h) followed by selection with 0.5 mg/ml G418. HEY expressing cells were produced by lentiviral transduction of 293tet cells with p199-FTH-hHey1-iEP, p199-Flag-mHey2-iEP constructs based on p199 plasmids [30] (for maps see Figure S1). For regulated HEYL expression pTol2-FS-mHeyL-iEins-WHE carrying insulator sequences (HS4ins) and the complete tet-regulatory module from pWHE459 [31] was introduced into HEK293 cells by Tol2-mediated transposition with pKate-N/Tol2 [32] followed by puromycin selection (1 µg/ml). The HEY1 knockdown was generated by lentiviral transduction of HEK293 cell with shRNA vectors (Open Biosystems clone ID V3LHS_404238). In all cases individual colonies were picked and validated separately. The HEY1-RK3 mutant was generated by PCR-mediated mutagenesis using primers spanning the altered sites. All constructs were verified by sequencing.
HL-1 cells were cultured in Claycomb medium (Sigma-Aldrich, Munich, Germany) containing 10% FCS, 100 µM norepinephrine, 4 mM L-glutamine, 50 U Penicillin and 50 µg/ml Streptomycin. For ChIP 5*106 HL-1 cells were transiently transfected with 8 µg plasmid DNA using 100 µl Ingenio Buffer (Mirrus, Madison, USA) and the Amaxa Nucleofector II electroporator (program T-20, Lonza, Basel, Switzerland). After 48 h of culture, cells were used for ChIP.
Hey1, Hey2 and HeyL knockout lines have been described before [3], [5]. The Act-Hey1 transgenic line expressing Hey1 under the control of the ß-actin promoter was obtained from M. Susa (NIBR, Basel) [9].
Total RNA was extracted either from cells or tissue samples using TriFast (peqGOLD, Peqlab, Germany, Erlangen) according to the manufacturer's protocol and quantified by OD260 nm measurements using a spectrophotometer (NanoDrop ND 1000, Peqlab).
Total RNA of control and dox-induced cells (1–2 µg/ml doxycycline for 48–72 h) was used for microarray analysis on Human Genome U133 Plus 2.0 Gene Arrays (Affymetrix, Santa Clara, CA). Labeling and washing were performed according to the standard Affymetrix protocol. The arrays were scanned using a GeneChip Scanner 3000 (Affymetrix). Data analysis and quality control was done using different R packages from the Bioconductor project (www.bioconductor.org). Probe sets were summarized using the RMA algorithm and resulting signal intensities were normalized by variance stabilization normalization [33].
2 µg RNA were reverse transcribed using the Revert Aid First-Strand cDNA synthesis Kit (Fermentas, Lithuania, Vilnius) with oligo(dT) primers. qRT-PCR was performed with an iCycler iQ5™ Real-Time PCR Detection System (BioRad, USA, Hercules). Primer sequences are listed in Table S6. Reactions contained 1/50 of the cDNA reaction and PCR was performed with annealing at 60°C and SybrGreen quantification. PCR products were confirmed by melting curve analysis and agarose gel electrophoresis. The housekeeping gene HPRT was used to normalize expression levels. All measurements were performed at least twice and mean values were calculated.
5*106 HEK293 cells were induced with 50 ng/ml doxycycline for 48 h to obtain low level overexpression of HEY proteins. The same amount of cells was kept uninduced as control. HL-1 cells were transiently transfected with pCS2p-Flag-Hey1, pCS2p-Flag-Hey2 [14] and pll3.7, which was used as control. Cells were harvested as described earlier [34]. Briefly, the cells were fixed with 1% paraformaldehyde for 10 min at room temperature. Fixation was stopped by adding glycine to 0.2 M and cells were washed three times with ice-cold PBS and harvested. All subsequent steps were done at 4°C. The cells were lysed in cell lysis buffer (50 mM Hepes-KOH pH 7.5, 150 mM NaCl, 1 mM EDTA, 1% TritonX-100, 0.1% Deoxycholate, 0.1% SDS) and spun down. The resulting pellet of nuclei was lysed in nuclei lysis buffer (cell lysis buffer containing 1% SDS) and sonicated using a Digital Sonifier W-250 D (Branson Ultrasonics, USA, Danbury). Debris was removed by centrifugation. For immunoprecipitation, Chromatin was diluted five-fold with ChIP buffer (0.01% SDS, 1.1% TritonX-100, 1.1 mM EDTA, 20 mM Tris pH 8.0, 167 mM NaCl) and 550 µl of diluted chromatin was mixed with 45 µl 1∶1 protein G agarose slurry (in cell lysis buffer) and 4 µg antibody (αFlag-M2 or rabbit IgG, Sigma-Aldrich) and incubated overnight. Then, the agarose beads were washed three times with cell lysis buffer, once with washing buffer (50 mM Hepes-KOH pH 7.5, 350 mM NaCl, 1 mM EDTA, 1% TritonX-100, 0.1% Deoxycholat, 0.1% SDS) and once with LiCl washing buffer (10 mM Tris-HCl pH 8.0, 250 mM LiCl, 1 mM EDTA, 0.5% Nonidet P-40, 0.5% SDS). Elution was performed with elution buffer (50 mM Tris pH 8.0, 10 mM EDTA, 1% SDS) at 68°C for 30 minutes. The eluted chromatin was incubated with 0.8 mg/ml Proteinase K and PFA fixation was reversed by incubation at 68°C overnight. The DNA was then purified using phenol-chloroform extraction, precipitated and quantified using PicoGreen (Invitrogen, USA, San Diego).
For ChIPseq the same protocol as for ChIP was used in principle, but 2.5*108 cells were employed and after lysis of nuclei, chromatin was spun down at 20.000 rpm (SW 41 TI rotor, Beckman, USA), washed twice with cell lysis buffer and sonicated in 1 ml cell lysis buffer per 100 µl chromatin pellet. For ChIP 10 µg antibody was used. 7–12 ng of ChIP DNA was subjected to sample preparation using the NEBNext ChIPseq sample preparation kit (New England Biolabs, Ipswich, USA) according to the manufacturer's instruction. Briefly, DNA was end-polished with T4 DNA polymerase and kinase. After column purification, Illumina adaptors were ligated to the ChIP DNA fragments. For HEY1 and HEY2 fragments were subjected to 15 cycles of PCR amplification and DNA with a length of 200–350 bp was excised from an agarose gel using Qiagen gel extraction kit. For HEYL 175–225 bp fragments were first excised and then amplified by 18 cycles of PCR. The DNA fragments were sequenced on an Illumina GAIIx platform (Illumina, USA, San Diego). 36 bp sequences were generated and mapped to the hg19 genome by bowtie 0.12.7 [35] with standard parameters. These raw sequencing data were further analyzed using the peak finding algorithm MACS 1.4.1 [36] using sequences from uninduced cells as control to identify the putative binding sites. All peaks with a minimum p-value of 10−5 and a minimum height of 10 were included. The uniquely mapping locations for each factor were used to generate the genome-wide intensity profiles, which were visualized using the USCS genome browser. PeakAnalyzer [37] was used to annotate peaks and to calculate overlaps between different bed files. Heat maps were generated using seqMiner 1.2.1 [38] with K-means raw clustering.
GO terms analysis was performed with DAVID 6.7 [39] using the functional annotation clustering method and allowing only biological processes. Clusters were named based on interpretation of enriched GO annotations.
The R-package motifRG (Bioconductor package motifRG, Zizhen Yao, manuscript in preparation) was used to identify binding motifs, using sequences +/−100 bp around the summit of the top 300 highest ranking peaks. Unrelated sequences with a similar distance towards transcription start sites of genes lacking ChIPseq peaks and with similar GC distribution were selected and used as control/background.
The JAG1 luciferase construct containing the potential Hey binding sequence tgaCGCGTGccc was mutated by cutting with MluI (ACGCGT, Fermentas), followed by Mung Bean Nuclease (Fermentas) treatment and religation using T4 DNA ligase (Fermentas). This results in a four base pair deletion.
For luciferase assays approximately 104 HEK293 cells were transiently transfected with 250 ng of the luciferase promoter construct and 50 ng of the regulatory HEY construct in a 24-well format. Cells were harvested after 48 h and lysed in 150 µl lysis buffer (25 mM Glycyl-Glycine pH 7.8, 15 mM MgSO4, 15 mM KPi, 4 mM EGTA, 1 mM DTT, 1% Trition-X100). After incubating for 10 min at room temperature cells were pelleted and 50 µl of the supernatant were measured in a GLOMAX 96 microplate luminometer (Promega, USA, Madison) using 150 µl assay buffer (lysis buffer with, 1 mM ATP, 0.1 µg/µl D-Luciferin). All measurements were done in triplicates.
EMSA was performed using binding buffer (20 mM Tris pH 7.6, 100 mM KCl, 0.5 mM EDTA, 0.1% Nonidet P-40, 1 mM MgCl2, 1 mM DTT, 10% glycerol) and 1 ng recombinant MBP-HEY1 protein, 1 µg poly-dAdT, 5 ng biotin-labeled probe and either 0, 25, 75 or 250 ng competing unlabeled probe in a total volume of 10 µl. After incubating on ice for 30 min, samples were loaded on a 6% polyacrylamide gel and later blotted onto Amersham Hybond N+ membranes (Amersham, UK). Detection was done using the PIERCE LightShift Chemiluminescent EMSA kit according to the manufacturer's recommendations (PIERCE, USA, Rockford).
HEK293T cells were transfected with either HEY1 or HEY1-RK3 expression plasmids 24 hours prior to fixation with 4% PFA. After blocking with 5% goat serum/0.3% Triton X-100/PBS, the Flag antibody (rabbit; Cell Signaling, Danvers, Massachusetts) was added (1∶800; o/n, 4°C). After washing in PBS at RT the secondary antibody Alexa488-rabbit (Bio-Rad) was used (1∶2000; 1 h, room temperature). Nuclei were stained using Hoechst33342 (1∶10000; Roth, Karlsruhe, Germany) and subsequently cover slides were mounted in Mowiol. Pictures were taken using a Leica AF6000 fluorescence microscope.
The ENCODE ChIPseq data for H3k4me3 was downloaded from “ftp://encodeftp.cse.ucsc.edu/pipeline/hg19/wgEncodeUwHistone/” (Producer: University of Washington) [17] and the ChIPseq data for PolII was downloaded from “http://www.ncbi.nlm.nih.gov/gds?term=GSE11892” [15].
|
10.1371/journal.ppat.1005058 | Utilizing Chemical Genomics to Identify Cytochrome b as a Novel Drug Target for Chagas Disease | Unbiased phenotypic screens enable identification of small molecules that inhibit pathogen growth by unanticipated mechanisms. These small molecules can be used as starting points for drug discovery programs that target such mechanisms. A major challenge of the approach is the identification of the cellular targets. Here we report GNF7686, a small molecule inhibitor of Trypanosoma cruzi, the causative agent of Chagas disease, and identification of cytochrome b as its target. Following discovery of GNF7686 in a parasite growth inhibition high throughput screen, we were able to evolve a GNF7686-resistant culture of T. cruzi epimastigotes. Clones from this culture bore a mutation coding for a substitution of leucine by phenylalanine at amino acid position 197 in cytochrome b. Cytochrome b is a component of complex III (cytochrome bc1) in the mitochondrial electron transport chain and catalyzes the transfer of electrons from ubiquinol to cytochrome c by a mechanism that utilizes two distinct catalytic sites, QN and QP. The L197F mutation is located in the QN site and confers resistance to GNF7686 in both parasite cell growth and biochemical cytochrome b assays. Additionally, the mutant cytochrome b confers resistance to antimycin A, another QN site inhibitor, but not to strobilurin or myxothiazol, which target the QP site. GNF7686 represents a promising starting point for Chagas disease drug discovery as it potently inhibits growth of intracellular T. cruzi amastigotes with a half maximal effective concentration (EC50) of 0.15 µM, and is highly specific for T. cruzi cytochrome b. No effect on the mammalian respiratory chain or mammalian cell proliferation was observed with up to 25 µM of GNF7686. Our approach, which combines T. cruzi chemical genetics with biochemical target validation, can be broadly applied to the discovery of additional novel drug targets and drug leads for Chagas disease.
| Chagas Disease, or American trypanosomiasis, is caused by the kinetoplastid protozoan Trypanosoma cruzi and is primarily transmitted to a mammalian host via a triatomine insect vector (the “kissing bug”) infected with T. cruzi parasites. Although discovered in 1909 by the physician Dr. Carlos Chagas, the disease gained recognition by the public health community only following a major outbreak in Brazil during the 1960s. Approximately eight million people (primarily in Central and South America) are infected with T. cruzi and cases are becoming more widespread due to migration out of the endemic regions. Current treatment options have severe problems with toxicity, limited efficacy, and long administration. Hence, discovery of new drugs for treatment of Chagas disease has become of prime interest to the biomedical research community. In this study, we report identification of a potent inhibitor of T. cruzi growth and use a chemical genetics-based approach to elucidate the associated mechanism of action. We found that this compound, GNF7686, targets cytochrome b, a component of the mitochondrial electron transport chain crucial for ATP generation. Our study provides new insights into the use of phenotypic screening to identify novel targets for kinetoplastid drug discovery.
| Chagas disease, or American trypanosomiasis, is a neglected disease caused by the kinetoplastid protozoan Trypanosoma cruzi (T. cruzi). Endemic to Latin America, Chagas disease is increasingly globalized due to population migration from endemic regions into developed countries, and the U.S. in particular. About eight million people are estimated to harbor the infection with 40,000 new cases added annually [1, 2].
In the 100+ years that have passed since the first description of Chagas disease by Carlos Chagas, only two drugs have been developed to treat this infection: nifurtimox and benznidazole [2, 3]. While both these drugs can clear T. cruzi from the infected mammalian hosts, they are both inadequate to address the medical need of millions of patients chronically infected today. The drug shortcomings include toxicity, prolonged treatment time, and high rate of treatment failure [2, 3].
T. cruzi is transmitted to mammalian hosts primarily via hematophagous triatomine bugs [4]. While in the vector insect, T. cruzi cells propagate as flagellated epimastigotes that transform into non-dividing infective trypomastigotes. As the T. cruzi-infected bug takes a blood meal from a host, it deposits motile T. cruzi trypomastigotes near the wound site. Following invasion of host cells in the bite wound or at mucous membranes, intracellular trypomastigotes undergo a morphological transformation into amastigotes and start to replicate [1, 4, 5]. After completing several rounds of intracellular division, the amastigotes transform into trypomastigotes that then leave the infected cell and initiate a new cycle of infection.
The acute phase of Chagas disease is often asymptomatic, characterized by readily detectable parasitemia, and usually resolves within a few weeks through control of parasite proliferation by the adaptive immune system [4]. In the chronic stage, infected individuals rarely display symptoms or evidence of the disease for decades. However, ~30% of these patients eventually go on to develop a severe cardiomyopathy and ~10% of patients progress with gastrointestinal complications [1, 4].
New drug discovery for Chagas disease is hampered by very limited number of validated T. cruzi drug targets. Drug discovery efforts have focused on the trypanosome ergosterol biosynthesis pathway and cruzain, a T. cruzi cysteine protease [6]. During the last decade, much attention has been paid to inhibitors of sterol 14 α-demethylase (CYP51), an essential enzyme in the ergosterol biosynthesis pathway [6, 7]. To a large degree, this interest has been fueled by the availability of drugs targeting fungal sterol 14 α-demethylase, such as posaconazole or ravuconazole [8, 9]. Both these drugs are exceptionally potent on T. cruzi in vitro and have been shown to effect radical parasitological cure in mouse models of Chagas disease [10]. Also, treatment with posaconazole cured T. cruzi infection in an immunosuppressed patient following benznidazole treatment failure [11]. However, a 60-day treatment with posaconazole, while transiently clearing the parasite from Chagas patients, did not prevent recrudescence of infection in a majority of patients (81%) as determined by PCR. A similar trial testing the efficacy of ravuconazole prodrug E1224 has recently reported failure to cure infection in a majority (~70%) of the treated Chagas patients [12, 13]. These failures in clinical phase 2 trials have been attributed to insufficient drug exposure or dosing duration [14].
In addition to inhibitors of the parasite ergosterol biosynthesis, several inhibitors of cruzain were reported as promising candidates for treating Chagas disease. Of these, the most advanced is K777, a vinyl sulfone peptidomimetic inhibitor [15, 16]. K777 has been shown to be safe and efficacious in animal models of acute and chronic Chagas disease [17, 18] and is currently undergoing preclinical development evaluation.
Identification of T. cruzi growth inhibitors by phenotypic screening represents a viable alternative to target-based Chagas disease drug discovery. The approach allows efficient discovery of small molecules that perturb new molecular targets. Limitations of this approach stem from ignorance of the molecular mechanism, which precludes the use of structure-assisted drug design and prevents early predictions of toxicity through inhibition of homologous host enzymes. Chemical optimization of hit molecules in the absence of target-based activity measurements can be confounded by complex structure-activity-relationships, as biochemical activity and cellular permeability cannot be distinguished [19, 20].
To overcome these limitations, we have established a chemical genetics approach to the determination of the mechanism of action of small molecule growth inhibitors in T. cruzi. Starting with a novel T. cruzi inhibitor GNF7686, we evolved resistant T. cruzi mutants in vitro, and then identified the resistance-conferring mutation by whole genome sequencing. Finally, we demonstrated inhibition of the putative target in a biochemical assay. An expansion of this approach to other T. cruzi growth inhibitors could lead to identification of many additional drug targets and associated lead inhibitors for Chagas disease, and is already underway in our laboratory. As with the case of the T. cruzi cytochrome b target reported in this study, such an approach could point to drugs and drug targets from other fields, and substantially accelerate the introduction of novel Chagas disease treatments into the clinic.
With the exception of decylubiquinone (MP Biomedicals), all chemicals were purchased from Sigma-Aldrich Corporation.
T. cruzi CL strain was propagated in NIH/3T3 fibroblast cells. NIH/3T3 cells were grown in RPMI-1640 media supplemented with 10% heat-inactivated fetal bovine serum (FBS, Hyclone) and 100 IU penicillin / 100 μg streptomycin (Hyclone) per mL at 37°C / 5% CO2, and passaged every three to four days. To establish infection, 6.25 × 105 of 3T3 cells were seeded in a T-175 flask. After attachment, cells were infected with 20–40 × 106 T. cruzi trypomastigotes. Following cell infection, parasites cycled between the trypomastigote and the intracellular proliferative amastigote forms and medium was changed biweekly.
T. cruzi epimastigotes were maintained in liver infusion tryptose (LIT) medium (9 g / L liver infusion broth, 5 g / L tryptose, 1 g / L NaCl, 8 g / L Na2HPO4, 0.4 g / L KCl, and 1 g / L glucose, pH = 7.2) supplemented with 10% heat-inactivated FBS and 15 μM hemin, and passaged every five days during middle to late logarithmic growth phase (maintained at 26°C / 0% CO2).
For differentiation of epimastigotes into trypomastigotes, saturated cultures of T. cruzi CL epimastigotes were harvested by centrifugation (1,000 × g for 10 min at 10°C), resuspended in artificial triatomine urine medium (TAU; 190 mM NaCl, 17 mM KCl, 2 mM MgCl2, 2 mM CaCl2, 0.035% NaHCO3, 8 mM phosphate buffer, pH 6.9) at a density of 5 x 108 cells / mL, and incubated at 26°C. Two hours later, the parasites were transferred to TAU 3AAG medium (TAU supplemented with 10 mM L-proline, 50 mM sodium L-glutamate, 2 mM sodium L-aspartate and 10 mM D-glucose) in T-25 culture flasks with a layer of culture medium approximately 1 cm in depth. This cell density was previously shown to be the optimal density for epimastigote differentiation [21, 22]. After 72 hour incubation, the mixture of epimastigotes and newly differentiated trypomastigotes was used for infection of mammalian host cells (NIH/3T3 mouse embryonic fibroblast line).
Briefly, NIH/3T3 cells were plated at a density of 0.025 million cells / mL into T-175 flasks and infected with 5 mL of pelleted epimastigote / trypomastigote mixture (collected from 25 mL of transformed culture) resuspended in RPMI-1640 medium supplemented with 10% FBS and 100 IU penicillin / 100 μg streptomycin per mL. After 24 hour incubation, non-internalized extracellular epimastigotes and trypomastigotes were removed, and infected NIH/3T3 cells were cultured for additional seven days. By then, newly formed trypomastigotes released from infected NIH/3T3 host cells were present in the medium and further used as described in the “T. cruzi in vitro efficacy assays” section.
To determine compound efficacy on T. cruzi intracellular amastigotes, NIH/3T3 cells were seeded (1,000 cells / well, 40 μL) into microscopy-grade, clear bottom, 384-well plates (Greiner) in RPMI-1640 medium containing 5% heat-inactivated fetal bovine serum and 100 IU penicillin / 100 μg streptomycin per mL. Plates were incubated overnight at 37°C / 5% CO2. Cells were infected with T. cruzi trypomastigotes at a multiplicity of infection (MOI) of 10 for the wild-type strain, and 20 for the GNF7686-resistant mutant strain. After six hours of infection (at 37°C / 5% CO2), the plates were washed by aspirating the medium and replacing with fresh screening medium to remove remaining extracellular trypomastigotes. The plates with infected cells were incubated overnight (37°C / 5% CO2) and compounds dissolved in DMSO were added to plate wells on the following day (0.2% DMSO final concentration). After 48-hour compound treatment, infected cells were fixed (4% paraformaldehyde in phosphate-buffered saline containing 0.5 mM CaCl2 and 0.5 mM MgCl2), permeabilized (0.1% Triton X-100 in PBS), and then stained using a 1:125,000 dilution (prepared in PBS) of SYBR green (Life Technologies). The plates were then scanned using the Evotec Opera High Content Screening System (Perkin Elmer) and amastigote proliferation was assessed by counting parasites within the 3T3 cells using CellProfiler version 2.1.0 cell image analysis software [23]. In some experiments, an alternative protocol for measurement of compound activity on intracellular T. cruzi was used [24].
To determine compound activity on the T. cruzi epimastigote form, epimastigotes (20 μL; 2.5 × 105 parasites / mL) were added to 384-well assay plates containing 20 μL of LIT media with pre-dispensed compounds (0.2% DMSO final concentration) and incubated for seven days at 26°C / 0% CO2. Parasite viability was assessed at the end of this incubation period using the CellTiter-Glo Luminescent Cell Viability Assay (Promega). Luminescence as a measure of parasite viability was measured on the EnVision plate reader.
To assess compound efficacy on trypomastigotes, parasites (30 μL; 2 × 106 trypomastigotes / mL) were added to 384-well plates containing 10 μL of RPMI-1640 without phenol red (Invitrogen) and supplemented with 10% FBS and 100 IU penicillin / 100 µg streptomycin per mL, and then treated with compounds (0.2% DMSO final concentration). Following a 48-hour incubation period, viability was assessed using the CellTiter-Glo Luminescent Cell Viability Assay (Promega).
Leishmania donovani (L. donovani) axenic amastigotes (strain MHOM/SD/62/1S-CL2D) were maintained at 37°C / 5% CO2 in RPMI-1640 medium containing 4 mM L-glutamine, 20% heat inactivated FBS, 100 IU penicillin / 100 μg streptomycin per mL, 23 μM folic acid, 100 M adenosine, 22 mM D-glucose, and 25 mM 2-(N-morpholino)ethanesulfonic acid (pH 5.5 adjusted at 37°C using 1M HCl).
For compound screening, axenic amastigotes were seeded into 384-well plates containing axenic amastigote medium with pre-dispensed compounds (0.25% final DMSO concentration) at a density of 9,600 cells / well. Plates were incubated for 48 hours at 37°C / 5% CO2. Cell viability was assessed using the CellTiter-Glo Luminescent Cell Viability Assay (Promega).
Trypanosoma brucei brucei (T. b. brucei) bloodstream form (strain Lister 427) was maintained in HMI-9 medium: IMDM (Iscove's Modified Dulbecco's Media), 10% heat-inactivated FBS, 10% Serum Plus medium supplement (SAFC biosciences), 1 mM hypoxanthine, 50 μM bathocuproine disulfonate.Na2, 1.5 mM cysteine, 1 mM pyruvate, 39 μg/mL thymidine, and 0.2 mM 2-mercapthoethanol) at 37°C / 5% CO2.
For compound screening, parasites were seeded into 384-well assay plates containing HMI-9 with pre-dispensed compounds (0.25% final DMSO concentration) at a density of 6,000 cells / well, and plates were then incubated for 48 hours at 37°C / 5% CO2. Cell viability was assessed using the CellTiter-Glo Luminescent Cell Viability Assay (Promega).
For conversion from the bloodstream form to the procyclic form, bloodstream form parasites were transferred from HMI-9 medium to Differentiating Trypanosome Medium (DTM, pH = 7.2), consisting of: 6.8 g / L NaCl, 400 mg / L KCl, 200 mg / L CaCl2, 140 mg / L NaH2PO4.H2O, 200 mg / L MgSO4.7H2O, 7.94 g / L HEPES, 2.2 g / L NaHCO3, 110 mg / L sodium pyruvate, 10 mg / L phenol red, 14 mg / L hypoxanthine, 1 mg / L biotin, 760 mg / L glycerol, 640 mg / L proline, 236 mg / L glutamic acid, 1.34 g / L glutamine, 7.5 mg / L hemin (in 50 mM sodium hydroxide), 1X MEM amino acid solution (Invitrogen),1X MEM non-essential amino acids solution (Invitrogen), 28.2 mg / L bathocuproine disulfonate.Na2, 182 mg/L cysteine, 0.2 mM 2-mercaptoethanol, 15% heat-inactivated FBS, and 5 mM sodium citrate and cis-aconitate [21, 22]. Following medium exchange, parasites were incubated at a lower temperature (27°C / 5% CO2), monitored for change in morphology to procyclic parasites, and sub-cultured for long-term cultivation.
For compound screening, 20 μliters (5,000 parasites) of T. b. brucei procyclic culture were added to 384-well plate wells filled with 20 μliters of DTM medium and pre-dispensed compounds (0.25% final DMSO concentration). Plates were then incubated for 72 hours at 27°C / 5% CO2. Cell viability was assessed using the CellTiter-Glo Luminescent Cell Viability Assay (Promega).
GNF7686 and cytochrome b inhibitors were assayed for activity on two Plasmodium falciparum (P. falciparum) lines: D10attB and yDHODH-D10attB. The D10attB line is reliant on the coenzyme Q-dependent malarial dihydroorotate dehydrogenase (PfDHODH), whereas the yDHODH-D10attB line expresses also the fumarate-utilizing Saccharomyces cerevisiae (S. cerevisiae) DHODH, which circumvents reliance on PfDHODH and renders this line fully resistant to cytochrome b inhibitors [25]. The use of these two lines allows for distinction of selection of potent cytochrome b inhibitors as described in detail in the Results section [26, 27]. Growth and viability of P. falciparum cell lines (in the presence or absence of drug) in infected erythrocytes were assessed using a SYBR Green-based proliferation assay exactly as described previously [28].
NIH/3T3 fibroblast cells were maintained in RPMI-1640 medium supplemented with 10% heat-inactivated FBS and 100 IU penicillin / 100 μg streptomycin per mL at 37°C / 5% CO2. For compound screening, cells were diluted to 4 × 104 cells / mL in assay medium (RPMI-1640, 5% FBS, and 100 IU penicillin / 100 μg streptomycin per mL) and seeded at 50 μL / well into 384-well plates. Following overnight incubation, compounds were added to each well (0.2% DMSO final concentration) and plates were further incubated for 96 hours. Cell viability was assessed using the CellTiter-Glo Luminescent Cell Viability Assay (Promega). Measured luminescence values were normalized to the value obtained for 0.2% DMSO, and plotted against the corresponding compound concentration for half maximal cytotoxic concentration (CC50) value determination.
The minimal inhibitory concentrations (MIC) of the compounds for inhibition of S. cerevisiae drug pump knock-out strain NF7061 (MATa his3Δ 1; leu2Δ 0; met15Δ 0; ura3Δ 0; snq2::KanMX; pdr5::KanMX; pdr1::NAT1; pdr3::KanMX; yap1::NAT1; pdr2::LEU2; yrm1::MET; yor1::LYS2) were determined by a modification of the microdilution technique described elsewhere [29]. Briefly, two-fold dilutions of the compound solution in DMSO were made. They were added to each well of 96-well assay plate containing 200 μL per well of either YPD (1% Difco Yeast Extract, 2% Difco Bacto Peptone and 2% Dextrose) or YPG (3% of glycerol in replacement of 2% Dextrose in YP) medium. Early stationary yeast NF7061 cells grown in either medium were collected and resuspended to 2 × 105 cells / mL. Ten microliters of the yeast cell suspension were inoculated to each well of the plates containing medium and compound to achieve a final inoculum of approximately 104 CFU / mL. The plates were incubated at 30°C for 24 hours (containing YPD) or 48 hours (containing YPG). The MIC end point was defined as the lowest compound concentration exhibiting no visual growth.
T. cruzi epimastigotes were initially treated with GNF7686 at EC20 value (0.01 μM, 0.2% DMSO) and continually passaged at the same concentration until the culture growth rate matched that of epimastigotes growing in the medium containing 0.2% DMSO. Parasites were subsequently passaged in a similar manner in the presence of increasing concentration of GNF7686 until significant resistance was achieved (~5-fold increase in the EC50 value). The time to generate resistance was approximately eleven months. Resistant epimastigotes were cloned by the limiting dilution technique.
Following expansion of GNF7686-resistant T. cruzi clones in LIT media, T. cruzi total DNA was isolated using Qiagen DNeasy Blood and Tissue Kit from 108 parasites per sample. Whole genome sequencing was performed using Ilumina HiSeq1000 next-generation sequencing platform.
Sequencing reads were aligned by Burrows-Wheeler Aligner (BWA, version 5.9.0) to the T. cruzi JR cl. 4 genome (version 1.0). Simple single nucleotide variants (SNVs) were called (using Samtools 1.19) looking for SNVs or small indels with an overall quality > 100 where the control was the drug-sensitive parental CL clone. Approximately 600900 reads called a ‘T’ and 520 reads called a “G” at L197F position resulting in L197F mutation. Heterozygous calls were determined by Samtools, and verified in the Integrated Genomics Viewer. Putative SNVs were manually checked in IGV [30, 31].
To further confirm the presence of L197F mutation, the T. cruzi cytochrome b gene was amplified by PCR (forward primer 5’-AGCTACTGTTCCTGTATTCGGC-3’ and reverse primer 5’-ACAAAAACAAAGTCGCTCACAA-3’) and cloned into the pCR2.1 vector. The insert DNA was sequenced using M13R and M13F (-21) primers (Genewiz).
GNF7686 and cytochrome b inhibitors (0.2% DMS0 final concentration) were added to 384-well assay plates containing 20 μL of assay buffer (250 mM sucrose, 15 mM KCl, 5 mM MgCl2, 1 mM EGTA, 30 mM K2HPO4, pH 7.4) and allowed to dissolve for two hours. Meanwhile, T. cruzi epimastigotes were harvested (800 × g for 5 min at 4°C), washed twice with buffer A (10 mM Tris-HCl, pH = 7.4, 0.23 M mannitol, 0.07 M sucrose, 0.2 mM EDTA, 0.2% bovine serum albumin, 0.5 mM phenylmethanesulfonyl fluoride), and finally resuspended in buffer A at a final concentration of 150 × 106 epimastigotes / mL. Next, 20 μL of T. cruzi epimastigote suspension (3 x 106 parasites) was added to each sample well, and then 15 μL of MitoXpress-Xtra probe (Cayman Chemicals) was added (final assay volume of 60 μL). Probe phosphorescence is quenched in the presence of oxygen and is inversely proportional to the amount of oxygen present in the solution. HS mineral oil (20 μL, Cayman Chemicals) was added to wells to prevent oxygen exchange between the assay buffer and air. Blank wells containing assay buffer without parasites, corresponding compound, and mineral oil were prepared in parallel and values were subtracted from sample values to specifically measure changes in oxygen concentration caused by parasite respiration. All sample and blank wells were prepared in duplicate. The assay plate was transferred to a Gemini XPS fluorescence plate reader (Molecular Devices) and read at 3 minute intervals at excitation and emission wavelengths of 380 nm and 650 nm, respectively, for 30 minutes at 37°C. The slope for the linear portion of time course of fluorescence increase (rate of oxygen consumption) was calculated after subtraction of blank well values. Obtained slope values were normalized to the slope obtained for 0.2% DMSO, and plotted against the corresponding compound concentration for half maximal inhibitory concentration (IC50) value determination.
T. cruzi epimastigotes (57 day old culture, density of 5 × 107 parasites / mL) were harvested by centrifugation (800 × g for 5 minutes at 4°C) and resuspended at 10 mg / mL of protein in buffer A containing 0.1 mg digitonin / mg protein. The parasites were then incubated with the detergent for 10 minutes at 26°C. The pellet fraction was collected by centrifugation (13,000 × g for 5 min at 4°C) and immediately used in the complex III assay.
Complex III activity was monitored using a coupled decylubiquinol / cytochrome c reaction [32, 33]. Decylubiquinone was first reduced to decylubiquinol as described [33]. The freshly reduced decylubiquinol (80 μM final concentration in the reaction) was added to a reaction buffer (25 mM potassium phosphate, 5 mM MgCl2, 2.5 mg / mL BSA, pH = 7.2) containing 1 mM KCN, 0.1 mM yeast cytochrome c, 0.6 mM n-D-B-maltoside, and 12 μg of digitonin-solubilized T. cruzi epimastigotes. The reduction of cytochrome c was monitored at 550 nm using the SpectraMax Plus384 absorbance microplate reader. Blank samples containing all components excluding decylubiquinol were processed in parallel and absorbance values were subtracted from sample absorbance values to specifically measure decylubiquinone-dependent reduction of cytochrome c. All sample and blank wells were prepared in triplicate. For IC50 determination, the slope of the linear portion of the corrected respiration trace was determined, and normalized to the slope obtained for 0.2% DMSO condition.
Sprague Dawley rats were euthanized and skeletal muscle (from hind legs) was removed and stored in CP1 buffer (0.1 M KCl, 0.05 M Tris, 2 mM EGTA, pH 7.4 at 4°C) on ice. Using the ‘Herb Mincer‘, tissue was minced 34 times and then transferred into CP1 buffer (on ice) to wash away fatty and connective tissues from muscle. Following two rounds of wash and decantation with CP1 buffer, the rinsed tissue was transferred to CP2 buffer (CP1 + 0.5% BSA, 5 mM MgCl2, 1 mM ATP, 250 units / 100 mL Protease Type VIII (Sigma P5380), pH 7.4 at 4°C) and incubated on ice for 5 minutes prior to homogenization using the Polytron PT3100. Following homogenization and centrifugation (500 × g for 10 min at 4°C), the supernatant was decanted using cheesecloth and further centrifuged (10,000 × g for 11 min at 4°C). The crude mitochondrial pellet was resuspended in 10 mL of CP1 buffer (carefully avoiding red blood cell pellet) and subjected to an additional centrifugation step (10,000 × g for 10 min at 4°C). The resulting pellet was again separated from the red blood cell pellet and centrifuged (600 x g for 6 min at 4°C). The supernatant containing resuspended mitochondria was subjected to one final round of centrifugation (5,000 × g for 11 min at 4°C) and the pelleted mitochondria were resuspended in CP1 buffer and stored at -80°C. Respiration in isolated rat mitochondria was measured using the same protocol as described for T. cruzi epimastigotes using 150 μg protein / sample.
GNF7686 (Fig 1A) was identified in a high throughput screen designed to find new small molecules with growth inhibition activity on L. donovani axenic amastigotes. A library of 700,000 compounds was assembled with a particular focus on drug-like properties and structural diversity and these compounds were tested for inhibitory activity on L. donovani at 4 μM concentration. The library has been previously profiled in more than 60 other high throughput screens, including biochemical and cell-based assays against human and pathogen targets. The screen history allowed rapid identification and elimination of compounds with a ‘frequent hitter’ property.
The screen yielded 2,306 primary hits (0.3% hit rate) that inhibited growth by > 50%. Data from more than 95% of the assay plates had Z′ > 0.7, using DMSO as the negative control and 5 μM pentamidine as the positive control. Primary hits from the screen were further characterized using a dose−response assay format to determine the EC50 values. In parallel, cytotoxicity of these compounds was determined against a proliferating mouse fibroblast cell line (NIH/3T3). The final set of condirmed hits consisted of compounds that had EC50 < 4 μM against L. donovani, as well as low or no 3T3 cytotoxicity (CC50 > 10 μM or SI > 10; SI = CC50/EC50). The final set of confirmed L. donovani hits consisted of 1003 inhibitors.
The confirmed hits were further assayed for activity on other two medically important kinetoplastid parasites T. cruzi and T. brucei. GNF7686 was selected for further investigation because of potent in vitro activity on all three T. cruzi morphological forms (intracellular amastigote EC50 = 0.15 μM, trypomastigote EC50 = 0.71 μM, epimastigote EC50 = 0.16 μM; Table 1 and Fig 1C). GNF7686 also inhibited the growth of L. donovani axenic amastigotes (EC50 = 0.46 μM) and promastigotes (EC50 = 0.46 μM), but not the growth of T. b. brucei bloodstream form trypomastigotes (EC50 > 25 μM). Curiously, GNF7686 was active on T. b. brucei procyclics (EC50 = 0.59 μM), the parasite form found in the tsetse fly vector that mediates T. b. brucei transmission [34]. GNF7686 did not inhibit growth of 3T3 cell line (CC50 > 20 μM).
To investigate the mechanism of action of GNF7686, we selected a population of drug-resistant T. cruzi epimastigotes through a long-term parasite culture in the presence of this compound. As tolerance for GNF7686 gradually increased over time, we periodically escalated the selection pressure by raising the inhibitor concentration. In the course of eleven months, the EC50 of the T. cruzi culture shifted ~ 4-fold from 0.16 µM to 0.73 μM (Table 1).
Populations of evolving microbes often comprise cells harboring alternative mutations that are derived from different mutation events [35, 36]. To simplify analysis of genomic changes accumulated during the selection by characterizing homogenous culture populations, we isolated three clones from the GNF7686-resistant culture. All three clones exhibited the same extent of GNF7686 resistance (EC50 ~ 1 μM) as the parent culture. Importantly, the sensitivity of clones to benznidazole remained at the same level as observed for the wild-type T. cruzi strain (Table 1), demonstrating that resistance to GNF7686 did not arise through a broadly pleiotropic mechanism. When cultured in medium lacking GNF7686, all three clones grew at a rate similar to the wild-type strain (mutant epimastigote doubling time = 5355 hours vs 60 hours for the wild-type strain), but reached stationary phase at a lower cell density (~60% of the wild-type strain, Fig 1B).
Whole genome sequencing identified the same set of five mutations in all three clones that included L197F in cytochrome b, L283M in the ATPase subunit of HsIVU protease, R75C in TCSYLVIO008926 hypothetical protein, and two mutations in non-coding regions. The observation that the three clones were identical at the genome sequence level suggests that they were siblings derived from one founding cell that expanded in the passaged culture during the selection. During whole genome sequencing, multiple sequence reads (up to 100 in total) from many independent DNA molecules are obtained for each nucleotide position in the genome. We uncovered that the L197F mutation in the cytochrome b gene and one of two mutations mapped to non-coding regions fully replaced the corresponding wild-type alleles, while the other three mutations remained heterozygous during the selection. Interestingly, the two mutations map both to the kinetoplast maxicircle DNA, which is present in 2050 copies per cell [37, 38]. This indicates that these two mutations, presumably appearing on one maxicircle copy at first, replaced the corresponding wild-type alleles during the selection to the point of achieving homoplasmy.
In addition to selection of T. cruzi mutants resistant to GNF7686, we subjected the inhibitor also to chemogenomic profiling in S. cerevisiae. The genome-wide deletion collections available for this eukaryotic model system provide powerful genetic tools for investigation of bioactive molecules [39, 40], and the approach was successfully applied to mechanism of action studies of various growth inhibitors in the past [41, 42]. In the haploinsufficiency profiling assay (HIP), complete collection of heterozygous yeast deletion strains, in which each strain has only one copy of a particular gene, is profiled for hypersensitivity to a compound. Gene deletions associated with increased compound sensitivity indicate pathways directly affected by the compound [43].
We observed that growth of S. cerevisiae is inhibited when the yeast was cultured in media containing glycerol but not glucose as the carbon source (see also below). We therefore conducted a HIP profiling experiment in medium containing glycerol. Testing of GNF7686 at its EC30 concentration in two independent HIP experiments resulted in a reproducible profile (Fig 2A). Identified hypersensitive heterozygous strains included those with deletions in genes involved in mitochondrial metabolism, such as CYT1 (cytochrome c1), HAP4 (a transcription factor involved in regulation of the respiratory chain including CYT1), CBP1 (a regulator of cytochrome b mRNA stability), and QCR6 (a subunit of the cytochrome c reductase complex) [44]. As all these hits pointed at inhibition of mitochondrial respiration by GNF7686, we performed additional HIP experiments with strobilurin, an inhibitor of cytochrome bc1 complex, and venturicidin, an inhibitor of F-type ATPase [45, 46]. In a control experiment, we also collected the HIP profile for benomyl, a microtubule binding inhibitor, which does not interfere with ATP production during oxidative phosphorylation.
While both venturicidin and benomyl yielded HIP profiles distinctly different from that of GNF7686 (Fig 2C and 2D), treatment of gene deletion strain collection with the cytochrome bc1 inhibitor strobilurin identified essentially the same set of sensitive, heterozygous mutants as GNF7686 and included CYT1, HAP4, CBP1 and QCR6 (Fig 2B). It is important to note that the gene coding for cytochrome b, which is the proposed direct target of strobilurin [47], is encoded by the mitochondrial genome in S. cerevisiae and not amenable to standard gene targeting protocols. Thus, the cytochrome b gene deletion strain is not present in the yeast heterozygous deletion pool and could not be directly identified by the HIP method.
The observed HIP compound profiles strongly suggested that GNF7686 directly interferes with function of the S. cerevisiae respiratory chain, possibly at the level of complex III.
Genomic analyses of GNF7686 resistance/sensitivity pointed to involvement of the T. cruzi cytochrome b in resistance to growth inhibition by GNF7686. Cytochrome b is a component of the multisubunit cytochrome bc1 complex, an asymmetric homodimer with two spatially separated catalytic sites QN and QP (Fig 3A). In concert, QN and QP catalyze oxidation of ubiquinol formed by preceding steps of the respiratory chain [48, 49].
Inhibitors of cytochrome b are already of interest as anti-parasitic drugs. Atovaquone, a hydroxy-naphthoquinone inhibitor of the QP site, is used in the treatment of malaria and fungal pneumonia [50]. Another hydroxy-naphthoquinone, buparvaquone, is used to treat cattle theileriosis, and potently inhibits growth of L. donovani [19, 51]. Surprisingly, the potential of cytochrome b inhibitors for treatment of Chagas disease has not been explored, even though cytochrome b inhibitors including antimycin A were shown to affect T. cruzi mitochondrial respiration, bioenergetics, and calcium homeostasis [52–55]. To assess the effect of this class of compounds on T. cruzi growth and survival, we tested prototypical QN and QP site inhibitors on intracellular amastigotes, trypomastigotes, and epimastigotes (inhibitor structures shown in Fig 1A) [46, 49].
The QN site inhibitor antimycin A potently inhibited the growth of epimastigotes and rapidly reduced viability of trypomastigotes. We also observed T. cruzi inhibition by compounds targeting the cytochrome b QP site. Two well-characterized QP site inhibitors, myxothiazol and strobilurin, blocked growth of epimastigotes, and reduced viability of trypomastigotes (Table 1). The effect of antimycin A, myxothiazol, and strobilurin on intracellular amastigotes could not be accurately determined because of the inhibitor toxicity on the host 3T3 cells.
T. b. brucei is a kinetoplastid parasite closely related to T. cruzi at the genomic level [56]. While the parasite bloodstream (mammalian) form of T. b. brucei relies primarily on the glycolytic pathway for ATP production, growth of the procyclic (insect) form requires activity of the conventional respiratory pathway, including cytochrome b [48]. In line with the hypothesis that GNF7686 is a cytochrome b inhibitor, GNF7686 inhibited the growth of the procyclic but not bloodstream T. b. brucei parasites (EC50 = 0.59 μM vs > 25 μM). We also observed similar differential activity on the two T. brucei forms with the other tested cytochrome b inhibitors such as antimycin A (EC50 = 0.03 μM vs > 25 μM; Table 2).
The inhibition of various morphological forms of T. cruzi by prototypical cytochrome b inhibitors is consistent with the hypothesis that the cytochrome b fulfills an essential function in parasite physiology and that GNF7686 inhibits its function. However, with the exception of antimycin A, the anti-parasitic potency of the other tested inhibitors is too weak to be of therapeutic significance.
Through a literature search, we identified a previously published report that described L198F mutation in S. cerevisiae cytochrome b, which is equivalent to the L197F mutation in T. cruzi cytochrome b (Fig 3B). The yeast L198F mutation confers resistance to ilicicolin H, a cytochrome b inhibitor with potent anti-fungal activity [46, 57]. Inspection of high resolution crystal structure of the yeast cytochrome bc1 complex further revealed that the Leu198 side chain is positioned in close proximity (< 5 Å) to ubiquinol bound inside the QN pocket and next to His197, which coordinates the iron atom in the bH heme. In accordance with the structure, L198F mutation also conferred resistance in yeast to other tested QN site inhibitors such as funiculosin and antimycin A [57, 58].
Additional characterization of the GNF7686-resistant T. cruzi mutants revealed that they were selectively resistant to antimycin A, a QN site inhibitor [46]. While antimycin A displayed very potent activity on wild type epimastigotes, a sharp decrease in potency (40-fold) was observed with GNF7686-resistant epimastigotes (EC50 = 1.8 μM, Table 1). Similarly, a steep shift in potency (20-fold) was observed between the wild-type and mutant trypomastigotes (Table 1). In contrast, QP inhibitors myxothiazol and strobilurin showed comparable activity on both wild-type and resistant strains (Table 1). In summary, the L197F mutation in the T. cruzi cytochrome b is likely located within the QN site and can interfere with binding of QN site inhibitors in a similar way as was previously described for the L198F mutation in the S. cerevisiae cytochrome b.
During aerobic respiration, the electron transport chain (ETC) conducts electrons derived from reduced carbon substrates through a series of redox reactions to the terminal electron acceptor, molecular oxygen, which is then reduced to water [33, 48]. Inhibition of electron flow through the ETC at any step, including cytochrome b, results in a block of oxygen consumption.
To evaluate whether GNF7686 disrupts the function of the T. cruzi ETC, oxygen consumption by intact T. cruzi epimastigotes was monitored in the presence of GNF7686 and prototypic cytochrome b inhibitors (Fig 4A and 4B). Antimycin A potently blocked respiration by wild-type epimastigotes (IC50 = 0.04 μM), but was ~7-fold less potent on GNF7686-resistant parasites (IC50 = 0.27 μM). Two QP site inhibitors employed in this report, myxothiazol and strobilurin, both inhibited epimastigote respiration, but, in contrast to antimycin A, the respiratory IC50 values of the QP inhibitors were comparable between wild-type and GNF7686-resistant T. cruzi.
GNF7686 inhibited respiration by wild-type parasites with an IC50 = 0.21 μM. A significant drop in inhibitor potency was observed with the GNF7686-resistant T. cruzi epimastigotes (oxygen consumption IC50 = 5.2 μM). As seen for growth inhibition, the results on respiration inhibition distinguish GNF7686 and antimycin A from the QP site inhibitors (Fig 4B).
We then asked whether GNF7686 inhibits T. cruzi cytochrome bc1 (complex III) directly (Fig 4C). Epimastigotes were permeabilized with digitonin, and KCN (complex IV inhibitor) was added to the permeabilized cells to block electron conductance by the parasite ETC. The reaction was then initiated by adding stoichiometric quantities of decylubiquinol (an electron donor for complex III) and oxidized yeast cytochrome c (an electron acceptor), and the catalytic activity of complex III was monitored through accumulation of reduced yeast cytochrome c. A control reaction with antimycin A confirmed earlier observations with intact epimastigotes (Table 1). Antimycin A potently blocked reduction of cytochrome c in the reaction with wild-type permeabilized epimastigotes, but a dramatic loss of inhibitor potency was observed (~100-fold) when GNF7686-resistant permeabilized parasites were used (Fig 4C). In a similar fashion, GNF7686 inhibited wild-type complex III activity with an IC50 of 0.40 μM, but a 20-fold loss of potency was observed with the mutant complex III (Fig 4C). These observations validate GNF7686 as a complex III inhibitor that likely targets the QN site of the T. cruzi cytochrome b.
The inhibitory effect of GNF7686 on mammalian cytochrome b function was assessed through monitoring oxygen consumption by mitochondria isolated from rat skeletal muscle cells. In the control reaction, antimycin A showed a potent inhibition (IC50 = 0.81 μM) of mitochondrial respiration (Fig 4D). In a parallel experiment, GNF7686 did not show any effect on oxygen consumption up to 25 μM concentration (Fig 4D). This observation validates GNF7686 as a highly selective inhibitor of the T. cruzi cytochrome b and a promising starting point for Chagas disease drug discovery.
We also examined effect of GNF7686 on cytochrome b in the malaria parasite, P. falciparum, and in yeast S. cerevisiae, the latter being used as a surrogate for pathogenic Pneumocystis jirovecii, a causative agent of a pneumocystis pneumonia [59]. Cytochrome b is a validated drug target in both organisms and atovaquone, an inhibitor of cytochrome b, is a clinical treatment for these diseases.
For the P. falciparum studies, all inhibitors were tested on two parasite lines—D10attB and yDHODH-D10attB (Table 2). In the wild-type D10attB line, the parasite de novo pyrimidine biosynthesis is dependent on a type 2 dihydroorotate dehydrogenase (PfDHODH) and requires a functional P. falciparum ETC, including cytochrome b, downstream from PfDHODH [26, 27]. In contrast, the yDHODH-D10attB cell line is modified with a type 1A dihydroorotate dehydrogenase from S. cerevisiae (yDHODH), which is cytosolic and utilizes fumarate as the terminal electron acceptor [25–27]. The assay was validated with antimycin A, which blocked growth of the D10attB line (EC50 = 0.7 μM), but was inactive on the yDHODH-D10attB parasite line (EC50 > 12.5 μM). Myxothiazol and strobilurin also showed a similar preferential activity on the D10attB line, whereas GNF7686 did not inhibit growth of either P. falciparum cell line. This result suggests that GNF7686 is not active on P. falciparum cytochrome b.
A similar, growth inhibition-based assessment of GNF7686 effect on the ETC function was also performed on S. cerevisiae (Table 2). Growth inhibition of a wild-type S. cerevisiae strain by compounds in two different media was monitored. The first medium contained glucose as the sole carbon source, which allows growth of yeast cells that lack a functional ETC [60]. In the second medium, glucose was replaced with glycerol, a non-fermentable carbon source. Under the latter condition, yeast growth is dependent on cellular respiration and functional cytochrome b [60]. All prototypic cytochrome b inhibitors used in this report potently inhibited yeast growth on medium with glycerol, but were inactive when yeast grew in medium with glucose. Following a similar pattern, GNF7686 weakly inhibited growth of yeast in glycerol medium (EC50 = 5.0 μM), but did not affect yeast growing in the medium with glucose.
In summary, GNF7686 selectively inhibits T. cruzi cytochrome b and does not affect respiration of mammalian mitochondria nor does it significantly inhibit respiration-dependent growth of P. falciparum and S. cerevisiae.
We have shown that cytochrome b is a possible target for new drug discovery efforts aimed at treating kinetoplastid diseases. The importance of this finding is underlined by the paucity of drug targets for these diseases. For Chagas disease, only sterol 14α-demethylase (CYP51) and cruzain have been explored in depth as possible targets. However, low parasitological cure rates that were observed (20–30%) during clinical testing of anti-fungal drugs targeting sterol 14α-demethylase (posaconazole and ravuconazole prodrug E1224) in Chagas disease patients have lessened enthusiasm for further work on repurposed CYP51 inhibitors as single agents [61, 62]. Additional clinical evaluation of this class of drugs partnered with benznidazole for combination treatments is still planned. It is also important to note that the failure of posaconazole and E1224 in phase 2 trials have been attributed to insufficient drug exposure or dosing duration [14], and work on T. cruzi-specific CYP51 inhibitors that could enter clinical development in the future is also ongoing [63, 64]. Finally, the anti-cancer drug BEZ235 has activity across kinetoplastid parasites, but it requires additional optimization (improvement of therapeutic index) before becoming a preclinical candidate [65]. Given this sparse landscape, new chemical starts and new drug targets are urgently needed to anchor drug discovery efforts for the kinetoplastid diseases.
Many groups have resorted to a ‘pre-genomic’ approach to drug discovery, in which compounds are screened to identify inhibitors of pathogen growth, without regard to mechanism of action. While this approach typically provides large number of chemical starting points and broad hit diversity, a lack of information on mechanism of action creates additional risk in chemical optimization, and in predicting possible toxicity liabilities. Next generation sequencing of evolved resistant pathogens has been used successfully to identify resistance mechanisms and, in many cases, target mechanisms, for several pathogens. In our own program, we have identified targets for malaria and tuberculosis [66, 67]. However, the approach has not been reduced to practice in kinetoplastid drug discovery until this study.
Several features of the results in this study bode well for future application of the approach. First, the number of mutations associated with the emergence of drug resistance was relatively low (five point mutations), which simplified subsequent target prediction and validation. Second, we were able to generate resistance in T. cruzi epimastigotes despite a relatively long doubling time and stable genome. Selection process required almost a year of drug pressure in this study but was ultimately successful. Finally, we found a mutation in a gene that is a ‘plausible’ drug target, where plausibility is supported by essentiality of the mutated gene in other cellular systems, or precedence from drugs targeting homologous targets in other organisms.
Drugs targeting cytochrome b are in clinical use for treatment of malaria and fungal pneumonia, and cytochrome b was also reported as a promising target for treatment of tuberculosis. The current study extends utility of cytochrome b as a drug target also to Chagas disease, and possibly leishmaniasis. Various structurally different cytochrome b inhibitors showed patterns of growth and biochemical inhibition that consistently confirmed that functional cytochrome b is essential for T. cruzi propagation. Based on the presented validation data, T. cruzi growth can be inhibited through targeting either QN or QP cytochrome b site. GNF7686 represents a new cytochrome b inhibitor, likely targeting the QN site, which has high selectivity for T. cruzi and does not show any effect on respiration of mammalian mitochondria. Crystal structures of cytochrome b from several sources (bovine, chicken, yeast, Rhodobacter, Paracoccus) were previously published [68, 69]. This opens opportunities for rational drug design based on homology modeling of T. cruzi cytochrome b structure. Finally, a counter-screen assay to measure inhibitory activity on the cognate human enzyme (as described here) can be used to guide chemical optimization away from host toxicity. While GNF7686 appears to be a promising starting point for kinetoplastid drug discovery, it does not inhibit the growth of P. falciparum or S. cerevisiae, and thus may not be a suitable starting point for anti-malaria or anti-fungal drug discovery. Further in vivo characterization of GNF7686 revealed that it has poor pharmacokinetic properties, including low oral bioavailability (F = 6%) and high in vivo mouse clearance (53 mL * min-1 * kg-1); thus, extension of the current studies to an in vivo model of Chagas disease will require identification of a GNF7686 analogue that has improved pharmacokinetic profile.
In summary, we have established an approach for identification of molecular targets of T. cruzi growth inhibitors that enables transition to target-based drug discovery for compounds with previously unknown mechanism of action. The first application of this approach resulted in identification of a highly selective inhibitor of T. cruzi cytochrome b, GNF7686, which can serve as an excellent starting point for discovery of new drugs for Chagas disease and leishmaniasis.
|
10.1371/journal.pntd.0003873 | Population Pharmacokinetics of an Indian F(ab')2 Snake Antivenom in Patients with Russell's Viper (Daboia russelii) Bites | There is limited information on antivenom pharmacokinetics. This study aimed to investigate the pharmacokinetics of an Indian snake antivenom in humans with Russell’s viper bites.
Patient data and serial blood samples were collected from patients with Russell’s viper (Daboia russelii) envenoming in Sri Lanka. All patients received Indian F(ab’)2 snake antivenom manufactured by VINS Bioproducts Ltd. Antivenom concentrations were measured with sandwich enzyme immunoassays. Timed antivenom concentrations were analysed using MONOLIXvs4.2. One, two and three compartment models with zero order input and first order elimination kinetics were assessed. Models were parameterized with clearance(CL), intercompartmental clearance(Q), central compartment volume(V) and peripheral compartment volume(VP). Between-subject-variability (BSV) on relative bioavailability (F) was included to account for dose variations. Covariates effects (age, sex, weight, antivenom batch, pre-antivenom concentrations) were explored by visual inspection and in model building. There were 75 patients, median age 57 years (40-70y) and 64 (85%) were male. 411 antivenom concentration data points were analysed. A two compartment model with zero order input, linear elimination kinetics and a combined error model best described the data. Inclusion of BSV on F and weight as a covariate on V improved the model. Inclusion of pre-antivenom concentrations or different batches on BSV of F did not. Final model parameter estimates were CL,0.078 Lh-1, V,2.2L, Q,0.178Lh-1 and VP,8.33L. The median half-life of distribution was 4.6h (10-90%iles:2.6-7.1h) and half-life of elimination, 140h (10th-90th percentilesx:95-223h).
Indian F(ab’)2 snake antivenom displayed biexponential disposition pharmacokinetics, with a rapid distribution half-life and more prolonged elimination half-life.
| Snake envenoming is a neglected tropical disease that affects hundreds of thousands of people in the rural tropics. Antivenom is the main treatment for snake bites but there is limited information on the pharmacokinetics and appropriate dosing regimen. Most studies have been done in animals and dosing guidelines are based on arbitrary and often irreversible clinical signs. In this study we measured serial antivenom concentrations in patients with Russell’s viper envenoming given antivenom. Using this data we modelled the pharmacokinetics of antivenom in the population and showed that antivenom concentrations had a bi-exponential decay with an initial decrease over 12 hours and then a slow decrease over days. There was significant variability in the dose given which was not affected by the particular antivenom batch given. The presence of venom did not appear to modify the pharmacokinetics of antivenom. Understanding the time course of antivenom in patients with snake envenoming will provide a better basis for antivenom dosing.
| Snake envenoming is a major health issue in South and South-eastern Asia [1]. Although antivenom is the most important treatment for snake envenoming, it can cause early systemic hypersensitivity reactions [2, 3], and there is limited evidence to support currently practiced dosing schedules. Dosing and assessment of the effectiveness of antivenom in human envenoming remains controversial and treatment protocols are not based on the kinetics of venom or antivenom. There are few studies of the pharmacokinetics of antivenom, and most of these are in animals [4].
Snake envenoming is a common problem in Sri Lanka and large amounts of antivenom are used throughout the country each year. A number of different Indian antivenoms are currently used and the initial dose ranges from 10 to 20 vials [5–7]. The initial dose is based on ED50 studies and clinical experience by titrating dose against the resolution of coagulopathy and neurotoxicity. However, the clinical effects of envenoming in these species are generally irreversible so determining if enough antivenom has been given and deciding to re-dose is often arbitrary and not based on whether all venom has been bound, or on the pharmacokinetics of antivenom. Measurement of venom and antivenom concentrations in patients with snake bite is required to improve effective initial and repeat dosing [8].
The pharmacokinetics of antivenom are expected to be similar to other intravenous drugs being delivered to the central compartment with zero order input kinetics (constant rate of infusion). Antivenom is then distributed throughout the body and is eliminated by the kidneys and/or the reticuloendothelial system [4]. Decreasing antivenom concentrations in the central compartment are therefore due to both distribution and elimination. Different types of antivenom have different pharmacokinetics due to the difference in their molecular masses [4]. Fab antivenoms have much larger volumes of distribution (VD) than F(ab’)2 or whole IgG [5, 9]. Most studies of antivenom pharmacokinetics show a biphasic (two-compartment) decline after intravenous administration of whole IgG and F(ab’)2 antivenoms, as a result of an initial rapid decline (distribution phase) and a slower decline (terminal elimination phase) [4, 9].
Most studies of the pharmacokinetics of antivenom are in animals [4, 10], and the pharmacokinetics appear to differ between animals making animal models problematic for defining the pharmacokinetics of antivenom in humans [10]. Although there have been several publications of antivenom concentrations in snake envenoming, there are only a few studies of the pharmacokinetics of antivenom in human snake envenoming [4, 5, 9, 11–14]. These studies were all in small numbers of patients using a classic two phase approach, without including input processes (i.e. delivery of the antivenom, usually via an infusion to the central compartment as a zero order process) and providing limited information on the pharmacokinetics and variation between patients.
A population approach to pharmacokinetic analysis is increasingly being used to define the pharmacokinetics of drugs in humans because it provides information about population variability and the need for individualisation of drug treatment. The traditional approach to pharmacokinetic analysis (two stage analysis) estimates the pharmacokinetic parameters for each individual patient and then provides summary statistics which only give a population average and standard error. In contrast the population approach estimates the typical value of each parameter for the population and the variability of the parameters simultaneously. This provides an estimate of unexplained random variation and allows the effects of covariates to be accounted for in the model (e.g. weight, renal function). There are no previously published population pharmacokinetic analyses of antivenom in humans or animals.
The aim of this study was to investigate the pharmacokinetics of antivenom in patients with snake bites using a population based analysis, including an investigation of the covariates that may influence the pharmacokinetics of antivenom.
This was a population pharmacokinetic analysis of an F(ab’)2 antivenom using data and serial antivenom concentrations collected in snake-bite patients admitted to a single hospital in Sri Lanka. The patients were recruited from within a large cohort of snakebites admitted to the Base Hospital Polonnaruwa in Central Eastern Sri Lanka.
The study was approved by the Ethical Review Committee, Faculty of Medicine, University of Peradeniya, Sri Lanka. All patients gave written and informed consent for the collection of clinical data and blood samples.
All patients (>15 years old) from October 2010 to March 2012 with a suspected snake bite who presented to the Base Hospital Polonnaruwa were recruited to a prospective cohort study. Those with coagulopathy were then entered in a dose finding randomised clinical trial of fresh frozen plasma. The entry criteria for the trial was a suspected Russell’s viper (Daboia russelii) bite with coagulopathy defined as an abnormal 20 minute whole blood clotting test (20WBCT). This resulted in a small number of patients being recruited where Russell’s viper (D. russelii) venom was not detected and on further testing, hump-nosed viper (Hypnale spp.) venom was detected (in some Hypnale bites the 20WBCT and coagulation studies may be abnormal [15, 16]).
In this pharmacokinetic study, patients were only recruited from the clinical trial and were included if they had serial serum collection for antivenom measurement and complete demographic details (including weight). All patients received the Indian polyvalent snake antivenom intravenously manufactured by VINS Bioproducts Limited (batch numbers: 1060 [MFD 2008], 1096 [MFD 2009], 1102 [MFD 2009], 01015/10-11 [MFD 2010], 01AS11112 [MFD 2011]). For a dose of antivenom, each of 10 vials of antivenom are reconstituted in 10ml of normal saline for a total of 100ml of antivenom. From a 500ml bag of normal saline 100ml volume is removed and replaced by the 100ml of antivenom so the 10 vials are administered in a total of 500ml of normal saline. This is given over 1 hour.
The following data were collected prospectively in all cases: demographics (age, sex and weight), time of the snake bite, clinical effects (local envenoming, coagulopathy, bleeding and neurotoxicity) and antivenom treatment (dose, time of administration and antivenom batch number). Blood samples were collected for research on admission and regularly throughout each patient admission. Blood was collected in serum tubes for venom-specific enzyme immunoassay (EIA) and antivenom EIA. All blood samples were immediately centrifuged, and then the serum aliquoted and frozen initially at -20°C, and then transferred to -80°C within 2 weeks of collection.
A sandwich enzyme immunoassay was used to measure antivenom in serum samples as previously described [8, 17]. The plate was first coated with Russell’s viper venom and then stored and blocked overnight. Serum was then added to the plates. The detecting antibodies were conjugated with horseradish peroxidase. Russell’s viper (D. russelii) and hump-nosed (Hypnale spp.) viper venoms were measured in samples with a venom specific enzyme immunoassay as previously described [6, 8, 17]. Briefly, polyclonal IgG antibodies were raised in rabbits against Russell’s viper (D. russelii) and hump-nosed viper (Hypnale spp.) venom. The antibodies were then bound to microplates and also conjugated to biotin for a sandwich enzyme immunoassay using streptavidin-horseradish peroxidase as the detecting agent. All samples were measured in triplicate, and the averaged absorbance converted to a concentration using a standard curve made up with serial dilutions of antivenom and using a sigmoidal curve. The limit of quantification for the antivenom enzyme immunoassay assay was 40μg/ml and for the venom enzyme immunoassay was 2ng/mL for Russell’s viper and 0.2ng/ml for hump-nosed viper.
Patient data was analysed using MONOLIX version 4.2 (Lixoft,Orsay, France. www.lixoft.com). MONOLIX uses the Stochastic Approximation Expectation Maximization algorithm (SAEM) and a Markov chain Monte-Carlo (MCMC) procedure for computing the maximum likelihood estimates of the population means and between-subject variances for all parameters [18]. One, two and three compartment models with zero order input and first order elimination kinetics were assessed and compared to determine the best structural model. Proportional and combined models were evaluated for the residual unexplained variability. Method M3 was used to deal with antivenom concentrations below the limit of quantification (BLQ) [19]. Between-subject variability (BSV) was included in the model and assumed to have log-normal distribution.
Models were parameterized in terms of volume of distribution (VD; V, VP, VP2), clearance (CL), inter-compartmental clearance (Q; Q1, Q2) and relative bioavailability (F) for either 1-, 2- or 3-compartment models. Initial estimates of parameters were taken from a previous pharmacokinetic study of anti-venom [9].
Uncertainty in antivenom dose was included in the model by allowing BSV on F to account for batch to batch variation in antivenom (five different batches) and for variation within batches. F was fixed to 1 and the BSV was estimated for each patient similar to including uncertainty on dose as previously described [18]. The BSV on F was plotted for each batch to determine if there was a difference between batches.
The effect of covariates, including age, sex, weight, and pre-antivenom concentrations in patients with detectable venom, were explored by visual inspection of the individual parameter estimates versus the covariate of interest. Age, sex and pre-antivenom concentrations were not included in the final model evaluation due to the absence of an association visually. The influence of weight (wt) on volume was included in the modelling process. Weight was assumed to be related to V by a power function. The covariate was centred to the average weight. Thus in the model the estimation of the effect of weight on volume is:
V = θV x (wt/wtav)∧fwt
Where θV is the typical value of volume of distribution, wt is the individual patient weight, wtav is the average weight and fwt accounts for the influence of wt on volume.
Model selection decisions were based on a decrease in the objective function (OFV), a decrease in residual error, clinical relevance of the pharmacokinetic parameters and goodness of fit plots. The log likelihood was computed for each model and used to discriminate through the difference in log likelihood (−2LL). A p-value of 0.05 was considered statistically significant, equivalent to a drop in OFV by 3.84.
From the final model we simulated 1000 patients using the individual predicted patient parameters from the final model with MatLab to explore different initial doses and repeat doses. The following scenarios were explored:
One dose (10 vials) of antivenom given with infusions rates of 20 minutes, 1 hour and 2 hours.
Two doses of antivenom given, each over 1 hour and 6 hours apart.
Two doses of antivenom given, each over 1 hour and 12 hours apart.
The median antivenom concentration versus time was plotted with 10% and 90% percentiles.
There were 75 patients with a median age of 38 years (16 to 64y) and 64 were male. Seventy one were Russell’s viper envenoming cases and 52 of these had detectable venom prior to the administration of antivenom. Four patients had hump-nosed viper envenoming (confirmed by detectable hump-nosed viper venom). In all four patients with hump-nosed viper envenoming there was a steady decline of venom concentrations despite the administration of antivenom consistent with the antivenom not being raised against this snake venom. In nineteen patients meeting the inclusion criteria venom was not detected prior to antivenom, most likely because the blood was collected prior to envenoming. The demographics of the patients are listed in Table 1.
There were 510 antivenom concentration data points but only 411 had detectable antivenom, the other 99 were serial samples after the disappearance of antivenom. There were 54 patients who had a single dose of antivenom who had 265 antivenom concentration measurements with a median of five antivenom concentrations in each patient (Range: 2 to 10), and a median antivenom concentration of 1607μg/ml (Range: 40 to 13673μg/ml). There were 21 patients who had multiple doses of antivenom who had 146 antivenom concentrations with a median of seven antivenom concentrations in each patient (Range: 3 to 11) and a median antivenom concentration of 2293μg/ml (Range: 40 to 12599μg/ml). The observed concentration versus time data is shown in Fig 1.
A two compartment model with zero order absorption and linear elimination kinetics and a combined error model best described the data. The final model incorporated BSV on F, which was fixed to 1 to allow variability between patients in dose. The model also incorporated weight as a covariate with a power effect on central volume, V. The inclusion of pre-antivenom concentrations on BSV of F did not improve the model. Plots of the BSV on F versus the batch number showed no relationship between the batch and BSV on F (S1 Fig). The final model parameter estimates were CL, 0.078 Lh-1, V, 2.2L, Q, 0.178Lh-1 and VP, 8.33L. The median half-life of distribution was 4.6h (10th-90th percentiles: 2.6 to 7.1h) and the half-life of elimination, 140h (10th-90th percentiles: 95 to 223h). There was no difference in the parameter estimates between those with Russell’s viper envenoming with detectable venom prior to antivenom (52), those with Russell’s viper envenoming and no detectable venom prior to antivenom (19) and those with hump-nosed viper envenoming (S2 Fig). S3 and S4 Figs shows the goodness-of-fit plots for the final model. The individual PK parameter estimates from the base models with modelling decisions and final model parameters are described in Table 2. There was also no difference in parameter estimates between patients given 1 dose of antivenom and those given 2 doses, or between patients with different initial venom concentrations (S5 Fig).
Simulations for one dose (10 vials) of antivenom given over 20 minutes, 1 hour and 2 hours shows there is a slightly lower and later peak antivenom concentration with slower infusions (Fig 2). Simulations for two doses of antivenom shows that antivenom concentrations decrease rapidly after each dose and there are low but persistent levels of antivenom after one dose and both two doses regimens (Fig 3).
The study adds to the limited information available on the pharmacokinetics of antivenom in humans supporting previous studies [4, 9]. Indian F(ab’)2 snake antivenom displayed biexponential disposition pharmacokinetics, with a rapid half-life of distribution and a much longer half-life of elimination. Weight accounted for some of the variability in the central volume, and the volumes of the central and peripheral compartment were consistent with a large molecule which does not have a large volume of distribution. Including variability on F improved the model showing that there was significant random variability in dose. The plots in Figs 2 and 3 show the expected antivenom concentration profiles in the first 24 hours after administration.
Previous human and most animal studies have also shown a biexponential decay in antivenom concentrations [4, 9, 20, 21], with similar values for the distribution half-life of 2 to 4 hours and much longer elimination half-life of 90 to 230h. Previous studies have been small with 10 or less patients in each analysis (for different antivenoms) and a classic two phase approach has been undertaken. Such an approach will over-estimate the error and not account for true random variability or covariate effects. In this study we have undertaken a population approach, which provides information on the variability of the pharmacokinetics in the population and an improved model by including weight and variability in dose. Previous studies have not shown why they chose particular models (2-compartment versus 3-compartment), with no statistical criteria or goodness of fit plots.
Some previous animal models and one human study have described the pharmacokinetics with a tri-exponential decay in animals [14, 22, 23]. These analyses have not included an input process in the analysis which will bias the estimation of the disposition parameters, particularly with three or more compartments when the initial very short half-life is similar to the time of the input phase. Ismail et al. estimated the initial rapid half-life in animals to be 0.2h and Vazquez et al estimated it to be 0.25h, which are both similar to the usual infusion rate of antivenom over 10 to 30 minutes. It is possible that there is only 2-compartmental disposition kinetics in these studies, and future pharmacokinetic analyses need to include an input phase in the model. A possible limitation of our study was that there may have been insufficient sampling in the initial period after antivenom administration to detect a third compartment. In contrast to this, Vazquez et al were likely to have taken samples in the input phase, since the first sample was taken 5min after antivenom administration, although they do not report the infusion time or rate [14].
One animal study of a F(ab’)2 has shown that the pharmacokinetics of antivenom are the same in envenomed and non-envenomed rabbits [24]. This is consistent with this study demonstrating that pre-antivenom venom concentrations did not influence the pharmacokinetics of antivenom, including different initial venom concentrations (S5 Fig). However, this may be different for Fab antivenoms where high molecular weight toxins may change the route of elimination from renal (for free Fab antivenom) to phagocytosis/reticulo-endothelial system for Fab-toxin molecules. The latter has been shown in rabbits with anti-Vipera Fab antivenom [25].
There has always been concerns about the variability between different batches of antivenom leading to potential differences in the dose administered between batches. The study did not support this concern and found that there was no difference in F on average between different batches (S1 Fig). However, the study found that including between subject variability on relative bioavailability did improve the model. This suggests there was random variability in the dose administered which is likely to be due to variable losses occurring during reconstitution of the individual freeze dried vials of antivenom. So, although there may be variability between batches, the variability in dosing errors appears to be larger than the differences between batches.
There are a number of limitations to the study including the fact that the sample collection was not optimally designed and sample times (windows) were based on timing of clinical samples and other research assays required for the clinical trial. This is unlikely to have a major influence on the analysis because a population approach will allow for both sparse and rich sampling in patients. Another issue is that antivenom is not a pure substance and consists of varying amounts of polyclonal antibodies to multiple toxins in the venom with varying affinities. However, the assay uses a single detecting antibody (anti-horse antibody), so will detect all antibodies against the snake toxins irrespective of their toxin target or affinity. Finally, the assay will only detect antibodies that bind to the snake toxins. In most antivenoms, specific antibodies to snake toxins make up only 10 to 20% of the total protein/immunoglobulin content. This is unlikely to have affected the pharmacokinetic analysis because only immunoglobulins binding to snake toxins are relevant to the analysis.
This population pharmacokinetic analysis demonstrates that Indian F(ab’)2 antivenom has biexponential disposition kinetics and following an initial decline in antivenom concentrations in the first 12 hours, low concentrations are present for days after administration. The study demonstrates that the antivenom concentrations were not affected by the initial venom concentrations suggesting that sufficient antivenom in excess of the venom was being administered. Understanding the pharmacokinetics of antivenom may assist in improving antivenom dosing by matching antivenom pharmacokinetics to the neutralisation of venom (pharmacodynamics), as well as clinical effects.
|
10.1371/journal.ppat.1003261 | Release of Luminal Exosomes Contributes to TLR4-Mediated Epithelial Antimicrobial Defense | Exosomes are membranous nanovesicles released by most cell types from multi-vesicular endosomes. They are speculated to transfer molecules to neighboring or distant cells and modulate many physiological and pathological procedures. Exosomes released from the gastrointestinal epithelium to the basolateral side have been implicated in antigen presentation. Here, we report that luminal release of exosomes from the biliary and intestinal epithelium is increased following infection by the protozoan parasite Cryptosporidium parvum. Release of exosomes involves activation of TLR4/IKK2 signaling through promoting the SNAP23-associated vesicular exocytotic process. Downregulation of let-7 family miRNAs by activation of TLR4 signaling increases SNAP23 expression, coordinating exosome release in response to C. parvum infection. Intriguingly, exosomes carry antimicrobial peptides of epithelial cell origin, including cathelicidin-37 and beta-defensin 2. Activation of TLR4 signaling enhances exosomal shuttle of epithelial antimicrobial peptides. Exposure of C. parvum sporozoites to released exosomes decreases their viability and infectivity both in vitro and ex vivo. Direct binding to the C. parvum sporozoite surface is required for the anti-C. parvum activity of released exosomes. Biliary epithelial cells also increase exosomal release and display exosome-associated anti-C. parvum activity following LPS stimulation. Our data indicate that TLR4 signaling regulates luminal exosome release and shuttling of antimicrobial peptides from the gastrointestinal epithelium, revealing a new arm of mucosal immunity relevant to antimicrobial defense.
| Exosomes are secreted membranous nanovesicles produced by a variety of cells. Exosomes shuttle various molecules to transfer them to neighboring or distant cells, and have been implicated as mediators in cell-cell communications to modulate physiological and pathological procedures. Here, we report that luminal release of exosomal vesicles is an important component of Toll-like receptor 4 (TLR4)-associated gastrointestinal epithelial defense against infection by Cryptosporidium parvum, an obligate intracellular protozoan that infects gastrointestinal epithelial cells. Activation of TLR4 signaling in host epithelial cells following C. parvum infection promotes luminal release of epithelial exosomes and exosomal shuttling of antimicrobial peptides from the epithelium. By direct binding to the C. parvum surface, exosomal vesicles reveal anti-C. parvum activity. Activation of TLR4 signaling in epithelial cells after LPS stimulation also increases exosomal release and exosome-associated anti-C. parvum activity. Therefore, we speculate that TLR4-mediated exosome release may be relevant to innate mucosal immunity in general, representing a new target for therapeutic intervention for infectious diseases at the mucosal surface.
| Eukaryotic cells release membrane vesicles into their extracellular environment under physiological and pathological conditions [1]. These vesicles mediate the secretion of a wide variety of proteins, lipids, mRNAs, and microRNAs (miRNAs), interact with neighboring cells, and thereby traffic molecules from the cytoplasm and membranes of one cell to other cells or extracellular spaces [1], [2]. There is increasing evidence that secreted vesicles play an important role in normal physiological processes, development, and viral infection and other human disease [3]–[6]. Exosomes represent a specific subtype of secreted membrane vesicles that are around 30–100 nm in size, formed inside the secreting cells in endosomal compartments called multi-vesicular bodies (MVBs) [2]. Exosomes are produced by a variety of cells (e.g., reticulocytes, epithelial cells, neurons, tumor cells) and have been found in bronchoalveolar lavage, urine, serum, bile, and breast milk [2], [7], [8]. The composition of exosomes is heterogenic, depending on the cellular origin of the exosome. Exosomes do not contain a random array of intracellular proteins, but a specific set of protein families arising from the plasma membrane, the endocytic pathway, and the cytosol, especially those of endosomal origin, such as CD63, ICAM-1, and MHC molecules [2], [9]–[13].
Secretion of exosomes is regulated by various stimuli, including the activation of P2X receptor by ATP on monocytes and neutrophils, thrombin receptor on platelets, and Toll-like receptor (TLR) 4 by LPS on dendritic cells [2], [14], [15]. Formation of intraluminal vesicles of MVBs and targeting of transmembrane proteins to these vesicles involve a complex intracellular sorting network, including the endosomal sorting complex required for transport (ESCRT) machinery [2], [15]. Fusion of MVBs with plasma membrane is an exocytotic process that requires the association of v-SNAREs (from the vesicles) and t-SNAREs (at the membrane) to form a ternary SNARE (SNAP receptor) complex. The SNARE complex brings the two membranes in apposition, a necessary step in overcoming the energy barrier required for membrane fusion [16]. Several Rab family proteins, including Rab11 and Rab27b, are key regulators of the exosome secretion pathway and are involved in MVB docking at the plasma membrane [17].
Epithelial cells along the mucosal surface provide the front line of defense against luminal pathogen infection in the gastrointestinal tract and are an important component of gastrointestinal mucosal immunity [18], [19]. TLRs recognize discrete pathogen-associated molecular patterns and activate a set of adaptor proteins (e.g., MyD88) and intracellular kinases (e.g., IKKs), leading to the nuclear translocation of transcription factors, such as NF-κB [20]. Activation of the TLR/NF-κB pathway initiates a series of host defense reactions against pathogens, including parasites. Exosomes derived from the apical and basolateral sides of gastrointestinal epithelial cells, including biliary epithelial cells, have recently been identified, but their physiologic and pathologic relevance is still unclear [21], [22]. These basolateral exosomes have been shown to modulate lymphocyte immune responses during mucosal infection [21]. Intestinal epithelial cell-derived exosomes containing αvβ6 integrin and food antigen induced the generation of tolerogenic dendritic cells in a model of tolerance induction [23]. The presence of these intestinal epithelial cell-derived exosomes impacted the development of antigen-specific T regulatory cells [23]. Release of exosomes into the bile has been shown to influence intracellular regulatory mechanisms and modulate biliary epithelial cell proliferation via interactions with epithelial primary cilia [24].
Cryptosporidium parvum is an obligate intracellular protozoan of the phylum Apicomplexa that infects intestinal and biliary epithelial cells [25]. Infection activates TLR4 signaling in host epithelial cells through direct parasite-host cell interactions [19], [25]. Due to the “minimally invasive” nature of infection, epithelial cells play a key role in activating and communicating with the host immune system against C. parvum infection. MicroRNAs are small regulatory RNAs that mediate either mRNA cleavage or translational suppression, resulting in gene suppression [26]. miRNAs can be seen as a fine-tuning for the cellular responses to external influences, and might be important players in the regulation of host immune response. In our previous studies, we demonstrated that activation of TLR4/NF-κB signaling in epithelial cells regulates transcription of miRNA genes to orchestrate host anti-C. parvum immune responses through modulation of miRNA-mediated posttranscriptional suppression [27], [28]. In the work described here, we found that C. parvum infection stimulates host epithelial monolayers to release apical exosomes through activation of TLR4 signaling, with the involvement of activation of the IKK2/SNAP23 exocytic sorting and let-7 miRNA-mediated gene regulation. Released exosomes shuttle several antimicrobial peptides, can bind to the C. parvum sporozoite surface, and display anti-C. parvum activity in vitro and ex vivo. Our results demonstrate, for the first time, that luminal release of exosomes is an important component of TLR4-associated epithelial immune reactions against C. parvum infection.
Non-malignant human biliary epithelial cells, H69 cells, were grown to confluence on Percoll inserts to form monolayers; we detected a few exosome-like microvesicles in the apical side under electron microscopy (EM) (Figure 1A). These microvesicles were cup-shaped, with a diameter size about 40–100 nm, typical for exosomes. When H69 monolayers were exposed to infection by C. parvum, abundant exosome-like microvesicles were found in the apical region by the epithelium (Figure 1A). These microvesicles were morphologically similar to the exosome-like microvesicles found in non-infected monolayers. Interestingly, MVBs were detected in the cytoplasm of infected cells. These MVBs usually contained several exosome-like microvesicles that were morphologically similar to those released to the apical region (Figure 1A). Using a well-established ultracentrifugation approach [29], we isolated and purified these exosome-like microvesicles from the apical supernatants of the H69 monolayers 24h after exposure to C. parvum infection. These purified microvesicles displayed cup-shape vesicular characteristics of exosomes, with a diameter size of 40–100 nm under scanning EM. Immunogold staining revealed that these exosome-like microvesicles were positive for exosome markers CD63 and ICAM-1 (Figure 1A), and, thus, they were subsequently referred to as exosomes. Interestingly, these apical exosomes released from H69 monolayers following infection were positive for MHC I and MHC II (Figure S1). We then quantified the release of apical exosomes from H69 monolayers at various time points after exposure to C. parvum. A time-dependent increase of apical exosomes was identified using multiple approaches, including Nanoparticle Tracking Analysis (NTA) [30], [31] (Figure 1B and Figure S1), EM (Figure 1B), and Western blot for CD63 (Figure S1).
Having determined the increment of apical exosome release from cell monolayers following C. parvum infection, we then tested the potential underlying mechanisms. Previous studies demonstrated that several intracellular signaling pathways are activated in host epithelial cells following infection, including the TLR4/IKK/NF-κB and PI-3K pathways [28], [32]. Transfection of epithelial cells with the TLR4-dominant mutant (TLR4-DN) or treatment of cells with the IKK2 inhibitor SC-514 [33] significantly inhibited C. parvum-induced exosome release (Figure 1C). In contrast, an inhibitor of PI-3K showed no effects. Similarly, exposure of H69 monolayers to LPS (a potent TLR4 ligand) increased release of apical exosomes (Figure S1). Release of apical exosomes induced by LPS stimulation or C. parvum infection was also noted in cultured mouse biliary epithelial monolayers (603B monolayers) and intestinal epithelial monolayers (IEC4.1 monolayers) (Figure S1).
To test C. parvum-induced exosome release in vivo, we applied a mouse model of biliary cryptosporidiosis through gallbladder injection of C. parvum oocysts into wild-type and TLR4-deficient mice [34]. Consistent with our previous results [35], a higher biliary infection burden was found in TLR4-deficient mice compared with the wild-type animals (Figure 2A). Intriguingly, abundant exosomes were detected in the lumen of the biliary tract from the wild-type animals following infection (Figure 2B and 2C). Only a few exosomes were detected in the biliary lumen in the infected TLR4-deficient mice. Taken together, these data suggest that C. parvum infection increases luminal release of exosomes from the biliary epithelium, probably through TLR4/IKK2-mediated activation of the MVB exocytotic pathway.
To test the contribution of MVB-associated exocytosis in C. parvum-induced apical exosome release, we performed experiments using confocal analysis on C. parvum-infected H69 cells. Using CD63 as the marker for MVBs [2], we detected an increase of MVBs in the cytoplasm of H69 cells following C. parvum infection (Figure 3A). The CD63-positive staining showed patchy distribution in the cytoplasm, with a diameter from 300 to 1,000 nm, consistent with characteristics of MVBs in cells as shown by EM in Figure 1A. In addition, these CD63-positive vesicles in the cytoplasm were overlaying with exosomes labeled with N-(Lissamine) rhodamine B sulfonyl dioleoylphosphatidylethanolamine (N-Rh-PE) [36] (Figure 3B), suggesting that they were MVBs. Immunostaining revealed positive reactions to Rab11, Rab27b, and SNAP23 co-localized to the CD63-positive MVBs (Figure 3C–3E). Of note, accumulation of SNAP23 was obviously around the MVBs in infected cells (Figure 3E), which was further confirmed in cells overexpressing SNAP23 (Figure S2). Interestingly, C. parvum-induced SNAP23 localization around MVBs was not detected in H69 cells stably expressing TLR4-DN (Figure 3F).
To further investigate the role of SNAP23 in C. parvum-induced exosome release, H69 cells were first treated with an siRNA to SNAP23, followed by exposure to C. parvum infection for 24 h. SNAP23 siRNA treatment resulted in a significant decrease in exosome release from infected H69 monolayers (Figure 4A). H69 cells were infected by C. parvum for various time points, followed by measurement of total SNAP23 by Western blot and phosphorylated SNAP23 by immunoprecipitation (IP). Increased levels of total SNAP23 (Figure 4B) and phosphorylated SNAP23 (Figure 4C) were detected in infected cells. C. parvum-induced SNAP23 expression and phosphorylation were not detected in H69 cells stably expressing TLR4-DN (Figure 4B). Induction of SNAP23 total protein was detected in mouse biliary epithelial 603B cells infected by C. parvum (Figure S3). Upregulation of SNAP23 and enhanced phosphorylation of SNAP23 were also detected in H69 cells following LPS stimulation (Figure S3). Previous studies demonstrated that SNAP23 is a substrate for IKK2, and phosphorylation of SNAP23 at Ser120 and Ser95 by IKK2 stimulates fusion of intracellular vesicles to the cell membrane, promoting exocytosis [37]. Indeed, a direct association between IKK2 and SNAP23 was evident from IP analysis (Figure 4D). Increased interaction between IKK2 and SNAP23 was detected in infected cells (Figure 4D). Such observations, however, do not prove that an IKK2-SNAP23 association is occurring in MVBs or exosomes or that it is just relevant to these processes. Together, these data suggest that TLR4 signaling activates IKK2 to phosphorylate SNAP23 to stimulate exosome release in cells following C. parvum infection.
miRNAs, such as the let-7 family members, have been implicated in the regulated exocytotic process [38]–[40]. Interestingly, several members of the let-7 miRNAs, including let-7i, let-7d, let-7f, let-7e, and miR-98, showed significant downregulation in H69 cells following C. parvum infection by array analysis (Figure 5A), consistent with our previous studies on H69 cells [27], [28]. Decreased expression of let-7i and miR-98 was further confirmed by Northern blot and real-time PCR analysis in H69 and 603B cells following C. parvum infection (Figure 5A). Consistent with results from our previous studies on LPS-suppressed expression of let-7 miRNAs in epithelial cells [41], we detected a decreased expression level of miR-98 in H69 cells following LPS stimulation (Figure S3). Using in silico database analysis, we found that there is conserved complementarity between SNAP23 3′UTR and miRNAs of the let-7 family (Figure 5A). To test the potential targeting of SNAP23 mRNA by let-7 miRNAs, we generated pMIR-REPORT luciferase constructs containing the SNAP23 3′UTR with the putative let-7 binding site (Figure 5B). In addition, constructs with the CCTC to GGAG for human and ACCT to TGGA for mouse mutation at the putative binding sites were also generated as controls. We then transfected H69 cells with these reporter constructs, followed by assessment of luciferase activity 24 h after transfection. As shown in Figure 5C, a significant decrease of luciferase activity was detected in cells transfected with the SNAP23 3′UTR construct containing the potential binding site compared with mutant control vector. No change in luciferase activity was observed in cells transfected with the mutant SNAP23 3′UTR construct, suggesting endogenous translational repression of the construct with the SNAP23 3′UTR. Anti-miRs (anti-miR miRNA inhibitors) and miRNA precursors are chemically modified RNA molecules designed to specifically inhibit and mimic, respectively, endogenous miRNAs [42]. Accordingly, anti-let-7i markedly increased SNAP23 3′UTR-associated luciferase reporter translation (Figure 5C). In contrast, the miR-98 precursor significantly decreased the luciferase activity (Figure 5C). Of note, anti-let-7i and miR-98 precursor only caused a modest alteration in the luciferase activity, probably due to the targeting of SNAP23 3′UTR by other members of the let-7 miRNA family in the cells. Similar results were also confirmed in 603B cells (Figure 5C).
To test whether miRNA-mediated suppression of SNAP23 is directly relevant to SNAP23 expression in biliary epithelial cells, we treated H69 cells with miR-98 precursor or anti-let-7i and then measured SNAP23 protein level (treated for 48 h) by Western blot or mRNA level (treated for 24 h) by real-time PCR. Treatment of cells with the anti-let-7i caused a significant increase in SNAP23 protein content (Figure 6A). Conversely, a decrease in SNAP23 protein level was detected in cells after treatment with the miR-98 precursor (Figure 6A). No significant changes in SNAP23 mRNA level were detected in cells after treatment with miR-98 precursor or anti-let-7i (Figure 6A). No significant changes in the phosphorylated SNAP23 level were detected in cells after treatment with the anti-let-7i (Figure 6B). These data suggest that let-7 family miRNAs target SNAP23 3′UTR, resulting in posttranscriptional suppression through translation suppression.
To test the impact of let-7 miRNAs on C. parvum-induced SNAP23 upregulation in H69 cells, we treated cells with the miR-98 precursor for 48 h and then exposed them to C. parvum for 24 h, followed by Western blot for SNAP23. The miR-98 precursor diminished C. parvum-induced SNAP23 expression (Figure 6C). Moreover, the miR-98 precursor partially blocked C. parvum-induced exosome release in H69 cells (Figure 6D). Interestingly, treatment of cells with the anti-let-7i showed no significant effects on C. parvum-induced exosome release (Figure 6D), presumably because of the low level of let-7i in the infected cells as shown in Figure 5A.
To investigate the potential effects of released apical exosomes on C. parvum, we incubated freshly excysted C. parvum sporozoites with a serial dilution of exosomes isolated from H69 monolayers after exposure to C. parvum for 24 h. Incubation of C. parvum sporozoites with isolated exosomes resulted in a decrease in C. parvum viability in a dose-dependent manner by the viability assay (Figure 7A). Sporozoites incubated with the exosomes isolated from non-infected monolayers showed a modest decrease in C. parvum viability (Figure 7A). Exosomes isolated from the basolateral side of infected H69 monolayers, or apical exosomes from infected H69 cells expressing TLR4-DN, showed no significant effects on C. parvum viability (Figure 7A). A decrease in C. parvum viability was also detected after incubation with apical exosomes isolated from H69 monolayer following LPS stimulation (Figure S3). We then incubated the same number of C. parvum sporozoites with apical exosomes isolated from infected H69 monolayers for 2 h; after extensive washing, these sporozoites were added to H69 cells for infection. A decreased infectivity to host cells for C. parvum sporozoites after pre-incubation with exosomes was detected, compared with the sporozoites after incubation with culture medium, as assessed by real-time PCR (Figure 7B) and immunostaining (Figure 7C). Incubation with exosomes from the apical side of non-infected H69 monolayers, or infected cells stably expressing TLR4-DN and MyD88-DN, showed no inhibitory effects on C. parvum sporozoite infectivity (Figure 7B and 7C). A decreased infectivity of C. parvum sporozoites to intestinal IEC4.1 cells after pre-incubation with exosomes was also detected by real-time PCR (Figure S4).
Notably, binding of exosomes to the C. parvum sporozoite surface after incubation with apical exosomes was detected by scanning EM. In H69 cell cultures after exposure to C. parvum for 2 h, we observed binding of exosomes to the parasite surface (Figure 8A). Some of the sporozoites with several exosomes binding to their surface were dying (showing high electrical density, as indicated by the asterisk in Figure 8A). To further confirm the direct binding of exosomes to the C. parvum sporozoite surface, we first labeled exosomes with the N-Rh-PE as previously reported [36]. Labeled exosomes were then incubated with freshly excysted C. parvum sporozoites, followed by observation under confocal microscopy. Fluorescent activity was detected in these sporozoites incubated with the labeled exosomes (Figure 8B). No detectable fluorescent activity was observed in the sporozoites incubated with the non-labeled exosomes.
Direct interactions between the Gal/GalNAc-containing glycoproteins on the epithelial cell apical membrane surface and Gal/GalNAc-ligand molecules on the C. parvum sporozoite surface have been implicated to mediate the attachment of C. parvum sporozoites to host cells [25]. The Gal/GalNAC-specific lectin PNA can markedly decrease the infection [43], [44]. We found that isolated apical exosomes carry Gal/GalNAc molecules (recognized by PNA), displaying positive reactions to PNA blotting at the epithelial apical membrane (Figure 8C). Interestingly, blotting of whole lysates of C. parvum sporozoites after pre-incubation with isolated exosomes revealed a positive reaction to exosome marker CD63 (Figure 8D). Moreover, pre-incubation of isolated exosomes with PNA (followed by extensive washing and ultracentrifugation to clear out unbound/free PNA) abolished the inhibitory effects of exosomes on C. parvum viability (Figure 8E). The above data indicate that apical exosomes released from C. parvum-infected biliary epithelium possess anti-C. parvum capacity. Interestingly, exosomes isolated from C. parvum-infected H69 monolayers showed no detectable inhibitory effects on the viability of the K12 strain E. coli after incubation in vitro (Figure S5).
Production of antimicrobial peptides, such as beta-defensins and cathelicidins, is a major element for TLR-mediated epithelial anti-C. parvum defense [19], [25], [45]. Increasing amounts of human beta-defensin 2 (HBD2) and cathelicidin-37 (LL-37) were detected by ELISA in apical exosomes isolated from C. parvum-infected H69 monolayers, compared with those isolated from non-infected cells (Figure 9A). In contrast, no increase in HBD2 and LL-37 contents was detected in the isolated exosomes from infected H69 cells stably expressing TLR4-DN (Figure 9A). Exosomal shuttling of HBD2 and LL-37 was further confirmed by immunogold staining, using the apical exosomes isolated from C. parvum-infected H69 monolayers (Figure 9B and 9C). Unfortunately, we have been unable to achieve efficient exosome preparations to demonstrate that the antimicrobial peptides in the exosomes are biochemically active. Therefore, to elucidate whether exosomal shuttle of HBD2 and LL-37 accounts for the anti-C. parvum activity of these isolated exosomes from C. parvum-infected cells, we performed the gain- and loss-of-function experiments through overexpression or knockdown of HBD2 and LL-37 in H69 cells. These cells were then infected with C. parvum for 24 h, and the released exosomes were isolated and then incubated with freshly excysted C. parvum sporozoites. Of note, pre-incubation of sporozoites with the exosomes isolated from cells overexpressing HBD2 and LL-37 resulted in a significant decrease in parasite infectivity (Figure 9D). This is probably due to an increased sorting of antimicrobial peptides into the exosomes in these cells at the basal condition. Moreover, after incubation with exosomes isolated from C. parvum-infected H69 cells overexpressing HBD2 and LL-37, C. parvum sporozoites showed a further decrease in infectivity (Figure 9D). In contrast, incubation with exosomes isolated from C. parvum-infected H69 cells with knockdown of HBD2 and LL-37 showed no significant effects on the infectivity of C. parvum sporozoites (Figure 9E and 9F). These data suggest that antimicrobial peptides HBD2 and LL-37 in apical exosomes that are released from epithelial cells contribute to mucosal immune defense in response to C. parvum infection.
One of the major findings of this study is that activation of TLR4 signaling increases luminal release of exosomes from the biliary epithelium during C. parvum infection. Whereas a basal level of exosomal luminal release exists in cultured biliary epithelial monolayers and in the murine biliary tract, a TLR4-dependent increase in luminal release of epithelial exosomes was detected following C. parvum infection. Release of exosomes to the extracellular environment involves fusion of MVBs with the plasma membrane. Intracellular trafficking and fusion of compartments classically require small GTPases of the Rab family [2]. Further supporting this concept and consistent with results from previous reports on induced exosome release [17], we detected a significant co-localization of Rab11 and Rab27b with MVBs in the cytoplasm of epithelial cells following C. parvum infection. Besides Rab11 and Rab27b, we identified that SNAP23 may regulate the fusion of MVBs with the plasma membrane required for exosome release. C. parvum infection increases the protein content of total SNAP23 and enhances phosphorylation of SNAP23 in infected cells. Intriguingly, activation of TLR4 may contribute to both events: TLR4 signaling increases SNAP23 protein expression through modulation of let-7-mediated gene regulation, and stimulates SNAP23 phosphorylation through activation of IKK2. We previously demonstrated TLR4/NF-κB-dependent downregulation of let-7 family miRNAs in C. parvum-infected biliary epithelial cells [27], [41]. Targeting of the 3′UTR of SNAP23 by let-7 family miRNAs resulted in translational suppression. Functional manipulation of let-7i caused reciprocal alterations in cellular SNAP23 protein content. Indeed, overexpression of let-7 miRNA members attenuated C. parvum-induced SNAP23 expression and exosome release from the biliary epithelium. In addition, C. parvum-induced phosphorylation of SNAP23 is dependent on TLR4 signaling. Knockdown of TLR4 blocked C. parvum-induced phosphorylation of SNAP23 in infected cells and, consequently, exosome release into the supernatants. Our data were supported by a recent report showing that SNAP23 is a substrate for IKK2 [37]. Phosphorylation of SNAP23 at Ser120 and Ser95 by IKK2 stimulates fusion of intracellular vesicles to the cell membrane, promoting exocytosis [37]. The next challenge will be to examine in greater depth the involvement of TLR4 signaling in C. parvum-induced co-localization of Rab11 and Rab27b with MVBs, and to determine the magnitudes of miRNA-mediated SNAP23 expression versus its phosphorylation in TLR4-regulated exosome release.
Another key point from this study is that released exosomes shuttle antimicrobial peptides and display anti-C. parvum capacity, thus contributing to mucosal anti-C. parvum defense. It has been reported that antimicrobial peptides, such as HBD2 and LL-37, are present in breast milk and provide protection for maternal breast tissue and the developing digestive tracts of newborns [46]. Additionally, antimicrobial peptides were identified in cervical-vaginal fluid, suggesting their involvement in extracellular immunology [47]. Notably, exosomes have been reported in breast milk and cervical-vaginal fluid [1], [2]. The anti-C. parvum activity of HBD2 was previously reported in intestinal and biliary cryptosporidiosis [19], [45]. Here, we identified shuttling of both HBD2 and LL-37 in the apical exosomes released from H69 monolayers. An increase in HBD2 and LL-37 content was detected in the apical exosomes from infected cells, and inhibition of TLR4 signaling decreased their exosomal content. Because these exosomes display a protein profile different from the whole cell lysate, targeting of cytoplasmic proteins to the MVBs within epithelial cells, such as HBD2 and LL-37, must be controlled through highly selective and regulated mechanisms. However, how TLR4 signaling regulates targeting of antimicrobial peptides to MVBs in epithelial cells remains largely unknown. One additional unanswered question is whether shuttling of MHC proteins in the apical exosomes contributes to antigen presentation.
Anti-C. parvum activity of apical exosomes released from the epithelium may involve direct binding to the C. parvum sporozoite surface. Morphologically, direct binding of exosomes to the C. parvum sporozoite surface was evident by scanning and transmission EM observations. Confocal analysis with exosomal labeling confirmed the delivery of exosomal content to C. parvum sporozoites after incubation. It is not clear what mechanisms are responsible for this exosome binding/targeting. The infection process of epithelial cells by C. parvum is initiated by the attachment of the parasite to the plasma membrane of host epithelial cells [25]. Unidentified specific molecules on the surface of both epithelial cells and C. parvum sporozoites mediate this attachment process [43]. Gal/GalNAc epitopes of glycoproteins on the epithelial apical membrane and Gal/GalNAc-specific sporozoite surface lectins are involved in the mechanism(s) of C. parvum attachment to intestinal and biliary epithelial cells [43], [44]. As expected, treatment of exosomes with Gal/GalNAc-specific PNA lectin diminished their anti-C. parvum activity, suggesting that these molecules may be involved, at least partially, in exosomal binding to the C. parvum sporozoite surface. The lifecycle of C. parvum, both in vitro and in vivo, has extracellular stages (i.e., sporozoites, merozoites, and microgametocytes) [25], and they are vulnerable to exosomal binding/targeting. Therefore, luminal release of exosomes from epithelial cells may represent an important element of TLR-mediated mucosal anti-C. parvum defense. Surprisingly, it appears that released exosomes have no significant effect on the viability of the K12 strain E. coli. One possibility is that the membrane structure of E. coli may not favor direct exosomal binding.
In addition, increased exosomal release and exosome-associated anti-C. parvum activity were detected in biliary epithelial cells after LPS stimulation. Therefore, TLR4-mediated exosome release may be relevant to innate mucosal immunity in general. Overall, our data suggest that activation of TLR4 signaling stimulates the biogenesis and luminal release of antimicrobial peptide-shuttling exosomes and contributes to gastrointestinal mucosal anti-C. parvum defense. Such a process may be explored for therapeutic intervention. Luminal release of exosomes may also mediate the function of distant cells along the gastrointestinal tract, or regulate the homeostasis of gut microbiota, through delivery signaling molecules.
This study was carried out in strict accordance with the recommendations in the Guide for the Care and Use of Laboratory Animals of the National Institutes of Health under the Assurance of Compliance Number A3348-01. All animal experiments were done in accordance with procedures (protocol number # 0868) approved by the Institutional Animal Care and Use Committee of the Creighton University School of Medicine. All surgery was performed under ketamine and xylazine anesthesia, and all efforts were made to minimize suffering.
C. parvum oocysts of the Iowa strain were purchased from a commercial source (Bunch Grass Farm, Deary, ID). Before infecting cells, oocysts were excysted to release infective sporozoites as previously described [19]. H69 cells (a gift from Dr. D. Jefferson, Tufts University,) are SV40-transformed normal human biliary epithelial cells originally derived from normal liver harvested for transplant [19], [27]. 603B cells are immortalized normal mouse biliary epithelial cells (a gift from Dr. Y. Ueno, Tohoku University School of Medicine, Japan). Murine intestinal epithelial cell line (IEC4.1) was a kind gift from Dr. Pingchang Yang (McMaster University, Hamilton, Canada).
TLR4-DN mutant was kindly provided by Dr. M. F. Smith (University of Virginia) and MyD88-DN was a gift from Prof. J. Tschopp (University of Lausanne, Switzerland). Cells stably expressing TLR4-DN or MyD88-DN were established as previously reported [19]. Primers used to amplify the open reading frame of human SNAP23 (NM_130798) were: 5′-GTCATGGATAATCTGTCATCAGAAGAAAT-3′ (forward) and 5′-TTAGCTGTCAATGAGTTTCTTTGCTC-3′ (reverse). PCR products were cloned into pcDNA3.3-TOPO vector (Invitrogen). Primers used to amplify the open reading frame of human beta-defensin 2 (NM_004942) and cathelicidin-37 (NM_004345) were: 5′-CAAGCTTCGATGAGGGTCTTGTATCTCCTCTTCTC-3′ (beta-defensin 2 forward) and 5′-CGGATCCTGGCTTTTTGCAGCATTTTG-3′ (beta-defensin 2 reverse) and 5′-CAAGCTTCGGCTATGGGGACCATGAAGACCC-3′ (cathelicidin-37 forward) and 5′-CGGATCCGGACTCTGTCCTGGGTACAAGATTC-3′ (cathelicidin-37 reverse). PCR products were cloned into the HindIII and BamH I sites of the pEGFPN3 vector (Invitrogen). WGA and PNA lectins were purchased from Sigma.
Models of human biliary cryptosporidiosis using H69 and 603B cells were employed as previously described [19], [27], [28], [41], [43]. Infection was done in a culture medium consisting of DMEM-F12, 100 U/ml penicillin, 100 µg/ml streptomycin, and freshly excysted sporozoites (from oocysts in a 3∶1 ratio with host cells). All excysted sporzoite preparations were tested to exclude LPS contamination using the Limulus Amebocyte Lysate (LAL) gel formation test [48]. Inactivated organisms (treated at 65°C for 30 min) were used for the non-infected controls and experiments were performed in triplicate.
For in vivo biliary infection, C57BL/6 wild-type mice and TLR4-deficient (C57BL/10ScNJ; Tlr4lps-del/Tlr4lps-del genotype) mice were obtained from Jackson Laboratories. C. parvum oocysts were directly injected into the gallbladder of C57BL/6J or TLR4-deficient mice, as previously reported [32], [33]. C. parvum infection and exosome release in the intrahepatic bile ducts were observed one week post-injection. Five animals from each group were sacrificed and livers were obtained for immunohistochemistry and electron microscopy.
Real-time PCR, immunofluorescence microscopy, and immunohistochemistry were used to assay C. parvum infection as previously reported [17], [27], [28], [41], [43]. Primers specific for C. parvum 18s ribosomal RNA (forward: 5′-TAGAGATTGGAGGTTGTTCCT-3′, and reverse: 5′-CTCCACCAACTAAGAACGGCC-3′) were used to amplify the cDNA specific to the parasite. For immunofluorescence microscopy, cells were fixed with 2% paraformaldehyde and incubated with a polyclonal antibody against C. parvum [19], followed by anti-rabbit FITC-conjugated secondary antibody and co-stained with 4,6-diamidino-2-phenylindole (DAPI) (Molecular Probes). The liver tissues were stained with H&E, and parasite burden in the biliary tree was performed by counting all intracellular parasite stages in identified bile ducts, as previously reported [35].
Supernatant medium from the apical reservoir of the Percoll inserts was collected at the indicated time points after infection. Medium was then harvested and centrifuged at 1000 rpm for 10 min to eliminate cells and again spun at 10,000 rpm for 30 min, followed by filtration through 0.22 µm filter to remove cell debris. Exosomes were pelleted by ultracentrifugation (Beckman Ti70 rotor) at 44,000 rpm for 70 min and precipitated using exosome precipitation solution (Exo-Quick; System Bioscience) following the manufacturer's instructions. Exosomes were further quantified by the NTA analysis using a Nanosight (model NS500), as previously reported [30], [31].
Cell cultures were fixed with 4% paraformaldehyde and incubated with antibodies (Santa Cruz) to SNAP23, CD63, Rab11, and Rab27, followed by treatment with Alexa Fluor 594–conjugated anti-rabbit secondary antibody (1∶100; Invitrogen). Cell cultures were then mounted in Slow Fade anti-fade reagent with DAPI (Molecular Probes). Images were captured using an inverted fluorescence microscope TE2000-E (Nikon).
EM was performed as previously described [24]. Briefly, for transmission EM, exosome pellets were resuspended in 2.5% glutaraldehyde, embedded with a mixture of 4% uranyl acetate and 2% methylcellulose (1∶9 ratio), and observed with a JEOL 1400 electron microscope (JEOL USA). For scanning EM, isolated exosomes were fixed immediately in 2.5% glutaraldehyde, dehydrated, dried in a critical point drying device, sputter coated, and examined with a Hitachi S-4700 microscope (Hitachi). For immunogold analysis, exosomes were fixed in 4% paraformaldehyde, blocked with 10% FCS-PBS for 20 min, and incubated overnight at 4°C with antibodies to CD63 (Santa Cruz), ICAM-1 (Santa Cruz), LL-37 (Hycult Biotech), or HBD2 (Alpha Diagnostic). After incubation with the secondary antibodies, samples were labeled with protein A-10-nm gold, embedded, and observed with a JEOL 1400 electron microscope. Transmission EM of mouse liver tissues was performed as previously reported [24].
siRNAs to SNAP23, LL-37 and HBD2, and negative control oligos (Dharmacon) were used at a concentration of 10 nM and transfected with Lipofectamine RNAimax according to the manufacturer's protocol (Invitrogen). Anti-miRs and miRNA precursors (Ambion) were used to manipulate miRNA function in cells, as previously reported [27], [28], [41].
Complementary 39 bp for human and 33 bp for mouse DNA oligonucleotides containing the putative let-7 family miRNAs target site within the 3′UTR of human SNAP23 (Human-Sense: 5′-ctag ACATGAATTCAGATTTACCTCAATGCTAAGAATTA-3′; Human-antisense:5′-agct TAATTCTTAGCATTGAGGTAAATCTGAATTCATGT-3′; Mouse-Sense: 5′-ctag CATGTGAATTCAGATTTACCTCAATACTA-3′; Mouse-antisense:5′-agct TAGTATTGAGGTAAATCTGAATTCACATG-3′) were cloned into the multiple cloning site of the pMIR-REPORT Luciferase vector (Ambion). Another pMIR-REPORT Luciferase construct containing mutant 3′UTR (CCTC to GGAG for human; ACCT to TGGA for mouse) was also generated as a control. Transfection and assessment of luciferase activity were performed as previously reported [27], [28], [41].
The mRNA sequence data for genes described in this study can be found in the NCBI under the following accession numbers: Homo sapiens CD63 (NM_001257389), Mus musculus CD63 (NM_001042580); Homo sapiens TLR4 (NM_001257389), Mus musculus TLR4 (NM_025817); Homo sapiens Rab27b (NM_004163), Mus musculus Rab27b (NM_030554); Homo sapiens MyD88 (NM_001172566), Mus musculus MyD88 (NM_010851); Homo sapiens IKK2 (NM_001190720), Mus musculus IKK2 (NM_001159774); Homo sapiens Rab11 (NM_014904), Mus musculus Rab11 (NM_017382); Homo sapiens SNAP23 (NM_003825), Mus musculus SNAP23 (NM_001177792); Homo sapiens let-7i (NR_029661), Mus musculus let-7i (NR_029527); Homo sapiens miR-98 (NR_029513), Mus musculus miR-98 (NR_029753); Homo sapiens HBD2 (NM_004942); Homo sapiens LL-37 (NM_004345), Mus musculus LL-37 (NM_009921).
|
10.1371/journal.pntd.0003402 | Golgi-Located NTPDase1 of Leishmania major Is Required for Lipophosphoglycan Elongation and Normal Lesion Development whereas Secreted NTPDase2 Is Dispensable for Virulence | Parasitic protozoa, such as Leishmania species, are thought to express a number of surface and secreted nucleoside triphosphate diphosphohydrolases (NTPDases) which hydrolyze a broad range of nucleoside tri- and diphosphates. However, the functional significance of NTPDases in parasite virulence is poorly defined. The Leishmania major genome was found to contain two putative NTPDases, termed LmNTPDase1 and 2, with predicted NTPDase catalytic domains and either an N-terminal signal sequence and/or transmembrane domain, respectively. Expression of both proteins as C-terminal GFP fusion proteins revealed that LmNTPDase1 was exclusively targeted to the Golgi apparatus, while LmNTPDase2 was predominantly secreted. An L. major LmNTPDase1 null mutant displayed increased sensitivity to serum complement lysis and exhibited a lag in lesion development when infections in susceptible BALB/c mice were initiated with promastigotes, but not with the obligate intracellular amastigote stage. This phenotype is characteristic of L. major strains lacking lipophosphoglycan (LPG), the major surface glycoconjugate of promastigote stages. Biochemical studies showed that the L. major NTPDase1 null mutant synthesized normal levels of LPG that was structurally identical to wild type LPG, with the exception of having shorter phosphoglycan chains. These data suggest that the Golgi-localized NTPase1 is involved in regulating the normal sugar-nucleotide dependent elongation of LPG and assembly of protective surface glycocalyx. In contrast, deletion of the gene encoding LmNTPDase2 had no measurable impact on parasite virulence in BALB/c mice. These data suggest that the Leishmania major NTPDase enzymes have potentially important roles in the insect stage, but only play a transient or non-major role in pathogenesis in the mammalian host.
| Nucleoside triphosphate diphosphohydrolases (NTPDases) are a family of enzymes expressed in many eukaryotes, ranging from single-celled parasites to mammals. In mammals, NTPDases can have an immunomodulatory role, while in pathogenic protists cell-surface and secreted NTPDases are thought to be important virulence factors, although this has never been explicitly tested. In this study we have investigated the function of two NTPDases, termed LmNTPDase1 and LmNTPDase2, in Leishmania major parasites. We show that LmNTPDase 1 and LmNTPDase 2 are differentially targeted to the Golgi apparatus and secreted, respectively. A Leishmania major mutant lacking the Golgi LmNTPDase1 exhibited a delayed capacity to induce lesions in susceptible mice when promastigote (insect) stages were used to initiate infection, but not when amastigote (mammalian-infective) stages were used. Loss of promastigote infectivity in the LmNTPDase1 null mutant was associated with the synthesis and surface expression of lipophosphoglycan (LPG), with shorter glycan chains and increased sensitivity to complement-mediated lysis. In contrast, a null mutant lacking the secreted LmNTPDase2 did not exhibit any difference in virulence. Our results suggest that Leishmania major NTPDases have specific roles in regulating Golgi glycosylation pathways, and nucleoside salvage pathways in the insect stages, but do not appear to be required for virulence of the mammalian-infective stages.
| Leishmania parasites cause a spectrum of diseases in humans, ranging from localized cutaneous lesions to disseminated mucocutaneous and lethal visceral infections. It is estimated that 1.5 to 2 million new cases of leishmaniasis occur annually and that more than 350 million people are at risk worldwide. Current first-line drug treatments are suboptimal due to high toxicity, cost, requirement for hospitalization and/or the emergence of drug-resistant strains, highlighting the need for the development of more effective therapeutics [1]. Leishmania parasites develop as extracellular promastigote stages in the digestive tract of the sandfly vector [2]. Following injection into the mammalian host during a sandfly bloodmeal, promastigotes are phagocytosed by a range of host cells (neutrophils, dendritic cells and macrophages) before differentiating to obligate intracellular amastigote stages that primarily proliferate within the phagolysosome compartment of macrophages. A number of surface molecules, including an abundant lipophosphoglycan (LPG) and several GPI-anchored glycoproteins, have been shown to be important for promastigote survival during these initial stages of infection [3]. In particular, LPG is thought to form a continuous surface glycocalyx that protects the promastigote stages of most Leishmania species from complement-mediated lysis and macrophage-induced oxidative stress during phagocytosis [3]–[5]. However, expression of LPG is down-regulated in amastigote stages and neither LPG nor GPI-anchored proteins are required for the long term growth and survival of this stage in macrophages. The potential role of other promastigote and amastigote secreted and surface proteins in the initiation and establishment of infection is less well defined.
A number of protozoan parasites have been shown to express nucleoside triphosphate diphosphohydrolase activities on their cell surface or in the extracellular milieu [6]–[9], and it has been suggested that hydrolysis of nucleotides may play a role in parasite pathogenesis [10]–[12]. Nucleoside triphosphate diphosphohydrolases (NTPDases, CD39_GDA1 protein superfamily) are a family of enzymes defined by the presence of five apyrase conserved regions (ACRs) and the ability to hydrolyze a wide range of nucleoside tri- and di-phosphates [13]. In mammals, surface-expressed NTPDases function in inflammation and immunity, vascular hemostasis and purine salvage [14], while in the intracellular bacterial pathogen, Legionella pneumophila, a secreted NTPDase is required for full virulence in a mouse model of disease [15], [16]. In Leishmania species, enzyme activity consistent with the presence of one or more surface-located NTPDases has been observed in both L. amazonensis and L. tropica, two species responsible for cutaneous leishmaniasis [17]–[19]. A number of lines of indirect evidence suggest that this surface NTPDase activity is important for virulence in the mammalian host. Specifically, surface NTPDase activity is elevated in virulent Leishmania strains and in the intracellular amastigote form of the parasite [17]–[19]; inhibition of surface NTPDase activity with chromium (III) adenosine 5′-triphosphate complex, reduced promastigote attachment and entry into mouse macrophages [20]; treatment of parasites with an antibody to the human NTPDase CD39 also reduced the interaction of Leishmania with mouse macrophages [19]; finally, polyclonal antibodies raised against synthetic peptides derived from the amino acid sequences of a putative L. braziliensis NTPDase caused significant cytotoxicity in cultured L. braziliensis promastigotes [21]. While these studies suggest roles for NTPDases in parasite nutrition, surface/secreted NTPDases could also contribute to pathogenesis by inducing host cell purinergic receptors. Purinergic receptors are upregulated in macrophages infected with L. amazonensis and these receptors display increased sensitivity to activation by nucleoside triphosphates (NTPs). As changes in the levels of extracellular NTPs and NDPs have been shown to alter purinergic receptor activity and the immune response [22], [23], it has been speculated that hydrolysis of host nucleotides by parasite ecto-NTPDases may restrict the immune response and facilitate parasite proliferation.
While these studies suggest NTPDases may function in Leishmania virulence and/or be essential for normal growth and development, they have relied heavily on techniques such as anti-NTPDase antibodies and/or chemical inhibition of enzyme activity to investigate the role of NTPDases in host-parasite interaction. Definitive genetic evidence of a relationship between a parasite NTPDase and parasite virulence is lacking. In this study, we show that L. major encodes two NTPDases, termed LmNTPDase1 and LmNTPDase2 (abbreviated to NTPD1 and NTPD2), and we generate null mutants in order to investigate their function during infection of mammalian cells. Our findings suggest that NTPD1 is primarily located to the Golgi apparatus, and plays an important role in regulating both the maturation of surface LPG and the capacity of L. major promastigotes to initially establish lesions. In contrast, NTPD2 was secreted, and was not required for lesion development, suggesting that its primary role is in the sandfly vector.
Use of mice in this study was approved by the Institutional Animal Care and Use Committee of the University of Melbourne (ethics number 1212647.1). All animal experiments were performed in accordance with the Australian National Health Medical Research council guidelines (Australian code of practice for the care and use of animals for scientific purposes, 8th Edition, 2013, ISBN: 1864965975).
Putative NTPDases were identified by BLAST [24] searching of the available Leishmania genomes, with subsequent manual identification of the conserved ACRs [25], [26]. Protein sequence alignments were performed using ClustalW [27], [28]. SMART [29], [30] was used to identify motifs within the protein sequences.
L. major substrain MHOM/SU/73/5-ASKH was used to create all mutant and transfected lines. Parasites were routinely cultured as axenic promastigotes in Medium-199 (M199, Gibco, Invitrogen, Australia) supplemented with 10% heat-inactivated foetal bovine serum (FBS, Invitrogen) at 27°C or, prior to mouse infection and LPG purification, in SDM-79 medium supplemented with 10% FBS. G418 (Invitrogen, 100 µg mL−1) or nourseothricin (Werner BioAgents, Germany, 100 µg mL−1) was used as appropriate to maintain selection pressure on parasites transfected with pXGFP+-derived plasmids or pIR1SAT-derived and pXGSAT-derived plasmids, while puromycin (Invitrogen, 20 µg mL−1), hygromycin (Boehringer Mannheim, 100 µg mL−1) and bleocin (Calbiochem, 10 µg mL−1) were used to select transformants during mutagenesis. Lesion amastigotes were isolated by disrupting murine lesions (diameter 5–10 mm) by passage through a 70 µm plastic sieve, followed by passage through a 27 G needle to lyse macrophages and release parasites [31]. Cell debris was removed by slow speed centrifugation (50×g, 10 min, 4°C) and the supernatant centrifuged (2000×g, 10 min, 4°C) to collect amastigotes. Amastigotes were washed once in PBS and counted using a haemocytometer prior to use in mouse infections.
Primer sequences used in genetic manipulation are detailed in supporting information (S1 Table). L. major NTPDase null mutants were created via sequential homologous gene replacement in a manner similar to that previously described [32], [33]. All L. major PCR products described below were obtained by amplification from genomic DNA. To delete ntpd1, an 854 bp 5′ untranslated region (UTR) containing a 5′ Asp718 site and a 3′ XhoI site was amplified, and a 805 bp 3′ UTR region containing a 5′ BamHI and a 3′ SacI site was amplified. These products were then sequentially cloned into the pBluescript II SK vector (Stratagene, CA, USA). Puromycin or hygromycin resistance cassettes were then excised from pXG-PAC and pXG-HYG [34] respectively and cloned into the XhoI/BamHI sites. To functionally delete ntpd2 a 688 bp fragment of the 5′ gene end was amplified with a 5′HindIII site and a 3′ BamHI/EcoRI/linker region, and an 1156 bp 3′ UTR region containing a 5′ BamHI/EcoRI/linker region and 3′ NotI site was amplified. An overlap PCR was then performed using these PCR products as template and the resultant product cloned into the HindIII/NotI sites of the pBluescript II SK vector (Stratagene, CA, USA). Puromycin and bleocin resistance cassettes were excised from pXG-PAC and pXG-PHLEO [34] respectively using BamHI and EcoRI, and cloned into the engineered BamHI/EcoRI sites. Deletion mutant constructs were verified by restriction digest profiles and DNA sequencing. Targeting constructs were then excised by KpnI/SapI (ntpd1) or HindIII/NotI (ntpd2) digest, gel purified and 5 µg of each sequentially electroporated into L. major as described previously [35]. Clonal transfectants resistant to both selection drugs were chosen and deletion of the target gene and integration of resistance cassettes confirmed via triplicate PCR. To generate the pIR1SAT-ntpd1 construct used in chromosomal complementation, full-length ntpd1 was excised from pXG-LmNTPDase1-GFP using BamHI and cloned into the BglII site of the pIR1SAT vector [36], [37]. SwaI digest was used to excise 5 µg of targeting DNA for electroporation into L. major Δntpd1. Clonal transformants were selected on basis of resistance to nourseothricin and incorporation into the ssu locus confirmed by PCR. To create the LmNTPDase-GFP fusion proteins, full length ntpd genes were individually cloned into pXG-GFP+ [38]. To express the LPG1-mCherry fusion protein, mCherry from pEGFP-mCherry-N1 [39] was amplified with a 5′SmaI/BglII site and 3′BamHI site and cloned into the SmaI/BamHI sites of pXGSAT, generating pXGSAT-mCherry. lpg1 [40] was amplified and then cloned into SmaI/BglII of pXGSAT-mCherry, creating pXG-LPG1-mCherry. The resulting constructs were confirmed via DNA sequencing and electroporated into wild type L. major as previously described [35].
Promastigotes were incubated in serum-free media for 24 hours before harvesting by high speed centrifugation (16000×g, 5 min). Supernatants were filtered through a 0.45 µM filter to remove intact parasites before supernatant proteins were precipitated with 10% trichloroacetic acid. The pellet and supernatant fractions were analyzed by standard SDS-PAGE and immunoblotting techniques, with LmNTPDase-GFP fusion proteins detected using anti-GFP antibody (clones 7.1 and 13.1, Roche, Germany) at 1∶1000 dilution. For microscopy studies live cells were immobilized on poly-L-lysine coated coverslips. Cells were visualized and images acquired using a Deltavision Elite fluorescent microscope and SoftWorx software.
Stationary phase promastigotes grown in SDM-79 supplemented with 10% FBS were harvested by centrifugation and LPG extracted from de-lipidated cells and purified using octyl-Sepharose chromatography, as described previously [41], [42]. The molecular weight of LPG was assessed via SDS-PAGE and silver staining using standard techniques. LPG was depolymerised with 40 mM trifluoroacetic acid (8 min, 100°C) and dephosphorylated with calf intestinal alkaline phosphatase. The repeat units were desalted by passage over a small column of AG 50-X12 (H+) over AG 4-X4 (OH-) (200 µL of each resin, Biorad) and chromatographed by high performance anion-exchange chromatography (HPAEC). The HPAEC system was equipped with a Dionex GP-50 gradient pump, a Carbo Pac PA-1 column (4×250 mm), with a PA-1 guard column and an ED50 integrated pulsed amperometric detector. The system was controlled and data analyzed by Chromeleon version 6.50 software (DIONEX). The eluents used in the system were 75 mM NaOH (E1) and 75 mM NaOH in 250 mM NaOAc (E2). Elution was performed by the following gradient: T0 = 0% (v/v) E2; T5 = 0% (v/v) E2; T40 = 100% (v/v) E2, T60 = 100% (v/v) E2, at a flow rate of 0.6 mL/minute. The phosphatidylinositol moiety of purified LPG was released by nitrous acid deamination (0.25 M sodium nitrite in 0.05 M sodium acetate buffer, pH 4.0; incubated at 40°C for 2.5 h), recovered by partitioning into water-saturated 1-butanol and analyzed using liquid chromatography mass spectrometry (LC/MS).
Washed stationary phase parasites (107 mL−1) were incubated with varying concentrations of peanut agglutinin (PNA) in PBS with 1% bovine serum albumin for 30 minutes at room temperature, and non-agglutinated parasites were counted using a haemocytometer (adapted from [43]).
Serum sensitivity assays were performed in a similar manner to those previously described [5]. Stationary phase promastigotes were washed and resuspended in PBS (107 cells in 500 µL PBS with 1 µg mL−1 propidium iodide) and incubated with varying concentrations of human sera for 30 minutes. Fluorescence (indicating cell lysis) was then measured by flow cytometry.
Virulence in mice was assessed using the tail base model of cutaneous leishmaniasis, as described previously [31]. Female BALB/c mice (6–8 week old, age-matched) were injected subcutaneously at the tail base. Lesion size was assessed weekly and scored 0–4, as described previously [44]. All parasite cell lines were passaged previously in mice to ensure no loss of virulence unrelated to the known genetic mutations. Parasites were re-isolated from mice as described in the “Parasite strains and culture conditions” section.
Unpaired, two-tailed t-tests were performed using Prism GraphPad software (version 6) and a P value less than 0.05 was considered significant. The exception was when more than two parasite strains were compared, in which case a two-way ANOVA, also using Prism GraphPad software, was performed to simultaneously compare the three different groups. A P value less than 0.05 was considered significant when comparing the differences between the three groups.
The L. major genome contains two putative NTPDase genes (LmjF15.0030 and LmjF10.0170), which are predicted to encode proteins with five ACR domains, the defining feature of all prokaryotic and eukaryotic NTPDase [45]. These genes are conserved amongst all sequenced Leishmania species, with homologues present in L. infantum, L. braziliensis, L. donovani and L. mexicana [46]. Importantly, a number of residues necessary for enzymatic activity of either CD39 or NTPDase3, the two best characterized mammalian NTPDases [47] are absolutely conserved within the Leishmania proteins (Fig. 1A). Using the nomenclature that we previously proposed for the parasite NTPDases [25], we refer to LmjF15.0030 as LmNTPDase1, and Lmj10.0170 as LmNTPDase2 (abbreviated to NTPD1 and NTPD2 in this study for succinctness). Homologues for NTPD1 and NTPD2 are present in T. brucei, but only NTPD2 exists in T. cruzi (Fig. 1B). Phylogenetic comparison with NTPDases found in other protozoa, mammals and yeast indicates that the trypanosomatid NTPDases are most closely related to mammalian NTPDase5 and NTPDase6, which are usually located intracellularly but can undergo secretion, and to the Golgi-located yeast NTPDase GDA1. Interestingly, the trypanosomatid NTPDases seem evolutionarily distinct from the NTPDases found in a range of apicomplexan parasites and Trichomonas protozoa (Fig. 1B), perhaps indicating divergent functions.
ntpd1 encodes for a protein (432 amino acids) with a putative N-terminal transmembrane domain (residues 17–36), while ntpd2 encodes for a longer protein (685 amino acids) with an N-terminal signal sequence (residues 1–20). To establish whether the two L. major NTPDases are secreted or targeted to the cell surface/intracellular compartment, wild type parasites were transfected with plasmids encoding NTPD1 and NTPD2 as fusion proteins containing C-terminal GFP. Western blot analysis of parasite cell pellets and culture supernatant showed that full-length proteins were expressed in each parasite line (Fig. 2A). Interestingly, while the NTPD1-GFP fusion protein was exclusively associated with the cell pellet, NTPD2-GFP fusion protein was secreted (Fig. 2A). The absence of detectable NTPD1 in the supernatant indicated that the presence of NTPD2 in the culture supernatant was not due to parasite lysis during culture, but represented active secretion (Fig. 2A). Furthermore, live cell fluorescence microscopy of promastigotes expressing NTPD2-GFP did not detect significant cell surface or intracellular fluorescence, consistent with NTPD2 being primarily a secreted protein. Interestingly, Western blot analysis detected a small pool of NTPD2-GFP within the cell pellet fraction (Fig. 2A), which is likely to represent newly synthesized NTPDase in transit to the cell surface, but below the level of detection of fluorescence microscopy. Because of the low abundance of this intracellular pool we can also not discount the possibility that NTPDase2 is directed to other intracellular organelles, such as the lysosome. In contrast, L. major promastigotes expressing NTPD1-GFP displayed a single, highly fluorescent punctate stain, at the anterior end of the parasite, proximal to the kinetoplast/flagellar pocket (Fig. 2B). This location is highly characteristic of the Golgi apparatus. L. major parasites expressing NTPD1-GFP were therefore co-transfected with a second plasmid encoding the known Golgi protein LPG1 [40] fused to mCherry. Parasites expressing both NTPD1-GFP and the Golgi marker displayed overlapping fluorescence indicative of co-localization (Fig. 2B). This co-localization was not seen in parasites transfected with either mCherry or GFP (both of which display cytoplasmic localization), indicating that NTPD1 is primarily located in the Golgi apparatus. Although yeast NTPDases have been localized to the Golgi apparatus [48], [49], this is the first time a parasite NTPDase has been identified in the Golgi apparatus, rather than being secreted from the parasite or located on the cell surface.
Previous transcript profiling studies have suggested that ntpd1 and ntpd2 are constitutively transcribed in both major developmental stages [50], [51], providing little information on potential stage-specific differences in function. To investigate the function of these enzymes we generated null mutants for each NTPDase gene, by sequential replacement of the two chromosomal alleles with drug resistance cassettes. ntpd1 was replaced with hygromycin and puromycin resistance cassettes, with gene deletion and correct integration of the resistance cassettes confirmed by triplicate PCR (S1 Fig.), demonstrating that ntpd1 is not essential under rich culture conditions. In a similar manner ntpd2 was replaced with puromycin and bleomycin cassettes, with PCR confirmation performed in triplicate (S1 Fig.), indicating that ntpd2 is also not essential in vitro. Both strains grew normally in routine culture medium.
To investigate whether LmNTPDase1 or 2 is required for virulence in the mammalian host, we tested the ability of L. major Δntpd1 and Δntpd2 to induce lesions in susceptible BALB/c mice. Promastigote stages of the L. major NTPD1 null mutant exhibited a marked and highly reproducible delay in lesion development. This delay was largely abrogated by complementation of the null mutant by insertion of a full-length ntpd1 gene in the highly-transcribed ribosomal ssu locus [52]. Interestingly, no delay in lesion development was observed when amastigote stages of the NTPD1 null mutant were used to initiate the infection (Fig. 3A–C). Together, these studies demonstrate that NTPD1 is required during the early stages of promastigote infectivity, but has limited function in production of lesions following amastigote infection.
In contrast to the NTPD1 null mutant, the NTPD2 null mutant exhibited a virulence phenotype in BALB/c mice that was indistinguishable from wild type parasites, regardless of whether promastigotes or amastigotes were used to initiate infection (Fig. 3D and 3E). Infections were repeated a number of times and it is possible that these parasites have adapted to loss of NTPD2. Regardless, these results suggest that NTPD2 is not required for virulence in the mammalian host. Lesion development within the mouse reflects both parasite replication and the host response, and our results do not rule out an alteration in parasite replication levels between wild type and the NTPD2 null mutant. However the ability to cause disease, as measured by lesion size, was unchanged between the two strains.
By analogy with the function of the Golgi-located yeast NTPDase, we predicted that NTPD1 may be involved in regulating the recycling of sugar-nucleotides in the Golgi lumen and hence glycosylation pathways [48], [49]. This hypothesis was further supported by the delayed lesion virulence phenotype of the NTPD1 null mutant, which is reminiscent of that seen previously for L. major mutant parasites that lack the major surface glycoconjugate, LPG [5], [53]. While LPG has multiple roles in the sandfly vector, it is only required for the early stages of promastigote infectivity in the mammalian host. LPG is not required for survival or growth of intracellular amastigotes, and LPG mutant parasites that survive the innate immune responses of the mammalian host can subsequently induce normal lesions [4], [5], as observed for the NTPD1 null mutant. To assess whether the L. major NTPD1 null mutant was defective in LPG biosynthesis, the de-lipidated wild type and mutant promastigotes were extracted in 9% 1-butanol and the lipoglycoconjugates purified by octyl-Sepharose chromatography [41]. The NTPD1 null mutant produced comparable levels of LPG as wild type parasites (Fig. 4A). As expected, both LPG preparations were visualized as smears on SDS-PAGE gels, reflecting heterogeneity in the length of the phosphoglycan chains that comprise the major portion of the LPG [42]. However, the LPG isolated from null mutant promastigotes reproducibly exhibited a lower average molecular weight on the SDS-PAGE gels (Fig. 4A) and eluted later from the octyl-Sepharose column (Fig. 4B), indicating shorter average chain length and/or reduced side chain branching. To distinguish between these possibilities, the LPG prepared from wild type and Δntpd1 promastigotes was depolymerized with mild acid treatment (40 mM TFA, 100°C, 8 min) and dephosphorylated prior to analysis by HPAEC. Both LPG preparations had essentially identical oligosaccharide repeat unit profiles (Fig. 4C). Furthermore, LC/MS analysis of the released PI lipid moieties showed that both wild type and mutant LPG contained identical very long chain (C24:0, C26:0) alkylglycerol moieties. Collectively, these structural analyses suggest that the faster SDS-PAGE mobility of LPG isolated from the NTPD1 null mutant reflects decreased phosphoglycan chain elongation, rather than altered side chain additions or increased hydrophobicity in the lipid anchor.
Expression of shorter LPG chains on the surface of the NTPD1 null mutant would be expected to lead to increased surface binding by the lectin, peanut agglutinin (PNA). PNA binds terminal β-Gal residues in the LPG side chains and intensity of binding is regulated by the abundance of β-Gal side chain, the extent to which these side chains are capped with arabinose and the overall length of the LPG [43]. Paradoxically, promastigotes expressing long LPG chains form surface aggregates in which LPG epitopes become cryptic and therefore bind less PNA. NTPD1 null mutant promastigotes were more effectively agglutinated than wild type promastigotes when harvested at the same stationary growth phase (Fig. 5A). Given that both wild type and mutant produce LPG with essentially identical side chain compositions (Fig. 4C), these results are consistent with the NTPD1 null promastigotes having a defect in LPG elongation.
To assess whether the defect in LPG chain elongation was physiologically significant, stationary phase wild type and NTPD1 null promastigotes were incubated with increasing concentrations of human serum. The complement resistance of L. major promastigotes has previously been shown to be highly dependent on LPG chain length and the formation of a thick protective surface glycocalyx [5]. NTPD1 null mutant promastigotes were significantly more sensitive to serum lysis than wild type parasites (Fig. 5B–D). In particular, FACS analysis of PI-stained parasites, showed ∼2-fold increased sensitivity at 5% serum concentrations (Fig. 5B). Collectively, these results provide strong evidence that loss of Golgi NTPDase results in less efficient elongation of LPG in virulent stationary phase promastigotes, leading to increased susceptibility to complement lysis and a marked delay in lesion development.
The genomes of many parasitic protozoa encode one or more NTPDases, which have been implicated in various host-parasite processes [6]–[9], [19]. However, the function of these enzymes in pathogenesis has not been rigorously defined using genetic approaches. In this study we have defined the subcellular localization and function of two clearly defined NTPDase enzymes in L. major. Both proteins are predicted to contain the five ACR domains that characterize NTPDases and to be constitutively transcribed in the two major life cycle stages. Based on analysis of GFP fusion proteins, we provide evidence that NTPD1 is primarily targeted to the Golgi apparatus, while NTPD2 is secreted into the extracellular milieu. We propose that NTPD1 has an important role in regulating glycosylation pathways in the Golgi apparatus as loss of NTPD1 resulted in a defect in LPG elongation in stationary phase promastigotes. Although the overall decrease in LPG chain length in the NTPD1 null mutant was modest, it was associated with significantly increased sensitivity to complement lysis and a conspicuous delay in lesion development when promastigotes were used to initiate infection. A similar lag in lesion development was not observed when NTPD1 null mutant amastigotes were used to initiate infection, consistent with the defect being associated with a promastigote-specific virulence factor such as LPG. The similarity between the virulence phenotype of the NTPD1 null mutant and previously generated L. major LPG mutants in which assembly of the entire phosphoglycan chain has been disrupted is striking [4], [53], and strongly suggests that LPG chain elongation during stationary phase is both critical for promastigote virulence, and likely to underlie the major function of this glycoconjugate during the early stages of infection in the mammalian host.
S. cerevisiae expresses two NTPDases, GDA1 and YND1, that are targeted to the Golgi apparatus with their catalytic domains orientated into the lumen [48], [49], [54]. These enzymes have been shown to hydrolyze NDP nucleotides to the corresponding NMP nucleotide, which is then used as the counter ion to import sugar nucleotides from the cytoplasm into the Golgi lumen. NTPDase-mediated hydrolysis of NDPs is thus critical for maintaining luminal levels of a range of sugar nucleotides that are used by Golgi glycosyltransferases [55]. In Leishmania, the Golgi apparatus contains enzymes required for the assembly and elongation of complex phosphoglycans on GPI anchor precursors, as well as a number of cell surface and secreted proteophosphoglycans (PPGs). All of these phosphoglycans contain the biosynthetic repeat unit, Galβ1-4Manα1-PO4, which is assembled by sequential transfer of Manα-1phosphate and galactose to the growing phosphoglycan chain by GDP-Man and UDP-Gal-dependent Golgi glycosyltransferases, respectively. The reactions catalyzed by the UDP-Gal dependent galactosyltransferases generate UDP, which would need to be converted to UMP by a NTPDase activity in order to sustain continued import of UDP-Gal into the Golgi lumen (Fig. 6). In contrast, the GDP-Man dependent Man-1-PO4-transferase(s) generate GMP, rather than GDP, and this NMP could be used to drive import of GDP-Man independent of the NTPDase activity. Thus the Golgi NTPDase is likely to be exclusively required for the galactosyltransferase-mediated reactions and not the GDP-Man-dependent Man-1-PO4 reactions. The fact that we see a specific defect in LPG chain elongation, but not in side chain modifications in the NTPDase mutant implies that β1-4-galactosyltransferase involved in assembly of the repeat unit backbone is more sensitive to depletion of UDP-Gal in the Golgi lumen than the β1-3galactosyltransferases that add additional galactose residues to the repeat unit backbone. At present, essentially nothing is known about the mechanisms that regulate LPG elongation, notwithstanding the importance of this process during the differentiation of rapidly dividing promastigotes to non-dividing, hypervirulent metacyclic promastigotes in culture and in the sandfly vector. Our findings raise the possibility that the changes in the availability of sugar nucleotides, either through changes in the activity/expression levels of Golgi membrane transporters or the luminal orientated NTPD1, could play an important role in this respect.
In contrast to NTPD1, deletion of NTPD2 had no measurable impact on the growth of L. major promastigotes in vitro or in vivo. As NTPD2 was secreted into the medium, it is unlikely that the absence of a detectable LPG or virulence phenotype in the NTPD2 mutant reflects redundancy between the two NTPDases. One possibility is that secreted NTPDase2 is primarily required for salvage of extracellular purines. Leishmania are purine auxotrophs but express a number of surface nucleotidases, acid phosphatases, nucleotide/nucleoside/purine base transporters, as well as intracellular enzymes involved in interconverting different purine intermediates [56]. This robust network of redundant purine salvage pathways could account for the absence of a conspicuous phenotype in the NTPD2 null mutant.
A recent study has suggested that L. braziliensis LbNTPDase1 is localized on the cell surface of promastigotes [21], and that opsonization with a polyclonal antibody directed to this protein was cytotoxic. Using this antibody, the authors also suggested that LbNTPDase1 may be additionally targeted to the mitochondria, cytoplasmic vesicles, kinetoplast and nucleus. It is possible that the Leishmania NTPDase1 homologues are targeted to different subcellular localizations in a species-specific manner and perform different functions. Further work to validate the specificity of the LbNTPDase1 polyclonal antibodies and/or determination of tagged proteins would be of interest.
Previous work demonstrated variation in the level of ecto-nucleotidase activity between Leishmania species [57]. Activity in L. major was lower than that observed for L. amazonensis, which was also more virulent in the mouse model used in the study, suggesting that the role of NTPDases in the disease process could differ between species of Leishmania. However, this study did not demonstrate that the observed ecto-nucleotidase activity was linked to ntpd gene expression, and the activity may relate to other enzymes. The same study also utilised Western blot analysis, using polyclonal antibody against T. cruzi NTPDase, to detect a band corresponding to the predicted size of NTPDase1 in L. amazonensis, but failed to identify a similar band in L. major. This may be due to failure of the antibody to recognize the L. major NTPDase, but could also suggest the natural level of expression of NTPDase1 in L. major is lower. However, in light of our findings that LmNTPDase1 localises to the Golgi apparatus, it is unlikely that lower expression of LmNTPDase1 would result in lower ecto-nucleotidase activity of L. major. Future studies taking defined genetic approaches to study NTPDases in other species of Leishmania would be extremely valuable in both defining their function, and in elucidating the value of this class of enzymes as a potential therapeutic target in Leishmania.
It is also important to recognize that a number of studies have implicated general surface-located hydrolysis of ATP, ADP (and sometimes other NTPs and NDPs) in the virulence of both Leishmania and a number of other parasites [18], [19], [58]–[62]. This observed activity has often been assumed to be due to the presence of NTPDases. However, our data raise the possibility that other classes of parasite enzymes are responsible for the observed activity and play a role in pathogenesis themselves. For example, a known NTPDase inhibitor, ARL67156, only inhibits 30% of observed ecto-ATPase activity of T. cruzi [6], suggesting that investigation of other classes of enzymes would also be worthwhile. It may be that a combinatorial approach is required, and that inhibition of two or more surface enzymes could be successful in treating disease.
In conclusion, this work considerably expands our knowledge of the role of Leishmania NTPDases in host-parasite interactions. We show for the first time that parasite NTPDases can be targeted to the Golgi, and play an important role in regulating the assembly of surface virulence factors. Unexpectedly, and notwithstanding previous studies suggesting that secreted NTPDases may have essential roles in purine acquisition, and/or host or parasite purinergic signalling, loss of the secreted NTPD2 had no discernible affect on promastigote or amastigote infectivity in mice. These studies highlight the importance of exploiting genetic approaches whenever possible in investigating the function of these enzymes in host-parasite interactions.
|
10.1371/journal.pgen.1006607 | The Arabidopsis bZIP11 transcription factor links low-energy signalling to auxin-mediated control of primary root growth | Plants have to tightly control their energy homeostasis to ensure survival and fitness under constantly changing environmental conditions. Thus, it is stringently required that energy-consuming stress-adaptation and growth-related processes are dynamically tuned according to the prevailing energy availability. The evolutionary conserved SUCROSE NON-FERMENTING1 RELATED KINASES1 (SnRK1) and the downstream group C/S1 basic leucine zipper (bZIP) transcription factors (TFs) are well-characterised central players in plants’ low-energy management. Nevertheless, mechanistic insights into plant growth control under energy deprived conditions remains largely elusive. In this work, we disclose the novel function of the low-energy activated group S1 bZIP11-related TFs as regulators of auxin-mediated primary root growth. Whereas transgenic gain-of-function approaches of these bZIPs interfere with the activity of the root apical meristem and result in root growth repression, root growth of loss-of-function plants show a pronounced insensitivity to low-energy conditions. Based on ensuing molecular and biochemical analyses, we propose a mechanistic model, in which bZIP11-related TFs gain control over the root meristem by directly activating IAA3/SHY2 transcription. IAA3/SHY2 is a pivotal negative regulator of root growth, which has been demonstrated to efficiently repress transcription of major auxin transport facilitators of the PIN-FORMED (PIN) gene family, thereby restricting polar auxin transport to the root tip and in consequence auxin-driven primary root growth. Taken together, our results disclose the central low-energy activated SnRK1-C/S1-bZIP signalling module as gateway to integrate information on the plant’s energy status into root meristem control, thereby balancing plant growth and cellular energy resources.
| Being in competition for reproductive success, plants use most of their photosynthetically produced energy resources to promote growth. However, under unfavourable environmental conditions plants also need to finance adaptive responses to ensure their survival. For this purpose a growth regulatory system is required to dynamically tune plant growth according to the plants’ prevailing energy status. Here, we characterize crucial components of this system that link plants’ energy management with root growth control. In detail, we demonstrate that a highly homologous group of energy-controlled regulators of the basic leucine zipper (bZIP) transcription factor family redundantly operate under energy deprivation to control expression of a determinant of hormonally-controlled meristematic root growth. By these means these regulators constitute a central hub to integrate detrimental environmental stress conditions, which converge on energy limitation, into plant growth. Understanding the interplay between the plants’ energy homeostasis and growth control are of major importance for future strategies to engineer efficient crop plants.
| Sustaining energy homeostasis is of crucial importance for all living organisms to ensure their fitness and survival. In this respect, especially plants that cannot evade ever-changing environmental conditions due to their sessile lifestyle need to balance their input into highly energy-demanding growth processes according to the prevailing energy supply [1, 2]. For this reason, plants possess an energy management system, which induces catabolic processes and represses anabolism and plant growth under energy-deprived conditions [3, 4]. Central regulators are the evolutionary conserved, low-energy activated SUCROSE NON-FERMENTING1 RELATED KINASES 1 (SnRK1α1 (At3g01090), SnRK1α2 (At3g29160)), which accomplish massive transcriptional and metabolic reprogramming under low-energy stress [3, 5, 6]. Part of the SnRK1 response has been proposed to be exerted by basic leucine zipper (bZIP) transcription factors (TFs) of group S1 (bZIP1 (At5g49450), -2 (At2g18160), -11 (At4g34590), -44 (At1g75390), -53 (At3g62420)) and group C (bZIP9 (At5g24800), -10 (At4g02640), -25 (At3g54620), -63 (At5g28770)) [3, 7, 8]. Due to preferential heterodimerization, these bZIPs constitute the functionally interlinked C/S1 heterodimerization network [9–13]. Recently, SnRK1-mediated in vivo phosphorylation of the group C member bZIP63 has been demonstrated to enhance dimerization with the group S1 member bZIP11 and positions SnRK1 directly upstream of the C/S1 bZIP network [8]. Besides post-translational regulation by the low-energy responsive SnRK1 kinases, transcription of several C and S1 bZIP genes has been found to be energy-controlled. In particular, expression of group S1 bZIP1 and -53 as well as that of group C bZIP9, -25 and -63 is strongly induced by energy deprivation [8, 10, 14]. Moreover, translation of all group S1 members, including the highly homologous bZIP2, -11 and -44, is negatively regulated by SIRT (Sucrose Induced Repression of Translation) [15]. In this context, it was demonstrated that translation of S1 bZIPs is controlled by an evolutionary conserved upstream ORF (uORF), which encodes for a small sucrose control peptide, inhibiting main ORF translation under high sucrose levels [16–18]. In line with their proposed function in low-energy signalling, S1 bZIP translation was found to be strongly de-repressed under energy deprived conditions [19]. Although functional analyses of the SnRK1/bZIP pathway have frequently been performed under pronounced starvation conditions procured by extended night treatment [3], the system has also been found to operate in response to naturally occurring stress situations [7] or progressive energy depletion mediated by low-light cultivation [4].
The C/S1 bZIP network has been implicated in orchestrating low-energy triggered catabolism [10, 20]. This includes, in particular, starvation induced breakdown of branched-chain amino acids to fuel primary metabolism [10]. In addition to their metabolic impact it has been demonstrated that specifically the highly homologous bZIP11-related TFs (bZIP2, -11, -44) quantitatively modulate auxin-responsive gene expression by recruiting the histone acetylation (HAT) machinery to target promoters [21]. In this respect, especially members of the Aux/IAA (AUXIN/INDOLE-3-ACETIC ACID) and GH3 (GRETCHEN HAGEN 3) gene families, which present crucial negative feedback regulators of auxin signalling and homeostasis, were found to be addressed.
Auxin signalling is well-known to play a key role in coordinating root growth and development [22]. Therefore, spatio-temporally controlled auxin transport is required to maintain and promote meristematic activity at the root apical meristem (RAM) or beneath the prospective sites of root organ formation [23]. By modulating expression of auxin transport facilitators of the PIN-FORMED (PIN) protein family, such as that of PIN1 (At1g73590) and PIN3 (At1g70940) [24–26], the phytohormones cytokinin and auxin antagonistically control polar auxin transport (PAT), thereby determining meristem size and consequently growth rates of the primary root [27]. Several preceding studies revealed that this dynamic process is mechanistically accomplished by a key intersection between both hormonal pathways, constituted by the cytokinin-responsive and auxin-labile INDOLE-3-ACETIC ACID INDUCIBLE 3 / SHORT HYPOCOTYL 2 (IAA3/SHY2) (At1g04240) repressor. Importantly, it could be demonstrated that this pivotal regulator directly represses PIN transcription and thereby controls PAT, RAM activity and primary root growth [28–30]. Recently, IAA3/SHY2 expression has been found to be controlled by bZIP11 via its HAT recruitment mechanism [21], suggesting that bZIP11 and closely related TFs constitute a gateway to integrate low-energy related stimuli into auxin-mediated root growth responses [21, 31–33].
In fact, root growth is well-known to show a high phenotypic plasticity in response to energy-demanding stress situations [1]. In particular, the primary root has been found to react rapidly to progressive energy depletion by arresting its growth [3, 34, 35]. The molecular players involved in this crucial adaptive response remain, however, unknown. In this study, we disclose the pivotal role of the low-energy activated bZIP11-related TFs as negative regulators of primary root growth under starvation conditions.
We propose that via controlling the starvation triggered expression of IAA3/SHY2 - a central component of root meristematic activity—the discussed bZIPs repress PIN expression and thereby interfere with PAT and, consequently, auxin-driven root growth. By this elegant mechanism bZIP11-related TFs provide means to adjust primary root growth in accordance to diverse detrimental environmental stimuli that converge on intracellular energy limitation.
Previous reports clearly demonstrated the impact of IAA3/SHY2 in restricting auxin driven meristematic root growth [28–30, 36]. As expression of this important root growth regulator was found to be triggered by starvation stimuli [33] and by the highly homologous, low-energy activated bZIP11-related TFs (bZIP2, -11, and -44) [21], we hypothesised that these bZIPs might interfere with primary root growth by inducing IAA3/SHY2 expression under energy deprivation. In order to address this question, we aimed at monitoring IAA3/SHY2 transcript abundance in bZIP mutant plants. As unfortunately (I) no T-DNA insertion mutants were available for bZIP11 or bZIP2, (II) recent studies disclosed functional redundancy in activating IAA3/SHY2 transcription by all three related bZIPs [21] and (III) individual bZIP knockdown approaches applying artificial microRNA (amiRNA) based techniques [37] were hampered by the high sequence homology between bZIP11 and its closely related factors, we generated an Estradiol (Est)-inducible amiRNA-bZIP2/-11/-44 line (XVE-ami2/11/44) utilizing the well-established XVE-system [38]. This approach enabled both, simultaneous transcript reduction of all three bZIPs and due to its inducibility the analysis of direct TF controlled responses. Moreover, it was straight-forward to circumvent putative lethality of a triple null-mutant. The efficiency and specificity of the transgenic knock-down line was determined in planta by quantitative real-time PCR (q-RT-PCR). By these means an Est-mediated reduction of the corresponding bZIP (bZIP2, -11 and -44) transcripts to roughly 20 to 40% [21] and not of closely related ones, such as those of bZIP1 or bZIP53 could be detected (S1 Fig).
Making use of the knock-down and respective WT plants, we analysed IAA3/SHY2 transcript abundance in presence of Est, in roots at defined time-points of the day, night and extended night applying q-RT-PCR (Fig 1A). Although no significant differences in IAA3/SHY2 expression were observed between the genotypes under a 16 h day / 8 h night growth regime (long day, LD), strong bZIP dependency became apparent under energy deprived conditions provoked by short-term (4 or 8 h) extended night treatment. In fact, IAA3/SHY2 expression, which continuously increased with duration of extended darkness in WT, was significantly alleviated in the multiple bZIP2/11/44 knock-down line. These data suggest, that bZIP activity is crucial to transduce low energy rather than light- or clock-related signals into IAA3 transcription. In order to address whether IAA3 expression is directly controlled by bZIP11-related TFs, we performed ChIP (Chromatin immunoprecipitation) using root material of XVE-bZIP11 plants. After Est-mediated bZIP11 induction, we observed a strong enrichment of precipitated IAA3 promoter fragments compared to the WT control (Fig 1B). In contrast, fragments corresponding to the ACTIN7 (At5g09810) promoter or to IAA3 coding and 3`UTR regions were only marginally enriched. The minor enrichment of coding and 3`UTR sequences was likely attributed to limited ChIP resolution of approximately 1000 bps. These findings, which are in line with previously published in vitro binding studies [39], suggest that bZIP11 is able to directly target the IAA3 promoter, presumably by binding to a cognate G-box cis-element located around -1800 bps apart from the translational start site. An additional ChIP signaI identified in the -600 bps promoter region could be explained by binding to a G-box like (TACGTG) motif. These findings support the view that the bZIPs under investigation interfere with the IAA3 controlled root growth regulatory system under starvation conditions.
Progressive energy deprivation has been found to rapidly result in primary root growth repression, which could be countervailed by exogenous sugars [34]. As bZIP11-related activity was found to be negatively correlated with intracellular sugar levels [16, 17, 19] and required to induce expression of the negative root growth regulator IAA3/SHY2 under starvation conditions, we hypothesized that bZIP11-related factors might control root growth depending on the prevailing sugar availability. To test this assumption, we monitored primary root growth of WT and XVE-ami2/11/44 plants under energy-deprived and high-energy conditions. In fact, energy depletion induced by cultivating the plants on MS medium without sugars, but in presence of the photosynthesis inhibitor DCMU (3-(3,4-dichlorophenyl)-1,1-dimethylurea) resulted in significantly reduced root growth, which could be by-passed by feeding external glucose (S2A Fig). Strikingly, this starvation-triggered root growth repression was much less pronounced in XVE-ami2/11/44 plants compared to WT, demonstrating that the knock-down plants failed to respond appropriately to energy deprivation by reducing root growth. These results could be confirmed using extended darkness as an alternative low-energy condition. Again, root growth was significantly less repressed in two independent XVE-ami2/11/44 knock-down lines compared to WT after 24 hours of extended night (Fig 2A and S2B Fig). However, it has to be noted that after 48 h of prolonged darkness both genotypes comparably showed severe root growth inhibition, which was likely caused by substantial energy exhaustion. Importantly, considering both alternative low-energy approaches, it has to be noted that no differences in primary root growth could be observed between WT and bZIP knock-down plants when cultivated in the presence of exogenous sucrose (S2A Fig) or under LD growth conditions (Fig 2A and S3B Fig). Finally, we also examined the response of iaa3 loss-of-function plants (shy2-24) to energy deprivation provoked by extended darkness (Fig 2B and S2C Fig). As iaa3 mutants revealed a comparably low responsiveness to the low energy situation as the bZIP knock-down lines we concluded that the respective bZIPs largely operate via the crucial root growth regulator IAA3/SHY2 to adjust root growth in accordance to energy availability. However, we cannot rule out at this point that other bZIP targets, particularly other Aux/IAAs might also contribute to a minor extend.
Expression of bZIP11, as well as that of its close homologs bZIP2 and bZIP44 is well-characterised to be repressed by sucrose [16, 17] and induced by low energy conditions [19]. In order to specifically analyse their mechanistic impact on root growth independent from the dark/starvation stimulus, which might affect several physiological plant responses, we made use of transgenic gain-of-function approaches of these bZIPs. Constitutive expression of bZIP11 [20] or its target IAA3/SHY2 [33] has been found to result in severe shoot and root growth repression. Hence, we generated Est-inducible over-expressers of HA-tagged bZIPs, enabling spatio-temporally controlled transgene expression. By this means, stable transgenic lines of bZIP2 (XVE-bZIP2.2), bZIP11 (XVE-bZIP11.3 and XVE-bZIP11.4) and bZIP44 (XVE-bZIP44.3 and XVE-bZIP44.9) were obtained, which showed an inducible moderate transgene expression as confirmed by Western Blot analysis (S3A Fig). To monitor root architecture of wild-type (WT) and transgenic XVE lines in presence and absence of Est, we cultivated 2-weeks old plants from each line for 1 week under LD regime on MS-medium supplemented with inducer or solvent, respectively. Whereas the WT did not show any noticeable response to Est treatment (Fig 3A, 3C and 3D), the XVE-bZIP2, -11 and -44 lines exhibited several low-auxin phenotypes at the most distal root part, such as strongly reduced primary root growth (Fig 3A–3C and S3B Fig), impaired auxin-induced root hair formation (S1 Table) and agravitropic root growth (Fig 3A, 3B, 3D and S3C Fig).
As these phenotypes are indicative of local auxin depletion at the root apex, we tested whether co-application of low to moderate levels of exogenous auxin (0.001 to 1 μM 1-naphthaleneacetic acid, NAA) were able to at least partially revoke the bZIP-associated root responses. In fact, we observed that application of low levels of NAA ranging from 0.01 to 0.1 μM antagonized bZIP-mediated primary root growth repression (S3D Fig), although they were shown to act (as a consequence of too high inner cellular auxin levels) slightly inhibitory on WT root growth [40]. More strikingly, root gravitropism could be largely rescued by moderate NAA concentrations (0.1 to 0.25 μM) (Fig 3D and S3E Fig). Individual or joint application of lower (0.001 μM) or higher NAA concentrations (up to 1 μM) had no significant effect or resulted in strong root growth inhibition, respectively (S3D and S3E Fig). In sum, these results strongly suggest that bZIP-mediated root growth inhibition is accomplished by repression of auxin signalling within and/or impairment of basipetal auxin transport to the distal meristematic root zone.
Members of the PIN gene family of auxin efflux carriers, such as PIN1, PIN2 (At5g57090), PIN3, PIN4 (At2g01420) and PIN7 (At1g23080), are well-known to implement cell-to-cell auxin transport in the root [25]. In particular, PIN1 and PIN3 were found to be the major auxin transport facilitators mediating polar auxin re-allocation from the shoot to the root tip [24, 25, 41]. As both PIN1 [42] and PIN3 (S4A Fig) have been shown to be repressed under energy deprived conditions and are known targets of the bZIP11-controlled IAA3/SHY2 repressor [32], we hypothesised that bZIP11 expression should result in reduced PIN transcription. In fact, this scenario would readily explain the low-auxin root growth phenotypes observed in bZIP over-expressors (Fig 3A–3D). Hence, we quantified PIN1 and PIN3 expression in XVE-bZIP11 lines applying q-RT-PCR. Remarkably, Est-mediated bZIP11 induction let to a rapid and strong repression of PIN transcription in roots (Fig 4A). In line with this observation, we found as early as 16 h after Est application a moderate to strong decline in root PIN1 and PIN3 protein abundance, as demonstrated by confocal laser scanning microscopy analysing XVE-bZIP11 plants, in a ProPIN1::PIN1:GFP or ProPIN3::PIN3:GFP background [43], respectively (Fig 4B and 4C and S4B Fig).
PIN1 and PIN3 show locally defined expression patterns in roots. Whereas both are strongly expressed in the vascular bundle [44, 45], PIN1 additionally shows high expression in endodermis cells of the root apex [46] and PIN3 in root columella cells [24]. IAA3/SHY2 expression has been found to largely resemble these PIN expression domains [30, 44, 45]. In order to examine if bZIP11 domains overlap with that of IAA3/SHY2 and PINs, we monitored genuine bZIP expression in the distal root. Therefore, Arabidopsis WT plants were stably transformed with a genomic bZIP11 fragment, composed of the ~2300 bp promoter, the entire 5’ UTR leader, which was found to confer Sucrose Induced Repression of Translation (SIRT) [17, 18], followed by the bZIP11 and GFP coding sequences.
Several reports demonstrated that bZIP11 functions in concert with the low-energy activated SnRK1 kinases in mediating metabolic reprogramming [3, 4, 7, 10]. Moreover its expression was found to be sucrose controlled [15, 16, 18, 19]. Hence, we analysed bZIP11 expression under progressive energy depletion at the middle of the night applying confocal laser scanning microscopy (Fig 5A and 5B). Reproducibly, we observed a highly specific expression pattern, which was restricted to the lateral and columella root cap, the root epidermis from root tip up to the elongation zone and within the endodermal cell layer from root tip to the differentiation zone. Moreover, a weak signal could be detected in the root stele. As bZIP translation was shown to be promoted by pronounced energy starvation [19] and IAA3/SHY2 as well as PIN expression was strongest in the central cylinder, we monitored stele-specific bZIP11 translation efficiency under short-term (2 h) extended night conditions. Using the well-established Translating Ribosome Affinity Purification (TRAP) method [47], an increase in stele-specific bZIP11 translation was observed under prolonged night when compared to expression at the middle of day (S5 Fig). In conclusion, genuine bZIP11 expression domains observed under low-energy situations largely overlap with those described for IAA3/SHY2 as well as that of PIN1 and/or PIN3.
IAA3/SHY2 mediated repression of PIN transcription has been found to result in impaired PAT from the shoot to the root tip [28]. As bZIP11, -2 or -44 induction was found to result in strong IAA3/SHY2 expression [21] and to low-auxin phenotypes at the basal root part, we analysed whether bZIP11 might affect basipetal auxin transport processes. In order to get first insights into auxin transport, we monitored expression of an auxin response reporter, consisting of the synthetic DR5 promoter fused to GFP (ProDR5::GFP) [48, 49] in the root using confocal microscopy (Fig 6A). Whereas strong, root tip localised GFP fluorescence was Est-insensitive in the WT control, a significant Est-dependent decline could be observed in the XVE-bZIP11 line, suggesting again that basipetal auxin transport to or local auxin signalling within the root tip are impaired by bZIP11 expression (Fig 6A and 6B). Similarly and consistent with bZIP11 function in low-energy signalling a short extension of the night period (8 hours) resulted in a strong decline in DR5-driven GFP expression in the root tip (S4A Fig).
In order to confirm the impact of bZIP11 on auxin transport and to characterize the bZIP11-mediated changes in auxin distribution, we quantitatively measured auxin concentrations in distinct segments of the plant. Applying liquid chromatography-tandem mass spectrometry (LC-MS/MS), we could indeed demonstrate that already 6 h after bZIP11 induction a 2-fold increase of auxin compared to the WT control became apparent in the upper proximal root parts including the hypocotyl. 30 h later an even more pronounced shift of auxin to the shoot could be found (Fig 6C). This reduced auxin translocation from shoot to root, which rapidly let to significant auxin depletion in the most distal root parts can most likely be explained by bZIP-mediated impairment of PIN driven polar auxin transport (PAT) and is highly consistent with the observed bZIP-mediated low-auxin phenotypes at the root apex (Fig 3A–3D and S3B and S3C Fig and S1 Table). Similar results were obtained analysing XVE-bZIP44 plants (S6A–S6C Fig), supporting functional redundancy among the bZIP11-related factors.
The antagonistic interplay between the phytohormones auxin and cytokinin is well-characterized to control root apical meristem size and consequently root growth rates [50, 51]. In particular, a high auxin to cytokinin ratio at the root tip is required to keep quiescent centre (QC) derived meristematic cells in their undifferentiated and rapidly dividing state. In contrast, decreasing auxin and increasing cytokinin levels in the more proximal root transition zone (TZ) drive cells to elongate and differentiate. By directly controlling PIN mediated basipetal auxin transport to the meristem, the bZIP11 target IAA3/SHY2 has been found to determine root meristem size and hence root growth rates. In consequence, we analysed the impact of bZIP11 expression on meristematic root growth. Therefore, we assessed the RAM size by counting the file of cortex cells beginning from the QC to the first elongated cortex cell in the TZ [36]. By these means, we microscopically analysed the RAM of XVE-bZIP11 plants in the presence and absence of Est and found that bZIP expression let to a highly significant reduction of meristem size (Fig 7A–7C). Notably, this was also found to be true for the highly related bZIP44 TF (S7A–S7C Fig), again supporting functional redundancy.
Plants need to invest most of their resources into growth to ensure their reproductive success in highly competitive habitats. However, a significant proportion of their resources are generally consumed by adaptive responses to a broad range of environmental stresses. To balance the input of resources into these essential but conflictive processes of survival and fitness, a growth regulatory system is required to adjust growth according to the prevailing energy availability.
In fact, it is well-documented that plants are equipped with a sophisticated energy management system to react rapidly and sensitively to changes in the availability of carbon skeletons by adjusting their metabolism and growth [34, 35, 52]. The associated extensive re-programming was largely ascribed to the low-energy activated SnRK1s [3] and the counteracting sugar-responsive TARGET OF RAPAMYCIN (TOR) (At1g50030) kinases [35]. Whereas TOR has been found to play a profound role in anabolic processes and root growth promotion in the presence of sugars [35, 53], starvation triggered SnRK1 activity represses TOR signalling and drives plant catabolism and root growth repression [3, 54]. In accordance with their proposed function, plants constitutively expressing SnRK1s phenocopy the starvation-mediated repression of primary root growth [3, 34]. Notably, Est-inducible expression of bZIP11 or the highly homologous bZIP2 and -44 TFs, which uncouples bZIP expression from repressive sugar regulation, results in similar primary root growth phenotypes. This suggests a partially redundant function among bZIP11-related factors and an interplay with SnRK1 kinases in energy-controlled primary root growth. A wealth of recent reports support this hypothesis showing that the group C/S1 bZIP heterodimerization network [11] exerts a significant proportion of SnRK1-mediated starvation responses [3, 4, 7, 8, 55]. In this respect, it has been highlighted that SnRK1s strongly enhance transcription mediated by bZIP11-related TFs [3] and promote bZIP11/bZIP63 heterodimerization by specific changes in the bZIP63 phosphorylation status [8]. Moreover, a significant overlap in global gene regulation provoked by moderate to severe energy depletion or expression of SnRK1 or bZIP11 has been observed [3, 4], indicating that expression of these regulators is sufficient to mimic natural starvation responses.
In order to address the impact of bZIP11-related factors in energy-controlled primary root growth, loss-of-function studies were conducted. Considering putative functional redundancy among the highly homologous bZIP11-related TFs, Est-inducible knock-down plants were generated, in which an efficient and specific amiRNA-guided transcript depletion of all three bZIPs was simultaneously achieved [21]. Consistent with their proposed role in low-energy signalling, knock-down of the respective bZIPs strongly impaired the plant’s ability to respond to energy limitation by reducing root growth. Most importantly, no effects on root growth could be observed between WT and bZIP knock-down under photosynthesis supporting growth conditions or in presence of exogenous sugars, accentuating the impact of these regulators in low-energy triggered root growth control. Accordingly, starvation-induced root growth repression has been found to resume after transferring dark cultivated plants into light [42] or after exogenous glucose application [34], characterizing the energy-controlled regulatory circuit as reversible and dynamic. In line with this, bZIP11 translation has been found to be tuned by sugar availability, being de-repressed by energy deprivation [19] and repressed by glucose and sucrose [17, 18]. Moreover, transcription mediated by bZIP11-related TFs was found to be promoted by the low-energy sensing SnRK1s [3]. As bZIP11-related activity is thus manifold controlled by intracellular sugar levels that reflect the endogenous energy status, bZIP11-related factors are proposed to act as a hub in energy signalling, thereby providing means to reversibly tune root growth in response to stress situations which converge on energy limitation (Fig 8).
In our recent studies we demonstrated that bZIP11-related TFs modulate auxin responsive gene expression of a specific set of negative feedback regulators of auxin signalling and homeostasis [21, 56]. However, their impact on auxin-related phenotypes have not been addressed, yet. In this work gain-of-function approaches of the respective bZIPs revealed that their expression redundantly repressed primary root growth, which could be partially rescued by exogenous auxin, indicative of altered root auxin signalling and/or auxin transport processes being at work.
Furthermore, results obtained by analysing endogenous PIN1 and PIN3 expression in XVE-bZIP11 plants suggested that the bZIP-mediated low-energy response on auxin-controlled primary root growth is mechanistically accomplished by transcriptional repression of major auxin transport facilitators. In line with this, genuine bZIP11 expression domains exhibited a significant overlap with that of PIN1 and/or PIN3 [24, 46]. However, as several reports highlighted bZIP11 as a strong activator of gene expression [20, 21], these data suggest that bZIP11-driven PIN repression is indirectly achieved. In fact, bioinformatics approaches on well-described auxin-responsive gene families emphasized the selective and conserved impact of bZIP-related cis-elements on promoters of Aux/IAA genes [31], which are potent negative regulators of auxin-controlled PIN expression [26]. In agreement with these results, we recently uncovered that bZIP11-related TFs quantitatively modulate expression of a specific subset of Aux/IAAs in Arabidopsis [21]. Importantly, expression of Aux/IAA3 (IAA3/SHY2), which is a well-documented negative regulator of PIN1 and PIN3 expression in roots [28, 29, 36, 50], was found to be dependent on bZIP11 expression and bZIP11-mediated recruitment of the histone acetylation machinery to its promoter [21]. Indeed, bZIP11 mutants lacking the respective recruitment domain fail to activate IAA3/SHY2 transcription [21]. Consistent with these results, we could confirm direct binding of bZIP11 to two G-box containing promoter regions, 1800 and 600 bps upstream of the IAA3 start codon. Importantly, these binding regions are in line with a recently published cistrome data-set, analysing genome-wide TF binding sites in vitro [39]. Although promoter binding was detected under non-starved conditions to minimise competition for cis-elements by endogenous bZIP11 (Fig 1B), analyses on starvation stimulated binding would be of interest to disclose if additional post-translational mechanism operate to modulate bZIP11 binding specificity and strength as it has been previously proposed [8].
Remarkably, studies on IAA3/SHY2 demonstrated that expression of gain-of-function variants of this repressor largely resemble the bZIP11-induced root growth responses (primary root growth and root hair repression as well as agravitropism) [57] and disclose its decisive role in tuning meristematic root growth by modulating PIN1- and PIN3- mediated PAT [28, 30, 36, 50]. As low-energy controlled expression of IAA3/SHY2 was found to mimic the bZIP11 expression profile - induced by starvation and repressed by sugars [33]- and was coherently found to be dependent on bZIP11-related TFs, IAA3/SHY2 constitutes a well-suited link between the energy stimulus integrating bZIP11-related TFs and the plants’ basic root growth regulatory system.
Importantly, root growth studies employing iaa3 and bZIP loss-of-function lines under energy deprived conditions, reveal a highly similar reduction in low-energy triggered repression of root growth for both genotypes. This suggests that bZIP11-related TFs largely exert their function on root growth via the root growth regulator IAA3. However, it has to be considered that TFs generally regulate several targets and hence further bZIP11-controlled mechanistic gateways might exist that interfere with RAM function. For instance, further starvation responsive genes that are involved in negatively regulating auxin signalling (e.g. IAA7) or auxin homeostasis (e.g. GH3.3) have been found to be bZIP11 controlled [21, 39].
Taken together, detailed genetic analyses and biochemical DNA binding studies propose a sequential activation of bZIP11, its target gene IAA3/SHY2, which in turn encodes the transcriptional repressor of PIN genes. As demonstrated in this study, this model is in line with the kinetic changes observed for transcripts and respective proteins of the involved players after Est-mediated bZIP induction. In fact, shortly after bZIP-mediated repression of PIN transcription (within 6 h), a successive reduction in PIN protein abundance (~ 16 h) as well as root tip auxin levels and primary root growth (~ 24 h) could be observed. Importantly, similar expression kinetics could be found under starvation conditions, supporting our assumption that the proposed signalling cascade is operating under physiological conditions. In this respect, it has been recently demonstrated that extended darkness rapidly activates bZIP11 via SnRK1-induced bZIP heterodimerisation [8]. Furthermore, we observed enhanced translation of bZIP11 RNA in the stele after 2 h of energy limiting conditions (S5 Fig). These findings suggest an interplay between several post-transcriptional mechanisms acting to control bZIP11 function during starvation. Along this line, induced IAA3/SHY2 transcription is observed within 4–8 h of extended night (Fig 1A), in which energy limitation increasingly becomes more severe. In accordance with enhanced expression of the IAA3/SHY2 repressor of PIN genes, a reduced level of PIN proteins should become visible after a short lag phase. Indeed, decreased PIN3-GFP levels were first observed after 8 h of extended night, steadily decreasing during the next 40 h. These dynamics of molecular events are in line with a decrease of free auxin in the root tip (S4A Fig) [58] and correlate with phenotypically reduced root growth 24 h after extended night (Fig 2A and 2B) [58], hence clearly supporting our working model.
Besides being expressed in roots, the bZIP11 promoter has also been found to be active in the shoot apical meristem (SAM) and young leaves [18]. As dark-treatment also results in altered PIN1 expression and auxin abundance in the SAM [59] as observed for the root meristem [42], it would be of great interest to analyse if bZIP11 mediates a more general starvation response on plant growth.
In summary, we disclosed in our work a crucial new aspect of bZIP11 function in plants’ energy management that is besides its well-documented impact on starvation triggered metabolic reprogramming, the control of meristematic root growth by modulating PAT. Hence, bZIP11 constitutes a central, energy-controlled hub, which provides means to integrate information on the plant’s energy status into plant metabolism and auxin-mediated growth responses. Understanding how plants balance growth in accordance to their energy supply is essential for developing future strategies to engineer plants for agricultural use.
Arabidopsis thaliana Columbia (Col-0) XVE-bZIP2.2, XVE-bZIP11.3, XVE-bZIP44.3 and XVE-ami2/11/44.4 lines were generated using the “Floral Dip Transformation” technique [60] applying the Agrobacterium tumefaciens strain GV3101 and are characterized in this work (S1 and S3A Figs). XVE-bZIP11.4, XVE-bZIP44.9 and XVE-amibZIP2/11/44.2 plant lines have been described before [21]. For visualization of PIN protein abundance, homozygous ProPIN1::PIN1:GFP and ProPIN3::PIN3:GFP [43] or, for mapping of auxin maxima, ProDR5::GFP plants [48] were crossed with the XVE-bZIP11.4 line, respectively. Concerning gene expression analyses and root morphology assays, surface-sterilized and stratified seeds were cultivated on ¼ MS [61] agar plates without sugars under long day growth conditions (16 h light / 8 h dark) at 21°C and a relative humidity of 60%. Gene expression was analysed in 2-weeks old plants that were treated with 7 μM Est (Sigma-Aldrich Chemie GmbH, Munich, Germany) dissolved in phosphate buffered saline (PBS) for up to 24 h. For root morphology assays, plants were grown for 2 weeks on ¼ MS agar plates in vertical position before they were transferred to media supplemented with or without 0.25 μM NAA and/or 10 μM Est and cultivated for another week. ProDR5::GFP expression in aseptically grown XVE-bZIP lines was monitored after 24 h of Est- or solvent-treatment, respectively. Starvation assays applying DCMU were performed in liquid medium. Seeds of the XVE-bZIP2/11/44 knock-down line were incubated in 1 ml ¼ MS medium without sugars but -/+ 10 μM of Est in a 24 well plate for 3 days in the dark at 20°C in a plant growth incubator. After 3 more days, DCMU (10 μM) and/or glucose (1–3%) was/were added to the medium and plants were transferred to long day growth conditions. Root growth was determined 3 days after treatment.
ChIP was described previously [21, 62] and was performed with minor modifications. In detail, 4 g of root material from aseptically grown XVE-bZIP11 plants was harvested at the middle of day after 8 h of Est treatment (10 μM) and incubated with cross-linking buffer (50 mM KH2PO4/K2HPO4 buffer (pH 5.8), 1% (v/v) formaldehyde) for 30 min under vacuum. Afterwards samples were incubated in glycine buffer (50 mM KH2PO4/K2HPO4 buffer (pH 5.8), 0.3 M glycine) for 15 min under vacuum and washed with ice-cold water. Samples were frozen in N2 and grinded. Nuclei extraction was performed at 4°C. Therefore, root material was resuspended in 24 ml ice-cold extraction buffer (1 M hexylenglycol, 50 mM PIPES-KOH (pH 7.2), 10 mM MgCl2, 5 mM ß-mercaptoethanol, one tablet per 10ml complete protease inhibitor cocktail tablets, Roche) and was cleared by filtrating it through two layers of miracloth. Afterwards 1 ml of 25% Triton X-100 was added dropwise to the extract. After incubation for 15 min, nuclei were isolated by density-gradient centrifugation using a 35% percoll cushion. The nuclei pellet was resuspended in sonication buffer (10 mM Tris-HCl (pH 7.4), 1 mM EDTA (pH 8.0), 0.25% SDS and protease inhibitor) prior to sonification for 20 times 30 s. Chromatin was cleared by centrifugation for 15 min at 11,000 g and 4°C and stored at -80°C. For each IP 10 μg chromatin and 3 μg ChIP grade a-HA antibody (ab9110) (Abcam, Cambridge, UK) were used. 70 μl of protein G-coated magnetic beads (Invitrogen, Karlsruhe, Germany) dissolved in ice-cold extraction buffer, supplemented with protease inhibitor (Roche, Mannheim, Germany) were applied to each sample. Antibody—antigen binding was achieved during a 4 h incubation step at 4°C and slow rotation on an Intelli-Mixer. Beads were washed five times with washing buffer supplemented with protease inhibitor, prior to elution of protein-DNA complexes using elution buffer (0.1 M glycine (pH 2.5), 500 mM NaCl, 0.05% (v/v) Tween-20). Precipitated DNA was quantified by q-RT-PCR using the oligonucleotide primers summarized in S2 Table. Data were normalized to DNA input, which was quantified by employing ACTIN8 (At1g49240) specific primers. Presented mean and SEM values were calculated from two independent ChIP experiments that were performed on each of two independently prepared chromatin samples from WT or XVE-bZIP11 plants, respectively.
To determine root morphology parameters, high resolution images of 40 individual plants per treatment were taken using the Camag reprostar 3 documentation system with a Canon G5 camera (CAMAG AG & Co. GmbH, Berlin, Germany). From these pictures, the root parameters of the differently treated plants were monitored. These are: the total primary root length before and after one week of treatment, the presence or absence of macroscopically visible root hairs and the abundance of roots with obvious agravitropic root growth. Root growth was considered to be agravitropic if roots showed at least one growth re-orientation of more than ~ 45° after inducer treatment. The root length was measured using the Image J 1.43u software (available at http://rsb.info.nih.gov/ij) whereas numbers of roots with/without root hairs and roots with agravitropic growth were determined by counting.
To determine GFP expression within the root, 200-fold enlarged bright-field and fluorescence images of individual root tips of ProPIN::PIN:GFP or ProDR5::GFP reporter lines were taken using the Leica SP8 confocal microscope. Fluorescence intensities were quantified as relative fluorescence intensity units using the Leica AF lite application suite 2.0.0. In order to quantify the root meristem size, 400-fold enlarged bright-field images of XVE-bZIP44 or XVE-bZIP11 plants were taken and analysed as previously described [36]. In detail, we focused on the QC cells and counted the expanding file of cortex cells originating from the QC up to the first strongly elongated cortex cell (at least twice in size) in the elongation zone. In the representative pictures in Fig 6A and 6B and S7A and S7B Fig we focused on the proposed boundary between meristem and elongation zone.
Free IAA was measured using liquid chromatography-tandem mass spectrometry (LC-MS/MS) as described previously [63]. Briefly, 25–50 mg (fresh-weight) of Arabidopsis seedlings was extracted in sodium-phosphate buffer (pH 7). To each extract, 10 pmol of [13C6] IAA was added as internal standard to check recovery during purification and to validate the quantification. Samples were purified using a combination of a reversed-phase and anion-exchange chromatography and were analyzed by the LC-MS/MS system consisting of an ACQUITY UPLC System (Waters) and Xevo TQ MS (Waters). Samples were dissolved in 15% acetonitrile, injected onto the ACQUITY UPLC BEH C18 column (100 x 2.1 mm, 1.7 μm; Waters) and eluted with a linear gradient (0–3 min, 15% B; 3–10 min, 20% B; 10–20 min, 30% B; flow-rate of 0.25 ml min-1) of 7.5 mM formic acid (A) and acetonitrile (B). Quantification was obtained using a multiple reaction monitoring (MRM) mode of [M-H]+ and the appropriate product ion. The limits of detection (signal to noise ratio 1:3) for all analytes ranged from 5 to 10 fmol. The linear range was at least over the five orders of magnitude with a correlation coefficient of 0.9991–0.9997. For each genotype, at least five independent replicates were performed.
TRAP method has been performed as described before [47]. In brief, 1.5 g of plant material were frozen in liquid nitrogen and homogenized to a fine powder using a mortar. The material was resuspended in twice the volume of ice-cold polysome extraction buffer (200 mM Tris-HCl, pH 9.0, 200 mM KCl, 25 mM EGTA, 36 mM MgCl2, 1% (v/v) octylphenyl-polyethylene glycol (Igepal CA-630), 1% (v/v) polyoxyethylene lauryl ether (Brig 35), 1% (v/v) Triton X-100, 1% (v/v) Tween-20, 1% (v/v) polyoxyethylene tridecyl ether, 1% (v/v) sodium deoxycholate, 1 mM dithiothreitol (DTT), 50 μg/ml cycloheximide, 50 μg/ml chloramphenicol) and incubated for 10 min on ice. The samples were centrifuged for 10 min at 4°C and 16.000 g. The obtained extract was again cleared by additional centrifugation (10 min at 4°C and 16.000 g) and filtration through two layers of miracloth. To precipitate FLAG-tagged ribosomes, 450 μl of magnetic beads coated with anti-FLAG antibodies (Sigma, Germany) were added to the extract. The mixture was incubated on an intellimixer for 3 hours at 4°C before the beads were collected on the side of the reaction tube using a magnet. The pellet was washed at least 5 times with 2 ml washing buffer (200 mM Tris-HCl, pH 9.0, 200 mM KCl, 25 mM EGTA, 36 mM MgCl2, 5 mM DTT, 50 μg/ml cycloheximide, 50 μg/ml chloramphenicol) before the ribosome-RNA complexes were eluted using anti-FLAG peptide (Sigma, Germany). The supernatant was collected and used to clean up the bound RNA using the RNEasy Micro Kit (Qiagen, Germany) following the manufacturer’s protocol. Isolated RNA was used for cDNA synthesis using oligo dT and random nonamer primers. To quantify bZIP11 transcripts, 4 μl of cDNA were used for subsequent q-RT-PCR analysis employing bZIP11 specific primers that are given in S2 Table.
Standard DNA techniques have been described, previously [64]. DNA sequence analyses were performed by LGC Genomics (Berlin, Germany). Western analysis was performed making use of a polyclonal α-HA antibody from rabbit (Santa Cruz, Santa Cruz, CA, USA) and an anti-rabbit IgG conjugated with a horseradish peroxidase (GE Healthcare, Freiburg, Germany). q-RT-PCR has been performed with SYBR green as described, previously [10]. Data were obtained from 3 replicates of at least 2 individual plant pools and were normalized to UBQ5 (At3g62250) transcript abundance. All q-RT-PCR oligonucleotide primers used are summarized in S2 Table.
Figures and statistical tests were performed applying the OriginPro 8.1G software. Significant differences between multiple constructs and treatments were determined using the One-way ANOVA test followed by a Tukey post-hoc test (p < 0.05) and are visualized by different letters. Significant differences between only two datasets were defined making use of the Student’s t-Test and labeled with asterisks (* p < 0.05; ** p < 0.01; *** p < 0.001).
|
10.1371/journal.pgen.1000225 | Diverse Splicing Patterns of Exonized Alu Elements in Human Tissues | Exonization of Alu elements is a major mechanism for birth of new exons in primate genomes. Prior analyses of expressed sequence tags show that almost all Alu-derived exons are alternatively spliced, and the vast majority of these exons have low transcript inclusion levels. In this work, we provide genomic and experimental evidence for diverse splicing patterns of exonized Alu elements in human tissues. Using Exon array data of 330 Alu-derived exons in 11 human tissues and detailed RT-PCR analyses of 38 exons, we show that some Alu-derived exons are constitutively spliced in a broad range of human tissues, and some display strong tissue-specific switch in their transcript inclusion levels. Most of such exons are derived from ancient Alu elements in the genome. In SEPN1, mutations of which are linked to a form of congenital muscular dystrophy, the muscle-specific inclusion of an Alu-derived exon may be important for regulating SEPN1 activity in muscle. Realtime qPCR analysis of this SEPN1 exon in macaque and chimpanzee tissues indicates human-specific increase in its transcript inclusion level and muscle specificity after the divergence of humans and chimpanzees. Our results imply that some Alu exonization events may have acquired adaptive benefits during the evolution of primate transcriptomes.
| New exons have been created and added to existing functional genes during eukaryotic genome evolution. Alu elements, a class of primate-specific retrotransposons, are a major source of new exons in primates. However, recent analyses of expressed sequence tags suggest that the vast majority of Alu-derived exons are low-abundance splice forms and represent non-functional evolutionary intermediates. In order to elucidate the evolutionary impact of Alu-derived exons, we investigated the splicing of 330 Alu-derived exons in 11 human tissues using data from high-density exon arrays with multiple oligonucleotide probes for every exon in the human genome. Our exon array analysis and further RT-PCR experiments reveal surprisingly diverse splicing patterns of these exons. Some Alu-derived exons are constitutively spliced, and some are strongly tissue-specific. In SEPN1, a gene implicated in a form of congenital muscular dystrophy, our data suggest that the muscle-specific inclusion of an Alu-derived exon results from a human-specific splicing change after the divergence of humans and chimpanzees. Our study provides novel insight into the evolutionary significance of Alu exonization events. A subset of Alu-derived exons, especially those derived from more ancient Alu elements in the genome, may have contributed to functional novelties during primate evolution.
| Alu is a class of primate-specific transposable elements that belongs to the short interspersed nuclear elements (SINE) family [1]. The rapid expansion of Alu during primate evolution has produced over one million copies of Alu elements in the human genome [2]. Until recently, Alu elements were considered as “junk DNA”, with no important functional or regulatory roles [1]. However, recent studies suggest a substantial influence by Alu elements on evolution of the human genome and regulation of gene expression [3].
Alu is a major source of new exons in primate genomes [4]–[6]. Alu elements have several sites resembling consensus splice sites in both sense and antisense orientations [7]. Therefore, the insertion of Alu elements into intronic regions may introduce new exons into existing, functioning genes. The evolutionary history of several such “exonization” events has been characterized in detail [8],[9]. For example, in p75TNFR, the insertion of an Alu element and a series of subsequent nucleotide substitutions created a new alternative first exon [8]. Sorek and colleagues investigated the splicing pattern of 61 Alu-containing exons using human mRNA and EST sequences [4]. All Alu-containing exons were alternatively spliced. The vast majority of these exons were included in the minor transcript isoforms, based on ESTs pooled from all tissues [4]. This is consistent with the hypothesis that the creation of a new minor-form alternative exon reduces the initial deleterious effects of exonization events [10]. However, due to the high noise in EST sequencing [11] and the low EST coverage for these Alu-derived exons [4], it was difficult to assess the splicing patterns of individual exons tissue by tissue. Regardless, there have been anecdotal reports for Alu-containing exons to have splicing patterns other than minor-form alternative splicing. Based on the tissue origins of human EST sequences, Mersch et al. predicted a few Alu-containing exons to be tissue-specific [12]. In another study, an Alu-containing exon of FAM55C was shown to be constitutively spliced in a neuroblastoma cell line [13]. These data suggest that the splicing profiles of exonized Alu elements may be more diverse than previously expected. In this study, we combined a genome-scale Exon array analysis with RT-PCR experiments to investigate the splicing profiles of exonized Alu elements in human tissues.
We collected a list of 330 Alu-derived exons, using annotations from the UCSC Genome Browser database [14] and Affymetrix human Exon 1.0 arrays (see details in Materials and Methods). We first analyzed the splicing signals of these exons as well as their evolutionary rates during primate evolution. For the purpose of comparison, we also analyzed 13103 constitutively spliced exons and 5389 exon-skipping cassette exons in the human genome, which were collected after applying a set of stringent filtering criteria to exons in the Alternative Splicing Annotation Project 2 (ASAP2) database (see Materials and Methods).
Our analysis showed that Alu-derived exons had significantly weaker splicing signals compared to constitutively spliced exons and typical cassette exons. For each exon, we scored its 5′ and 3′ splice site using models of consensus splice sites in MAXENT [15]. The median 5′ splice site score of Alu-derived exons was 7.35, compared to 8.27 for cassette exons and 8.88 for constitutive exons, a statistically significant difference (P = 3.0e-6 for Alu-derived exon vs cassette exons; P<2.2e-16 for Alu-derived exons vs constitutive exons; Wilcoxon rank sum test). We observed the same trend for the 3′ splice site. The median 3′ splice site score of Alu-derived exons was 6.79, significantly lower than the scores of cassette exons (7.86) and constitutive exons (8.87). In addition, Alu-derived exons had a lower density of exonic splicing regulatory elements (ESRs). We used two sets of ESRs from the studies of Goren et al [16] and Fairbrother et al [17]. For each exon, we calculated the density of ESRs as the number of nucleotides covered by ESRs divided by the total length of the exon. The average ESR density of Goren et al was 0.484 on Alu-derived exons, compared to 0.500 on cassette exons and 0.532 on constitutive exons (P = 0.04 for Alu-derived exon vs cassette exons; P = 6.5e-14 for Alu-derived exons vs constitutive exons). The same trend was observed for ESRs of Fairbrother et al: the average density was 0.144 on Alu-derived exons, which was significantly lower than the density on cassette exons (0.268) and constitutive exons (0.328).
We also found that Alu-derived exons had much higher evolutionary rates during primate evolution, compared to constitutive exons and cassette exons. Recently, the genome sequences of several non-human primates have become available. Therefore, we can study the sequence evolution of Alu-derived exons in primates after the initial Alu insertion events. To determine the evolutionary rate of different classes of exons, we analyzed the pairwise alignments of the human genome to the genomes of chimpanzee, orangutan, macaque and marmoset, which were increasingly distant from humans [18]. For exons present in both human and chimpanzee genomes, the overall nucleotide substitution rate of Alu-derived exons was 1.34%, compared to 0.73% for cassette exons and 0.52% for constitutive exons (P≤2.2e-16 in Alu vs cassette exons and Alu vs constitutive exon comparisons, Wilcoxon rank sum test). Similarly, between human and orangutan genomes, the overall nucleotide substitution rates of Alu-derived exons, cassette exons and constitutive exons were 3.69%, 1.81%, and 1.31% respectively. The same trend was also observed in pairwise comparisons of human-macaque and human-marmoset genomes (see Table 1). We also obtained similar results when we restricted our analysis to exons smaller than 250 nt (data not shown). These comparative analyses span the last ∼50 million years of primate evolution [18].
Taken together, these data are consistent with the hypothesis that the majority of primate-specific human exons derived from Alu elements are evolutionary intermediates without established functions [4],[6]. The high evolutionary rate of Alu-derived exons observed in primate genome alignments probably reflects the combined effect of reduced negative selection pressure on non-functional Alu exons as well as positive selection pressure on Alu exons with adaptive benefits. However, distinguishing the effect of positive selection from that of the reduced negative selection is a difficult task in general [19],[20]. Identifying the subset of Alu exonization events that have undergone positive selection using sequence-based approaches is particularly difficult for some practical reasons. Most Alu-derived exons are short (median length of the 330 exons is 121 nucleotides). They are too new to have homologous sequences from distantly related species – homologous sequences of these exons may only exist in non-human primates. Thus, for most exons the number of nucleotide differences between homologous sequences is small, which significantly decreases the power of statistical tests. Although SNP-based approaches have been applied to genome-wide scans of positive selection on the human genome [21]–[26], the regions identified by these studies are typically very large, making it a major challenge to locate the causal allele for positive selection [27]. In addition, SNP-based methods are sensitive to the temporal phases of positive selection [28], influenced by the ascertainment bias [29], and confounded by demographic factors [19], [30]–[32]. For example, the Alu-derived exon of ADAR2 (ADARB1) is a well-known case of functional exonization. This exon inserts an in-frame peptide segment into the catalytic domain of ADAR2, altering its catalytic activity [33]. Using HapMap (I+II) SNP data [21],[34], we tested for the reduction of SNP heterozygosity, the skewed allele frequency spectrum with Tajima's D [35] and Fay and Wu's H [36], and the increased population differentiation (Fst) [26],[37] (see details of the analysis in Text S1). We did not observe evidence of positive selection on this ADAR2 exon using these metrics (see Figure S1A). Similarly, SNP-based tests did not indicate evidence of positive selection for the alternative first exon of p75TNFR (see Figure S1B), the result of another well-known functional exonization event [8]. These data show the limitation of using sequence-based approaches to identify functional Alu exonization events.
A direct approach to assess the impact of individual Alu-derived exons on mRNA and protein products is to examine the splicing patterns of these exons in human tissues. Therefore, we proceeded with a large-scale splicing analysis of Alu-derived exons, using Affymetrix Exon array data of 330 exons in 11 human tissues and RT-PCR experiments of 38 exons, described in detail below.
To examine the splicing patterns of Alu-derived exons, we used a public Affymetrix Exon 1.0 array data set on 11 human tissues (breast, cerebellum, heart, kidney, liver, muscle, pancreas, prostate, spleen, testes, thyroid) [38], with three replicates per tissue. The Affymetrix human Exon 1.0 array is a high-density exon-tiling microarray platform designed for genome-wide analysis of pre-mRNA splicing, with over six million probes for well-annotated and predicted exons in the human genome [39],[40]. Most exons are targeted by a probeset of four perfect-match probes.
We compiled a list of 330 Exon array probesets targeting the 330 Alu-derived exons (see details in Materials and Methods). In each of the 330 probesets, we had at least three probes to infer the splicing profile of the exon, after we filtered probes showing abnormal intensities (Materials and Methods). Using a series of statistical methods that we developed for Exon array analysis [41],[42], for each probeset targeting an Alu-derived exon, we calculated the background-corrected intensities of its multiple probes and the overall expression levels of the gene in 11 tissues. These data were used to infer the splicing patterns of the exon.
A large fraction of the 330 Alu-derived exons had low probe intensities in all surveyed tissues. Using a presence/absence call algorithm we developed for Exon array analysis, which compares the observed intensity of a probe to its predicted background intensity, we summarized a probeset-level Z-score for each exon in individual tissues as in [41]. A high Z-score suggests that the target exon is expressed. 174 (53%) Alu exons had a Z-score of greater than 6 in at least one tissue, including 119 (36%) exons whose Z-score was greater than 10 in at least one tissue. We also applied the same Z-score calculation to 37687 “background” probes on Exon array. These probes do not match any known genomic and transcript sequence in mammalian genomes [43], so we can use their Z-score to estimate the false positive rate of the analysis. 5% of the background probes had Z-score greater than 6 in at least one tissue, including 3% whose Z-score was greater than 10 in at least one tissue. Based on these false positive rate estimates, at the Z-score cutoff of either 10 or 6, we estimated that 33%–48% of the 330 Alu-derived exons in our study were expressed in some of the tissues. The remaining exons were not expressed at all or were expressed at very low levels in these 11 adult tissues. Of course, this is only a rough estimate, because the Z-score of individual probesets could be affected by a variety of microarray artifacts such as low probe-affinity or cross-hybridization [44],[45]. Overall, these data are consistent with the observation that most Alu-derived exons had low transcript inclusion levels in EST databases [4]. Such Alu-derived exons may represent non-functional evolutionary intermediates that are rarely incorporated in the transcripts [9]. It is also possible that some of these exons are indeed expressed in other tissues or developmental states.
Despite the low transcript abundance of many Alu-derived exons, a small fraction of exons showed highly correlated probe intensities with the overall expression levels of their corresponding genes across the surveyed tissues, suggesting stable exon inclusion. We found 19 Alu-derived exons where three probes or more correlated with gene expression levels, including the well-characterized Alu-derived exon in ADAR2 (ADARB1) that inserts an in-frame peptide segment to ADAR2's catalytic domain [5]. Detailed descriptions of these 19 exons are provided in Table 2. Among the 19 “correlated” exons, 12 were in the 5′-UTR. One exon was in 3′-UTR and one exon was part of a non-coding transcript. The remaining five exons were in coding regions, including two that introduced premature termination codons. This distribution is consistent with the hypothesis that most functional Alu exonization events do not contribute to the proteome but may play a role in regulating gene expression [46],[47]. Similar to the finding by a recent study of species-specific exons [48], we observed an excess of Alu-derived internal exons in 5′-UTR as compared to 3′-UTR. This may reflect stronger negative selection pressure against exon creation in 3′-UTR because such exons could trigger mRNA nonsense-mediated decay. The 5′-UTR Alu exons may influence the transcriptional or translational regulation of their host genes, as suggested by Goodyer and colleagues [49].
Several types of splicing patterns could explain the observed correlation between probe intensities and estimated gene expression levels. These “correlated” exons could be constitutively spliced, or alternatively spliced at similar levels across tissues, or alternatively spliced but with certain variations in exon inclusion levels from tissue to tissue. However, we could not distinguish these situations based on Exon array data alone, since uncertainties in microarray probe affinity [44] prevent estimations of the absolute transcript abundance of individual exons.
To uncover the exact splicing patterns of the “correlated” exons we analyzed all 19 exons by RT-PCR, using RNAs from all available tissues surveyed by Exon array (purchased from Clontech, Mountain View, CA) except breast tissue. For each exon, we designed RT-PCR primers targeting its flanking constitutive exons. The identities of all PCR products close to the expected sizes of exon inclusion or skipping forms were further confirmed by sequencing (Materials and Methods). We discovered three major categories of splicing patterns in these 19 exons (Table 2). Six exons (in FAM55C, NLRP1, ZNF611, ADAL, RPP38, RSPH10B) were constitutively spliced. For example, the four probes of an Alu-derived exon in NLRP1 had a minimal correlation of 0.86 with the expression levels of NLRP1 in the Exon array data (Figure 1A). Our RT-PCR analysis showed a single isoform corresponding to the exon inclusion form in all surveyed tissues (Figure 1B). In FAM55C, an Alu-derived exon was shown previously to be included in the only isoform product in a human neuroblastoma cell line [13]. We found all four probes of this FAM55C exon had a minimal correlation of 0.78 with the overall gene expression levels (Figure 1C). Our RT-PCR experiments showed that this exon was constitutively spliced (Figure 1D). In another three tested genes (SLFN11, NOX5, B3GALNT1), the Alu-derived exons were alternatively spliced, but the transcript inclusion levels varied in individual tissues. For example, the SLFN11 exon was included in the major transcript product in most tissues but appeared as the minor form in pancreas. We observed Alu exon inclusion isoforms of varying lengths that resulted from alternative splice site usages of the Alu-derived exon and its upstream alternative exon (Figure 1E). In NOX5, a single exon-inclusion isoform was detected in most tissues, but an additional exon-skipping isoform was detected in liver, pancreas and testes (Figure 1F). In the remaining 10 tested genes, the exons were alternatively spliced with varying levels of transcript inclusion, but no exon showed evidence of tissue-specificity in our semi-quantitative RT-PCR analyses (see Table 2 and Figure S2).
We also conducted RT-PCR analyses of 11 “uncorrelated” exons (Table S1). The lack of correlation between probe intensities of an exon and overall gene expression levels can be due to a number of reasons. If the target Alu-derived exon has very low transcript inclusion levels, or if the probes have poor binding affinity to the target exon, the intensities of the microarray probes could be largely saturated by microarray noise, resulting in poor correlation with the overall gene expression levels. It is also possible that the correlation pattern of a highly expressed Alu-derived exon is obscured due to microarrray artifact (such as cross-hybridization) in a subset of samples. Thus, by analyzing “uncorrelated” exons, especially those with high probeset-level Z-scores in individual tissues, we may discover additional Alu-derived exons with high transcript inclusion levels. Indeed, among six RT-PCR tested “uncorrelated” exons whose probeset-level Z-score was greater than 7 in at least three tissues, we found two constitutive exons, three exons with medium to high transcript inclusion levels, as well as one exon in the minor transcript isoform (see Table S1 and Figure S3). By contrast, among five exons whose probeset-level Z-score was smaller than 3 in all 11 tissues (suggesting weak exon inclusion), four exons had very weak exon-inclusion transcripts in all surveyed tissues. The exon in FAM124B had medium transcript inclusion levels (see Figure S3).
Taken together, our RT-PCR analysis of 19 “correlated” exons and 11 “uncorrelated” exons indicates that a subset of Alu-derived exons have acquired strong splicing signals, so that they are included in the transcript products at high levels. Moreover, while prior EST-based analyses suggested all Alu-derived exons to be alternatively spliced [4], we provide experimental evidence that some Alu-derived exons are constitutively spliced in a broad range of normal human tissues.
Our analysis of the “correlated” Alu exons revealed that some exons had varying transcript inclusion levels in different tissues. It is possible that exons with strong tissue-specific splicing patterns do not have highly correlated intensities with the overall gene expression levels, and were missed by the above analysis. Therefore, we combined computational analysis and manual inspection of Exon array data to specifically search for tissue-specific exons (see Materials and Methods). We selected three exons (in ICA1, ZNF254, FAM79B/TPRG1) that appeared to exhibit strong tissue-specific splicing patterns for RT-PCR. We also selected five other Alu-derived exons with prior experimental evidence for exon inclusion in at least one tissue or cell line [9],[12],[50],[51], regardless of whether reliable Exon array probes existed for these exons (Table 3). Our RT-PCR experiments detected four exons with tissue-specific splicing patterns (also see Figure S4 for the other four exons with no tissue-specificity). In ICA1, the Exon array data suggested testes-specific exon inclusion (Figure 2A). The RT-PCR analysis detected a strong band corresponding to the exon inclusion form specifically in the testes (Figure 2B). In ZNF254, the RT-PCR analysis indicated strong exon inclusion in cerebellum, which was consistent with the Exon array profile (Figure 2C–D). We also found that this exon was almost completely skipped in pancreas, although this pattern was not observed in the Exon array data. In PKP2, the exon inclusion form was shown to be the minor isoform in HT29, a colon cancer cell line [9]. Our RT-PCR result showed that this exon was skipped in all other surveyed tissues but was included in the minor transcript product in the pancreas (Figure 2E).
Some tissue-specific Alu-derived exons have interesting functional implications. For example, SEPN1 encodes selenoprotein N, 1, which is expressed in skeletal muscle and has been suggested to play a role in protection against oxidant damage [50]. Mutations in SEPN1 were linked to a form of congenital muscular dystrophy [50]. SEPN1 is expressed as two alternatively spliced isoforms. The full-length isoform contains an Alu-derived exon, which is predicted to be the minor isoform based on EST data. The Alu-derived exon contains a second in-frame TGA selenocysteine residue. However, the protein product corresponding to the exon inclusion isoform was not detected by Western blot in the HeLa cell [52]. Our RT-PCR result indicated a strong muscle-specific increase in the inclusion level of this Alu-derived exon (Figure 2F). It will be interesting to investigate whether this splicing pattern represents a mechanism for modulating SEPN1 activity in muscle.
To further elucidate the evolution of this muscle-specific Alu exon in SEPN1, we obtained matching macaque and chimpanzee tissues and analyzed the splicing pattern of this exon in primate tissues using semi-quantitative RT-PCR as well as realtime quantitative PCR (see Materials and Methods). RT-PCR analysis of this exon in macaque tissues showed no exon inclusion (see Figure 3B), consistent with the fact that this Alu exon was absent from the corresponding SEPN1 region in the rhesus macaque genome. In chimpanzees, both exon inclusion and skipping forms were produced, but the exon inclusion levels were significantly lower compared to human tissues based on the RT-PCR gel pictures (Figure 3B). The splicing difference of this SEPN1 exon between humans and chimpanzees was further confirmed by realtime qPCR using isoform-specific primers (Figure 3C–D). These data depict the evolutionary history during the creation of an Alu-derived primate-specific exon and the establishment of its tissue-specific splicing pattern. Our results suggest that the strong transcript inclusion and muscle-specificity of the human SEPN1 exon was acquired after the divergence of humans and chimpanzees.
In this study, we conducted RT-PCR analysis of 38 Alu-derived exons in 10 human tissues. 26 of the 38 exons had at least medium inclusion levels in certain tissues. These exons are in genes from a wide range of functional categories (see the complete list in Table S2). Analyses of these 26 exons revealed several interesting characteristics. 23 of the 26 exons were derived from the antisense strand of Alu elements, among which 14 were from the right arm of the antisense Alu (see Figure S5), consistent with a recent report that the right arm of Alu antisense strand is a hotspot for exonization [53]. Moreover, of these 26 exons, 23 were from AluJ class and 3 were from the AluS class. By contrast, in the total set of Alu-derived exons in our study, 211 were from AluJ and 111 were from AluS, a 4-fold shift in the ratio of AluJ to AluS (7.7 in the “substantially included” set versus 1.9 in the total set; P = 0.01, one-tailed Fisher exact test). In the human genome, AluJ is outnumbered by AluS at a ratio of 1 to 2.3 [14] (Figure 4). The similar trend was also found in the 19 “correlated” exons; 16 were from the AluJ class and 3 were from the AluS class. Taken together, these data are consistent with the fact that AluJ is the oldest Alu subclass in the human genome [54], so that exons derived from AluJ elements had more evolutionary time to accumulate nucleotide changes that strengthened exon inclusion in the transcript products.
We did not observe a significant difference in the splice site score and ESR density of the 26 substantially included Alu exons compared to other Alu-derived exons (data not shown). This could be due to the lack of statistical power. Alternatively, it may reflect the current lack of knowledge of the complete set of cis-elements that regulate splicing [55],[56]. Future experimental studies (such as mini-gene experiments) are needed to dissect the exact regulatory elements important for strong transcript inclusion and/or tissue-specific splicing of individual Alu-derived exons.
Our study reveals diverse splicing patterns of exonized Alu elements in the human transcriptome. Most new exons originated from Alu elements probably represent non-functional splice forms that are included in the transcripts at low frequencies [4],[6]. However, a small subset of exonization events, in particular those associated with more ancient Alu elements, could evolve strong splicing regulatory signals to become constitutive or tissue-specific, possibly driven by positive selection. The analysis of high-density exon tiling array data across a broad range of tissues provides an efficient approach to identify such exons. Considering the incomplete coverage of Exon 1.0 arrays on human transcribed regions, and the high noise in the observed intensities of probes targeting individual exons [57],[58], we expect that many constitutive or tissue-specific Alu-derived exons are missed by this study. Also, while we focus on primate-specific exons derived from Alu repeats, a recent study by Alekseyenko and colleagues identified nearly 3000 human-specific exons created by de novo substitution in intronic regions during primate evolution [59]. With improved exon microarray platforms and analysis algorithms in the future, more species-specific exons with regulatory roles are likely to be discovered.
Our data provide novel insight into the evolutionary impact of newly created exons in eukaryotic genomes. During evolution, new exons are frequently added to existing functioning genes via a variety of mechanisms, such as exonization of transposable elements, exon duplication, and de novo exonization from intronic regions [6]. Modrek and Lee found that the birth of new exons was strongly coupled with widespread occurrence of alternative splicing in eukaryotic genes [60]. Through pairwise comparisons of human and rodent genomes, they showed that nearly 75% of human alternatively spliced exons with low transcript inclusion levels were absent from the corresponding genomic sequence of the rodent orthologs. By contrast, the number was less than 5% for constitutive exons [60]. This pattern was corroborated by subsequent analyses of exon creation events in vertebrates using multiple genome alignments [48],[59]. Based on these observations, Modrek and Lee proposed an evolutionary model that alternative splicing can facilitate the evolution of new exons – the creation of a new exon in the minor transcript isoform keeps the original gene product intact, which reduces the negative selection pressure against the new exon, allowing it to evolve towards an adaptive function [10],[60]. On the other hand, this evolutionary model also predicts that the vast majority of new exons found by comparative genomics analyses are non-functional evolutionary intermediates. In fact, most previous genomic studies have focused on the low transcript inclusion levels of new exons [4],[6],[48],[59],[60]. It is unclear to what extent new exons could have produced functional and regulatory novelties. In this study, based on a large-scale splicing analysis of human tissues, we show that a number of primate-specific exons derived from Alu retrotransposons have a major impact on their genes' mRNA/protein products in a ubiquitous or tissue-specific manner. In SEPN1, the strong transcript inclusion and muscle-specificity of the Alu derived exon represents a human-specific splicing change after the divergence of humans and chimpanzees. These data suggest that some new exons may contribute to species-specific differences between humans and non-human primates.
Our study has discovered a large list of Alu-derived exons with substantial transcript inclusion levels. This exon list can be valuable for a variety of further investigations. These exons provide candidates for detailed mechanistic analyses and can be used to characterize the splicing regulatory mechanisms of Alu-derived exons. If suitable tissue samples from closely or distantly related primate species are available, it will be possible to precisely reconstruct the evolutionary events preceding the emergence of constitutive or tissue-specific Alu-derived exons. Further experimental studies will be needed to elucidate the functional significance of individual exonization events (e.g. the muscle-specific inclusion of the Alu-derived exon in SEPN1).
We downloaded a public Affymetrix Exon 1.0 array data set on 11 human tissues (breast, cerebellum, heart, kidney, liver, muscle, pancreas, prostate, spleen, testes, thyroid) [38], with three replicates per tissue (http://www.affymetrix.com/support/technical/sample_data/exon_array_data.affx).
We compiled a list of Exon array probesets targeting exonized Alu elements. The locations of Alu elements in the human genome were downloaded from RepeatMasker annotation of the UCSC Genome Browser database [14]. The locations of internal exons (i.e. exons flanked by both 5′ and 3′ exons) in human genes were taken from the UCSC KnownGenes database [14]. This database combines transcript annotations from multiple sequence databases [14]. To eliminate long exonic regions likely resulting from intron retention events, we removed probesets whose probe selection regions were over 250 bp as in [61]. We then defined an exon as Alu-derived if the Alu element covered at least 25 bp of the exon and over 50% of the total length of the Exon array probe selection region. We collected 526 Exon array probesets targeting such Alu-derived exons. Since microarray probes targeting Alu repeats may cross-hybridize to off-target transcripts, we used a conservative approach to identify and remove individual probes showing abnormal intensities (see “Analysis of Exon array data” below). After probe filtering, we collected a final list of 330 Exon array probesets, with at least three reliable probes in each probeset to infer the splicing profiles of Alu-derived exons.
We collected 13103 constitutively spliced exons and 5389 exon-skipping cassette exons in the human genome, after applying stringent filtering criteria to exons in the Alternative Splicing Annotation Project 2 (ASAP2) database [62]. ASAP2 determined the splicing patterns of human exons based on the analysis of mRNA/EST sequences [62]. Constitutive exons were defined as those without any evidence of exon skipping in mRNA/EST data. To ensure that no skipping form was missed due to incomplete transcript sampling in EST databases, each constitutive exon included in our study was required to have at least 50 exon inclusion ESTs. We obtained 13103 high-confidence constitutive exons using this criterion. For exon skipping cassette exons, we collected 5389 ASAP2 exons with at least 3 inclusion ESTs and at least 3 skipping ESTs.
For each exon, we scored its 5′ and 3′ splice sites using consensus splice site models in MAXENT [15]. For 5′ splice site, we analyzed 3 nucleotides in exons and 6 nucleotides in introns. For 3′ splice sites, we analyzed 3 nucleotides in exons and 20 nucleotides in introns. We also calculated the density of exonic splicing regulatory elements (ESRs). Two sets of elements were used separately: (i) 285 exonic splicing regulatory elements from Goren et al [16]; (ii) 238 exonic splicing enhancers from Fairbrother et al [63]. For each exon, the ESR density was calculated as the number of nucleotides covered by ESRs, divided by the total length of the exon.
To determine the nucleotide substitution rate of exons in primates, we downloaded and analyzed the UCSC pairwise genome alignments of the human genome (hg18) to the genomes of chimpanzee (panTro2), orangutan (ponAbe2), rhesus macaque (rheMac2) and marmoset (calJac1) [64]. In each pairwise alignment, we defined an exon to be conserved in a non-human primate if there was at least one homologous region that covered at least 80% of the human exon with at least 80% sequence identity. We included a conserved exon in the nucleotide substitution rate analysis if there was a single (unambiguous) homologous region in the genome alignment. For such exons, we calculated the nucleotide substitution rate between the human genome and the genome of a non-human primate as the number of conserved nucleotide within the aligned region, divided by the total length of the aligned region. The alignment analysis was performed using Pygr [65], a python bioinformatics library that provided efficient access to alignment intervals in the UCSC genome alignments.
Briefly, we first predicted the background intensities of individual Exon array probes, using a sequence-specific linear model [41],[66] trained from “genomic” and “anti-genomic” background probes on the Exon 1.0 array [43]. For every probe, the predicted background intensity was an estimate for the amount of non-specific hybridization to the probe. This background intensity was subtracted from the observed probe intensity before downstream analyses [41]. Second, for each gene we used a correlation-based iterative probe selection algorithm to construct robust estimates of overall gene expression levels, independent of splicing patterns of individual exons [42]. Third, since oligonucleotide probes for Alu-derived exons may be more likely to cross-hybridize than typical Exon array probes, we used two independent methods to identify and remove individual probes with abnormal probe intensities. We searched all 25mer oligonucleotide probes against all RefSeq-supported exon regions, allowing up to 3 bp mismatches. Once a potential off-target gene was found for a probe, we calculated the Pearson correlation coefficient between the probe's intensities and the off-target gene's estimated expression levels across the 11 tissues [45]. We defined a probe to be cross-hybridizing if there was an off-target gene within 3 bp mismatches, and if the computed Pearson correlation coefficient was above 0.55. Such probes were removed from further analyses. We also detected probes whose intensities were higher than 95% of all other probes for RefSeq-supported exons of the same gene in at least 3 of the 11 tissues. Such probes were regarded as outlier probes and were also removed. After probe filtering, we collected a final list of 330 Exon array probesets, with at least three reliable probes in each probeset to infer the splicing profiles of Alu-derived exons.
For each Alu-derived exon, using a presence/absence call algorithm that compares the observed intensity of a probe to its predicted background intensity, we summarized a probeset-level Z-score for exon expression in individual tissues as in [41]. We also calculated the Pearson correlation co-efficient of individual probes' intensities with the overall gene expression levels in 11 tissues (estimated from all exons of a gene, see [41],[42]). We defined a probe to be “correlated” with gene expression levels if the Pearson correlation co-efficient was above 0.6. We defined an exon to be “correlated” if it had at least three probes correlated with gene expression levels.
We used a two-step approach to identify strong tissue-specific exons, by combining computational analysis and manual inspection of Exon array data. For each probe of an exon in a tissue, we calculated a “splicing index”, defined as the background-corrected probe intensity divided by the estimated gene expression level [40]. We used a Z-score method used by Graveley and colleagues [67] to test whether the splicing index of a particular tissue was an outlier compared to other tissues . A highly positive Z-score suggests tissue-specific exon inclusion. After this initial computational screening, we manually inspected the Exon array data of potential tissue-specific exons.
Total RNA samples from 10 human tissues were purchased from Clontech (Mountain View, CA). Single-pass cDNA was synthesized using High-Capacity cDNA Reverse Transcription Kit (Applied Biosystems, Foster City, CA) according to manufacturer's instructions. For each tested Alu-derived exon, we designed a pair of forward and reverse PCR primers at flanking constitutive exons using PRIMER3 [68]. Primer sequences and positions are described in Table S3. Two µg of total RNA were used for each 20 ul cDNA synthesis reaction. For each candidate Alu exonization event, 1 µl of cDNA were used for the amplification in a 25 µl PCR reaction. PCR reactions were run for 40 cycles in a Bio-Rad thermocycler with an annealing temperature of 62°C. The reaction products were resolved on 2% TAE/agarose gels. All of the candidate DNA fragments corresponding to exon inclusion and exon skipping forms were cloned for sequencing using Zero Blunt TOPO PCR Cloning Kit (Invitrogen, Carlsbad, CA).
Total RNA samples from rhesus macaque tissues (brain, skeletal muscle, pancreas) were purchased from Biochain Inc (Hayward, CA). Frozen tissue samples (cerebellum, skeletal muscle, liver, kidney) of two chimpanzees were generously provided by Southwest National Primate Research Center (San Antonio, TX). RNA was prepared using TRIzol (Invitrogen) according to the manufacturer's instructions. Single-pass cDNA was synthesized using the High Capacity cDNA Reverse Transcription Kit (Applied Biosystems, Foster City, CA). The quantitative real-time polymerase chain reaction (qRT-PCR) was performed using Power SYBR Green PCR Master Mix (Applied Biosystems, Foster City, CA). The following primers were used in qRT-PCR: SEPN1 Exon 3 skipping form: forward: 5′-GGGACAGATGGCCTTTTTCT-3′; reverse: 5′-AGTTGACCCTGTTAGCTTCTCAG-3′ ; SEPN1 Exon 3 inclusion form: forward 5′- GGAGTTCAAACCCATTGCTG -3′; reverse: 5′- AATTGAGCCAGGGAAGTTGA -3′. These qPCR primers match perfectly to their transcript targets in human and chimpanzee. Using a mathematical method described by Pfaffl [69], we calculated and presented the SEPN1 exon 3 inclusion level as a ratio to the exon 3 skipping level in each sample.
|
10.1371/journal.ppat.1000357 | Attachment and Entry of Chlamydia Have Distinct Requirements for Host Protein Disulfide Isomerase | Chlamydia is an obligate intracellular pathogen that causes a wide range of diseases in humans. Attachment and entry are key processes in infectivity and subsequent pathogenesis of Chlamydia, yet the mechanisms governing these interactions are unknown. It was recently shown that a cell line, CHO6, that is resistant to attachment, and thus infectivity, of multiple Chlamydia species has a defect in protein disulfide isomerase (PDI) N–terminal signal sequence processing. Ectopic expression of PDI in CHO6 cells led to restoration of Chlamydia attachment and infectivity; however, the mechanism leading to this recovery was not ascertained. To advance our understanding of the role of PDI in Chlamydia infection, we used RNA interference to establish that cellular PDI is essential for bacterial attachment to cells, making PDI the only host protein identified as necessary for attachment of multiple species of Chlamydia. Genetic complementation and PDI-specific inhibitors were used to determine that cell surface PDI enzymatic activity is required for bacterial entry into cells, but enzymatic function was not required for bacterial attachment. We further determined that it is a PDI-mediated reduction at the cell surface that triggers bacterial uptake. While PDI is necessary for Chlamydia attachment to cells, the bacteria do not appear to utilize plasma membrane–associated PDI as a receptor, suggesting that Chlamydia binds a cell surface protein that requires structural association with PDI. Our findings demonstrate that PDI has two essential and independent roles in the process of chlamydial infectivity: it is structurally required for chlamydial attachment, and the thiol-mediated oxido-reductive function of PDI is necessary for entry.
| Chlamydia is a large burden on global health. It is the most common cause of infectious blindness, and the CDC (Centers for Disease Control and Prevention) estimates that in the United States alone there are more than 2 million people with sexually transmitted Chlamydia infections. Chlamydia is an obligate intracellular bacteria; thus, attachment and subsequent invasion of cells are key steps in Chlamydia pathogenesis. While strides have been made in understanding the molecular mechanism of Chlamydia infection, fundamental aspects of this process still remain elusive. We have identified a host protein, protein disulfide isomerase (PDI), that is essential for Chlamydia attachment as well as for entry into cells. Cell-surface PDI-mediated disulfide reduction is required for Chlamydia entry into cells, whereas bacterial attachment is independent of PDI enzymatic activity. Although PDI is necessary for Chlamydia attachment, the bacteria apparently does not bind directly to cell-associated PDI, suggesting that Chlamydia attaches to a host protein(s) associated with PDI. This study advances our understanding of Chlamydia pathogenesis by the characterization of a host factor essential for independent stages of bacterial attachment and entry.
| Fundamental to understanding of intracellular bacterial pathogenesis is knowledge of the mechanism of bacterial attachment and subsequent entry into cells. There are two main processes by which bacteria stimulate their entry into nonphagocytic cells: by bacterial contact mediated activation of a cell surface receptor (the “zipper” mechanism) or by injecting bacterial proteins into the cell cytosol (the “trigger” mechanism) [1],[2]. Once the bacterial and host factors involved in the invasion process are identified this knowledge can be employed to devise antimicrobial strategies that target cellular infection. Blockade of this first step of bacterial infection is ideal for intracellular bacteria as these pathogens are able to avoid a number of host defenses by residing within cells.
Chlamydia is an obligate intracellular bacteria that can infect a number of different eukaryotic cells. Human chlamydial infection causes a wide range of pathologies. Chlamydia is the most common bacterial sexually transmitted disease [3], the leading cause of infectious blindness [4], and a community acquired respiratory pathogen [5]. Chlamydia infects cells as a metabolically inactive elementary body (EB) and then once within cells differentiates into the metabolically active but noninfectious form known as the reticulate body (RB). The EB are small (0.3-µm) and have a rigid outer membrane consisting of a mesh of disulfide cross-linked cysteine-rich proteins [6]. This membrane structure causes the EB to be osmotically stable and thus resistant to the stresses of the extracellular environment [7]. The RB, which is much larger (1-µm), is not osmotically stable owing to a decrease in disulfide cross-linked envelope proteins. Following replication the RB condense back into EB in a process that involves the expression of EB-specific disulfide-rich proteins and oxido-reductive processing. These EB can then infect neighboring cells or new hosts.
Attachment and entry into cells are key steps in chlamydial development and pathogenesis, yet the mechanism governing these interactions is still unknown. A number of bacterial ligands, including the major outer membrane protein [8], glycosaminoglycans [9],[10], heat shock protein 70 [11], and OmcB [12],[13] have been implicated in the process. It is likely that a host proteinacious factor(s) is involved in Chlamydia attachment as infectivity is lost following mild trypsin treatment of cells [14]. Several host proteins including: epithelial membrane protein 2 [15], mannose 6-phosphate receptor [16], the estrogen receptor complex [17], platelet-derived growth factor receptor [18], and protein disulfide isomerase (PDI) [17],[19] influence Chlamydia attachment. However, only one mammalian protein, PDI, has been demonstrated to be involved in attachment of multiple species and serovars of Chlamydia [19].
The role for PDI in Chlamydia infection was originally elucidated by proteomic analysis of CHO6 cells. CHO6 cells were generated by chemical mutagenesis of Chlamydia-susceptible CHOK1 cells and the mutant cells are resistant to attachment of C. trachomatis, C. pneumoniae, and C. psittaci [20]. CHO6 have a defect in PDI processing and express two forms of the protein. CHO6 express full length PDI, which is found in the parental cell line CHOK1, as well as a truncated form lacking the N-terminal signal sequence that is unique to CHO6 cells [19]. This truncated form of PDI present in CHO6 is not the result of a mutation in PDI itself but is instead likely due to an unidentified processing defect in CHO6 cells [19]. It was determined that ectopic expression of full length PDI in CHO6 rescues chlamydial attachment and consequently infectivity [19].
PDI is a multi-functional protein: it catalyzes the reduction, oxidation, and isomerization of disulfide bonds, it can act as a chaperone or anti-chaperone [21], and PDI is a subunit of collagen prolyl 4-hydroxylase and microsomal triglyceride transferase [22]. PDI consists of five domains. It contains two thioredoxin-like catalytic domains (a and a′) separated by two non-catalytic domains (b and b′) and has a small C-terminal domain (c). The two catalytic domains contain characteristic CGHC active-site motifs, the cysteines in these sites are essential for PDI enzymatic activity [23]. An ER retention signal (KDEL) is located at the C-terminal of PDI, PDI also has an N-terminal signal sequence. PDI is highly enriched in the endoplasmic reticulum but is also found in the cytosol, nucleus, and on the cell surface [24]. PDI reductive function at the cell surface has been shown to be essential for entry of a number of different viruses [25]–[28], as well as being necessary for activating diphtheria toxin [29].
It is known that PDI is involved in chlamydial infection [19], but the precise role of PDI is not clear as PDI has a number of diverse functions in cells. In this study we evaluated the role of PDI enzymatic activity in Chlamydia infectivity. We have determined that although cellular PDI is required for both Chlamydia attachment and entry the requirement is mechanistically different in the two processes. PDI cell surface enzymatic activity was necessary for entry of bacteria into cells. In contrast Chlamydia attachment to host cells required PDI but was independent of cell surface PDI enzymatic activity.
CHO6 cells, which have a mutation that affects PDI processing [19], are resistant to attachment of multiple species of Chlamydia [20]. It has previously been shown that chlamydial infectivity of CHO6 can be rescued by expression of recombinant PDI [19], suggesting that PDI is involved in bacterial attachment. The mutation in CHO6 leading to differential processing of PDI and consequent lack of Chlamydia attachment is unknown. PDI is essential for cell viability, thus gene disruption approaches cannot be used to test if PDI is necessary for Chlamydia attachment or if additional mutations play a role in the phenotype of the CHO6 cell line. In contrast, siRNA has been used to transiently knockdown expression of PDI in mammalian cells [30]. siRNA-mediated downregulation of PDI was used to test whether PDI alone is required for Chlamydia infectivity. HeLa cells were transfected with PDI-targeting siRNA and PDI knockdown was assessed by immunoblot. Approximately 80–90% reduction in PDI expression was achieved (Figure 1A). When PDI was depleted from HeLa cells bacteria were no longer able to attach to and thus infect cells (Figure 1B). The degree of chlamydial attachment to siRNA treated cells was quantified by counting the number of bacteria attached to cells. Following PDI depletion there were only 0.69±1.04 C. psittaci and 1.27±1.39 C. trachomatis per cell, a 97.8% and 97.7% reduction in attachment respectively (Figure 1C). These results demonstrate that cellular PDI is essential for attachment of Chlamydia to the surface of host cells, making PDI the only host protein identified that is required for Chlamydia attachment.
To further characterize the nature of the PDI requirement in Chlamydia infectivity, we tested the hypothesis that the oxido-reductive or cysteine isomerase roles of PDI were required for restoration of Chlamydia attachment and entry to CHO6 cells. The catalytic domain of PDI is well characterized and specific amino acids involved in PDI isomerase activity have been identified [31]. The role of PDI enzymatic activity in chlamydial infectivity was analyzed by generating recombinant PDI with disabled catalytic domains. This was accomplished by converting the four key cysteine residues within the catalytic domains to serine (PDI-4CS). Although PDI-4CS is no longer able to reduce, oxidize, or rearrange disulfide bonds it can still fulfill PDI chaperone, anti-chaperone, and structural roles [31].
Prior to examination of Chlamydia infectivity restoration in cells expressing PDI-4CS, cell surface PDI enzymatic activity was evaluated. Plasma membrane PDI activity can be tested by cellular sensitivity to diphtheria toxin (DT) as cell surface PDI enzymatic activity is required for DT-mediated killing of cells. PDI reduces interchain disulfide bonds in DT triggering chain separation that then allows for translocation of DT chain A across the membrane and subsequent toxin-induced cell death [29]. CHO6 cells have aberrant cell surface PDI activity and are resistant to DT. If native PDI is expressed in CHO6 cells toxin sensitivity is restored [19]. DT sensitivity of CHO6 expressing native PDI (CHO6+PDI) or CHO6 expressing enzymatically disabled PDI (CHO6+PDI-4CS) were tested. As previously reported, CHO6+PDI became sensitive to toxin (Figure 2), whereas CHO6 cells expressing PDI lacking enzymatic function (CHO6+PDI-4CS) were largely resistant to DT (Figure 2). These results demonstrated that cell surface PDI enzymatic activity is rescued in CHO6 following expression of native PDI but not in cells expressing PDI lacking enzymatic function. These data establish a model that can be used to specifically address the role of plasma membrane PDI enzymatic activity in Chlamydia infectivity.
CHO6 cells were transfected with vectors expressing native PDI or PDI-4CS and both attachment and entry were separately evaluated. Despite lacking PDI enzymatic activity at the cell surface, CHO6 cells transfected with PDI-4CS showed equivalent Chlamydia attachment as cells transfected with native PDI (Figure 3A). The level of attachment recovery was quantified by counting the number of bacteria associated with cells (Figure 3B). These data show that while PDI is necessary for chlamydial attachment, the function of PDI in attachment is independent of the protein's enzymatic activity.
Because PDI that is enzymatically nonfunctional could restore bacterial attachment to the cell, the ability of Chlamydia to establish a productive infection in cells expressing the parental or enzymatic mutant PDI protein was evaluated (Figure 3C). The development of Chlamydia laden vacuoles was only observed in the parental CHOK1 and CHO6 cells expressing PDI (Figure 3C). Infection rescue was quantified by determining the number of inclusion per field of view (Figure 3D), no inclusion were seen in CHO6 cells or CHO6 cells expressing PDI-4CS. It was noted that despite the lack of productive infection of CHO6+PDI-4CS, many of the bacteria remained persistently attached to the cells throughout the 24 h course of the experiment (Figure 3C).
Enzymatically nonfunctional PDI was capable of rescuing attachment but not infectivity. This outcome could be the result of an enzymatic role for PDI in either cellular entry or in other chlamydial developmental processes soon after entry that may prevent growth. Chlamydial entry was evaluated by allowing attached Chlamydia to enter cells by shifting them to 37° C for 2 h. Entry was analyzed by comparing the number of surface bound bacteria by immunofluorescence staining of unpermeabilized cells between CHO6+PDI versus CHO6+PDI-4CS. After 2 h at 37°C, entry was restored in CHO6+PDI with 95.7% of bacteria internalized following the 2 h incubation (Figure 3E and 3F). In contrast, there was no significant internalization of Chlamydia that had attached to CHO6+PDI-4CS (Figure 3E and 3F). Although PDI, independent of its enzymatic function was sufficient to restore chlamydial attachment, enzymatically functional PDI was required for uptake of bacteria into host cells.
From these results it can be concluded that PDI serves two distinct yet essential functions for Chlamydia attachment and entry. The role of PDI in attachment is not enzymatic but perhaps limited to a structural or chaperone function, whereas subsequent bacterial entry requires PDI enzymatic activity.
Evaluation of Chlamydia infectivity in CHO6 cells expressing enzymatically nonfunctional PDI indicate that cell surface PDI activity is required for entry of bacteria but not for attachment. However, from those experiments it is equivocal if the PDI enzymatic activity necessary for bacterial entry is occurring at the cell surface or intracellularly. To specifically test if the required PDI enzymatic activity was occurring at the cell surface, Chlamydia attachment and entry were evaluated in CHOK1 cells in the presence of bacitracin. Bacitracin is a membrane impermeable PDI-specific inhibitor that has been used to test the role of plasma-membrane-specific function of PDI in mammalian cells [32]–[34]. Bacitracin is considered to be a PDI inhibitor because it does not inhibit thioredoxin mediated reduction [32]–[34]. The precise mechanism of bacitracin inhibition of PDI function is not known. Bacitracin has previously been used with Chlamydia. Davis et al. [17] found that addition of the inhibitor during infection led to a 16 to 36% decrease in bacterial infectivity. Given the dual role of PDI in Chlamydia infection illuminated by genetic approaches, we revisited these early findings by examining the effect of bacitracin-mediated inhibition of cell surface PDI on not only Chlamydia infection but also attachment.
CHOK1 cells were inoculated in the presence of bacitracin and attachment was then evaluated. In the presence of bacitracin Chlamydia attached to cells at similar levels as untreated cells (Figure 4A). This demonstrates that extracellular PDI enzymatic function is not required for chlamydial attachment and supports the results of our experiments with CHO6 expressing enzymatically nonfunctional PDI (Figure 3A and 3B). The role of plasma membrane PDI function in Chlamydia entry was examined by inoculating cells and cultivating them in media containing bacitracin for 2 h. When bacitracin was present during initial infection and for the 2 h following attachment Chlamydia were unable to enter cells and EB remained persistently attached to cells (Figure 4B). This confirmed that there is a requirement for plasma membrane PDI activity for chlamydial entry and supports our previous analyses. The effect of bacitracin on bacterial attachment and infection was quantified using a quantitative real-time PCR assay as described previously [35]. When bacitracin was present the level of attachment was 92.9% that of untreated cells, indicating that cell surface PDI enzymatic activity is not necessary for Chlamydia attachment (Figure 4C). In contrast when the level of infection was quantified 24 h after bacterial attachment, infection of bacitracin treated cells was reduced by 73.4% as compared to untreated cells (Figure 4C).
Along with being a PDI-specific inhibitor, bacitracin is also a bactericidal antibiotic that targets gram-positive bacteria. The mechanism of bacitracin-mediated bacterial death is that it inhibits bacterial cell wall synthesis by inhibiting dephosphorylation of lipid pyrophosphate. Chlamydia is a gram negative-like bacteria, but it remained important to ensure that the block of bacterial entry by bacitracin was not simply due to damage to chlamydial organisms by the inhibitor treatment. We determined that treating cells with the inhibitor only prior to infection and followed by washing had no effect on Chlamydia attachment or entry (Figure 4D). Normal bacterial development was also seen if the inhibitor was removed or added after the first 8 h of the infection (Figure 4D). The reversibility of the inhibition demonstrates that the observed effects were not due to damage to Chlamydia or the cell by the inhibitor treatment. To ensure that the requirement for PDI enzymatic activity was not unique to CHOK1 cells attachment and infectivity analysis with bacitracin was also conducted in HeLa cells. As in CHOK1 cells Chlamydia attached to HeLa cells in the presence of bacitracin but were unable to enter (Figure S1).
Using genetics we determined that PDI lacking enzymatic function was able to complement attachment but not bacterial entry and subsequent development. DT analysis demonstrated that following complementation with enzymatically nonfunctional PDI there was a defect in cell surface PDI activity. The ability of bacitracin to inhibit bacterial entry, in addition to our results following PDI expression in CHO6, define an essential role at the cell surface for PDI enzymatic function in Chlamydia entry.
Having established that cell surface PDI enzymatic activity was required for Chlamydia entry into cells we next sought to determine the molecular mechanism of that activity. PDI is able to reduce, oxidize, and rearrange disulfide bonds and all three of these activities are arrested in the presence of bacitracin. Because of the role of PDI-mediated reduction in viral entry [25]–[28], we evaluated if it was a reductive function that was necessary for Chlamydia entry into cells. CHOK1 cells were infected with Chlamydia in the presence of bacitracin, arresting infectivity of the bacteria at the cell surface. The membrane impermeant disulfide reducing agent TCEP (Tris(2-carboxyethyl)phosphine hydrochloride) was then added and the cells were incubated at 37° C for 2 h. The addition of TCEP was able to overcome bacitracin inhibition of Chlamydia entry, confirming that the stimulus necessary for bacterial entry into cells is a PDI-mediated reduction occurring at the cell surface (Figure 5A). Following entry and TCEP removal Chlamydia were able to establish a productive infection in cells demonstrating that the bacteria were not damaged by the 2 h incubation with TCEP (Figure 5B).
The mechanism of bacitracin mediated inhibition of PDI enzymatic activity is not known, making it possible that TCEP was simply inhibiting the interaction between PDI and bacitracin and in that manner restoring bacterial entry. To control for this possibility the effect of TCEP on entry of Chlamydia into CHO6 cells expressing enzymatically nonfunctional PDI (CHO6+PDI-4CS) was analyzed. Chlamydia are able to attach to CHO6 cells expressing PDI-4CS but the bacteria are unable to enter (Figure 3). When TCEP was added the bacteria were then able to enter the CHO6 cells expressing non-enzymatically functional PDI (Figure 5C). From these experiments we can conclude that it is cell surface PDI mediated reduction that is required for Chlamydia uptake into cells.
The functional role of PDI in the bacterial attachment stage was next examined. The most direct hypothesis is that Chlamydia bind PDI as a receptor. PDI is not an integral membrane protein. PDI is secreted from cells and then maintained at the surface through electrostatic interactions with other cell surface proteins [24]. To test for Chlamydia binding directly to cell surface PDI we generated a PDI protein that was tethered to the plasma membrane by a C-terminal gpi anchor. A similar strategy has been used to study the role of PDI in HIV entry [28]. This plasma membrane anchored PDI (PDI-gpi) was expressed in CHO6 cells and resulted in high-level cell surface PDI expression that was readily detectable by PDI-specific antibody (Figure 6A).
Prior to analysis of bacterial attachment to CHO6 expressing PDI-gpi, it was first determined if the gpi-anchored PDI was able to interact with other cell surface proteins in the same manner as unanchored PDI. This was tested by evaluating DT sensitivity of cell expressing PDI-gpi. PDI interaction with DT bound to its cell surface receptor (heparin-binding epidermal growth factor) is necessary for DT mediated cell death [36]. Whereas toxin sensitivity is recovered by expression of unanchored PDI (Figure 2), the gpi-anchored PDI was unable to normally interact with other cell surface proteins and could not restore DT sensitivity to CHO6 cells (Figure 6B). The high-level of PDI-gpi expression did not interfere with normal cell surface interaction, as it did not reduce CHOK1 toxin sensitivity (Figure 6B). From these results we could conclude that CHO6 expressing PDI-gpi functioned as a model for testing if Chlamydia was able to directly attach to PDI independent of PDI interactions with other cell surface proteins. Chlamydia attachment to CHO6 cells expressing PDI-gpi was evaluated, no recovery of bacterial attachment was observed (Figure 6C). These results suggest there is a lack of Chlamydia binding directly to PDI.
The experiments with the gpi-anchored PDI are indicative of a lack of direct attachment on Chlamydia to PDI, but the possibility that the gpi-anchor was causing a structural change in PDI that led to the lack of bacterial attachment could not be ruled out. To address this possibility the effect of polyclonal PDI-specific antibody of Chlamydia attachment was analyzed. The effect of antibody to PDI on Chlamydia infection has previously been evaluated by Davis et al. [17], they observed a temperature dependent decrease in Chlamydia infection. We further developed those findings by specifically addressing inhibition of bacterial attachment as well as entry. When PDI antibody was present prior to and during bacterial attachment no significant change in the number of bacteria attached to cells was observed (Figure 7A), suggesting that Chlamydia was not directly binding PDI. Bacterial entry in the presence of PDI antibody was also evaluated. The antibody significantly reduced bacterial entry (Figure 7B) and led to a persistent attachment phenotype similar to what was seen in the presence of bacitracin (Figure 4). The inhibition of Chlamydia entry by PDI antibody corroborated our previous determination that cell surface PDI reductive function was required for bacterial uptake.
A polyclonal PDI antibody was used in these experiments, and it is likely that this antibody would inhibit PDI enzymatic activity by steric interference. To directly assess the level of PDI enzymatic activity a turbidimetric assay of insulin disulfide reduction was performed (Protocol S1). We determined that addition of PDI antibody to the reaction significantly reduced the rate and level of insulin reduction, indicating that the antibody had an inhibitory effect of PDI enzymatic activity (Figure S2). We conclude that while cellular PDI is necessary for Chlamydia attachment to cells the bacteria does not initiate attachment through a direct interaction with PDI. Much like HIV and diphtheria toxin, Chlamydia likely binds to a cell surface protein(s) that is associated with PDI.
Cell surface PDI-mediated disulfide bond reduction is involved in the infectious entry of a number of viruses. Upon binding of human immunodeficiency virus (HIV) envelope protein to CD4 receptor and co-receptor (CCR5 or CXCR4) PDI reduces disulfide bonds in HIV gp120, exposing the gp41 fusion peptide [28]. The fusion peptide then inserts into the target cell surface triggering viral and cell membrane fusion. For Newcastle disease virus, cell surface PDI enzymatic activity is required for the conformational changes in the viral fusion protein that are necessary for cell-viral membrane fusion [27]. Similarly PDI-mediated reduction of Sindbis virus envelope is required for membrane fusion and release of the viral genome into cells [25]. We have experimentally established that the initial stages of chlamydial infectivity, attachment and entry, each require host cell surface PDI; however, the functional participation of PDI was mechanistically unique at each stage. Cell surface PDI enzymatic activity was required for Chlamydia entry. Independent of PDI enzymatic function, PDI was additionally essential for bacterial attachment to host cells.
The role for protein disulfide exchange in chlamydial infection has been previously explored using inhibitors. Davis et al. [17] and Raulston et al. [11] showed that inhibition of cell surface reductive function by addition of bacitracin or dithio-bis-2-nitrobenzoic acid (DTNB) adversely affected infectivity of C. trachomatis serovar E. These results can now be understood in terms of the direct enzymatic role for PDI in bacterial entry. Raulston et al. [11] found no inhibition of bacterial attachment using DTNB similar to our observations with bacitracin. When Davis et al. [17] and Raulston et al. [11] evaluated the effect of bacitracin or DTNB they reported a slight decrease in infectivity, whereas we observed a near complete loss of bacterial entry and subsequent infection. These differences are likely due to the fact that PDI is constitutively trafficked to the cell surface [24], making bacitracin inhibition functionally reversible. Both Davis et al. [17] and Raulston et al. [11] removed the inhibitor after initial bacterial attachment allowing newly exported PDI to stimulate uptake. In our experimental design, the inhibitor was present throughout the course of the infection and if the inhibitor was removed following attachment, our results were similar to those of Davis et al. [17] and Raulston et al. [11]. Our data also significantly expand the previous observations as this now extends the requirement for disulfide exchange in the process of bacterial infection from one strain to multiple species of Chlamydia.
A requirement for PDI in chlamydial attachment was implicated by complementation of the PDI gene in attachment-deficient mutant CHO6 cells [19]. We anticipated that the oxido-reductive enzymatic activity of PDI would be required for bacterial attachment, similar to what is characterized for viral attachment [25],[27],[28]. It was a surprise that chlamydial attachment to CHO6 cells was rescued by complementation with the enzymatically inactive PDI-4CS. Further evaluation using PDI inhibitors confirmed that Chlamydia attachment is independent of cell surface PDI enzymatic activity. Testing of gpi-anchored PDI tethered to the cell surface suggested that chlamydiae do not directly bind PDI as the sole target for attachment and this implicates an interaction with other proteins that require PDI. Consistent with characterized functions of PDI, it may be PDI's function as a chaperone or structural component of a host protein or protein complex [37] that is required for chlamydial attachment. PDI chaperone activity is independent of the protein's two catalytic domains (a, a′) and PDI functions as a chaperone outside of the ER [38],[39]. PDI could serve to stabilize a host receptor protein in the correct orientation or context for bacterial binding. There are also examples of PDI functioning as a structural subunit of proteins. PDI is the β-subunit of tetrameric enzyme collagen prolyl 4-hydroxylase, and PDI also makes up half of the heterodimeric protein complex microsomal triglyceride transfer protein [40],[41]. The receptor that Chlamydia binds may be a multiprotein complex that includes PDI.
In addition to a requirement of PDI for chlamydial attachment, it was shown that PDI enzymatic activity was necessary and sufficient to stimulate chlamydial entry into the host cell. Given the ability of simple chemical reduction to replace the enzymatic function of PDI for cell-adherent chlamydiae, it appears that reduction rather than disulfide exchange is minimally required for chlamydial entry. It is not known whether PDI functions to reduce a host or bacterial component to initiate entry. It has been previously shown that reduction of C. trachomatis L2 EB prior to infection leads to a decrease in inclusion forming units [42]. This experiment illustrates the problem of differentiating the effect of reduction on the infectious process. One can pre-reduce host cells or bacteria prior to infection, but due to the rapid rate of reoxidation following the reducing agent removal step and unknown detrimental effects, especially following pan-blocking of disulfides by chemical agents, make interpretation of outcomes confounded and uncertain.
One can speculate that following initial bacterial attachment, PDI enzymatic activity mediates the establishment of a functional contact between the chlamydial organism and the host cell leading to bacterial uptake. It is known that plasma membrane PDI interacts with a number of host surface proteins. For example, PDI modifies the adhesion receptors integrin α2β1 and L-selectin resulting in receptor activation and ligand binding [33]. As well as modifying cellular surface proteins, PDI binds and modifies various viral proteins and toxins [25]–[29]. Thus, there is biological precedent supporting PDI modification of either host or microbial factors required for pathogenesis.
Multiple cell surface proteins have been previously implicated in Chlamydia infectivity, these include epithelial membrane protein 2 [15], mannose 6-phosphate receptor [16], estrogen receptor complex [17], and platelet-derived growth factor receptor [18]. Unlike PDI, these proteins appear to be involved in the infectivity of only a subset of chlamydial species or biovars. PDI interacts with a broad array of host proteins within the endoplasmic reticulum and at the cell surface allowing for the possibility that PDI may be an underlying requirement for several independent Chlamydia-host cell receptor interactions. PDI is involved in protein folding in the endoplasmic reticulum and PDI expression can be correlated to the level of secretion of a number of proteins [43]–[45]. We established using PDI-specific chemical inhibitors, anti-PDI antibodies, and chemical reduction of bound organisms that the function of PDI in chamydial infectivity requires surface accessible PDI. Proteomic analysis using 2-dimensional gel electrophoresis of biotin-labeled surface proteins failed to show any detectible and consistent difference between CHO6 and CHOK1 cell lines other than for PDI (19 and data not shown). This suggests that the lack of bacterial attachment to CHO6 cells is not due to a general defect in plasma membrane protein composition. If the enzymatic role of PDI is targeted to a host protein it seems likely that this occurs at the cell surface and not in a secretion pathway.
Alternative to acting on a host protein, PDI could be targeting the chlamydial organism. The highly disulfide cross-linked structure of the chlamydial EB surface proteins that are only present in the infectious form of the chlamydial organism [7], suggest that these could be the target for PDI. It has been shown that reduction of the EB surface is necessary for surface display of bacterial Hsp70 [11] and it is possible that PDI activates changes in chlamydial surface proteins that are required to initiate cellular entry. The chlamydial protein Tarp is translocated into host cells following bacterial attachment [46]. This translocation occurs via type III secretion and triggers actin recruitment to the site of bacterial attachment [46]. Tarp is localized within EB and is not surface exposed until the commencement of bacterial entry [46]. PDI-mediated reduction of EB surface proteins may be essential to activate Chlamydia type III secretion into the cell membrane and subsequent Tarp translocation. The type III secretion system needle protein of C. trachomatis, CdsF, was recently identified [47]. CdsF is conserved among the chlamydial species, and it is one of the few bacterial needle proteins that contains cysteine residues. One can speculate that PDI-mediated reduction of the EB surface is involved in CdsF function.
While the full intricacies of the Chlamydia-host cell interaction remain enigmatic these findings illuminate important new details of the molecular mechanisms involved. Chlamydia attachment and entry into cells are separable processes that both have a unique requirement for PDI. Determination of the target of PDI enzymatic activity that leads to chlamydial entry may be expected to provide targets for generation of new anti-Chlamydia therapies and illuminate fundamental cell biological processes exploited by Chlamydia to mediate pathogen attachment and infectivity.
CHOK1 and CHO6 cells were maintained in RPMI 1640 (Invitrogen, Carlsbad, CA) supplemented with 10% heat-inactivated fetal bovine serum (FBS) (Hyclone, Logan, UT), 2 mM glutamine (Invitrogen), and 1 mM HEPES (Invitrogen). HeLa and L929 cells were maintained in RPMI 1640 supplemented with 10% FBS. Cells were grown at 37°C in an atmosphere containing 5% CO2. C. trachomatis L2/434/Bu EB and C. psittaci PF6 BC were purified from L929 cells on 30% and 30–44% discontinuous Renografin gradients (E. R. Squibb and Sons, Cranbury, NJ) [48] and stored at −80°C until use.
On the night prior to analysis 1×105 cells were plated on 12 mm glass coverslips in 24 well plates. Cells were washed with Hanks buffered saline solution (HBSS) (Invitrogen) and then incubated with C. trachomatis L2/434/Bu or C. psittaci PF6 BC in RPMI 1640 with 10% FBS for 1 h at 24°C [19], [49]–[52]. Following infection cells were washed 4 times with HBSS and methanol fixed for attachment analysis. For entry and infectivity analysis fresh cell culture media was added and cells were incubated 2 h for entry and 24 h for infectivity at 37°C in an atmosphere containing 5% CO2. For entry analysis cells were fixed with methanol or with 4% paraformaldehyde, for infectivity assessment cells were fixed with methanol. Following fixation coverslips were incubated 15 min in blocking solution (HBSS+2.5% bovine serum albumin (BSA)) (Fisher, Fair Lawn, NJ). Blocking solution was removed and coverslips were incubated 1 h with mouse anti-Chlamydia MOMP antibody for C. trachomatis or anti-Chlamydia LPS antibody (Santa Cruz Biotechnology, Santa Cruz, CA) for C. psittaci and then washed with HBSS. The wash was followed with a 30 min incubation with goat anti-mouse AlexaFluor 488 (Invitrogen) and Evans Blue. Coverslips were mounted in VectaShield Hard Set mounting media (Fisher) and evaluated on a Nikon Eclipse E800 microscope. When used bacitracin (Sigma, St. Louis, MO) was added 20 min prior to infection at 3 mM, infection was performed in media containing 3 mM bacitracin and following infection cells were maintained in culture media with 3 mM bacitracin. When used 50 mM TCEP (Pierce, Rockford, IL) in cell culture media was added following bacterial attachment. For attachment and entry inhibition analysis cells were incubated for 20 min with polyclonal rabbit anti-PDI antibody (Stressgen Victoria, BC) or goat anti-bovine IgG antibody (Pierce, Rockford, IL), the bacterial inoculum was then added to the antibody containing media and cell were incubated for 1 h at 24°C. When Stressgen rabbit anti-PDI antibody was used for PDI visualization, PDI staining was performed similarly to Chlamydia staining and goat anti-rabbit AlexaFluor 594 (Invitrogen) was used as a secondary.
For quantification the number of cell-associated apple-green fluorescing particles of size and shape consistent with 300 nM organisms were counted in two planes in 8 separate fields of view containing at least 10 cells. Fluorescent particles that appeared to be Chlamydia but were larger due to aggregation were enumerated separately when separate organisms could be discerned or when distinction was not possible were counted as one to provide a conservative estimate of bound organisms per cell.
The gene encoding protein disulfide isomerase was cloned from CHOK1 cDNA and inserted into the expression vector pBICEP-CMV-1 (Sigma) yielding PDI with a N-terminal FLAG tag. PDI protein lacking enzymatic function (PDI-4CS) was generated by mutating 4 essential cysteine residues to serine with the Stratagene (La Jolla, CA) QuikChange Kit and primers Cys-55,58-Ser F (5′ - TGC CCC GTG GTC TGG CCA CTC CAA AGC TCT GG - 3′), Cys-55,58-Ser R (5′ - CCA GAG CTT TGG AGT GGC CAG ACC ACG GGG CA - 3′), Cys-399,402-Ser F (5′ - TAT GCC CCC TGG TCT GGC CAC TCC AAG CAG CT - 3′), and Cys-399,402-Ser R (5′ - AGC TGC TTG GAG TGG CCA GAC CAG GGG GCA TA - 3′). The gpi-anchor for PDI was amplified from human folate receptor 1 using primers GPI F (5′ - ATT GCC CGG GGC TGC AGC CAT GAG TGG - 3′) and GPI R (5′ – CAA TGT CGA CTC AGC TGA GCA GCC ACA GCA – 3′). The gpi-anchor was than ligated to PDI lacking a stop codon and the PDI-gpi construct was ligated into the pBICEP-CMV-1 vector. PDI vectors were transfected into CHO cells with Effectene (Qiagen, Valencia, CA) following manufacturer's instructions. Chlamydia attachment and entry were evaluated 48 h after transfection.
PDI was silenced in HeLa cells with PDI-specific siRNA (5′- GAC CTC CCC TTC AAA GTT GTT –3′) (Dharmacon, Layafette, CO) [30]. Dharmacon siControl non-targeting siRNA #1 was used at as a negative control. Cells were transfected as described in Hybiske and Stephens, 2007 with slight modification [35]. Cells were first transfected in 6 well plates with 10 µl of 40 µM siRNA duplexes with 5 µl of Olgiofectamine (Invitrogen) in OptiMem (Invitrogen). 24 h after transfection cells were transfected again. 90 h after the first transfection cells were replated on fibronectin (Sigma) coated coverslips in 24 well plates, and attachment, infectivity, or protein expression was evaluated 6 h after replating.
24 h prior to toxin treatment cells were plated in 96-well plates at 1×104 cells per well. Cells were washed once with HBSS and than incubated for 4 h at 37°C with diphtheria toxin (Biomol, Plymouth Meeting, PA) (0 or 100 ng ml−1) in HBSS containing 15% FBS. Following incubation cells were washed three times with HBSS and were than maintain in RPMI 1640 without phenol red supplemented with 10% FBS, 2 mM glutamine, and 1 mM Hepes at 37°C in an atmosphere containing 5% CO2. 72 h after toxin treatment cell viability was determined by Promega (Madison, WI) CellTiter 96 Aqueous One Cell Proliferation Assay as directed by the manufacturer's instructions. Cell viability is linearly proportional to absorbance.
The number of Chlamydia per cell was determined using a quantitative PCR based method previously described by Hybiske and Stephens, [35]. Genomic DNA from CHOK1 cells infected with C. trachomatis was isolated using the High-Pure PCR template preparation kit (Roche, Indianapolis, IN). Purified genomic DNA was used as a template in quantitative PCR to determine the relative levels of chlamydial (16S) and CHOK1 (ß-globin) genomic equivalents.
The NCBI Entrez (http://www.ncbi.nlm.nih.gov/sites/entrez?db=protein) accession number for genes discussed in this paper are Homo sapiens PDI (NP_006840) and Cricetulus griseus PDI (AAM_00284).
|
10.1371/journal.ppat.1002252 | Robust Antigen Specific Th17 T Cell Response to Group A Streptococcus Is Dependent on IL-6 and Intranasal Route of Infection | Group A streptococcus (GAS, Streptococcus pyogenes) is the cause of a variety of clinical conditions, ranging from pharyngitis to autoimmune disease. Peptide-major histocompatibility complex class II (pMHCII) tetramers have recently emerged as a highly sensitive means to quantify pMHCII-specific CD4+ helper T cells and evaluate their contribution to both protective immunity and autoimmune complications induced by specific bacterial pathogens. In lieu of identifying an immunodominant peptide expressed by GAS, a surrogate peptide (2W) was fused to the highly expressed M1 protein on the surface of GAS to allow in-depth analysis of the CD4+ helper T cell response in C57BL/6 mice that express the I-Ab MHCII molecule. Following intranasal inoculation with GAS-2W, antigen-experienced 2W:I-Ab-specific CD4+ T cells were identified in the nasal-associated lymphoid tissue (NALT) that produced IL-17A or IL-17A and IFN-γ if infection was recurrent. The dominant Th17 response was also dependent on the intranasal route of inoculation; intravenous or subcutaneous inoculations produced primarily IFN-γ+ 2W:I-Ab+ CD4+ T cells. The acquisition of IL-17A production by 2W:I-Ab-specific T cells and the capacity of mice to survive infection depended on the innate cytokine IL-6. IL-6-deficient mice that survived infection became long-term carriers despite the presence of abundant IFN-γ-producing 2W:I-Ab-specific CD4+ T cells. Our results suggest that an imbalance between IL-17- and IFN-γ-producing CD4+ T cells could contribute to GAS carriage in humans.
| Group A streptococcus (GAS) causes many different conditions, ranging from strep throat, flesh eating disease to post infectious complications involving the heart. Here, we used a novel technique to study the CD4+ T cell immune response against GAS infection in a mouse model. We first generated a recombinant GAS strain that expresses a specific epitope (2W) - M protein fusion and used this to intranasally inoculate mice. Peptide specific CD4+ T cells were concentrated and analyzed using 2W-MHC-II tetramers. This technology allowed us to probe the antigen specific CD4+ T cell response to new depths and certainty. Infection induced a robust 2W-specific Th17 cell response, which was dependent on the route of infection, IL-6, and was independent of superantigens. IL-6-/- mice were exquisitely susceptible to intranasal infection. However, those that survived became immune carriers, unable to clear streptococci from NALT. Further, multiple infections generated an IL-17+ IFN-γ+ double positive population of CD4+ T cells that are known to be associated with autoimmune disease in humans and directly responsible for autoimmune pathology in rodent models. Our results provide a new direction for understanding two important consequences of streptococcal pharyngitis, the very common immune carrier state, and the rarer state involving autoimmune complications.
| Group A streptococcus (GAS, Streptococcus pyogenes) is an important bacterial pathogen that causes many different clinical conditions ranging from pharyngitis, impetigo, toxic shock syndrome, necrotizing fasciitis to post infectious autoimmune sequelae like acute rheumatic fever and glomerulonephritis [1], [2]. GAS is an important cause of morbidity and mortality worldwide. Estimates are that 500,000 deaths occur each year due to severe GAS infections [3]. GAS produces a wide variety of virulence factors that play important roles in adhesion, immune evasion, dissemination, tissue destruction and toxicity [1], [4]. Remarkably, one third of those treated with antibiotics continue to shed streptococci and a significant number of these carriers have recurrent disease by the same strain [5], [6]. Carrier state is an important public health problem because it maintains the cycle of disease in a community. Tonsils are known to harbor and shed streptococci, even after intense antibiotic therapy. Osterlund et al found that 93% of tonsils excised from children retained intracellular GAS, and others reported isolation of GAS from excised tonsils, confirming that this secondary lymphoid tissue is an important reservoir [7].
The capacity of GAS to survive in lymphoid tissue is not understood. Although GAS induces a humoral immune response in most individuals, streptococci can still persist. GAS also induces a CD4+ T cell response [8]. CD4+ T cells express T cell antigen receptors (TCR) that recognize short peptides bound to MHCII molecules (pMHCII) expressed by host antigen-presenting cells. During primary infection, the rare naïve CD4+ T cells, which by chance express TCRs complementary to bacterial pMHCII complexes, proliferate and differentiate into Th1, Th2, or Th17 effector cells that produce cytokines such as IFN-γ, IL-4 or IL-17, respectively, which help to eliminate the pathogen. Which of these effector cell types will develop during infection is determined by cytokines produced by cells of the innate immune system. For example in mice, the differentiation of Th17 cells in vitro depends on IL-6 and Transforming Growth Factor-beta (TGF-β) [9]. We recently demonstrated that GAS infection of the NALT, the murine equivalent of the tonsils, generates Th17 cells, which contribute to immune protection [10]. Lu et al have shown that Th17 cells protect against other streptococcal infections as well [11].
Previous studies used adoptive transfer of monoclonal TCR transgenic T cells to study T cell responses to infection with antigen tagged GAS [8], [12]. A drawback of this approach is that it results in an abnormally large number of naïve precursors, which experience inefficient activation due to competition for limiting pMHCII [13]-[15]. Therefore, the nature of the CD4+ T cell response to GAS infection under physiological conditions is still unknown.
To avoid these limitations, we used a new cell enrichment method based on fluorochrome-labeled pMHCII tetramers and magnetic beads [16] to characterize the endogenous polyclonal CD4+ T cell response to GAS. This approach depended on the availability of a tetramer of the I-Ab MHCII molecule of the preferred C57BL/6 mouse strain bound to a peptide from the GAS proteome. Unfortunately, no such peptide has been identified to date. We therefore produced a recombinant GAS strain that expresses an immunogenic peptide (EAWGALANWAVDSA) called 2W [17] fused to the M1 protein on its surface. 2W:I-Ab tetramer staining and magnetic bead enrichment was used to characterize 2W:I-Ab+ CD4+ T cells from NALT and other lymphoid tissues after intranasal GAS-2W infection. Our results demonstrate that an intranasal infection is critical for mounting an effective IL-6-dependent pMHCII-specific Th17 response. A lack of this response led to a preponderance of Th1 cells and failure to control GAS infection. This work defines the Th17 response to GAS infection, and may shed light on the basis for the carrier state and autoimmune complications in humans.
An M1 GAS strain 90–226 was genetically engineered to express the 14 amino acid 2W peptide (EAWGALANWAVDSA) as a cell wall surface hybrid M1 fusion protein. The hybrid emm1.0::2W gene was constructed in plasmid pFW5 in E. coli and then introduced into the chromosomal emm1.0 gene by allelic replacement [18]. The corresponding chimeric protein is composed of the 14 amino acid 2W peptide inserted in frame after the first five amino acids of the mature M1 surface protein (Fig. 1A). The strain, designated 90–226 emm1.0::2W (GAS-2W), is genetically stable without spectinomycin selection due to replacement of the wild-type gene in the chromosome. The anti-phagocytic property of the M1-2W fusion protein was assessed to test whether insertion of the 2W peptide altered the function of the M protein. GAS-2W was as resistant to phagocytosis as the wild-type 90-226 in whole blood bactericidal assays (Fig. 1B). An M- variant 90-226 emm1.0::km (GAS-ΔM) [18] was included as a negative control and was susceptible to phagocytic killing as expected. Furthermore, the GAS-2W strain colonized mouse NALT as efficiently as the wild type (Fig. 1C).
Spleen and lymph node cells from mice were pooled for quantification of 2W:I-Ab-specific CD4+ T cells by pMHCII-based cell enrichment at various times after infection with GAS-2W. Each C57BL/6 mouse had between 200 and 250 2W:I-Ab-specific CD4+ T cells before infection, and these cells expressed small amounts of CD44 (CD44Lo) on the surface as expected for naïve cells (Fig. 2A) [16], [19]. In mice that were inoculated intranasally with GAS-2W, this naïve 2W:I-Ab-specific CD4+ T cell population expanded to ∼105 cells in seven days (Fig. 2A). Most of these cells expressed large amounts of CD44 (CD44Hi), indicative of pMHCII-dependent activation. Expansion of 2W:I-Ab-specific CD4+ T cells was first detected three days after primary infection with GAS-2W and increased rapidly to a peak seven days after inoculation (Fig. 2B). 2W:I-Ab-specific CD4+ T cells then began to decline and reached 10% of the maximum level by day 20-post infection. Expansion of these CD4+ T cells was proportional to the dose of GAS-2W used for infection (Fig. 2C).
In vivo lymphokine production was measured by rechallenging mice with heat-killed GAS-2W (HK-GAS-2W) and then measuring intracellular lymphokines produced by 2W:I-Ab-specific T cells 3 hours later [20]. 2W:I-Ab+ cells in mice infected intranasally with live GAS-2W one week earlier produced IL-17A, but not IFN-γ after challenge with HK-GAS-2W (Fig. 3A). 2W:I-Ab+ cells in mice inoculated intranasally with HK-GAS-2W one week earlier also produced IL-17A after challenge with HK-GAS-2W (Fig. 3B). Therefore, exposure of NALT to either live or dead GAS induced bacterial pMHCII-specific Th17 cells. In order to compare the approximate percentage of 2W:I-Ab-specific Th17 cells to total GAS induced Th17 cells, B6 Mice were inoculated once intranasally with 2x108 CFU of GAS-2W. Ten days after the infection mice were restimulated in vivo by IV injection of heat killed GAS-2W. 2W:I-Ab+ CD4+ T cells from spleen were stained and enriched as described in the methods. Both bound (2W:I-Ab+ CD4+ T cells) and unbound (2W:I-Ab- CD4+ T cells or flow through) fractions were collected and analyzed for intracellular cytokines IL-17A and IFN-γ. Total numbers of CD4+2W:I-Ab+IL-17A+ and CD4+2W:I-Ab-IL-17A+ cells were calculated for the entire spleen. The approximate ratio of 2W:I-Ab+IL-17A+ cells to total 2W:I-Ab- IL-17A+ cells ranged from 1:7 to 1:12 (Fig S1).
Li et al reported that the Streptococcal superantigen, which contaminates commercial peptidoglycan preparations induced human lymphocytes to produce IL-17A [21]; therefore, experiments were performed to determine whether 2W:I-Ab-specific CD4+ T cell clonal expansion was pMHCII-specific. B6 mice were inoculated three times at weekly intervals with either 2×108 CFU of GAS-2W or wild type 2W- GAS (GAS-WT). Three days after the last infection, mice were euthanized and enriched CD4+2W:I-Ab+ T cells from spleen, cervical lymph nodes (CLN), and NALT were analyzed (Fig. 4.). Expansion of 2W:I-Ab-specific CD4+ T cells only occurred in mice inoculated with GAS-2W (Fig. 4A), and not in mice inoculated with GAS-WT that lacked the 2W epitope. The number of CD4+2W:I-Ab+ T cells in tissue from mice infected with GAS-WT was less than 200, similar to that of naïve mice. This indicates that the clonal expansion of 2W:I-Ab-specific CD4+ T cells is a pMHCII-specific response and not due to superantigens produced by this strain of GAS (Fig. 4A). The same samples were restimulated in vitro with pharmacologic TCR mimics, PMA and ionomycin, and stained for intracellular cytokines in order to evaluate their cytokine phenotype. Most of the 2W:I-Ab-specific CD4+ T cells from NALT, CLN and spleen produced IL-17A (Fig. 4B); greater than 75% of the 2W:I-Ab-specific CD4+ T cells from all three tissues had a Th17 phenotype (Fig. 4B). The response was more robust in NALT, which is a preferred target of intranasal GAS infection. In NALT pMHCII-specific IL-17A producers reached more than 90% of the total 2W:I-Ab-specific CD4+ T cells.
Even though initial priming and expansion of CD4+2W:I-Ab+ T cells is independent of superantigen as shown above, it was possible that cytokine activation in T cells from infected mice was mediated by superantigens associated with heat-killed bacteria. To address specificity of the recall response, B6 mice that were inoculated twice with GAS-2W streptococci and then were restimulated in vivo by intravenous injection of either HK-GAS-2W or heat killed wild-type 2W-GAS (HK-GAS-WT) bacteria. CD4+CD44Hi2W:I-Ab+ cells from mice that were primed with GAS-2W had expanded significantly, produced IL-17A 21 days later following restimulation with HK-GAS-2W, but not with 2W- HK-GAS-WT bacteria (Fig. 4C). Thus both priming of CD4+ T cells and recall of cytokine expression by GAS is antigen-specific and independent of superantigens in this animal model.
Children often experience multiple episodes of GAS pharyngitis before reaching the age of 15. Therefore, we investigated whether repeated infection amplified, shifted or dampened the Th17 response. Following multiple infections, most of the CD4+CD44Hi2W:I-Ab+ T cells had elevated levels of intracellular IL-17A (Fig.4B). The response was more robust in NALT, the major infection site following intranasal GAS infection.
Recently, Pepper et al showed that the intranasal route was important for the generation IL-17-producing 2W:I-Ab-specific T cells in response to a recombinant strain of Listeria monocytogenes that expressed the 2W peptide (LM-2W) [20]. Most GAS infections naturally involve oro-pharyngeal tissue and preferably colonize tonsils in humans and NALT in mice. Experiments were therefore performed to assess the impact of the intranasal route of inoculation of GAS on the CD4+ T cell response. B6 mice were inoculated with HK-GAS-2W intranasally, intravenously or subcutaneously. HK-GAS-2W bacteria were used to avoid spread of infection from the site of inoculation. 2W:I-Ab-specific T cells expanded in the spleens of these mice in response to all routes of inoculation. Intranasal infection with GAS-2W induced IL-17A-producing 2W:I-Ab-specific T cells; whereas intravenous and subcutaneous inoculations induced IFN-γ-producing cells (Fig. 5). The Th17/Th1 ratio was more than 40 times higher in cells from intranasally inoculated mice compared to that from mice inoculated intravenously or subcutaneously. Thus, the intranasal route of infection was critical for the generation of Th17 cells during GAS infection.
To compare the relative magnitude of Th17 induction by GAS and LM intranasal infections, mice were inoculated either intranasally or intravenously with either GAS-2W or LM-2W. Intranasally both GAS-2W and LM-2W induced primarily a Th17 response. Notably, however, a larger fraction of the 2W:I-Ab-specific T cell population produced IL-17A following intranasal GAS-2W infection than following intranasal LM-2W infection (Fig. 5B) [20] ; Furthermore, both GAS and LM primarily induced a Th1 response to intravenous inoculation but LM induced a significantly greater Th1 response than GAS (Fig. 5B). Therefore, the PAMPs expressed by or nature of GAS infections creates an environment more conducive to priming the Th17 phenotype than those of LM. Moreover, these data suggest that the 2W epitope does not influence the Th phenotype per se, since the same naïve 2W:I-Ab-specific CD4+ T cell population has expanded and differentiated into different phenotypes in response to alternative routes of infection.
Although IL-6 is known to be critical for some Th17 responses in mice, it is not known whether this is the case for streptococcal infection. We therefore compared the primary T cell response to GAS infection of B6 IL-6 knockout mice (IL-6 -/-) mice to age matched wild type B6 mice, inoculated intranasally with 2x108 cfu of GAS-2W. There was no significant difference in the colonization of both groups of mice 24 hrs after inoculation with this sublethal dose of bacteria. However, over the next 7 days nearly a third of the IL-6 -/- mice succumbed to infection (*P = 0.0473, Logrank Test) (Fig. 6). In contrast, wild-type B6 mice had no casualties and most had reduced their bacterial load in NALT. IL-6-/- mice that survived infection continued to harbor high bacterial counts for up to 60 days; whereas, all wild-type mice completely cleared the bacteria from NALT by day 7.
IL-6-/- mice that survived the first week of infection returned to health even though they retained significant numbers of GAS-2W organisms in NALT. T cells from both wild-type and IL-6-/- mice were restimulated in vivo after one week by intravenous injection of HK-GAS-2W. The 2W:I-Ab+ cells from IL-6-/- mice (Fig. 6A) had proliferated in response to infection with GAS-2W streptococci but failed to produce IL-17A. Instead, many of these cells produced IFN-γ (Fig. 6A). These results confirm that IL-6 is required to promote a Th17 response after GAS infection. Long-term persistence of GAS-2W in NALT from IL-6-/- mice is consistent with the possibility that clearance from NALT and perhaps human tonsils requires a vigorous Th17 adaptive response and that IFN-γ+ T cells lack potential to eliminate streptococci from lymphoid tissue.
The Th17 phenotype is less stable than the Th1 phenotype [20], [22], [23]. For example, IL-17A+ T cells were observed to acquire the capacity to produce IFN-γ in vitro by varying the cytokine environment [24]. The regulatory program and precursor to these double positive T cells is unclear, as is the impact of infection on this phenotype. These considerations prompted experiments to test the influence of multiple intranasal infections on the Th17 population in NALT. Wild-type B6 mice usually clear GAS from NALT 5-7 days after an intranasal inoculation [25]. For recurrent infection, B6 mice were inoculated intranasally with 2×108 GAS-2W CFU at weekly intervals for 12 weeks. One week after the last inoculation cells were harvested from NALT, CLN and spleen, activated in vitro with PMA and ionomycin, and analyzed for intracellular cytokines IL-17A and IFN-γ. After repeated exposure to GAS-2W, IL-17A+ IFN-γ+ 2W:I-Ab-specific CD4+ cells preferentially accumulated in the NALT, whereas the primary phenotype of CD4+CD44Hi2W:I-Ab+ cells in spleen and CLN from the same mice retained a Th17 phenotype (Fig. 7A and 7B). More than 60% of CD4+CD44Hi2W:I-Ab+ cells were positive for both cytokines; whereas, fewer than 12% from spleens and cervical lymph nodes were double positive. The implications of this change will be discussed below.
We recently discovered that GAS induces a polyclonal, TGF-β1 dependent Th17 response in intranasally infected mice. Moreover, adoptive transfer of CD4+ IL-17A+ T cells imparted partial protection to naïve mice against intranasal infection. That investigation raised several questions, which could be more definitively answered using pMHCII tetramers to quantify antigen specific T cells that recognize a streptococcal expressed antigen. In lieu of identifying immunodominant bacterial specific T cell epitopes, the surrogate 14 amino acid peptide 2W was fused to M1 protein and expressed on the streptococcal surface. The recombinant serotype M1 strain, GAS-2W, was constructed and M protein function confirmed by phagocytosis assays using human blood. Moreover, the GAS-2W strain colonized mouse NALT as efficiently as wild type 90-226 streptococci.
For unknown reasons, humans often fail to develop protective immunity after a single episode of pharyngitis or tonsillitis, and persistence of GAS in tonsils even after antibiotic intervention is common [7]. This problem could be explained by failure to develop a sufficient Th17 immune response. However, we found that a single intranasal infection with GAS-2W rapidly induced a robust 2W:I-Ab-specific Th17 population, which paralleled the clearance of bacteria in mice. The 2W:I-Ab-specific Th17 population represented ∼10% of the total GAS specific Th17 induced (Fig. S1). Superantigens are powerful mitogens that induce massive TCR V beta chain-specific clonal expansion of T cells. Group A streptococci, including the GAS-2W strain are known to produce many superantigens, which have been implicated in highly lethal streptococcal toxic shock and autoimmune sequelae [1], [26]. Such superantigens do not account for the expansion of 2W:I-Ab-specific CD4+ T cells following GAS-2W infection, despite the fact that some of these T cells express the relevant TCR V beta chains [19]. This conclusion may not reflect the situation during human infections because murine T cells are known to be relatively refractory to bacterial superantigens. Dominance of the Th17 response to GAS is impressive relative to that observed for other bacterial pathogens like Listeria. After a single infection 20-39% of the 2W-specific T cells produced IL-17A. By comparison, only about 5–12% of 2W:I-Ab-specific CD4+ cells produced IL-17A in response to primary LM-2W intranasal infection (Fig. 5, [20]). Furthermore, both GAS and LM primarily induced a Th1 response to intravenous inoculation and as expected LM induced a more robust Th1 response than GAS (Fig. 5B). Therefore, GAS might have other features that make for more efficient Th17 induction than LM. Furthermore, these results suggest that the 2W epitope per se does not influence the out come of the CD4+ T cell response. The more robust Th17 response could reflect the GAS tropism for NALT [8] and human tonsils [4], the potential for these bacteria to induce TGF-β1 production in these sites [10], expression of a unique pathogen-associated molecular patterns (PAMPs), or a combination of these factors. GAS was shown to induce TGF-β1 and IL-6 in NALT [10], but the specific PAMPs that lead to expression of these cytokines are unknown. The substantial hyaluronic acid capsule or its degradation products are known to induce TLR2- and TLR4-mediated inflammation [27], [28], and TLR2 agonists are known to promote Th17 differentiation [29]. M1 protein is also a TLR2 agonist [30] and known to induce inflammatory cytokines, including IL-6 [31] and TGF-β1 [32], and is, therefore, another potential PAMP that could direct a Th17 response. The finding that intranasal inoculation, even with heat-killed GAS-2W cells induces an antigen-specific Th17 response, while other routes induce a Th1 response also has important implications for development and delivery of vaccines to protect against GAS and other mucosal pathogens. The pathogenic potential of IL-17A+ T cells and their short half-life questions the potential safety and utility of intranasal vaccines.
We previously observed that single cell suspensions of NALT, spleen, and cervical lymph nodes from intranasally infected mice secrete TGF-β1, IL-6 and IL-17 upon ex vivo restimulation [10]. Although IL-17 secretion was dependent on TGF-β1 receptor signals, dependence on IL-6 was not tested. As predicted [33], IL-6-/- mice failed to mount a Th17 response and instead developed IFN-γ+ T cells following intranasal infection. As reported by Diao et al, we found high mortality in IL-6-/- mice infected with GAS [34], perhaps as a consequence of exuberant TNF-α expression. Nearly a third of IL-6-/- mice, inoculated intranasally, developed lethal systemic infections; however, survivors fully recovered without subsequently developing systemic infections even though they retained large numbers of GAS in NALT (Fig. 6). This suggests that protection is compartmentalized, ie. opsonic antibody is required to clear bacteria from blood, but does not efficiently remove them from NALT or tonsils. On the other hand a robust Th17 cellular response would be required to eliminate streptococci from those lymphoid organs. Secretory antibody that baths mucosal surfaces may also protect against pharyngitis, but may not be able to reach bacteria sequestered within NALT. These findings are also consistent with studies of Th17 protection induced by intranasal infections of mice with encapsulated S. pneumoniae [11]. It is theoretically possible that local activation of antigen specific T cells could recruit phagocytes with potential to ingest other unrelated bacteria at that site, but recruitment of phagocytes to those infectious foci would still be antigen specific. In addition, IL-6-/- mice developed a robust antigen specific Th1 response (Fig. 6) but failed to clear the bacteria from NALT. This finding is consistent with the observation that IFN-γ-/- mice [35] cleared streptococci from NALT more rapidly than normal mice. We postulate that efficient elimination of GAS depends on IL-17A production by Streptococcal specific CD4+ T cells to attract neutrophils to infected NALT, a capacity that IFN-γ-producing T cells lack. Persistence of large numbers of GAS in NALT of IL-6-/- mice is reminiscent of the situation in human immune carriers who may also fail to develop an adequate Th17 response.
The Th17 response is known to contribute to immune protection against infections by several pathogens [10], [11], [36], [37], [38]; however, these T cells can also be pathogenic, as demonstrated in murine autoimmune models, such as experimental encephalomyelitis (EAE) [39], [40]. A potentially important discovery from our experiments is that repeated intranasal exposure to GAS-2W results in the accumulation of an IL-17A+ IFN-γ+ (double positive) 2W:I-Ab-specific CD4+ T cells in NALT. In contrast, 2W:I-Ab-specific CD4+ T cells in CLN and spleens from the same mice predominantly retained the Th17 phenotype. Preferential accumulation of double positive T cells in NALT could be explained by a unique homing potential, by more rapid expansion or a longer half-life of double positive cells relative to those only able to produce IL-17A. Whether the large population of antigen-specific IL-17+ IFN-γ+ double positive cells observed in NALT following recurrent GAS infection is associated with an autoimmune response is unknown: however, repeated intranasal infection of mice was reported to occasionally produce endocarditis and valvular lesions[41], suggesting a link to human autoimmune disease.
This study was carried out in strict accordance with the recommendations in the Guide for the Care and Use of Laboratory Animals of the National Institutes of Health. All animal experiments were conducted under University of Minnesota Institutional Animal Care and Use Committee (IACUC) approved protocol number 0806A36362.
Streptococci were grown in Todd-Hewitt broth supplemented with 2% neopeptone (THB-Neo) or on solid media containing Difco blood agar base and sheep blood at 37°C in 5% CO2. All growth media were purchased from Difco Laboratories, Detroit, MI. Strain 90–226 (serotype M1) was originally isolated from the blood of a septic patient [18].
Five to six weeks old female C57BL/6 mice were purchased from Taconic farms (Germantown, NY) and used at 7–8 weeks of age. IL-6-/- mice in C57BL/6 background were purchased from the Jackson Laboratory (Bar Harbor, ME) [42]. Mice were housed under specific pathogen-free conditions at RAR facilities of University of Minnesota. Mice inoculated with GAS were housed in biosafety level 2 facilities.
An M1 GAS strain 90–226 was genetically engineered to express the 14 amino acid 2W peptide (EAWGALANWAVDSA) on the bacterial surface as a fusion protein with streptococcal M protein (Fig. 1). The corresponding chimeric protein is composed of the 2W epitope inserted in-frame after the first five amino acids of the mature M1 surface protein. The hybrid emm1.0::2W gene was constructed using standard molecular biology techniques and then introduced into the chromosomal emm1.0 gene locus by allelic replacement. Briefly, the emm1.0 gene of strain 90–226 is contiguous with mga at the 5′ end and with the sic gene at the 3′ end, and has its own promoter. Plasmid pFW5 has two multiple cloning sites on either side of a spectinomycin resistance gene [8], [18]. The entire sic gene with its promoter was PCR amplified and inserted into the multiple cloning site downstream of the spectinomycin resistance gene in pFW5. A fragment containing the C-terminal half of mga through the entire emm1.0 gene was PCR amplified by a two-step method to insert the sequence of the 2W epitope into the emm1.0 gene. In the first step, DNA fragments, each coding for part of the 2W epitope sequence (overlapping) were amplified by PCR. In the second step these overlapping fragments were annealed, extended and further amplified to generate a 2.5-kb fragment, which was inserted into the multiple cloning site upstream of the spectinomycin resistance gene in pFW5. This plasmid was transformed into strain 90–226 emm1.0::km for gene replacement, and transformants were selected on spectinomycin (100 µg/ml). The spectinomycin-resistant transformants were then screened for kanamycin sensitivity. Positive clones (SpecRKanS) were screened for gene replacement and the entire region including the 2W insertion was amplified and sequence verified. The resulting strain was designated Streptococcus pyogenes 90-226 emm1.0::2W (GAS-2W). A whole blood phagocytosis assay was used to test whether GAS-2W streptococci are M+ [43].
C57BL/6 mice were anesthetized with isoflurane/oxygen mixture for 1 min and inoculated intranasally with GAS-2W by placing appropriate doses in a total volume of 15 µl PBS (7.5 µl/nostril). Fractionated droplets were placed with a pipette tip near the entrance of the nostril and the inoculum was involuntarily aspirated into the nasal cavity.
NALT, Cervical lymph nodes (CLN) and Spleen from C57BL/6 mice were harvested by dissection, disrupted on a nylon screen in 1–2 ml of complete EHAA medium (cEHAA). Resulting single cell suspensions were filtered over a nylon screen and washed with cEHAA before invitro stimulation and staining. Freshly made single cell suspensions in cEHAA were stimulated in vitro with PMA and ionomycin for 4 hours. PMA and ionomycin were added at 50 ng/ml and 500 ng/ml final concentrations, respectively. Cells were incubated at 37C for 1 hour and Brefeldin A (BrA) was added to disrupt golgi and prevent cytokines from being secreted into the medium (1∶1000 of 10 mg/ml DMSO stock) and incubated for three additional hours before proceeding to tetramer enrichment.
2W:I-Ab tetramers were generated as described in detail by Moon et al [16], [19]. Briefly, 2W:I-Ab was expressed in S2 cells. S2 cells were co-transfected with plasmids encoding the I-Ab alpha chain, the I-Ab beta chain and BirA gene. Transfected cells were selected, passaged in serum-free media, and cultures scaled up to one liter cultures. Expression was induced with copper sulfate and biotinylated 2W:I-Ab heterodimers were purified from supernatants with a Ni++ affinity column. Bound 2W:I-Ab molecules were eluted and purified. Tetramers were created by mixing biotinylated 2W:I-Ab molecules with Phycoerythrin (PE) or Allophycocyanin (APC)-conjugated streptavidin.
2W:I-Ab-specific CD4+ T cells were enriched as described previously [16], [19], [20]. Briefly mice were euthanized and NALT, cervical lymph nodes and Spleen were harvested separately. Single cell suspensions were prepared in 200 µl of Fc block (supernatant from 2.4G2 hybridoma cells grown in serum-free media + 2% mouse serum, 2% rat serum, 0.1% sodium azide) from each tissue. Phycoerythrin (PE) or Allophycocyanin (APC) conjugated tetramer was added at a final concentration of 10 nM and the cells were incubated at room temperature for 1 hr, followed by a wash in 15 ml of ice-cold FACS buffer (PBS + 2% fetal bovine serum, 0.1% sodium azide). The tetramer-stained cells were then resuspended in a volume of 200 µl of FACS buffer and mixed with 50 µl of anti-PE or anti-APC antibody conjugated magnetic microbeads (Miltenyi Biotech, Auburn, CA) and incubated on ice in dark for 30 min, followed by one wash with 10 ml of FACS buffer. The cells were resuspended in 3 ml of FACS buffer and passed over a magnetized LS column (Miltenyi Biotech). The column was washed with 3 ml of FACS buffer three times and then removed from the magnetic field. The bound cells were eluted by pushing 5 ml of FACS buffer through the column with a plunger. The resulting enriched fractions were resuspended in 0.1 ml of FACS buffer, and a small volume was removed for cell counting while the rest of the sample was stained with a cocktail of fluorochrome-labeled antibodies.
GAS-2W was grown in THB-Neo media to O.D.600 of 0.5, washed once with PBS, pelleted and resuspended in appropriate volume of PBS and incubated at 60C for 30 minutes. Viability of bacteria was confirmed by plating out on blood agar plates. Heat-killed GAS-2W was stored in aliquots at -20C until use. To induce cytokine production by 2W:I-Ab-specific CD4+ T cells in vivo, 2 X 108 CFU heat-killed GAS-2W in 200 µl of PBS was inoculated intravenously through tail vein. Mice were sacrificed after 3–4 hrs and single cell suspensions of spleen were made in cEHAA medium supplemented with BrA.
Intracellular cytokine staining for IL-17A and IFN-γ was done using BD Cytofix/Cytoperm fixation-permeabilization solution and anti-IL-17A-PE and anti-IFN-γ-PE-Cy7 antibodies according to manufacturer's protocols. Briefly cells stimulated in vivo with heat-killed GAS-2W or in vitro with PMA-ionomycin were subjected to 2W-MHCII tetramer enrichment. Enriched cells were stained for surface markers, washed and fixed with 250 µl of Fixation/Permeabilization solution (BD Biosciences) per sample for 20 min on ice. All samples were washed with 1 ml of 1X Perm/Wash buffer (BD Biosciences) and anti IL-17A PE and anti IFN-γ-PE-Cy7 antibodies were added in 100 µl of 1X Perm/Wash buffer and incubated overnight at 4C in dark. The following day cells were washed with 1 ml of 1X Perm/Wash buffer and resuspended in 400 µl of FACS buffer and transferred to a micro tube within a FACS tube for analysis in LSR-II flow cytometer.
A Becton Dickinson LSR-II flow cytometer was used to collect and analyze events that have the light scatter properties of lymphocytes and are CD4+ and 2W:I-Ab+. Pacific Blue-conjugated anti-B220, CD11b, F4/80 (Caltag), CD11c (eBioscience, San Diego, CA); Pacific Orange-conjugated CD8 (eBioscience); PerCP-conjugated CD4 (BD PharMingen); PE-Cy7-conjugated IFN-γ (eBioscience); PE conjugated IL-17A; APC-AF780 conjugated CD4 and AlexaFluor 700-conjugated CD44 (eBioscience); PE-conjugated CD4; FITC-conjugated CD4; PB-conjugated CD4; APC-conjugated CD4; AF700-conjugated CD4; PerCP-Cy5.5-conjugated CD4; APC-AF750-conjugated CD4. Antibodies were purchased from the indicated sources. FACS acquisition was performed using FACSDiva software and analysis was done using FlowJo (Tree Star, Ashland, OR).
All values are expressed as mean ± SEM. Differences between groups were analyzed using Student's t-test by GraphPad Prism (Version 4.03 for Windows, GraphPad Software, San Diego, CA). Differences are considered significant at P<0.05.
|
10.1371/journal.pgen.1007331 | Molecular basis of hemoglobin adaptation in the high-flying bar-headed goose | During the adaptive evolution of a particular trait, some selectively fixed mutations may be directly causative and others may be purely compensatory. The relative contribution of these two classes of mutation to adaptive phenotypic evolution depends on the form and prevalence of mutational pleiotropy. To investigate the nature of adaptive substitutions and their pleiotropic effects, we used a protein engineering approach to characterize the molecular basis of hemoglobin (Hb) adaptation in the high-flying bar-headed goose (Anser indicus), a hypoxia-tolerant species renowned for its trans-Himalayan migratory flights. To test the effects of observed substitutions on evolutionarily relevant genetic backgrounds, we synthesized all possible genotypic intermediates in the line of descent connecting the wildtype bar-headed goose genotype with the most recent common ancestor of bar-headed goose and its lowland relatives. Site-directed mutagenesis experiments revealed one major-effect mutation that significantly increased Hb-O2 affinity on all possible genetic backgrounds. Two other mutations exhibited smaller average effect sizes and less additivity across backgrounds. One of the latter mutations produced a concomitant increase in the autoxidation rate, a deleterious side-effect that was fully compensated by a second-site mutation at a spatially proximal residue. The experiments revealed three key insights: (i) subtle, localized structural changes can produce large functional effects; (ii) relative effect sizes of function-altering mutations may depend on the sequential order in which they occur; and (iii) compensation of deleterious pleiotropic effects may play an important role in the adaptive evolution of protein function.
| During adaptive phenotypic evolution, some of the associated genetic changes may contribute directly to changes in the selected trait (causative mutations) and other changes may ameliorate the negative side-effects of the causative changes (compensatory mutations). To assess the nature of such changes and their relative prevalence, we used a protein engineering approach to characterize the molecular basis of a well-documented biochemical adaptation: the increased hemoglobin-oxygen affinity in the bar-headed goose (Anser indicus), a champion of high-altitude flight. The experiments revealed the contributions of specific substitutions to the adaptive increase in hemoglobin-oxygen affinity in bar-headed goose and demonstrated that compensatory interactions may play an important role in adaptive protein evolution due to trade-offs between different functional properties.
| During the adaptive evolution of a given trait, some of the selectively fixed mutations will be directly causative (contributing to the adaptive improvement of the trait itself) and some may be purely compensatory (alleviating problems that were created by initial attempts at solution). Little is known about the relative contributions of these two types of substitution in adaptive phenotypic evolution and much depends on the prevalence and magnitude of antagonistic pleiotropy [1–9]. If mutations that produce an adaptive improvement in one trait have adverse effects on other traits, then the fixation of such mutations will select for compensatory mutations to mitigate the deleterious side effects, and evolution will proceed as a ‘two steps forward, one step back’ process. In systems where it is possible to identify the complete set of potentially causative mutations that are associated with an adaptive change in phenotype, key insights could be obtained by using reverse genetics experiments to measure the direct effects of individual mutations on the selected phenotype in conjunction with assessments of mutational pleiotropy in the same genetic background.
To investigate the nature of adaptive mutations and their pleiotropic effects, we used a protein engineering approach to characterize the molecular basis of hemoglobin (Hb) adaptation in the high-flying bar-headed goose (Anser indicus). This hypoxia-tolerant species is renowned for its trans-Himalayan migratory flights [10–12], and its elevated Hb-O2 affinity is thought to make a key contribution to its capacity for powered flight at extreme elevations of 6000–9000 m [13–20]. At such elevations, an increased Hb-O2 affinity helps safeguard arterial O2 saturation, thereby compensating for the low O2 tension of inspired air. This can help sustain O2 delivery to metabolizing tissues because if environmental hypoxia is sufficiently severe, the benefit of increasing pulmonary O2 loading typically outweighs the cost associated with a lower O2 unloading pressure in the systemic circulation [21–24].
The Hb of birds and other jawed vertebrates is a heterotetramer consisting of two α-chain and two β-chain subunits. The Hb tetramer undergoes an oxygenation-linked transition in quaternary structure, whereby the two semi-rigid α1β1 and α2β2 dimers rotate around one another by 15° during the reversible switch between the deoxy (low-affinity [T]) conformation and the oxy (high-affinity [R]) conformation [25–28]. Oxygenation-linked shifts in the T↔R equilibrium govern the cooperativity of O2-binding and are central to Hb’s role in respiratory gas transport.
The major Hb isoform of the bar-headed goose has an appreciably higher O2-affinity than that of the closely related greylag goose (Anser anser), a strictly lowland species [13,29]. The major Hbs of the two species differ at five amino acid sites: three in the αA-chain subunit and two in the βA-chain subunit [30,31]. Of these five amino acid differences, Perutz [32] predicted that the Pro→Ala replacement at α119 (αP119A) is primarily responsible for the adaptive increase in Hb-O2 affinity in bar-headed goose. This site is located at an intersubunit (α1β1/α2β2) interface where the ancestral Pro α119 forms a van der Waals contact with Met β55 on the opposing subunit of the same αβ dimer. Perutz predicted that the single αP119A mutation would eliminate this intradimer contact, thereby destabilizing the T-state and shifting the conformational equilibrium in favor of the high-affinity R-state. Jessen et al. [33] and Weber et al. [34] tested Perutz’s hypothesis using a protein engineering approach based on site-directed mutagenesis of recombinant human Hb, and their experiments confirmed the predicted mechanism.
As a result of these experiments, bar-headed goose Hb is often held up as an example of a biochemical adaptation that is attributable to a single, large-effect substitution [35,36]. However, several key questions remain unanswered: Do the other substitutions also contribute to the change in Hb-O2 affinity? If not, do they compensate for deleterious pleiotropic effects of the affinity-enhancing αP119A substitution? Given that the substitutions in question involve closely linked sites in the same gene, another possibility is that neutral mutations at the other sites simply hitchhiked to fixation along with the positively selected mutation. Since the other substitutions in bar-headed goose Hb have not been tested, we do not know whether αP119A accounts for all or most of the evolved change in O2 affinity. Moreover, the original studies tested the effect of αP119A by introducing the goose-specific amino acid state into recombinant human Hb [33,34]. One potential problem with this type of ‘horizontal’ comparison–where residues are swapped between orthologous proteins of contemporary species–is that the focal mutation is introduced into a sequence context that is not evolutionarily relevant. If mutations have context-dependent effects, then introducing goose-specific substitutions into human Hb may not recapitulate the phenotypic effects of the mutations on the genetic background in which they actually occurred (i.e., in the ancestor of bar-headed goose). An alternative ‘vertical’ approach is to reconstruct and resurrect ancestral proteins to test the effects of historical mutations on the genetic background in which they actually occurred during evolution [37,38].
Here we revisit the functional evolution of bar-headed goose Hb, a classic text-book example of biochemical adaptation. We reconstructed the αA- and βA-chain Hb sequences of the most recent common ancestor of the bar-headed goose and its closest living relatives, all of which are lowland species in the genus Anser. After identifying the particular substitutions that are specific to bar-headed goose, we used a combinatorial approach to test the functional effects of each mutation in all possible multi-site combinations. To examine possible pleiotropic effects of causative mutations, we also measured several properties that potentially trade-off with Hb-O2 affinity: susceptibility to spontaneous heme oxidation (autoxidation rate), allosteric regulatory capacity (the sensitivity of Hb-O2 affinity to modulation by anionic effectors), and various secondary and tertiary structural properties. Measuring the direct and indirect effects of these mutations enabled us to address two fundamental questions about molecular adaptation: (i) Do each of the mutations contribute to the increased Hb-O2 affinity? If so, what are their relative effects? And (ii) Do function-altering mutations have deleterious pleiotropic effects on other aspects of protein structure or function? If so, are these effects compensated by mutations at other sites?
Using globin sequences from bar-headed goose, greylag goose, and other waterfowl species in the subfamily Anserinae, we reconstructed the α- and β-chain sequences of the bar-headed goose/greylag goose ancestor, which we call ‘AncAnser’ because it represents the most recent common ancestor of all extant species in the genus Anser (Fig 1A). The principle of parsimony clearly indicates that all three of the α-chain substitutions that distinguish the Hbs of bar-headed goose and greylag goose occurred in the bar-headed goose lineage (Gα18S, Aα63V, and αP119A), whereas each of the two β-globin substitutions occurred in the greylag goose lineage (βT4S and βD125E)(Fig 1A and 1B).
It is often implicitly assumed that the difference in Hb-O2 affinity between bar-headed goose and greylag goose is attributable to a derived increase in Hb-O2 affinity in the bar-headed goose lineage [14,35,36,39]. In principle, however, the pattern could be at least partly attributable to a derived reduction in Hb-O2 affinity in the greylag goose lineage, even if αP119A does account for the majority of the change in bar-headed goose. To resolve the polarity of character state change, we synthesized, purified, and functionally tested recombinant Hbs (rHbs) representing the wildtype Hb of bar-headed goose, the wildtype Hb of greylag goose, and the reconstructed Hb of their common ancestor, AncAnser. Functional differences between bar-headed goose and AncAnser rHbs reflect the net effect of three substitutions (αG18S, αA63V, and αP119A) and differences between greylag goose and AncAnser reflect the net effect of two substitutions (βT4S and βD125E; Fig 1B).
Since genetically based differences in Hb-O2 affinity may be attributable to differences in intrinsic O2-affinity and/or changes in sensitivity to allosteric effectors in the red blood cell, we measured O2-equilibria of purified rHbs under four standardized treatments: (i) in the absence of allosteric effectors (stripped), (ii) in the presence of Cl- ions (added as KCl), (iii) in the presence of inositol hexaphosphate (IHP, a chemical analog of the endogenously produced inositol pentaphosphate), and (iv) in the simultaneous presence of KCl and IHP. This latter treatment is most relevant to in vivo conditions in avian red blood cells. In each treatment, we measured P50, the partial pressure of O2 (PO2) at which Hb is 50% saturated. To complement equilibrium measurements on the set of three rHbs and to gain further insight into functional mechanisms, we also performed stopped-flow kinetic experiments to estimate apparent O2 dissociation rates under the same conditions.
The O2-equilibrium measurements confirmed the results of previous studies [13,29] by demonstrating that the wildtype rHb of bar-headed goose has a higher intrinsic O2-affinity than that of greylag goose (as revealed by the lower P50 for stripped Hb)(Fig 2A, Table 1). This difference persisted in the presence of Cl- ions (P50(KCl)), in the presence of IHP (P50(IHP)), and in the simultaneous presence of both anions (P50(KCl+IHP))(Fig 2A, Table 1). All rHbs exhibited cooperative O2-binding, as indicated by Hill coefficients (n50’s) >2 in the presence of IHP. The difference in Hb-O2 affinity between bar-headed goose and greylag goose is mainly attributable to differences in intrinsic affinity, as there were no appreciable differences in sensitivities to allosteric effectors (Table 1). This is consistent with a previous report that native Hbs of bar-headed goose and greylag goose have similarly high binding constants for inositol pentaphosphate [29]. Pairwise comparisons between each of the two modern-day species and their reconstructed ancestor (AncAnser) revealed that the elevated Hb-O2 affinity of the bar-headed goose is a derived character state. O2-equilibrium properties of greylag goose and AncAnser rHbs were generally very similar (Fig 2A). The triangulated comparison involving rHbs from the two contemporary species (bar-headed goose and greylag goose) and their reconstructed ancestor (AncAnser) revealed that the observed difference in Hb-O2 affinity (P50(KCl+IHP)) between bar-headed goose and greylag goose is mainly attributable to a derived increase in Hb-O2 affinity in the bar-headed goose lineage, but it is also partly attributable to a derived reduction in Hb-O2 affinity in the greylag goose lineage (Fig 2A). This demonstrates the value of ancestral protein resurrection for inferring the direction and magnitude of historical evolutionary changes in character state.
Kinetic measurements demonstrated that the increased O2-affinity of bar-headed goose rHb is associated with a lower apparent rate of O2 dissociation, koff (Fig 2B) relative to the rHbs of both greylag goose and AncAnser.
In combination with the inferred history of sequence changes (Fig 1A and 1B), the comparison between the rHbs of bar-headed goose and AncAnser indicates that the derived increase in Hb-O2 affinity in bar-headed goose must be attributable to the independent or joint effects of the three substitutions at sites α18, α63, and α119. To measure the effects of each individual mutation in all possible multi-site combinations, we used site-directed mutagenesis to synthesize each of the six possible mutational intermediates that connect the ancestral and descendant genotypes (Fig 1B). In similar fashion, we synthesized each of the two possible mutational intermediates that connect AncAnser and the wildtype genotype of greylag goose (Fig 1B).
The analysis of the bar-headed goose mutations on the AncAnser background revealed that mutations at each of the three α-chain sites (αG18S, αA63V, and αP119A) produced significant increases in intrinsic Hb-O2 affinity (indicated by reductions in P50(stripped))(Fig 3, Table 1). The Pα119A mutation had the largest effect on the ancestral background, producing an 18% reduction in P50(stripped) (increase in intrinsic Hb-O2 affinity). On the same background, αG18S or αA63V produced 7% and 14% reductions in P50(stripped), respectively. In the set of six (= 3!) possible mutational pathways connecting the low-affinity AncAnser genotype (GAP) and the high-affinity bar-headed goose genotype (SVA), the αP119A mutation produced a significant increase in Hb-O2 affinity on each of four possible backgrounds (corresponding to the first step in the pathway, two alternative second steps, and the third step; Fig 3). When tested on identical backgrounds, αP119A invariably produced a larger increase in intrinsic Hb-O2 affinity than either αG18S or αA63V. Nonetheless, of the six possible forward pathways connecting GAP and SVA, αP119A had the largest effect in four pathways and αA63V had the largest effect in the remaining two. The two pathways in which αA63V had the largest effect were those in which it occurred as the first step. In fact, αG18S or αA63V only produced significant increases in Hb-O2 affinity when they occurred as the first step. The effects of these two mutations were always smaller in magnitude when they occurred on backgrounds in which the derived Ala α119 was present. In addition to differences in average effect size, αP119A also exhibited a higher degree of additivity across backgrounds than the other two mutations. For example, the affinity-enhancing effect of αP119A on the AncAnser background is mirrored by a similarly pronounced reduction in O2-affinity when the mutation is reverted on the wildtype bar-headed goose background (αA119P). By contrast, forward and reverse mutations at α18 and α63 do not show the same symmetry of effect (S1 Fig).
Comparison of crystal structures for human and bar-headed goose Hbs [40] revealed that each of the three bar-headed goose α-chain substitutions have structurally localized effects. In the major bar-headed goose Hb isoform, Ser α18 and Ala α119 are located at the edges of the α1β1 intradimer interface. As noted by Jessen et al. [33], the αP119A mutation has very little effect on the main-chain formation and appears to exert its functional effect via the elimination of side chain contacts and increased backbone flexibility. With regard to the αA63V mutation, the introduction of the valine side chain causes minor steric clashes with Gly 25 and Gly 59 of the same subunit (Fig 4). This interaction may alter O2-affinity by impinging on the neighboring His α58, the ‘distal histidine’ that stabilizes the α-heme Fe-O2 bond [41–46].
Given that the AncAnser and greylag goose rHbs exhibit similar equilibrium and kinetic O2-binding properties (Fig 2), the two greylag goose substitutions (βT4S and βD125E) do not produce an appreciable net change in combination. Interestingly, however, each mutation by itself produces a slightly reduced sensitivity to IHP (Table 1), such that values of P50(IHP) and P50(KCl+IHP) for the single-mutant intermediates were lower than those for AncAnser and the wildtype genotype of greylag goose.
Since amino acid mutations often affect multiple aspects of protein biochemistry [47–50], it is of interest to test whether adaptive mutations that improve one aspect of protein function simultaneously compromise other properties. Amino acid mutations that alter the oxygenation properties of Hb often have pleiotropic effects on allosteric regulatory capacity, structural stability, and susceptibility to heme loss and/or heme oxidation [51–58]. Accordingly, we tested whether mutational changes in intrinsic O2-affinity are associated with potentially deleterious changes in other structural and functional properties.
Analysis of the full set of bar-headed goose and greylag goose rHb mutants revealed modest variability in autoxidation rate (S2A Fig, Table 2). This property is physiologically relevant because oxidation of the ferrous (Fe2+) heme iron to the ferric state (Fe3+) releases superoxide (O2-) or perhydroxy (HO2•) radical, and prevents reversible Fe-O2 binding, rendering Hb inoperative as an O2-transport molecule. Although mutational changes in intrinsic O2 affinity (∆log P50(stripped)) were not significantly correlated with changes in autoxidation rate in the full dataset (r = -0.311), analysis of the bar-headed goose rHb mutants revealed a striking pairwise interaction between mutations at α18 and α63 (residues which are located within 7 Å of one another). The αA63V mutation produced a significant >2-fold increase in the autoxidation rate on backgrounds in which the ancestral Gly is present at α18 (Fig 5, Table 2). The adjacent Val α62 is highly conserved because it plays a critical role in restricting solvent access to the distal heme pocket, thereby preventing water-catalyzed rupture of the Fe-O2 bond to release a superoxide ion [58–61]. An increase in side chain volume at α63 may compromise this gating function, resulting in an increased susceptibility to heme oxidation. The increased autoxidation rate caused by αA63V is fully compensated by αG18S (Fig 5), a highly unusual amino acid replacement because glycine is the only amino acid at this site (the C-terminal end of the A helix) that permits the main chain to adopt the typical Ramachandran angles (S3 Fig). Introduction of the serine side chain at α18 in bar-headed goose Hb forces this residue to undergo a peptide flip relative to human Hb, so the carbonyl oxygen points in the opposite direction. This unusual replacement at α18 may be required to accommodate the bulkier Val side chain at α63, thereby alleviating conformational stress. Site-directed mutagenesis experiments on mutant Hbs and myoglobins have documented a positive, linear correlation between log(P50) and log(kauto) [58–61]. The αG18S and αA63V mutations are therefore unusual because reductions in Hb-O2 affinity are not invariably coupled with increases in autoxidation rate.
Aside from the compensatory interaction between mutations at α18 and α63, we observed no evidence for trade-offs between O2-affinity and any of the other measured functional or structural properties. There were no significant correlations between ∆log P50(stripped) and changes in allosteric regulatory capacity (Table 1), as measured by sensitivity to Cl- (r = -0.534), IHP (r = -0.137), or both anions in combination (r = -0.300). The goose rHbs revealed no appreciable variation in α-helical secondary structure as measured by circular dichroism spectroscopy (S2B Fig, S1 Table) and there were no significant correlations between Δlog P50(stripped) and changes in secondary structure over the physiological range (pH 6.5, r = -0.357; pH 7.5, r = -0.052). Likewise, the rHbs exhibited very little variation in the stability of tertiary structure as measured by UV-visible spectroscopy (S2C Fig, S2 Table) and there were no significant correlations between Δlog P50(stripped) and changes in stability over the physiological range (pH 6.5, r = -0.511; pH 7.5, r = -0.338). In summary, we found no evidence for pleiotropic trade-offs between intrinsic O2-affinity and any measured properties of Hb structure or function other than autoxidation rate.
We now return to the two questions we posed at the outset:
(1) Do each of the bar-headed goose substitutions contribute to the increased Hb-O2 affinity?
It depends on the order in which the substitutions occur. Our experiments demonstrated that the αP119A mutation always produced a significant increase in intrinsic Hb-O2 affinity regardless of the background in which it occurred. As documented previously [33,34], the αP119A mutation also produces a significant affinity-enhancing effect on the far more divergent background of human Hb (which differs from bar-headed goose Hb at 89 of 267 amino acid sites in each αβ half-molecule [33% divergence in protein sequence]). By contrast, αG18S or αA63V only produced significant affinity-enhancing effects when they occurred as the first step in the pathway (on the AncAnser background). If it was advantageous for the ancestor of today’s bar-headed geese to have an increased Hb-O2 affinity, our experiments suggest that any of the three α-chain mutations alone would have conferred a beneficial effect, but only αP119A would have produced the same effect after the other two had already fixed. This illustrates an important point about distributions of mutational effect sizes in adaptive walks: in the presence of epistasis, relative effect sizes may be highly dependent on the sequential order in which the substitutions occur.
(2) Do function-altering mutations have deleterious pleiotropic effects on other aspects of protein structure or function?
On the AncAnser background, the affinity-enhancing mutation, αA63V, produces a pronounced increase in the autoxidation rate. This is consistent with the fact that engineered Hb and myoglobin mutants with altered affinities often exhibit increased autoxidation rates [54,56,58,62]. In the case of bar-headed goose Hb, the increased autoxidation rate caused by αA63V is completely compensated by a polarity-changing mutation at a spatially proximal site, αG18S. This compensatory interaction suggests that the αG18S mutation may have been fixed by selection not because it produced a beneficial main effect on Hb-O2 affinity, but because it mitigated the deleterious pleiotropic effects of the affinity-altering αA63V mutation. Alternatively, if αG18S preceded αA63V during the evolution of bar-headed goose Hb, then the (conditionally) deleterious side effects of αA63V would not have been manifest.
Our experiments revealed no evidence to suggest that the affinity-altering αP119A mutation perturbed other structural and functional properties of Hb. Data on natural and engineered human Hb mutants have provided important insights into structure-function relationships and the nature of trade-offs between different functional properties [52,54,56–58,63]. An important question concerns the extent to which function-altering spontaneous mutations are generally representative of those that eventually fix and contribute to divergence in protein function between species. There are good reasons to expect that the spectrum of pleiotropic effects among spontaneous mutations or low-frequency variants may be different from the spectrum of effects among evolutionary substitutions (mutations that passed through the filter of purifying selection and eventually increased to a frequency of 1.0) [64]. The affinity-altering mutations that are most likely to fix (whether due to drift or positive selection) may be those that have minimal pleiotropic effects and therefore do not require compensatory mutations at other sites.
We took sequence data for the α A- and βA-globin genes of all waterfowl species from published sources [30,31].
After optimizing nucleotide sequences of AncAnser αA- and βA-globin genes in accordance with E. coli codon preferences, we synthesized the αA-βA-globin cassette (Eurofins MWG Operon). We cloned the globin cassette into a custom pGM vector system [65,66], as described previously [67–74], and we then used site-directed mutagenesis to derive globin sequences of greylag goose, bar-headed goose, and each of the mutational intermediates connecting these wildtype sequences with AncAnser. We conducted the codon mutagenesis using the QuikChange II XL Site-Directed Mutagenesis kit (Agilent Technologies) and we verified all codon changes by DNA sequencing.
We carried out recombinant Hb (rHb) expression in the E. coli JM109 (DE3) strain as described previously [66]. To ensure the complete cleavage of N-terminal methionines from the nascent globin chains, we over-expressed methionine aminopeptidase (MAP) by co-transforming a plasmid (pCO-MAP) along with a kanamycin resistance gene (48). We then co-transformed the pGM and pCO-MAP plasmids and subjected them to dual selection in an LB agar plate containing ampicillin and kanamycin. We carried out the over expression of each rHb mutant in 1.5 L of TB medium.
We grew bacterial cells at 37°C in an orbital shaker at 200 rpm until absorbance values reached 0.6 to 0.8 at 600 nm. We then induced the bacterial cultures with 0.2 mM IPTG and supplemented them with hemin (50 μg/ml) and glucose (20 g/L). The bacterial culture conditions and the protocol for preparing cell lysates were described previously [66]. We resuspended bacterial cells in lysis buffer (50 mM Tris, 1 mM EDTA, 0.5 mM DTT, pH 7.0) with lysozyme (1 mg/g wet cells) and incubated them in an ice bath for 30 min. Following sonication of the cells, we added 0.5–1.0% polyethyleneimine solution, and we then centrifuged the crude lysate at 13,000 rpm for 45 min at 4°C.
We purified the rHbs by means of two-step ion-exchange chromatography. Using high-performance liquid chromatography (Äkta start, GE Healthcare), we passed the samples through a cation exchange-column (SP-Sepharose) followed by passage through an anion-exchange column (Q-Sepharose). We subjected the clarified supernatant to overnight dialysis in Hepes buffer (20 mM Hepes with 0.5mM EDTA, 1 mM DTT, 0.5mM IHP, pH 7.0) at 4°C. We used prepackaged SP-Sepharose columns (HiTrap SPHP, 5 mL, 17–516101; GE Healthcare) equilibrated with Hepes buffer (20 mM Hepes with 0.5mM EDTA, 1 mM DTT, 0.5mM IHP pH 7.0). After passing the samples through the column, we eluted the rHb solutions against a linear gradient of 0–1.0 M NaCl. After desalting the eluted samples, we performed an overnight dialysis against Tris buffer (20 mM Tris, 0.5mM EDTA, 1 mM DTT, pH 8.4) at 4°C. We then passed the dialyzed samples through a pre-equilibrated Q-Sepharose column (HiTrap QHP, 1 mL, 17-5158-01; GE Healthcare) with Tris buffer (20 mM Tris, 0.5mM EDTA, 1 mM DTT, pH 8.4). We eluted the rHb samples with a linear gradient of 0–1.0 M NaCl. We then concentrated the samples and desalted them by means of overnight dialysis against 10 mM Hepes buffer (pH 7.4). We then stored the purified samples at -80° C prior to the measurement of O2-equilibria and O2 dissociation kinetics. We analyzed the purified rHb samples by means of sodium dodecyl sulphate (SDS) polyacrylamide gel electrophoresis and isoelectric focusing. After preparing rHb samples as oxyHb, deoxyHb, and carbonmonoxy derivatives, we measured absorbance at 450–600 nm to confirm the expected absorbance maxima.
Using purified rHb solutions (0.3 mM heme), we measured O2-equilibrium curves at 37°C in 0.1 M Hepes buffer (pH 7.4) in the absence (‘stripped’) and presence of 0.1 M KCl and IHP (at two-fold molar excess over tetrameric Hb), and in the simultaneous presence of KCl and IHP. We measured O2-equilibria of 5 μl thin-film samples in a modified diffusion chamber where absorption at 436 nm was monitored during stepwise changes in the equilibration of N2/O2 gas mixtures generated by precision Wösthoff mixing pumps [75–77]. We estimated values of P50 and n50 (Hill’s cooperativity coefficient) by fitting the Hill equation Y = PO2n/(P50n + PO2n) to the experimental O2 saturation data by means of nonlinear regression (Y = fractional O2 saturation; n, cooperativity coefficient). Standard errors of the mean P50 were based on triplicate measurements of independently purified rHbs, and the nonlinear fitting of each curve was based on 5–8 equilibration steps. Free Cl- concentrations were measured with a model 926S Mark II chloride analyzer (Sherwood Scientific Ltd, Cambridge, UK).
We determined apparent O2 dissociation constants (koff) of purified oxy rHbs at 37°C using an OLIS RSM 1000 UV/Vis rapid-scanning stopped flow spectrophotometer (OLIS, Bogart, CA) equipped with an OLIS data collection software. Briefly, rHb (10 μM heme) in 200 mM Hepes, pH 7.4, was mixed 1:1 with N2-equilibrated 200 mM Hepes, pH 7.4, containing 40 mM freshly dissolved sodium dithionite [78]. We monitored absorbance at 431 nm as a function of time. All traces exhibited the best fit to a monoexponential function (r2 > 0.99).
To estimate autoxidation rates, we treated purified rHb samples with potassium ferricyanide (K3[Fe(CN)6]), and we then reduced rHbs to the ferrous (Fe2+) state by treating the samples with sodium dithionite (Na2S2O4). We removed the dithionite by means of chromatography (Sephadex G-50). For each rate measurement, we used 200 μl of 20 μM oxyHb in 100 mM potassium phosphate buffer, pH 7.0, containing 1 mM EDTA and 3 mM catalase and superoxide dismutase per mole oxyHb. To measure the spontaneous conversion of ferrous (Fe2+) oxyHb to ferric (Fe3+) metHb we recorded the absorbance spectrum at regular intervals over a 90 h period. We collected spectra between 400nm and 700nm using a BioTek Synergy2 multi-mode microplate reader (BioTek Instruments). We estimated autoxidation rates by plotting the A541/A630 ratio (ratio of absorbances at 540nm and 630nm) vs time, using IGOR Pro 6.37 software (Wavemetrics). We used the exponential offset formula in IGOR to calculate the 50% absorbance per half-life (i.e., 0.5AU/half-life). Standard errors of the mean autoxidation rate were based on triplicate measurements of independently purified rHbs.
We assessed the pH-dependent stability of the rHbs by means of UV-visible spectroscopy. We prepared 20 mM filtered buffers spanning the pH range 2.0–11.0. We prepared 20 mM glycine-HCl for pH 2.0–3.5; 20 mM acetate for pH 4.0–5.5; 20 mM phosphate for pH 6.0–8.0; 20 mM glycine-NaOH for pH 8.5–10.0; 20 mM carbonate-NaOH for pH 10.5 and phosphate-NaOH for pH 11.0. We diluted the purified rHb samples in the pH-specific buffers to achieve uniform protein concentrations of 0.15 mg/ml. We incubated the samples for 3–4 h at 25°C prior to spectroscopic measurements, and maintained this same temperature during the course of the experiments. We measured absorbance in the range 260–700 nm using a Cary Varian Bio100 UV-Vis spectrophotometer (Varian) with Quartz cuvettes, and used IGOR Pro 6.37 (WaveMetrics) to process the raw spectra. For the same set of rHbs, we tested for changes in secondary structure of the globin chains by measuring circular dichroism spectra on a JASCO J-815 spectropolarimeter using a quartz cell with a path length of 1 mm. We assessed changes in secondary structure by measuring molar ellipticity in the far UV region between 190 and 260 nm in three consecutive spectral scans per sample.
We modelled structures of goose Hbs and the various mutational intermediates using the program COOT [79], based on the crystal structures of bar-headed goose Hb (PDB models 1hv4 and 1c40)[40,80], greylag goose Hb (PDB 1faw)[81], and human deoxyHb (PDB 2dn2).
|
10.1371/journal.pgen.1005706 | Trm9-Catalyzed tRNA Modifications Regulate Global Protein Expression by Codon-Biased Translation | Post-transcriptional modifications of transfer RNAs (tRNAs) have long been recognized to play crucial roles in regulating the rate and fidelity of translation. However, the extent to which they determine global protein production remains poorly understood. Here we use quantitative proteomics to show a direct link between wobble uridine 5-methoxycarbonylmethyl (mcm5) and 5-methoxy-carbonyl-methyl-2-thio (mcm5s2) modifications catalyzed by tRNA methyltransferase 9 (Trm9) in tRNAArg(UCU) and tRNAGlu(UUC) and selective translation of proteins from genes enriched with their cognate codons. Controlling for bias in protein expression and alternations in mRNA expression, we find that loss of Trm9 selectively impairs expression of proteins from genes enriched with AGA and GAA codons under both normal and stress conditions. Moreover, we show that AGA and GAA codons occur with high frequency in clusters along the transcripts, which may play a role in modulating translation. Consistent with these results, proteins subject to enhanced ribosome pausing in yeast lacking mcm5U and mcm5s2U are more likely to be down-regulated and contain a larger number of AGA/GAA clusters. Together, these results suggest that Trm9-catalyzed tRNA modifications play a significant role in regulating protein expression within the cell.
| Here we present evidence for a more complicated role for transfer RNAs (tRNAs) than as mere adapters that link the genetic code in messenger RNA (mRNA) to the amino acid sequence of a protein during translation. tRNAs have long been known to be modified with dozens of different chemical structures other than the 4 canonical ribonucleosides, though the role of these modifications in controlling translation is poorly understood. By quantifying the expression of thousands of proteins in the yeast S. cerevisiae, we identified a mechanistic link between modified ribonucleosides located at the wobble position of two tRNAs, tRNAArg(UCU) and tRNAGlu(UUC), and the translation of proteins derived from genes enriched with codons read by these tRNAs: AGA and GAA. In cells lacking the enzyme that inserts these modifications, tRNA methyltransferase 9 (Trm9), we found a significant reduction in proteins from genes enriched with AGA and GAA codons and with runs of these codons. Also, mRNAs enriched with runs of AGA and GAA codons are subject to stalled translation on ribosomes in yeast lacking mcm5U and mcm5s2U. Together, these results reveal a distinct role for Trm9-catalyzed tRNA modifications in selectively regulating the expression of proteins enriched with AGA and GAA codons.
| A striking feature of tRNA molecules is the large number of post-transcriptional modifications, representing up to 10% of the ribonucleoside content [1,2]. Ranging from simple base methylation to complex modifications involving multiple enzymatic steps, modified ribonucleosides are phylogenetically widespread and have long been recognized to play crucial roles in tRNA functions [1,3–5]. Modifications in or around the anticodon loop of tRNA affect translation rate and fidelity through stabilization of codon-anticodon pairing, while other modifications remote from the anticodon loop have specific roles in regulating tRNA stability and folding [1,3,4,6–10]. These observations fuel the hypothesis that tRNA modifications play a broader role in regulating global protein expression, with a focus here on wobble uridine modifications catalyzed by tRNA methyltransferase 9 (Trm9) in budding yeast.
Modification of the wobble uridine in tRNAArg(UCU), tRNAGly(UCC), tRNALys(UUU), tRNAGln(UUG) and tRNAGlu(UUC) requires a number of key activities (Fig 1). The Elongator complex (ELP1-ELP6) uses uridine as a substrate and catalyzes the formation of 5-carboxymethyluridine (cm5U). In association with Trm112, Trm9 will use cm5U as a substrate and catalyze the formation of 5-methoxycarbonyl-methyluridine (mcm5U) at the wobble position of tRNAArg(UCU), tRNAGly(UCC), tRNALys(UUU), tRNAGln(UUG) and tRNAGlu(UUC) (Fig 1) [11–14]. The wobble position of tRNALys(UUU), tRNAGln(UUG) and tRNAGlu(UUC) is further thiolated by an enzyme cascade involving Urm1, Uba4, Ctu1, Ncs2 and Ncs6 to yield mcm5s2U (Fig 1) [11,12,14].
We and others have performed studies in budding yeast, which suggest that stress-induced reprogramming of wobble modifications in tRNA leads to enhanced translation of codon-biased mRNAs for critical stress response genes. [15–19]. For example, we showed that deficiencies in Trm9 and another anticodon loop tRNA methyltransferase, Trm4, cause sensitivity to DNA alkylating agents (methylmethane sulfonate, MMS) and reactive oxygen species (H2O2), respectively [16,20–23]. The roles of the tRNA modifications catalyzed by Trm4 and Trm9 in the cell response were found to involve stress-induced increases in wobble m5C in tRNALeu(CAA) and wobble mcm5U in tRNAArg(UCU) and tRNAGln(UUG), respectively. With Trm9, these changes were directly linked, by Western blots, to enhanced translation of several stress response proteins enriched with AGA and GAA codons [20]. These results not only support the idea that dynamic changes in tRNA wobble modifications facilitate translation of critical stress response proteins [19,20], but they also raise the possibility of an alternative or accessory genetic code involving selective use of degenerate codons as an adaptation for translational regulation by tRNA modifications.
Although there are several examples of individual genes supporting this notion [8,10,17,20,24], what is lacking in this model for translational control of stress response is the larger view of the role of tRNA modifications in regulating global protein expression. Regulation of protein expression occurs at a variety of different levels [25–27], for example, by regulating transcription activity, splicing efficiency, and mRNA export and stability [25,28–33]. Moreover, different stages of protein synthesis are also subject to regulation to ensure efficiency and to preserve fidelity [25,26,34]. Among the variety of factors regulating gene expression, very little is known about the role of tRNA modifications as determinants of global protein translation. Recent studies have shown that loss of Ctu1 or ELP3 result in a moderate reduction in the global protein expression [10], while urm1 and elp3 knockout impairs translation of proteins with high usage of AAA, CAA, and GAA codons [9]. Using ribosomal footprinting, two recent genome-based studies measured average ribosome occupancy on each codon type in several yeast U34 modification mutants [24,35]. While there were striking discrepancies between the two similar studies, the consistent results suggested that loss of wobble uridine modifications in ncs2Δ, ncs6Δ, elp3Δ and elp6Δ mutants leads to changes in ribosome occupancy for some codons associated with tRNAArg(UCU), tRNAGly(UCC), tRNALys(UUU), tRNAGln(UUG) and tRNAGlu(UUC). Of particular note, however, is that the effects caused by loss of U34 modifying enzymes on overall ribosome occupancy on each codon type were averaged over the genome and thus did not reflect codon usage patterns in individual genes, which greatly limits their regulatory conclusions. To address these problems and identify gene-specific regulatory rules, we performed an integrated analysis of proteome, transcriptome and gene-specific ribosome footprinting to investigate the role of Trm9-catalyzed tRNA modifications, mcm5U and mcm5s2U, in regulating global protein expression.
With the overall goal of assessing the effects of loss of Trm9 and its products mcm5U and mcm5s2U on global protein translation, we first verified that modified ribonucleosides mcm5U and mcm5s2U were absent in the trm9Δ cells (S1A and S1B Fig) while the abundance of the hypomodified tRNA species were not significantly affected (S1C Fig) under both normal and stress conditions. These results corroborate previous studies [4,11,12,16,20,36] and establish the trm9Δ cells as a well-controlled model system for analyzing the influence of tRNA modification on global protein expression.
We then used a SILAC-based quantitative proteomic analysis to assess global protein expression in unexposed and MMS-exposed wild-type (WT) and trm9Δ yeast [37]. Proteins derived from lys1Δ yeast grown with [13C6,15N2]-L-lysine were used as an internal standard that was added to protein extracts in each sample, with quantitation of proteins relative to this standard accomplished by LC-MS/MS analysis of protein digests [37]. Protein coverage was maximized by extensive peptide fractionation using an off-gel isoelectric focusing system [38]. Using this approach, we achieved high-confidence identification of 2,408 proteins with a false-discovery rate of 0.46% (S1 Table) and good reproducibility for the three biological replicates analyzed for each cell type and treatment condition (S1D and S1E Fig). Interestingly, protein expression patterns showed greater similarity between the same yeast strains under different treatment conditions than between different yeast strains under the same conditions, which indicates that loss of Trm9 has a stronger influence on global protein expression than MMS treatment. Altogether, we identified 231 proteins that were significantly down-regulated and 95 up-regulated proteins in trm9Δ cells compared with WT cells during normal growth (p <0.05, Student’s t-test and fold-change >1.2) (S2 Table). We also identified 195 significantly down-regulated proteins and 137 significantly up-regulated proteins in trm9Δ cells in response to MMS treatment (S2 Table).
We first examined whether changes in protein expression are highly selective given the assumption that, after controlling for protein length, proteins with enhanced usage of codons dependent on these wobble modifications are more likely to be down-regulated in trm9Δ cells. In this case, we would expect to see genes enriched with mcm5U-dependent codons AGA and GGA, as well as mcm5s2U-dependent codons, CAA, GAA and AAA, to be selectively down-regulated. Moreover, it has been shown that the mcm5 side chain facilitates wobble decoding for G-ending codons [4]. Accordingly, proteins enriched with AGG and GGG codons may likewise be affected by loss of mcm5U.
To this end, we analyzed gene-specific codon usage patterns for all 5886 genes in the yeast genome to determine groups of proteins that are significantly enriched with each codon. A Z-score was calculated to indicate whether a certain codon is over- or under-represented in each individual gene compared to the genome average. Hierarchical clustering analysis of Z-scores of all genes revealed clusters of codons with relatively similar patterns of usage across different genes (Fig 2A; codon usage data for individual genes is presented in S3 Table). The heat map in Fig 2A shows clustering of CAA, AGA and GAA codons, which is distinguished from clustering of GGG, AGG and GGA codons, while the AAA codon was separated from the others.
We then asked whether proteins enriched with these wobble modification-dependent codons were selectively down-regulated in trm9Δ cells. To this end, we calculated the number of significantly down-regulated proteins in each group of proteins enriched with one of the 61 codons. Moreover, to control for the size of different groups and the randomness of changes in protein expression, the percentage of down-regulated proteins, as well as the ratio of the number of down-regulated proteins to the number of up-regulated proteins (D/U) in each group were calculated. For example, in our proteomic dataset, 148 proteins overrepresented with AGA codon were identified and quantified, of which 54 (36.5%) were significantly down-regulated, while only 10 were significantly up-regulated in trm9Δ cells (D/U = 5.4). Among the 196 proteins with high GAA usage, 45 (23.0%) were significantly down-regulated and 10 were significantly up-regulated in trm9Δ cells (D/U = 4.5). In contrast, for all 2408 proteins identified, only 231 (9.6%) were down-regulated while 95 were up-regulated (D/U = 2.4). As shown in Fig 2B and 2C, the percentage of down-regulated proteins and D/U ratios were significantly enhanced in AGA- and GAA-enriched groups as compared with the genome average. In addition, the CAA group showed a high D/U ratio but the percentage of down–regulated proteins showed no difference from the genome average. This suggested that proteins from genes enriched with AGA/GAA codons were preferentially down-regulated in trm9Δ cells. In contrast, we observed no evidence that expression of proteins from genes enriched with GGA, GGG, AGG or AAA were selectively inhibited in trm9Δ cells. One possible explanation for lack of effect is that these codons are all non-optimal codons with low overall usage in the genome (see Discussion).
However, several codons independent of the modifications were also associated with a high proportion of down-regulated proteins, which could be explained by co-enrichment of these codons with AGA and GAA. To investigate the influence of codon co-enrichment, we removed proteins enriched with both AGA and another codon from the group of proteins enriched with AGA, and vice versa. As shown in Fig 2D, this analysis revealed that high usage of the AGA codon, to the exclusion of any other codon, remained the single best predictor for protein down-regulation in trm9Δ cells. In contrast, as expected, the percentages of down-regulated proteins were reduced after removing proteins whose reduction could be better explained by co-enrichment of AGA codon. A similar result was observed for the GAA enriched group (Fig 2E).
In response to MMS treatment, changes in global protein expression in trm9Δ cells were likewise skewed as a function of high usage of AGA and GAA codons, but not the other codons dependent on the wobble modifications (S2A–S2D Fig). Taken together, these results support a role for mcm5U and mcm5s2U modifications in regulating proteins enriched with AGA and GAA codons under both normal growth and stress conditions, establishing that Trm9-catalyzed tRNA modifications play a significant role in regulating protein expression.
We then examined whether proteins enriched with AGA or GAA codons were more likely to be down-regulated than expected by chance in trm9Δ cells and whether the results could be better explained by, for example, changes in mRNA level or biased protein expression. To this end, for proteins enriched with AGA codon (n = 148), we performed 100,000 random samplings of 148 proteins from the proteins that are not enriched with AGA codon, and calculated the percentage of down-regulated proteins and D/U ratio for each sampling. This analysis demonstrated that groups of proteins from genes enriched with AGA were more likely to possess a higher proportion of down-regulated proteins (Fig 3A) as well as a greater D/U ratio (Fig 3C) than expected by chance. Similarly, proteins enriched with GAA codons (n = 196) were more likely to be down-regulated than genome average in trm9Δ cells (Fig 3B and 3D), but not for those enriched with the other codons dependent on the wobble modifications (S3A–S3J Fig). Taken together, these results suggest that depletion of mcm5U and mcm5s2U represses expression of proteins enriched with AGA and GAA codons in a highly selective manner.
However, this analysis could be misleading without controlling for changes in mRNA levels, which may be a major contributor to the changes in protein expression. To this end, we combined the proteomic data with our previous microarray data from trm9Δ cells [19]. We found that changes in protein level and changes in mRNA expression were not correlated, with only 4% (13 out of 326) of the significantly changed proteins explained by changes in mRNA level. We then repeated the analysis after removing these proteins from the dataset. As expected, proteins enriched with AGA and GAA codons were still preferentially down-regulated in trm9Δ cells (S4A and S4B Fig).
Another feature in question was a bias caused by protein abundance. As shown in S5A and S5B Fig, proteins from genes with high usage of AGA and GAA codons were skewed toward highly expressed proteins. Accordingly, these proteins may be more dramatically down-regulated because they were present at higher levels in WT cells. To control for this, we repeated the analysis by randomly selecting a group of proteins with the same expression level as proteins enriched with AGA or GAA codon, respectively, in each sampling. As shown in S5C and S5D Fig, after controlling for protein abundance, proteins from genes enriched with AGA or GAA were still more likely to be down-regulated in trm9Δ cells than expected by chance.
We then examined the possibility that a specific protein is down-regulated in trm9Δ cells as a function of increased usage of AGA/GAA codons in all proteins, regardless of whether they are enriched with AGA/GAA codons or not. As shown in Fig 3E and 3F, we found that reduced protein expression in trm9Δ cells was significantly correlated with enhanced usage of AGA (rs = -0.17, p = 2.8E-09) and GAA (rs = -0.17, p = 7.4E-08), respectively. Furthermore, we binned proteins into seven groups based on their usage of AGA and GAA codons and calculated the D/U ratio in each group. As shown in Fig 3G and 3H, increased usage of AGA and GAA codons additively enhanced the chance of down-regulation in trm9Δ cells on a genomic scale. These correlations held when we examined the data for MMS-treated cells (S6 Fig). Together, these results establish that depletion of mcm5U and mcm5s2U selectively repressed expression of proteins with high usage of AGA and GAA codons.
In addition to overall codon usage, certain features such as codon clustering (i.e., close spacing of codons along a gene sequence) may also regulate the rate of translation along a transcript. We thus scanned each mRNA sequence with a sliding window searching for physical clustering of AGA and GAA codons. As shown in Fig 4A and 4B, for genes enriched with AGA and GAA codons, we observed non-random distributions of these codons along the transcripts. We then tested whether such clustering occurred more frequently than expected by chance. To this end, we counted the number of triplet runs (3-mer) of AGA and GAA combinations in each gene. Maintaining codon composition, we shuffled the codons of each gene and counted the number of triplet runs. After performing this shuffling 10,000 times for each gene, we found that the actual number of codon runs observed was significantly higher than randomization (Fig 4C, p = 2.5E-187, Mann-Whitney U test). The results remained robust when quadruple or quintuple codon combinations were used (S7A and S7B Fig).
In the absence of mcm5U and mcm5s2U modifications, these codon runs may generate a local sequence unfavorable for translation by enhancing the chance of ribosomal pausing. In line with this notion, we showed that genes for down-regulated proteins in trm9Δ cells contained a significantly higher number of AGA and GAA runs than the other proteins (Fig 4D, p = 7.4E-6, Mann-Whitney U test). However, this could be potentially explained if genes for down-regulated proteins contained more AGA and GAA codons, and as a result, a higher number of codon runs. We controlled for this scenario by limiting our analysis to proteins from genes enriched with AGA and GAA codons. We found no significant difference in the usage of these codons between proteins with codon runs (n = 144) and those without codon runs (n = 154) (mean frequency: with = 11.2% versus without = 11.0%; p = 0.84, Mann-Whitney U test). Controlling for codon usage, we found that down-regulated proteins still had a significantly higher number of codon runs than the other proteins (Fig 4E, p = 0.017, Mann-Whitney U test). These results support the notion that clustering of certain codons imposes an additional layer of regulation on translation efficiency and provide independent evidence for selective inhibition of proteins from genes enriched with AGA and GAA codons in trm9Δ cells.
Ribosome footprinting analysis provides an opportunity to quantify the rate of translation of specific mRNA sequences in vivo based on the assumption that the slower a ribosome travels along a specific region of a transcript, the more likely that the ribosomal density in that region will be enhanced. Zinshteyn and Gilbert [24] used ribosome footprinting to assess the effect of mcm5U and mcm5s2U on translation rates in elp3Δ yeast cells lacking these modifications and found ribosome accumulations at AAA, CAA, and GAA codons. However, their results were based on genome-average ribosomal occupancy on each codon type, and cannot be used to predict altered expression of individual proteins. We thus integrated this ribosomal footprinting data with our proteomic data to examine whether there is a link between ribosomal pausing and reduced protein expression in cells lacking mcm5U and mcm5s2U.
After controlling for differences in sequencing depth and changes in mRNA expression, we calculated the changes in stringently mapped ribosomal densities that occur within a single transcript between elp3Δ cells and WT cells [24]. As shown in Fig 5A, proteins whose transcripts have enhanced ribosomal density (pausing) are preferentially down-regulated compared to those without increase in ribosomal density (i.e., no pausing) (p = 3.56E-05, K-S test). Specifically, as shown in Fig 5B, 75 out of the 292 (26%) transcripts with pausing were found among the significantly down-regulated proteins in trm9Δ cells, while only 156 out of the 930 (17%) proteins without pausing were significantly down-regulated in our proteomics dataset (p = 6.92E-4, chi-square test).
We then asked whether ribosomal pausing is associated with enhanced usage of AGA and GAA codons. We note that all genes with ribosome footprinting information were analyzed, regardless of whether they were identified in our proteomic study. As expected, groups of transcripts with pausing tended to contain a higher proportion of genes with high usage of AGA and GAA codons (Fig 5C, p = 7.16E-11, K-S test). Specifically, a significantly higher rate of genes enriched with AGA and GAA codons was observed in genes with pausing (226/1251, 18.1%) than in genes without pausing (598/4353, 13.7%) (Fig 5D, p = 1.39E-4, chi-square test).
We further asked whether clustering of AGA and GAA codons could enhance ribosomal pausing. As shown in Fig 5E, the genes prone to pausing were skewed toward those with more runs of AGA and GAA codons (p = 4.87E-13, K-S test) and the frequency of AGA and GAA codon runs in stalled genes is significantly higher than those without pausing (Fig 5F, p = 2.62E-16, Mann-Whitney U test). However, this bias toward codon runs could simply result from an association of ribosomal pausing with transcripts possessing high AGA and GAA codon usage. To control for this bias, we limited our analysis to proteins enriched with AGA and GAA codons in both groups. As shown in Fig 5G, genes prone to pausing are still skewed toward higher numbers of codon runs (p = 6.86E-05, K-S test) and the number of codon runs is significantly higher in transcripts on stalled ribosomes (Fig 5H, p = 3.17E-6, Mann-Whitney U test). Taken together, these data provide independent evidence that loss of mcm5U and mcm5s2U selectively reduces translation of genes enriched with AGA and GAA codons by causing ribosomal pausing.
The fact that loss of Trm9-catalyzed tRNA modifications disrupts expression of proteins from AGA- and GAA-enriched genes led us to explore the Trm9-dependent proteome for a molecular mechanism underlying the associated phenotype of MMS sensitivity. To this end, we analyzed the biological processes associated Trm9-dependent proteins, with comparison of normally growing and MMS-treated trm9Δ cells. Using the David program [36], we find that most of the defects in protein expression occurring in response to MMS exposure are readily observed under normal growth conditions in trm9Δ cells (S8 Fig; S4 and S5 Tables). Notably, down-regulated proteins with a unique codon usage pattern linked to Trm9 are heavily enriched in translation machinery. For example, 18 out of the 20 components of the 60S ribosomal subunit and all 15 components of the 40S ribosomal subunit are significantly down-regulated under normal and/or stress condition, indicating impaired function of this basic translation machinery (Fig 6A). Paralleling the reduction in ribosomes, we also observed a down-regulation of proteins involved in different steps of translation (Fig 6B), including eIF2, eIF4A, eIF4g and DED1 involved in translation initiation, 15 out of 17 proteins involved in translational elongation, and SSB1, YEF3 and RPL10 involved in translation termination. Moreover, six out of seven proteins involved in protein folding were significantly down-regulated (Fig 6C). Notably, 11 of the 12 aminoacyl-tRNA synthetases were likewise significantly down-regulated. These results revealed an unexpected role for Trm9-catalyzed tRNA modifications in regulating translation, which is consistent with our previous observation that loss of Trm9 impaired expression of proteins involved in translation elongation (YEF3) and DNA damage repair (RNR1 and RNR3), and leaded to translational infidelity, protein errors and activation of protein stress response pathways [19,20]. This is held up to explain the observation that some proteins without high usage of AGA and GAA codons were also down-regulated in trm9Δ cells. However, we also suggest that the effect, if any, could not far surpasses that induced by codon usage bias, otherwise we should not observe the selective repression of proteins enriched with AGA and GAA codons. In addition to translation components, we also observed significant down-regulation of proteins involved in DNA damage repair (MPH1, RPL40A and DEF1) and cell cycle control (NBP1, YRB1, CMD1 and MYO1) pathways (Fig 6D). This is consistent with our previous observations that trm9Δ cells display delayed transition into S-phase following DNA damage [19].
The variety and conservation of modified ribonucleosides in tRNA support the idea that they must play an important role in regulating protein expression [2,3,8,17,19], though the evidence remains largely circumstantial without testing their influence on a global scale in vivo. To this end, we integrated proteome, transcriptome and ribosome footprinting data to elucidate the role of mcm5U and mcm5s2U in regulating global protein expression. After controlling for various confounding factors, such as protein abundance, changes in mRNA levels, and potential influence of other codons, we found robust evidence that expression of proteins enriched with AGA and GAA codons, and to a lesser extent with CAA, are preferentially repressed in cells lacking mcm5U and mcm5s2U under both normal and stress conditions. Consistent with this result, we previously examined expression of several TAP-tagged endogenous proteins, and found that loss of Trm9 only affected expression of proteins overrepresented with AGA and GAA codons [20]. Moreover, we re-engineered a Trm9-depedent gene, ribonucleotide reductase 1 (RNR1), to replace all Trm9-dependent codons with Trm9-independent synonymous codons. In striking contrast to the wild-type gene, the mutant RNR1 gene was largely resistant to trm9Δ-induced repression of protein expression [19]. A combined consideration of the proteomic results presented here and previous genetic studies [39] reveal a highly important role of wobble uridine modifications in regulating global gene expression.
Our data clearly revealed that loss of Trm9 and its wobble modifications causes a significant shift to reduced expression of AGA- and GAA-enriched genes. However, it is important to point out that this regulation is not an “all or none” effect—that is, the loss of Trm9 causes a significant shift in translation but not a complete down-regulation of all AGA- and GAA-enriched genes. The observation that not all AGA- and GAA-enriched genes are affected by loss of Trm9 illustrates the fact that gene expression in general and translation in particular are regulated by a complex interplay of different factors that control the efficiency and fidelity of different steps of protein synthesis. It is also important to point out that we compared groups of proteins enriched or not enriched with a single codon as an unbiased test of the hypothesis that the effects of Trm9 loss should be more pronounced for proteins enriched with Trm9-dependent codons. This proved to be the case, but does not imply that genes enriched with AGA and GAA codons are the only ones affected by loss of Trm9, or that all genes enriched with AGA and GAA must be affected by Trm9 loss. However, we did not see any evidence that genes enriched with other Trm9-dependent codons, including GGA, GGG, AGG and AAA, were preferentially down-regulated in cells lacking Trm9. This could be explained by, for example, lower usage of these codons in yeast genes relative to AGA and GAA codons, or that the effect was counterbalanced by poor usage of AGA, GAA and CAA codons, which is supported by the codon usage clustering result in Fig 2A. Nonetheless, the key point is that our results clearly established, as a proof of concept, that Trm9-dependent tRNA modifications play a significant role in regulating protein expression in vivo.
An interesting feature of AGA- and GAA-enriched genes is the observation that the codons are more likely to cluster together than expected by chance. Such clustering has been found to affect local translation rate, which has emerged as a mechanism to fine-tune protein expression and minimize protein folding errors, thus providing an additional layer of translational control. For example, biased combinations of codon runs differ in their propensity to cause mistranslation or ribosome pausing [40–42]. Furthermore, large codon clusters could have a greater effect on protein production than an equivalent number of randomly scattered codons, while clustering of rare codons could play an important role in regulating tissue-specific protein expression [40,43,44]. Here we found that Trm9-dependent proteins from genes enriched with AGA and GAA codons showed a significantly increased frequency of AGA and GAA codon runs. Our results provide evidence that 1) proteins with AGA/GAA codon runs, after controlling for codon usage, are more likely to be down-regulated in trm9Δ cells than those without codon runs, and that 2) ribosomal pausing in yeast cells lacking mcm5U and mcm5s2U is more likely to occur with transcripts possessing a larger number of AGA/GAA codon runs. These results support the idea that codon clusters add another layer of translational control to protein production.
Since our proteomic analyses provide new insights into the functional complexity of wobble uridine modifications in regulating translation, it is important to place our results in the context of published studies that address U34 modifications from other perspectives and reveal a highly complicated system of regulatory control. Here we compare our results for Trm9-dependent modifications with the genetic studies (tRNA over-expression rescue) of Bjork and coworkers in ncs6Δ and elp3Δ mutants [39], as well as with ribosome footprinting studies of Zinshteyn and Gilbert [24] and, more recently, Nedialkova and Leidel [35] in ncs2Δ, ncs6Δ, elp3Δ and elp6Δ mutants. S6 Table summarizes these studies. Most notably, the footprinting data showed that the loss of ncs2Δ, ncs6Δ and elp3Δ caused ribosomal pausing on codons GAA, CAA, AAA and GAG, but not AGA, and the authors speculated that the effect was too small to influence protein production [24]. In addition to discordant conclusions based on the same yeast mutants [24,35] and misinterpretation of oxidative stress affects on U34 modifications [35], we point to several problems with these footprinting studies in terms of the uncertainty of both RNA-seq technology and data analytics, as detailed in a recent review of ribosome profiling technology [45], problems that obviate stringent comparisons of the ribosome footprinting data sets with the proteomic data. In terms of data analytics, the two studies analyzed the RNA-seq footprinting data in terms of genome-average effects of ribosomal pausing on each codon type and did not specify ribosomal pausing on individual genes at single-codon resolution [24,35]. Ribosomal pausing on the same codon can vary dramatically among genes and even along the same transcript, depending upon the structural and physiochemical properties of the local protein sequence. So the approach to data analysis used in the two ribosome footprinting studies precludes drawing conclusions about ribosomal pausing on individual transcripts or changes in expression of individual genes. To address these problems, we reanalyzed the elp3Δ ribosome footprinting data of Zinshteyn and Gilbert [24] in terms of individual genes and found a significant association between enhanced ribosomal pausing and high usage of AGA/GAA codons, as well as the number of AGA/GAA runs along the transcripts. Another feature of the footprinting data, which involves a focus on short mRNA fragments protected by a single ribosome, could explain the apparent absence of AGA codons among the codons associated with ribosome pausing. Close spacing of paused ribosomes has been shown to produce longer protected RNA fragments that are ignored in most ribosomal footprinting methods [45]. Such close spacing could occur by strong pausing at a high density of AGA codons, with a possible contribution from the drag produced when positively charged Arg residues (coded by AGA) interact with the negatively charged ribosomal exit tunnel [46]. This could explain why overall ribosomal occupancy on AGA codons actually decreased in elp6Δ cells in the studies of Nedialkova and Leidel [35]. Lastly, all mutants used in the footprinting studies have modification deficiencies that result in the presence of wobble U or s2U on specific tRNAs, while the trm9 mutant used here leaves wobble cm5U or cm5s2U and the nsc6 mutant used in Bjork et al. [39] leaves a wobble mcm5U. These different wobble modification structures further confound the comparison of the results of the various studies and highlight the complexity of U34 modification effects.
Our analysis of proteins regulated by mcm5U and mcm5s2U modifications provides several insights into the molecular mechanisms underlying the Trm9-dependent alkylation stress response in budding yeast. First, our proteomic analysis revealed 153 of the proteins found to confer sensitivity to MMS in a phenotypic screening study [22], with 30 of these down-regulated and 7 up-regulated by loss of Trm9 (S7 Table). These proteins may thus play a role in the MMS stress response phenotype. An analysis of the functional categories of proteins down-regulated in trm9Δ cells provides insights into the molecular mechanisms. One striking feature is that a majority of down-regulated proteins in trm9Δ cells are ribosome related or involved in different steps of translation (Fig 6). This behavior has strong parallels with Trm4-dependent translation of ribosomal protein paralogs from TTG-enriched genes during the response to oxidative stress [17]. As one of the central control points for gene expression, protein synthesis is regulated at multiple levels for translation efficiency and error reduction [19,26], with ribosomal protein variants allowing control of ribosome structure and function for plasticity in the cell response to environmental changes and stress [47]. So it is not surprising that loss of Trm9 causes in large-scale changes in protein expression as a result of perturbations of the translational machinery, even for proteins not enriched with AGA and GAA codons. The additional changes in expression of elongation factors could alter local translation rates, leading to mis-folding and impaired protein function [41], which would be compounded by the down-regulation of chaperone proteins observed in trm9Δ cells. Consistent with the idea that cells lacking Trm9 suffer from translational inadequacy, loss of Trm9 was found to perturb polysome profiles of Trm9-dependent transcripts [20], to cause a mis-folded protein stress due to multiple translational errors including mis-incorporations and frame shifts [19], and to disrupt translation of AGA, GAA, CAA, GAG codons [19]. The sum of these dysregulations thus induces accumulation of mis-translated and mis-folded proteins that activates protein stress response pathways in trm9Δ cells, which is consistent with the fact that loss of Trm9 recapitulates the stress response associated with exposure to the protein- and nuclei acid-damaging agent, MMS.
Another feature of the trm9Δ phenotype involves up-regulation of energy production. Since both protein synthesis and rescue of mis-folded proteins by chaperones are energy-dependent processes (S8 Fig; S4 and S5 Tables) [41], it is likely that energy demands are elevated in trm9Δ cells to maintain translation activities and to activate degradation pathways for errors in protein translation and folding. This model is supported by our proteomics data and Gene Ontology analysis. In comparison with WT cells, we find that proteins involved in glucose metabolism and the tricarboxylic acid cycle are coordinately up-regulated in trm9Δ cells under both normal and MMS stress conditions (S8 Fig; S4 and S5 Tables), possibly to keep up with the increased requirement for energy consumption due to translational errors caused by loss of Trm9 [19]. Oxidative stress response proteins were also activated in trm9Δ cells (S8 Fig; S4 and S5 Tables), suggesting that loss of Trm9 confers a state of oxidative stress with elevated reactive oxygen or nitrogen species that are harmful to the cell.
The observed proteomic changes all reflect a phenotype characterized by cell stress, which is consistent with the up-regulation of proteins involved in apoptosis and cell death in trm9Δ cells (S8 Fig). Indeed, loss of Trm9 partially recapitulates the MMS-induced stress response in budding yeast [20]. In the absence of Trm9-catalyzed tRNA modifications, cells experience a dysregulated uncoupling of modified tRNAs from codon-biased translation, which leads to a highly regulated but unfavorable steady-state of altered protein synthesis, energy metabolism and cell death. In this regard, we propose a novel model in which regulation of the spectrum of modified ribonucleoside levels at wobble positions in the system of tRNAs fine-tunes global protein expression by codon-biased translation of mRNAs and reprogramming of translational machinery. Trm9 activity thus illustrates a systems-level mechanism for translational control of cell behavior, with mechanistic parallels in other tRNA modification enzymes.
Urea, sodium chloride, Tris, sodium fluoride, β-glycerophosphate, sodium orthovanadate, sodium pyrophosphate, dithiothreitol, iodoacetamide were purchased from Sigma Chemical Co. (St. Louis, MO). Endoproteinase Lys-C was purchased from Wako (Richmond, VA). All chemicals and reagents were of the highest purity available. All strains of S. cerevisiae BY4741 were purchased from American Type Culture Collections (Manassas, VA).
Yeast strain BY4741 was used in this study. WT and trm9Δ yeast cells were grown in yeast nitrogen base (YNB) liquid medium containing 30 mg/l normal L-lysine. MMS at a concentration of 0.0125% was used to treat log-phase yeast cells (OD600 0.7) for 1 h, which caused a lethality of <5%. lys1Δ yeast cells were grown in YNB medium containing U-[13C6,15N2]-lysine (Sigma) at 30 mg/l for at least 10 generations to log-phase. Cells were harvested by centrifugation for 10 min at 1,500 × g at 4°C and washed twice with cold water.
Modified ribonucleosides in cytoplasmic tRNA were identified and quantified as reported previously [16,17]. Briefly, WT and trm9Δ yeast cells were lysed by lyticase treatment in the presence of deaminase inhibitors (5 μg/ml coformycin, 50 μg/ml tetrahydrouridine) and antioxidants (0.1 mM desferrioxamine, 0.1 mM butylated hydroxytoluene). Total RNA were extracted by the Trizol-chloroform method following manufacturer's instructions, and tRNA-containing small RNA species were enriched using the PureLink miRNA Isolation Kit (Invitrogen). The quantity of tRNA was then determined by UV-vis spectrophotometer (absorbance at 260 nm) and Bioanalyzer analysis (Agilent Bioanalyzer Small RNA Kit). Using identical quantities of tRNA from WT and trm9Δ cells, each tRNA sample was mixed with [15N]5-2-deoxyadenosine as an internal standard and the tRNA was enzymatically hydrolyzed with nuclease P1 and RNase A, followed by dephosphorylation by alkaline phosphatase [16,17]. Proteins and other large molecules were removed by ultrafiltration, and the digested ribonucleosides were then resolved by reversed-phase HPLC (Agilent 1100) with a mobile phase of 1~100% acetonitrile in 8 mM ammonium acetate at a flow rate of 300 μl/min for 1 h. Eluted ribonucleosides were analyzed on an Agilent 6410 Triple Quadrupole mass spectrometer. Modified ribonucleosides were identified by HPLC retention time and collision-induced dissociation (CID) fragmentation pattern. The signal intensity for each ribonucleoside was normalized with the signal intensity of [15N]5-dA and abundance of the modified ribonucleosides was compared for WT and trm9Δ cells in the presence and absence of MMS.
DNA probes complementary to 5sRNA (5'-TGGTAGATATGGCCGCAACC-3'), tRNAArg(UCU) (5'-CACGGCTTAGAAGGCCGTTG-3'), tRNAGlu(UUC) (5'-CTCCGCTACGGGGAGTCGAAC-3'), tRNAGln(UUG) (5'- GGTCGTACTGGGAATCGAACCCAGG-3'), tRNALys(UUU) (5'-CTCCCACTGCGAGATTCGAACTCGC-3') and tRNAGly(UCC) (5'-GTGTAGTGGTTATCATCCCACCCTTC-3') were end labeled with T4 polynucleotide kinase (NEB) in the presence of [32P]-ATP according to the manufacturer’s instructions. Total RNA (10μg) extracted by the Trizol-chloroform method was separated on a 12% polyacrylamide/8M urea/1×TBE gel followed by semi-dry electroblotting onto Hybond N+ nylon membranes (GE Healthcare). Membranes were cross-linked and pre-hybridized for 1 h in hybridization buffer (50% formamide, 0.5% SDS, 5×SSPE, 5×Denhardt’s solution, and 20 μg/ml sheared, denatured, salmon sperm DNA). Then the membrane was hybridized in the same solution containing 10 pmol radio labeled probes overnight at 42°C. The membrane was washed 3-times with 4×SSC at ambient temperature for 10 min. The membrane was wrapped in plastic wrap (Saran) and placed in a cassette with Kodak MS film at -80°C overnight.
Light SILAC-labeled WT and trm9Δ yeast cells as well as heavy-labeled lys1Δ yeast cells were disrupted in an alkaline buffer (2 M NaOH, 8% v/v 2-mercaptoethanol), and proteins were isolated by TCA precipitation. Yeast proteins were pelleted by centrifugation for 15 min at 15,000 × g at 4°C, and resuspended in lysis buffer (8 M urea, 75 mM NaCl, 50 mM Tris, pH 8.2, 50 mM NaF, 50 mM β-glycerophosphate, 1 mM sodium orthovanadate, 10 mM sodium pyrophosphate, 1 mM PMSF). Protein concentration was determined using the Bradford assay. This lys1Δ yeast protein extract was used as global internal standard and mixed (1:1, w/w) with WT and trm9Δ proteins separately.
The protein mixture was reduced for 2.5 h at 37°C in 1 mM dithiothreitol (DTT), alkylated for 40 min by 5.5 mM iodoacetamide (IAA) at ambient temperature in the dark, and then digested with 50:1 (w/w) endoproteinase lys-C overnight at 37°C. Peptide mixtures were fractionated into 24 fractions according to their isoelectric point using Agilent 3100 OFFGEL Fractionator (Agilent). Each peptide fraction was acidified by adding 0.1% formic acid, and loaded onto a C18 trap column (200Å Magic C18 AQ 5μm, 150 μm × 10 mm) at flow rate of 5 μl/min, with subsequent elution from a coupled analytic column (200Å Magic C18 AQ 5 μm, 75 μm × 150 mm) at 200 nl/min using a 2–98% acetonitrile gradient (180 min) in 0.1% formic acid. Eluted peptides were analyzed on a QSTAR-XL (Applied Biosystems/MDS Sciex) mass spectrometer. Three technical replicates were performed for each sample.
Acquired MS/MS spectra were parsed by Spectrum Mill (Agilent) and searched against the Saccharomyces Genome Database (SGD). SILAC peptide and protein quantitation was performed with DEQ (Differential Expression Quantitation). SILAC protein ratios are determined as the average of all peptide ratios assigned to the protein, and the proteins quantified in at least two replicates of the sample are recruited for further study. Differential protein expression was determined by Student’s t-test between different samples.
Using protein-coding sequences for 5886 Saccharomyces cerevisiae genes downloaded from the SGD database (http://www.yeastgenome.org/), we applied custom scripts to determine the gene-specific codon frequency in terms of the number of a particular codon per thousand codons in the open reading frame. Whether a gene was over- or under-represented with a specific codon relative to the genome average was determined by calculating a Z-score based on a hypergeometric distribution with a cut-off of p < 0.01. Gene-specific codon usage data were analyzed by hierarchical clustering using Cluster 3.0, and visualized as a heat map using Treeview [48]. Simulation and ORF shuffling were performed in R project using custom scripts. A spreadsheet containing the gene-specific codon usage data is presented in
We investigated gene-specific clusters of AGA and GAA by calculating the frequency of AGA/GAA codons over a sliding window which was then subtracted the mean value of that gene. Similar to previous studies [49,50], a window size of 15 nt was used in this study. The data were plotted as a histogram, with positive peaks indicating clusters of AGA and GAA codons in these regions. The number of short codon runs in the form of triplets of AGA, GAA or their combinations in each gene was counted using custom scripts in R project.
Ribosomal footprinting data of the wildtype and elp3Δ cells as well as the RNA-seq data of the corresponding samples were obtained from a previous study (GSE45366) [24]. The reads per kilobase per million mapped reads (rpkms) of ribosomal footprinting data (FP) and the matched RNA-seq data (T) were used for comparison after normalization for library size. FP/T ratio for each gene was calculated to indicate ribosomal density on each transcript. A cut-off of >1.2 fold increase in FP/T ratio was used to determine whether ribosomal density on a certain gene was enhanced in cells lacking mcm5U and mcm5s2U.
|
10.1371/journal.pntd.0000167 | BCG Revaccination Does Not Protect Against Leprosy in the Brazilian Amazon: A Cluster Randomised Trial | Although BCG has been found to impart protection against leprosy in many populations, the utility of repeat or booster BCG vaccinations is still unclear. When a policy of giving a second BCG dose to school children in Brazil was introduced, a trial was conducted to assess its impact against tuberculosis, and a leprosy component was then undertaken in parallel. Objective: to estimate the protection against leprosy imparted by a second dose of BCG given to schoolchildren.
This is a cluster randomised community trial, with 6 years and 8 months of follow-up. Study site: City of Manaus, Amazon region, a leprosy-endemic area in Brazil. Participants: 99,770 school children with neonatal BCG (aged 7–14 years at baseline), of whom 42,662 were in the intervention arm (revaccination). Intervention: BCG given by intradermal injection. Main outcome: Leprosy (all clinical forms). Results: The incidence rate ratio of leprosy in the intervention over the control arm within the follow-up, in schoolchildren with neonatal BCG, controlled for potential confounders and adjusted for clustering, was 0.99 (95% confidence interval: 0.68 to 1.45).
There was no evidence of protection conferred by the second dose of BCG vaccination in school children against leprosy during the trial follow-up. These results point to a need to consider the effectiveness of the current policy of BCG vaccination of contacts of leprosy cases in Brazilian Amazon region.
| BCG is a vaccine developed and used to protect against tuberculosis, but it can also protect against leprosy. In Brazil, children receive BCG at birth, and since 1996 a trial has been conducted to find out if a second dose of BCG administered to schoolchildren gives additional protection against tuberculosis. We use this trial to find out if such vaccination protects against leprosy. The trial was conducted in the Brazilian Amazon, involving almost 100,000 children aged 7–14 years who had received neonatal BCG. Half of them received a second dose of BCG at school, and the other half did not. We followed the children for 6 years and observed that there were as many new cases of leprosy in the vaccinated children as in the unvaccinated children. Therefore, we concluded that a second dose of BCG given at school age in the Brazilian Amazon offers no additional protection against leprosy.
| BCG vaccination is given routinely to neonates to prevent tuberculosis in Brazil and in most of the world. BCG also protects against leprosy, with estimates of protection ranging from 20% to 90% [1],[2]. In Brazil, in addition to routine BCG vaccination at birth to prevent tuberculosis, BCG is officially recommended for household contacts of leprosy cases. In 1994 the Brazilian Ministry of Health expanded its tuberculosis control policy to recommend the routine BCG vaccination of school age children (around 7–14 years old). Given the high coverage of neonatal vaccination, this was effectively revaccination for most children. A large cluster randomised trial (BCG-REVAC) was started in 1996 to assess the effectiveness against tuberculosis of BCG vaccination of schoolchildren [3],[4]. One of the trial sites was the city of Manaus, which is also endemic for leprosy. In this city the trial objective was then expanded to estimate the effectiveness on leprosy. This paper reports the results of the BCG-REVAC trial in preventing leprosy based on follow-up from January 1999 to August 2006.
Details of the methodology of the BCG-REVAC trial and of the leprosy component have been published elsewhere (regarding trial co-ordination, screening to detect leprosy cases before the trial, and sample size) [4],[5]. A CONSORT checklist is available in Text S1. We summarise here relevant methodological aspects.
The main objective of the leprosy component of the BCG-REVAC trial was to estimate the protection against all forms of leprosy given by one dose of BCG under routine conditions to schoolchildren aged 7–14 years in a population with coverage of neonatal BCG of about 89%. Our original hypothesis was that BCG revaccination would cause 50% reduction in incidence, based on the estimate observed in the trial in Malawi [6]. This is a vaccine effectiveness, pragmatic trial [7], rather than an efficacy trial. The study design attempted to reproduce the routine implementation of the policy of BCG vaccination of schoolchildren according to the 1994 recommendation.
The study site was the city of Manaus, in Amazonas State of Brazil, with about 1,500,000 inhabitants in 2002. The city is divided into 56 administrative districts, which in turn are grouped into 6 geographical areas (North, East, South, West, Centre West and Centre South). The new case detection rate (NCDR) has been around 6.5 cases per 10,000 per year of leprosy since the 1990s. In 1997 the NCDR was 6.6 (814 cases) in the total population and 4.9 (110 cases) in children aged 7–14 years. The trial study population was schoolchildren residing in the city, aged 7–14 years and attending state schools at the time of the trial implementation in 1998 (year of birth between 1984 and 1991). No child was excluded on the basis of previous history of tuberculosis or leprosy, mirroring the official recommendation for BCG vaccination to schoolchildren.
Randomisation. There are several reasons to do randomisation at cluster level in studies on infectious diseases [8]. In this trial, the main reason was operational: a list of schools with approximate numbers of students, but not names, was available. Without a list of student's names, individual randomization would be much harder given the very large number of children involved. The decision also considered the following advantages of cluster randomisation in this case: intervention (vaccination) more likely to be acceptable, as all schoolchildren within the same school would be allocated to receive or not to receive vaccination; simpler execution; large number of randomisation units (schools) expected to result in similar comparable allocation groups. Furthermore, since the recommendation was to vaccinate schoolchildren, schools represent the settings where this intervention would naturally be implemented in a real campaign.
There was no previous study on leprosy in which intra-class correlation (ICC) had been estimated, and ICC estimated by this study before the trial follow-up resulted in a negative value [5], which was interpreted as suggestive of no effect of clustering [9]. Therefore, although this was a cluster randomised trial, the initial sample size estimation made no allowance for clustering, and it was estimated 50,000 children in each allocation arm [5], using formulae in chapter 7 of Friedman et al. [10].
The randomisation followed several steps and was conducted using a list of schools provided by the local education department, with estimates of the number of schoolchildren in each school. Only schools with more than 50 schoolchildren in the target age group and in the main urban area of Manaus were included. First, the 56 districts were classified into strata according to the incidence of leprosy and tuberculosis in each district before the trial (1996). If the leprosy incidence (NCDR) in 1996 in a district was above the rate of the city as a whole, then it was categorised as “above” = 1, otherwise (below the city rate) as “below” = 0. The same procedure was used for tuberculosis, “above” or “below” the city rate. The 56 districts were thus grouped into 5 strata of incidence of leprosy and tuberculosis: four combinations of rate of tuberculosis and leprosy (above/above, above/below, below/above and below/below the rate of the city, respectively), and a fifth category with the districts with unavailable data on leprosy and tuberculosis. Second, the schools in the list were sorted by a) greater geographical areas, b) the 5 strata and then c) on the estimate of the number of schoolchildren at the target age group. Third, within the same geographical areas and 5 strata, the schools with the closest number of schoolchildren were then taken as a pair. Fourth, random numbers were generated by computer for each school, and in each pair the school with the smallest random number was allocated to the control arm, and the other was allocated to the intervention arm. When a school had no pair (odd number of schools in the stratum by geographical area and category on leprosy/tuberculosis), it was allocated at random (odd number was control, even number was intervention). The randomisation process was implemented by two researchers, S.S.C and S.P.
Three hundred and forty five (345) schools from the original list were used in the randomisation, selected based on number of students and being located in urban area. The number of schoolchildren was approximately estimated to be 161,736. However, subsequent field work showed that some of the information in the list was inaccurate, and schools in the list were excluded for several reasons: number of schoolchildren smaller than 50; school closed during the time of trial implementation; school mostly included children with special needs. Also, a single school was listed as two schools because it was based over two sites. Finally, from the original 345 schools, only 286 were eventually entered in the trial.
After randomisation, visits to these 286 schools were conducted between July and October 1998 to collect children's data, including BCG scar reading and BCG vaccination. Data were transcribed from school records, and children were examined to identify those who had received neonatal BCG vaccination, based on BCG scar.
Children in the schools allocated to vaccination received 0.1 ml of lyophilised BCG produced in Brazil, Moreau strain, administered by intradermal injection [4]. Four different batches were used but vaccination in any one school was done with a single batch. Vaccination began in September and finished in December 1998. Children in the control arm did not receive a placebo.
Data were collected from 156,331 schoolchildren (nearly 70% of the estimated population of Manaus aged 7–14 years in 1998), and 3,893 children were later excluded because they were outside this age range (see Figure 1). The analysis plan originally proposed to estimate the vaccine effect in children with either no or one BCG scar as observed at baseline, consisting of 110,218 children: 51,207 in the intervention arm (allocated to vaccination) of whom 46,997 were vaccinated (92%). Reasons for not vaccinating the remaining 4,210 children included refusals, withdrawals, and changes in school [5].
Because most leprosy in Manaus is tuberculoid, which has a shorter incubation period [11], and the protection of BCG against leprosy was observed very soon (after 1 year) in some trials [12],[13], the decision was made to start the follow-up from 1st January 1999 (and to end in August 2006, see analysis). First January 1999 corresponded to more than 2 months after vaccination for nearly 50% of the vaccinated children (range 22–100 days).
Allocation concealment refers to the process of making the investigators not able to know the randomisation sequence between the time it is generated and the time a particular code is allocated to study unit [14]. In our trial, the allocation was done immediately after the sequence generation, so the allocation was in effect concealed.
Most cases were diagnosed at the leprosy reference centre in Manaus, where the local leprosy control programme and its surveillance system are located. In the study site, would be very unusual for a suspected case of leprosy to be identified at school and referred to the health services for diagnosis and treatment. Leprosy cases normally go spontaneously to routine health services, or are referred from primary services to the leprosy reference centre, as several services have dermatologic clinics. Therefore, detection rates are not expected to differ systematically between control and intervention schools.
In order to give blind diagnosis at the reference centre, most patients (85 out of 91) up to 2001 had their right deltoid area covered by an adhesive tape on entering the medical offices for investigation of leprosy the diagnosis, however, after 2001 this procedure was found to be little used and was then definitely discontinued. Physicians were also continually asked to refrain from inquiring about the BCG status until a definitive diagnosis of leprosy was made, unless the physician has judged it was necessary to know the BCG status for good clinical practice. Therefore, the trial can not be considered as being blind, and furthermore the absence of placebo meant that neither those administering the interventions, nor the participants were blind to their assignment.
Cases were thus classified into multibacillary (MB) and paucibacillary (PB) as reported in medical records and histopathology exams, and diagnosis made following the routine procedures used in the reference service, based mostly on combination of clinical signs, baciloscopy and histopathology and on WHO criteria [15]. Bacillary index and biopsies are routine procedures for all suspect cases, unless there are contraindications such as in young children, facial lesions, and refusal by the patients. Data on anaesthesia, aspects of skin lesions and thickening nerve are routinely collected during the clinical examination. Such data on the diagnosis and classification were thus periodically retrieved as described in the medical records soon after the diagnosis, but any doubt on classification and diagnosis was discussed with the professionals responsible by the patient or exam. Cases are routinely classified according to grade of disability (0 = no anaesthesia or deformity in hands or feet, and no eye problem; grade 1 = with anaesthesia but no deformity, or with eye problem but vision not severely affected; grade 2 = visible deformity or vision severely affected [16]), and this was information was also retrieved. The difficulty of diagnosing leprosy is well known, and one way to overcome such difficulties is to categorise the leprosy cases in “certainty levels” based on typical signs [17]. There was an initial attempt during the first/second year of follow-up to register relevant signs and symptoms into a standardised questionnaire to be completed by the doctors responsible by the assistance of each suspect case, as well as histopathology exams should follow specific procedures and findings annotated into a standardised form. However, the completion rate was low and the information from these forms was not used. There was no independent review panel for deciding on the final diagnosis of leprosy cases. Given that it is an effectiveness (pragmatic) trial, which means it was aimed at assessing vaccine effect under routine conditions, it was decided that all leprosy cases reported in the surveillance system should be included in the analysis, but that the analysis should include a comparison of cases classified into 2 levels of certainty based on laboratory and clinical presentation: those with confirmed histopathology, or positive baciloscopy or thickening nerve, versus those without any of these data. Therefore, we performed sub-group analysis for these two levels.
Leprosy cases detected by the local surveillance system were linked to the trial population by matching information from the notification to records in the trial data-base (date of birth, name of case and name of the case's mother) [5]. There were 650 leprosy cases detected by the local surveillance from 1999 to August 2006 in the target age population and residing in Manaus, of whom 253 cases were identified in the 156,331 total trial population, and 117 cases (out of 253) were among those 92,770 children with one BCG scar (see Figure 1). In this group, 91.6% of children in intervention arm (n = 42,662) actually received BCG in the trial, and only 2 children in the control arm (n = 50,108) were wrongly vaccinated (originally enrolled in school in the control arm but actually attending school in intervention arm).
Surveillance for adverse events. The routine passive surveillance of adverse events was enhanced. A letter containing information on BCG adverse events was distributed to all children on the day of vaccination to motivate parents to take their children to a health facility if they had a health problem following vaccination. Teachers in the trial schools and health workers in the reference medical centres for tuberculosis and leprosy were made aware of the trial and alerted to possible BCG adverse events. Suspect adverse events were diagnosed in the health facilities and treatment provided. This vaccine safety surveillance continued for 4 months after the end of vaccination.
The BCG-REVAC trial received ethical approval by the Brazilian National Ethical Committee (CONEP, Comissão Nacional de Ética em Pesquisa) [5]. The trial is registered with an International Standard Randomised Controlled Trial Number, ISRCTN07601391 (http://www.controlled-trials.com/ISRCTN07601391).
The main outcome of the leprosy component of this trial consisted of leprosy cases diagnosed in the health facilities in Manaus, expressed as the NCDR of leprosy per 10,000 person years in the two arms during the trial follow-up. Baseline characteristics of the population are presented at individual and cluster levels separately for the intervention and control arms, and for those excluded and those included in the vaccine effect estimation. No significance test was used to assess differences on baseline characteristics [18]. All calculations were based on rates and rate ratios estimated by Poisson regression. The 95% confidence intervals were based on robust (“sandwich”) variance estimator, specifying school (not pair of schools) as cluster to allow for clustering within schools [19]. Rates and rate ratios were also adjusted for covariates to correct for any imbalance between intervention groups, and covariate adjustments are stated below when describing each analysis.
BCG vaccine protection was estimated as (1-RR)×100, RR being the ratio of the rate in the intervention arm over the rate in the control arm, among those 92,770 individuals with one BCG scar, with intention-to-treat analysis. Statistical analysis was done in STATA version 7.0.
Interim analysis of the leprosy results was not planned. Originally, the analysis was planned for all leprosy cases involving those with no or one BCG scar, which was planned to have a power of 80% to detect a vaccine protection of 50%. This analysis was conducted in 2003, but not published [20]. However, it was subsequently recognised that the most important estimate would be for those with neonatal BCG, rather than for all children regardless of previous BCG status, because Brazil currently achieves a high neonatal BCG coverage rate, which means that, in the near future, most individuals will have received neonatal BCG. In addition, there was not enough power to assess heterogeneity of vaccine effect according to one or zero BCG scar. The decision was therefore made to redo the analysis, among only those children with one BCG scar at entry, at a time when a study power of 80% was projected to have been achieved for this sub-population. Based on the hypothesised rate ratio of 0.5, and 91% coverage in the vaccine arm, this was achieved with the 117 cases detected up to August 2006 [21]. Hence the decision was made to analyse the cases accumulated up to this time. This paper therefore reports the estimate for vaccine protection among children with one BCG scar, that is, effect of re-vaccination. At the moment of this analysis, the number of cases among those with no BCG scar was not sufficient for a study power of 80% and thus vaccine protection was not estimated in this group.
The number of leprosy cases detected in the trial and number of children are shown in Figure 1. The baseline characteristics of the two allocation arms were similar regarding gender, age at entry into the trial (Table 1). A higher proportion of children in intervention arm were in schools located in areas that had higher incidence of tuberculosis and leprosy (NCDR) before the trial (bold numbers in Table 1).
Among those excluded from the analysis for having no BCG scar reading, or with no scar or >1 scar, the leprosy rates (per 10,000) were 3.07 (95% C.I.: 2.43 to 3.89; 69 cases in 224,605 person years) in the control arm, and 2.80 (95% C.I.: 2.20 to 3.57; 65 cases in 232,016 person years) in the intervention arm. All results below are restricted to those with one BCG scar.
Description of cases. Among the 117 cases in the population with 1 prior BCG scar, 30 cases were MB and 87 PB. Most cases (112) had grade 0 (no disability); in the intervention arm there were 3 cases with grade 1 and 2 cases with grade 2. Nineteen cases had nerve thickening (16.2%): 8 (13.1%) in control arm, and 11 (19.6%) in intervention. Fifty seven cases (48.7%) were confirmed by histopathology: 29 (47.5%) in control arm, and 28 (50.0%) in intervention. The mean age at diagnosis was 15.6 years (sd 3.1) in control and 14.2 (sd 3.0) in intervention. There were 47.5% of male cases in the control arm (29/61) and 51.8% (29/56) in the intervention arm.
The rates (per 10,000 person-years) of MB cases were 0.36 in the control arm (14 cases/383,754 person years) and 0.49 in the intervention arm (16/326,673 person years). For PB the rate was 1.22 in both arms. The rates by calendar year separately and allocation arm are shown in Table 2. There was a borderline statistically significant increase in the rate in the intervention arm in the first year of follow-up (1999), the rate ratio being 2.50 (95% C.I.: 0.99 to 6.35) (Table 2). This increase in the first year was observed for PB and MB cases: the rate ratio was 2.63 (95% C.I.: 0.31 to 22.00) for MB cases and 2.91 (95% C.I.: 0.87 to 7.20) for PB cases, adjusted as in Table 2.
The rate ratio between allocation arms is shown in Table 3, separately for MB and PB, controlled for study variables and adjusted for effect of clustering. There was no evidence of protection by the second dose of BCG during the follow-up period. For the whole study period, the robust standard error of the intervention-over-control rate ratios (controlled for the variables as in Table 3) arms were 0.18987 adjusted for clustering and 0.18581 not adjusted for clustering. We estimated design effect as (0.18987/0.18581)2 = 1.0442. Given that the design effect is 1+(n–1)×ICC, where ICC is the intra-class correlation coefficient and n is the mean cluster size ( = 327), the ICC was estimated as 0.00013568.
The rate ratio between those vaccinated and not vaccinated, regardless of allocation arm (using on-treatment analysis) for the whole trial follow-up was, for all leprosy cases 0.97 (95% C.I.: 0.67 to 1.41), for MB cases 1.08 (0.54 to 2.15) and PB cases 0.92 (0.59 to 1.43), after controlling for study variables as in Table 3. After excluding the first year of follow-up, the rate ratio for all leprosy cases was 0.81 (0.54 to 1.21), for MB cases was 1.09 (0.50 to 2.37), and for PB cases was 0.71 (0.43 to 1.15).
The rate ratio intervention over control arms, based on cases with confirmed histopathology, positive baciloscopy or thickening nerve (n = 60) was 1.51 (95% C.I.: 1.04 to 2.19), after control for sex, BCG scar, year of birth, and the previous rates of leprosy and tuberculosis in the districts where schools were located. For those with neither confirmed histopathology, positive baciloscopy nor thickening nerve, but based on clinical judgement on typical skin lesion and presence of anaesthesia (n = 57), it was 1.03 (95% C.I.: 0.76 to 1.39).
Among the total 152,438 individuals enrolled in the study, 47,307 individuals were vaccinated, and 18 cases were reported with adverse events related to BCG (risk of 3.80 per 10,000). Eight cases was due to ulcer greater than 1 centimetre, 7 cases had cold abscess, and the 3 remaining cases had axillary lymph node enlargement without suppuration, “hot” abscess with suppuration and nodule in vaccination site. Children without previous BCG scar had a risk of 3.67 (3 cases/8,176 individuals), and those with 1 BCG scar (revaccination) the risk was of 3.84 (15 cases/39,067), this corresponded to a risk ratio of 1.05 (95% C.I.: 0.31 to 3.53), after adjustment for effect of clustering, sex and year of birth.
This study found no evidence of protection of the second dose of BCG against all forms of leprosy among school children within 6 years and 8 months of follow-up. This remained after controlling for potential confounders and adjusting for effect of clustering. The confidence interval of 0.72 to 1.58 (for the all cases and the whole period, controlled for covariates and adjusted for clustering) is consistent with a vaccine protection of up to 28% and an increase in leprosy in those vaccinated up to 58%. We conclude that the results did not support the original hypothesis of a vaccine protection of 50%, and additional follow-up is thus not planned. There was no statistically significant vaccine protection against MB and PB cases, but this analysis was not powered to evaluate protection separately for clinical forms at this duration of follow up. We can speculate some alternative hypotheses why the expected protection by revaccination was not observed.
First, differential detection rate. The diagnosis of most leprosy cases was blind to vaccination status in the first two years, but after 2001 the procedures recommended for blind diagnosis were not used. However, diagnosis was based on routine procedures as before the trial, and it is unlikely that special attention was paid to patients' revaccination status: BCG history tends to not be considered in routine procedures in Manaus. It is also unlikely that patients themselves sought medical attention differentially according to whether they were vaccinated in the trial. So a subjective assessment would suggest that distortion of the estimates due to differential detection rate because of lack of blindness seems unlikely.
Second, the result could be due to misdiagnosis. The difficulty of diagnosing leprosy is well known [17]. If patients with other skin lesions were wrongly diagnosed as having leprosy and were included in the study, this would lead to an underestimation of the vaccine effect. However, the quality of diagnosis in the study was high: most leprosy cases had the typical signs or laboratory findings of leprosy. This is strengthened by the fact that this study was based on a reference centre for leprosy diagnosis and treatment, with very experienced clinicians; and protection did not increase with certainty of diagnosis. Therefore, it is unlikely that false positive diagnoses would be responsible for lack of vaccine effect.
Third, linkage of cases. The linkage was done blind to vaccination status, and so it is unlikely that any failure in linkage of cases would be differential. Fourth, this could be due to imbalances in the baseline characteristics in the comparison groups or selection bias when defining the study population. Indeed, the incidence of leprosy (NCDR) in the geographical areas where vaccinated and control schools were located were unbalanced, despite the randomisation process. This imbalance must have been caused from the inaccuracy of the list of schools used in the randomisation process and the exclusion after allocation. This unbalance could potentially distort the vaccine effect, which is why we controlled for sex, age, year of birth, and previous incidence of tuberculosis and leprosy in the analysis.
Fifth, poor vaccine administration and vaccine storage. The vaccination was done following routine procedures, however, temperature was regularly checked and no problems were detected. Furthermore, vaccine strain, the staff and procedures used in the trial were similar to used in neonatal vaccination, and neonatal BCG vaccination was shown to be protective against leprosy in the cohort study nested on the trial [22]. Therefore, vaccine administration and vaccine storage were at worst similar to those which have resulted in protection of 90% in the nested cohort study for neonatal vaccination.
Sixth, the lack of vaccine effect could not be due to a distortion caused by vaccination by the trial of individuals in the control arm, because very few individuals in the control arm among those with one BCG scar were vaccinated (n = 2 of 50,108). BCG vaccination in schoolchildren was suspended in the study site because of the trial during the study follow-up, and definitely in the whole country in 2006. Nevertheless, 312 schoolchildren in Manaus were reported as wrongly vaccinated in routine in 2002 although none belonged to the trial population. No other child in the target age group was reported vaccinated by public health services [23], and the number of children who received BCG in private services is negligible, given that it is offered free of charge by public services.
Eighth, the presence of HIV infectionin Manaus. There are no data on infection, but the annual average number of reported AIDS cases aged 15–19 years between 2001 and 2004 was 13 cases, with a estimated population in this age group of 182,745 in 2004, therefore HIV infection is expected to be very low and not have any effect on the vaccine effect observed [23].
The increase in the leprosy rate during the first year of follow-up in the intervention arm deserves comment. There is some prior evidence of BCG vaccination increasing the risk of leprosy in the initial follow up period, probably due to change in disease progression (“negative effect”) [24]–[26]. However, revaccination was also not protective when the first year of follow-up was excluded, and the increase in the first year of follow up was not responsible for the absence of effect observed for the whole period.
How does this result compare to previous studies? Among trials, in the study in Papua New Guinea the participants received different number of doses during the trial. However, the reports did not present vaccine protection or data separately for the number of doses received, although it was stated that the number of doses did not affect the vaccine protection [13]. In the last trial in India (1991), the study participants included those with and without previous vaccination, but separate data for previous vaccination were not shown, although it was reported that previous vaccination did not affect the protection conferred by the vaccine given by the trial [27]. In contrast, the trial in Malawi estimated the vaccine protection given by a second dose, and showed a statistically significant vaccine protection of 50% [6], but there were several different characteristics, including: different BCG strain (from Glaxo); screening to remove leprosy cases before vaccination; randomisation by individual rather than cluster; a mix of passive and active case detection; and broader age range from infants to adults. Among case-control studies, two assessed vaccine effect by number of doses and both showed additional protection with more than one dose [28],[29], but only in one study was a statistically significant trend of higher protection with increased number of doses [28]. A recent meta-analysis of BCG vaccination against leprosy concluded that additional doses offer additional protection [30]. This meta-analysis also found strong evidence of heterogeneity between studies. In our oppinion, although there is indeed some evidence for additional protection, the results are variable and do not support an unequivocal conclusion for additional protection by more than one dose in all sites.
Had the follow-up period been longer, would it be possible to observe vaccine protection in the years to come? There has been a recent report of BCG protection against tuberculosis lasting for at least 20 years in Brazil [31]. Indeed, in the last trial in India (1991) no vaccine protection was observed in the first years, but a statistically significant result was observed afterwards [24]. Therefore it is theoretically possible that continued follow up will demonstrate protection in coming years, but it is uncertain that this would be of public health importance.
BCG revaccination is currently recommended to contacts of leprosy patients in Brazil [32]. The results of this trial are not directly applicable to the setting of contacts, as protection of revaccination might be different given the close exposure in contacts, but we suggest that the effectiveness of revaccination in contacts must be evaluated to inform a review of such recommendation. |
10.1371/journal.ppat.0030105 | A Mycobacterium ESX-1–Secreted Virulence Factor with Unique Requirements for Export | Specialized secretion systems of pathogenic bacteria commonly transport multiple effectors that act in concert to control and exploit the host cell as a replication-permissive niche. Both the Mycobacterium marinum and the Mycobacterium tuberculosis genomes contain an extended region of difference 1 (extRD1) locus that encodes one such pathway, the early secretory antigenic target 6 (ESAT-6) system 1 (ESX-1) secretion apparatus. ESX-1 is required for virulence and for secretion of the proteins ESAT-6, culture filtrate protein 10 (CFP-10), and EspA. Here, we show that both Rv3881c and its M. marinum homolog, Mh3881c, are secreted proteins, and disruption of RD1 in either organism blocks secretion. We have renamed the Rv3881c/Mh3881c gene espB for ESX-1 substrate protein B. Secretion of M. marinum EspB (EspBM) requires both the Mh3879c and Mh3871 genes within RD1, while CFP-10 secretion is not affected by disruption of Mh3879c. In contrast, disruption of Mh3866 or Mh3867 within the extRD1 locus prevents CFP-10 secretion without effect on EspBM. Mutants that fail to secrete only EspBM or only CFP-10 are less attenuated in macrophages than mutants failing to secrete both substrates. EspBM physically interacts with Mh3879c; the M. tuberculosis homolog, EspBT, physically interacts with Rv3879c; and mutants of EspBM that fail to bind Mh3879c fail to be secreted. We also found interaction between Rv3879c and Rv3871, a component of the ESX-1 machine, suggesting a mechanism for the secretion of EspB. The results establish EspB as a substrate of ESX-1 that is required for virulence and growth in macrophages and suggests that the contribution of ESX-1 to virulence may arise from the secretion of multiple independent substrates.
| A major mechanism used by pathogenic bacteria for disabling host defenses is secretion of virulence proteins. These effectors are often transported by specialized secretion machines. One such pathway, present in Mycobacterium and other Gram-positive genera, is ESX-1 (early secretory antigenic target 6 system 1). Although ESX-1 is required for multiple phenotypes related to the pathogenesis of infection, only three substrates of the secretion machine have been identified to date, and the mechanism by which these substrates are exported is not understood. In our efforts to understand this virulence-related secretion mechanism, we identified a novel substrate and found that its delivery to the ESX-1 machine requires different protein interactions than previously identified substrates. Finally, we present data that the various ESX-1 substrates contribute additively to virulence. These data are incorporated into a model of ESX-1 function.
| The cell surface–associated and secreted proteins of pathogenic bacteria promote the uptake of nutrients; facilitate attachment to specific surfaces, cells, or proteins; function in cell wall maintenance and cell division; and offer protection from harsh environmental conditions, including the host immune system. In Mycobacteria, there are at least four pathways to secrete proteins—Sec, SecA2, twin-arginine translocase, and the early secretory antigenic target 6 (ESAT-6) system 1 (ESX-1). Much attention has been focused on the ESX-1 pathway because it is required for virulence and for the secretion of ESAT-6 and culture filtrate protein 10 (CFP-10), two major targets of the immune response in infected individuals.
M. tuberculosis ESX-1 is required for virulence in mice, growth in macrophages, and the suppression of macrophage inflammatory and immune responses, including the arrest of phagosome maturation and the reduced expression of IL-12 and TNF-α [1–6]. The homologous M. marinum ESX-1 is required for virulence in zebrafish, growth in macrophages, cytolysis and cytoxicity, and cell-to-cell spread, in addition to ESAT-6 and CFP-10 secretion [7,8]. In zebrafish embryo infections, M. marinum ESX-1 is required for macrophage aggregation and granuloma formation [9]. In M. smegmatis, ESX-1, in addition to being required for secretion of ESAT-6 and CFP-10, modulates conjugal DNA transfer [10,11]. In contrast, most strains of M. ulcerans, which is closely related genetically to M. marinum and M. tuberculosis, but persists in extracellular locations during mammalian infection, lack most of the ESX-1 components as well as orthologs of the genes extending from Rv3879c thru Rv3883c [12,13]. Although the ESX-1 secretion machinery (Rv3870, Rv3871, and Rv3877) is required for the arrest of phagosome maturation by M. tuberculosis during an infection of macrophages, the known ESX-1 substrates are dispensable [6]. The multiple phenotypes and host responses dictated by the ESX-1 secretory apparatus suggest that there may be additional substrates, components, and regulatory molecules yet to be identified.
Recently, a third ESX-1 substrate, EspA (Rv3616c), was identified [14]. Unlike ESAT-6 and CFP-10, EspA is encoded at a locus distant from the ESX-1 machine, yet this substrate is codependent with both ESAT-6 and CFP-10 for secretion. The mechanism for this interdependence has not been determined, but the interaction between ESAT-6 and CFP-10 in the bacterial cytosol appears to be required for secretion of the heterodimer [15–18]. Presumably, the stable heterodimer is also required for the secretion of EspA.
The M. tuberculosis region of difference 1 (RD1) locus (Rv3871-Rv3879c) and the neighboring genes encode the ESX-1 substrates ESAT-6 and CFP-10, as well as core components of the secretion machine [1–3]. These core components include at least two putative SpoIIIE/FtsK ATPase family members (Rv3870 and Rv3871), a proline-rich predicted chromosome-partitioning ATPase (Rv3876), and a putative transporter protein with 12 transmembrane domains (Rv3877). The non-RD1 gene cluster Rv3616c-Rv3614c also is required for secretion of the known substrates [14,19]. Additional proteins are likely to be necessary for the assembly of the ESX-1 machinery, because in M. smegmatis, genes extending from homologs of Rv3866 through Rv3883c have been shown to be required for ESX-1–mediated secretion [11]; an M. bovis mutant disrupted for the expression of the genes homologous to Rv3867 through Rv3869 fails to secrete ESAT-6 and CFP-10 [20]; and in M. marinum, the locus required for ESX-1–mediated secretion extends at least from the homolog of Rv3866 (Mh3866) to the homolog of Rv3881c (Mh3881c), which in this work we rename espB (see below) [7].
Although these studies have identified multiple genes required for ESX-1 function, the biochemical interactions necessary for assembly of the secretion machine and for transport of substrates are still not understood. A model for CFP-10 secretion is that the carboxyterminus of the CFP-10 substrate is recognized by Rv3871, which in turn interacts with the integral membrane protein Rv3870 to direct CFP-10 through the secretion pore [15]. The interaction of CFP-10 with Rv3871 is also required for secretion of ESAT-6, suggesting that this is a requisite step in secretion of the ESAT-6/CFP-10 heterodimer by the ESX-1 machine.
Here, we show that Rv3881c and its M. marinum homolog, Mh3881c, are substrates for secretion by ESX-1. For this reason, we have named the gene product of this locus ESX-1 substrate protein B (EspB). In both species, espB encodes a gylcine-rich protein with a predicted molecular weight of ∼47 kDa, without any region of apparent similarity to the secretion signal of CFP-10 or other known secretion signals. Although a substrate of ESX-1, we find that the specific genes required for secretion of EspB differ from those required for the secretion of CFP-10. Biochemical investigation demonstrates that EspB forms a complex with Rv3879c and that Rv3879c interacts with Rv3871, the same component of ESX-1 that interacts with the ESAT-6/CFP-10 complex during its secretion. These data support a model that different substrates are delivered to the ESX-1 machine by molecularly distinguishable pathways. Moreover, each of these pathways for ESX-1–mediated secretion contributes to mycobacterial virulence.
A previous genetic screen for M. marinum mutants that fail to cause hemolysis led to the isolation of eight mutants in the extended RD1 (extRD1) locus, Mh3866-Mh3881c [7]. Of the eight mutants, espBM::tn (Mh3881c::tn) was the most attenuated for virulence to zebrafish, growth in macrophages, and cytotoxicity to J774 cells. Thus, we decided to investigate the espBM–encoded protein (EspBM) and its M. tuberculosis homolog (EspBT) in detail. The gene, espBM, is the first in a two-gene operon. Using quantitative RT-PCR, we found that the mutation disrupts the expression of both genes in the operon (unpublished data). We then sought to determine the genetic requirements for restoration of intracellular growth to the mutant. Introduction of a non-integrating plasmid, expressing either EspBM from the espBM promoter or EspBT from its native promoter, was sufficient to appreciably restore growth in macrophages to espBM::tn (Figure 1A). The non-integrating plasmids expressing both EspBM and Mh3880c or both EspBT and Rv3880c were not superior in restoration of intracellular growth. Thus, EspB is necessary and sufficient to appreciably complement espBM::tn, and the M. tuberculosis homolog functions equally well in M. marinum, demonstrating conservation of function.
While the expression of EspBM from a non-integrating plasmid appreciably restored growth in macrophages to espBM::tn, the complementation was not complete. Among possible explanations are that the transposon insertion exerted a polar effect on the operon upstream, Rv3883c-Rv3882c, which also might have a role in intracellular growth, or that a proper stoichiometry between EspB and ESX-1 is required for complete complementation. Therefore, integrating plasmids encoding either espBM along with the espBM promoter, espBM-Mh3880c along with the espBM promoter, or the entire locus Mh3883c-Mh3880c along with the Mh3883c promoter, were introduced into espBM::tn. The locus Mh3883c-Mh3880c, along with the Mh3883c promoter, was also introduced into espBM::tn on a non-integrating plasmid. Of these constructs, only the integrating plasmids encoding espBM-Mh3880c or Mh3883c-Mh3880c fully complemented the growth defect of espBM::tn (Figure 1B). Similarly, espBM alone appreciably restored a rough colony morphology to the espBM::tn mutant, but espBM-Mh3880c or Mh3883c-Mh3880c fully restored the rough colony morphology to espBM::tn (Figure S1). These results suggest that Mh3880c can contribute to M. marinum growth in macrophages when it is expressed along with espBM from the bacterial chromosome. In contrast, espBM contributes equally well to bacterial virulence whether expressed episomally or on the chromosome, suggesting that its contribution is more independent of its stoichiometry with respect to other virulence components.
As a first step toward understanding the role of EspB in virulence and growth in macrophages, we determined its localization in Mycobacteria grown in broth culture. The cell lysate and culture filtrate fractions of M. tuberculosis H37Rv, wild-type M. marinum, and M. marinum espBM::tn were probed with a mouse polyclonal antibody raised against a 100 amino acid fragment of EspBT extending from amino acid 234 to 333 (Figure 2A). EspB was detected in both the cell lysate and the culture filtrate fractions of M. tuberculosis, as well as in both the cell lysate and the culture filtrate fractions of wild-type M. marinum. EspB was not detected in either fraction of the espBM::tn culture, verifying the specificity of the antibody. GroEL, a non-secreted bacterial cytoplasmic protein, was found exclusively in the cell lysate, demonstrating that EspB did not appear in the culture filtrate as a result of cell lysis.
The EspB in the cell lysate had an Mr of 55 kDa on SDS-PAGE, while the EspB in the culture filtrate of both species ran at a slightly lower molecular weight. A lower molecular weight of EspB in the culture filtrate was also observed in a prior proteomic analysis of M. tuberculosis H37Rv proteins [21], in which EspB in the cell lysate was observed on a 2-D gel as a single spot with an apparent molecular weight of 55.6 kDa, while the EspB in the culture filtrate was observed as two spots with apparent molecular weights of 49.7 kDa and 48.4 kDa. Therefore, EspB might be cleaved either during or after secretion. To test this possibility, a V5 epitope tag was fused to the N-terminus of EspBM and a His6x epitope tag was fused to the C-terminus. The resulting construct, V5-EspBM-His6x, was expressed in the espBM::tn mutant. Like the native protein, V5-tagged EspB was detected in the cell lysate as a single band and as a doublet in the culture filtrate (Figure 2B). In contrast, His-tagged protein was only detected in the cell lysate fraction, suggesting that EspBM in the culture filtrate is C-terminally truncated.
To assess which ESX-1 genes are required for EspB secretion, its compartmentalization between cell lysate and culture filtrate was determined for several M. marinum ESX-1 mutants (Figure 2C). Although EspBM was found in both the cell lysate and culture filtrate fractions of most mutants, EspBM was not detected in the culture filtrates of MmΔRD1, Mh3868::tn, Mh3879c::tn, or Mh3871::tn.
The Mh3868::tn mutants failed to accumulate protein in the pellet, suggesting that Mh3868 protein could be involved in EspB synthesis or stability. Thus, of the ESX-1 genes tested, only Mh3879 and Mh3871 were clearly involved in EspBM secretion. In contrast, none of the mutants secreted ESAT-6 [7], and only Mh3879c::tn and Mh3878c::tn secreted CFP-10 normally. This difference in secretion requirements for ESAT-6 and CFP-10 in M. marinum has been noted previously [7]. Complementation of Mh3879::tn and espB::tn restored EspBM secretion. GroEL was absent from culture filtrates of all strains, and secretion of the fibronectin attachment protein (FAP), a protein secreted in a Sec-dependent manner [22], was not disturbed in any of the extRD1 mutants. Thus, the product of the espBM gene is a secreted protein that requires Mh3871, a core component of the ESX-1 secretion machine, for export; we have therefore named it ESX-1 substrate protein B (EspB). However, EspB, ESAT-6, and CFP-10 differ with respect to the extRD1 genes required for their secretion. EspBM secretion depends on Mh3879c, but is independent of Mh3866 and Mh3867, while CFP-10 shows the inverse pattern.
To demonstrate the importance of the ESX-1 machine in EspB secretion in another strain of M. marinum, we examined the 1218R strain and an isogenic mutant in which the Mh3871 gene had been disrupted. The M strain, used for the previous experiments, is a human isolate, whereas 1218R was originally isolated from an infected fish. Wild-type 1218R secreted EspB, but the Mh3871 mutant did not (Figure S2), confirming the importance of ESX-1 in the secretion of this protein by M. marinum. Complementation of the mutant with either the M. marinum or M. tuberculosis homolog of Mh3871 restored secretion of EspBM to this mutant, suggesting parallel functions for the genes in the two species.
To test directly whether ESX-1 was required for EspB secretion by M. tuberculosis, we examined culture filtrates from M. tuberculosis Erdman and the isogenic mutants Rv3870::tn, Rv3871::tn, and ΔCFP-10 (Figure 2D). Secretion of EspBT by wild-type M. tuberculosis was abrogated in the Rv3870 and Rv3871 mutants, but not in the ΔCFP-10 mutant. Thus, EspB is a secreted protein in both M. marinum and M. tuberculosis, and its secretion requires core ESX-1 components in both species of Mycobacteria. Importantly, EspB is the first ESX-1 substrate in M. tuberculosis whose secretion is not disrupted in the ΔCFP-10 mutant.
Of the ten M. marinum extRD1 mutants we examined, MmΔRD1, espBM::tn, and Mh3871::tn were disrupted for the secretion of all three substrates: ESAT-6, CFP-10, and EspBM. In contrast, the Mh3879::tn mutant was disrupted only for the secretion of ESAT-6 and EspBM, while the Mh3866::tn and Mh3867::tn mutants were disrupted only for ESAT-6 and CFP-10 secretion. To assess the importance of the multiple ESX-1 substrates for growth in macrophages, we infected murine bone marrow–derived macrophages (BMDMs) with wild-type M. marinum, with strains lacking one secreted effector, or with strains lacking secretion of all the known ESX-1 substrates. As shown in Figure 3, MmΔRD1, espBM::tn, and Mh3871::tn, which fail to secrete all substrates, are more attenuated for growth in macrophages than Mh3866::tn, which still secretes EspBM, or Mh3879c::tn, which still secretes CFP-10. Therefore, we conclude that the various substrates of ESX-1 each contribute to virulence.
To learn more about the involvement of ESX-1 in EspB secretion, we tested whether EspB would interact with other ESX-1 genes by bacterial two-hybrid analysis (Figure 4). An advantage of the bacterial two-hybrid system is that it can allow detection of interactions of membrane-bound proteins [23]. In this assay, potential protein–protein interactions are assessed by determining the ratio of colonies that grow on selective medium to the number grown on non-selective medium. For each of the bait plasmids, co-transformation with an empty target resulted in a ratio of colonies on selective to non-selective medium of less than 0.1%, as did co-transformation of the EspBT target with an empty bait. In contrast, the Rv3879c bait and EspBT target resulted in a ratio of 7.6%, an increase of more than 75-fold. An Rv3876 bait also showed interaction above background with EspBT, but since the M. marinum Mh3876::tn mutant showed significant EspBM secretion (Figure 2C), any interaction between Rv3876 and EspBT is not likely to be required for EspB secretion and thus was not pursued.
To test for an analogous interaction between EspBM and Mh3879c and to confirm the potential interaction between EspBT and Rv3879c suggested by the two-hybrid assay, we performed in vitro pull-down assays. All of the proteins used were expressed in Escherichia coli as GST- or V5-epitope-tagged fusions. Controls for nonspecific interactions included GST alone, as well as GST-syntaxin2, and GST-Shp1. As shown in Figure 5A, GST-tagged EspBM, but none of the GST controls, bound specifically to V5-tagged Mh3879c. In the reciprocal experiment, GST-tagged Mh3879c bound specifically to V5-EspBM. Similarly, as shown in Figure 5B, GST-tagged EspBT bound specifically to V5-tagged Rv3879c, and GST-tagged Rv3879c bound specifically to V5-tagged EspBT. These data demonstrate that recombinant EspBT and Rv3879c, as well as their M. marinum homologs, interact in vitro.
Since Rv3871 mutants in both M. tuberculosis and M. marinum fail to secrete EspB, we used GST pulldowns to test whether Rv3871 interacts with either EspBT or Rv3879c. GST-tagged Rv3879c bound to V5-tagged Rv3871, whereas the GST controls and GST-EspBT did not bind to Rv3871. This suggests that Rv3879c may facilitate EspBT secretion through an interaction with Rv3871.
To identify whether EspB, like CFP-10, requires its carboxyterminus for secretion, we constructed a series of EspBM deletion mutants with N-terminal V5 tags and expressed them in the espBM::tn mutant strain using the espBM promoter. As shown in Figure 6A, V5-tagged full-length EspBM was secreted. This N-terminally tagged protein, like native EspBM, underwent C-terminal truncation either during or after secretion. EspBM deletion mutant constructs Δ(2–31), Δ(264–271), and Δ(400–454) were stably expressed in M. marinum, but only EspBM Δ(400–454) accumulated in the culture filtrate. The secreted EspBM Δ(400–454) had a higher apparent molecular weight than the secreted full-length EspBM, presumably because deletion of the C-terminal 55 amino acids inhibits some of the carboxyterminal proteolytic processing. This result demonstrates that the C-terminus of EspBM is dispensable for secretion, but N-terminal and internal amino acids are required. Next, we tested how these EspBM mutants interacted with Mh3879c. Lysates of E. coli that express V5-tagged EspBM mutants were incubated with GST-Mh3879c. While full-length EspBM and EspBM Δ(400–454) bound to GST-Mh3879c, the stably expressed but non-secreted EspBM Δ(2–31) and EspBM Δ(264–271) constructs did not bind to GST-Mh3879c (Figure 6B). These data support a model in which EspB interacts with Rv3879c, which in turn interacts with Rv3871, to facilitate the secretion of EspB.
Because CFP-10 and ESAT-6 are secreted as a heterodimer, we assessed whether Mh3879c and EspB might be secreted similarly. The fusion constructs V5-Mh3879c, Mh3879c-His6x, and V5-Mh3879c-His6x were expressed from the endogenous Mh3879c promoter on non-integrating plasmids in both wild-type M. marinum and in the Mh3879c::tn mutant. Introduction of V5-Mh3879c fully complemented the EspBM secretion defect of the Mh3879c::tn mutant, but Mh3879c-His6x and V5-Mh3879c-His6x failed to complement the secretion defect (Figure S3A). In wild-type M. marinum, V5-Mh3879c and V5-EspBM were expressed at nearly identical levels in the cell lysate, but only V5-EspBM was detected in the culture filtrate (Figure S3A). To determine whether failure of secretion reflected inefficient competition of V5-tagged protein with native protein, the V5-EspBM secretion was also analyzed in the Mh3879::tn mutant. In this strain as well, V5-Mh3879c was found only in the cell lysate. Thus, V5-tagged Mh3879c, while fully competent to mediate EspB secretion, was not itself secreted, suggesting that Mh3879c and EspB are not secreted as a heterodimer. The C-terminally His6x-tagged Mh3879c, which did not restore EspB secretion to the Mh3879::tn mutant, also was detected only in the cell lysate. Since Mh3879c-His6x failed to complement the EspBM secretion defect of the Mh3879c::tn mutant, we hypothesized that the carboxyterminus of Rv3879c might be required for interaction with EspB. To test this hypothesis, lysates of E. coli that express V5-tagged Rv3879 mutants were incubated with GST alone, GST-Rv3871, or GST-EspBT. While full-length Rv3879 and Rv3879 Δ(1–166) bound to GST-EspBT, Rv3879 Δ(564–729) failed to bind to GST-EspBT (Figure S3B). None of the constructs bound to GST alone, and all three constructs bound to GST-Rv3871. Thus, the carboxyterminal 166 amino acids of Rv3879 are required for EspB secretion, but not for interaction with the ESX-1 machine.
In this study, we identified EspB as a novel substrate of the ESX-1 secretion system and demonstrated a requirement for the Mh3879c and Mh3871 genes in the secretion of EspBM. Further, we showed protein complex formation between EspBM and Mh3879c, as well as identical behavior of their M. tuberculosis homologs. Two mutants of EspBM that were stable after synthesis but failed to bind Mh3879c were not secreted, while a large carboxylterminal deletion did not interfere with either Mh3879c binding or secretion. Additionally, the carboxyterminus of Rv3879c/Mh3879c is required for interaction with and secretion of EspB. These results suggest that the EspB/Mh3879c protein complex is required for EspBM secretion. While complex formation between ESAT-6 and CFP-10 is required for their secretion as a heterodimer by M. tuberculosis, Mh3879c appears not to be secreted. Our data, though, do not exclude the possibility that the aminoterminus of Mh3879c is quantitatively removed during or immediately after secretion, since we do not have and could not probe with antibodies to the native protein. We hypothesize that Mh3879c acts as a cytosolic chaperone to deliver EspBM to the secretion machine. We showed that Rv3879c interacts directly with Rv3871 and that Rv3871, in addition to being required for the secretion of ESAT-6/CFP-10, is required for the secretion of EspB. Although our work does not reveal precisely how EspB is delivered to the ESX-1 machine, our data demonstrate that Rv3879c can interact with Rv3871 as well as with EspBT, suggesting that EspB may be targeted to Rv3871 in this way. We propose that the mechanisms of EspB and CFP-10 secretion intersect at binding to Rv3871 (Figure 7).
We also found that disruption of Mh3868 leads to loss of accumulation of EspB in the bacterial cytosol. We previously observed that disruption of Mh3868 prevents bacterial accumulation of ESAT-6 and CFP-10 [7]. Mh3868 and its M. tuberculosis homolog Rv3868 are predicted to be AAA ATPases, which suggests that they may function as chaperones for the translocation of ESX-1 substrates, but little is known about this key protein. We have found that CFP-10 and espBM mRNAs are expressed in the Mh3868::tn mutants (B. McLaughlin and E. Brown, unpublished data), suggesting that the Mh3868 gene product affects either the translation or stability of the ESX-1 substrates. Characterizing the function of Mh3868 will certainly be important to better understand ESX-1–mediated secretion.
Like ESAT-6 and CFP-10, EspA is secreted by the ESX-1 machine. Whether any of the M. marinum genes with sequence similarity to espA are functional orthologs has not yet been determined. Loss of either EspA or EspB inhibits secretion of ESAT-6 and CFP-10, but the reason for their requirement is unknown. It may be that as substrates reach the final common pathway for secretion, they interact in a manner that leads to cooperative secretion. Clearly, though, the secretion of EspB is quite distinct from that of EspA. While EspA requires CFP-10 for its secretion, EspB secretion is independent of CFP-10. EspB secretion is not disrupted in the M. marinum mutants Mh3866::tn and Mh3867::tn, neither of which secrete CFP-10, nor is EspB secretion disrupted in the M. tuberculosis ΔCFP-10 mutant. These data are consistent with the model that EspB, unlike either ESAT-6 or EspA, is targeted to the ESX-1 machine independently of CFP-10.
These studies beg the question of whether it is possible to determine which ESX-1 substrates are most important for virulence. This has been a difficult task because of the apparent codependence of the various substrates on each other for secretion. However, our results allowed a somewhat different approach. We used a set of extRD1 mutants in which some (Mh3866::tn and Mh3867::tn) failed to secrete CFP-10, but did secrete EspB; while another mutant (Mh3879::tn) secreted CFP-10 but failed to secrete EspB; while mutants that disrupted the core secretion machinery (MmΔRD1 and Mh3871::tn) and espBM::tn itself failed to secrete all substrates. We found that mutants lacking secretion of both substrates had a more marked growth defect in macrophages than the mutants lacking secretion of only one substrate. This suggests that the different substrates make distinct, and potentially additive, contributions to virulence. Although we cannot say that the defects in intracellular growth of the various mutants are caused by the substrates we have identified, our work does support the hypothesis that ESX-1 secretes more than one substrate that contributes to the virulence of Mycobacteria and that different substrates may have independent contributions to bacterial pathogenesis.
In summary, this work has identified a novel substrate for ESX-1–dependent secretion and has demonstrated interactions of this substrate with a protein encoded within RD1, expanding our understanding of how genes within this locus contribute to this novel secretion pathway. Furthermore, we have demonstrated that secretion of distinct ESX-1 substrates follows variable pathways to interaction with the core secretion machinery, and that the different substrates may contribute independently to intracellular survival and growth of the bacteria. These data extend the understanding of a major virulence mechanism of Mycobacteria.
All strains and plasmids used in this study are listed in Table 1. M. marinum strains were grown as previously described [24]. The designations assigned by the Sanger Institute in the annotation of the M. marinum genome and the corresponding DNA sequences are available at http://www.sanger.ac.uk/Projects/M_marinum/.
The transposon insertion in the mutant espBM::tn lies between the 175th and 176th base pairs of the espBM gene, and the kanamycin gene within the transposon is transcribed opposite to the direction of transcription of the espBM gene. The strains M. marinum M attB::hygr and espBM::tn attB::hygr were constructed by transforming the strains M. marinum M WT and espBM::tn with the plasmid pMV306.hyg. The strains espBM::tn attB:: espBM hygr and espBM::tn attB:: espBM -Mh3880c hygr were constructed by ligating 250 bp upstream of espBM along with espBM or espBM -Mh3880c into pMV306.hyg and then transforming the resulting plasmids, pBM264 and pBM262, into espBM::tn. The strain espBM::tn attB:: Mh3883c-Mh3880c hygr was constructed by ligating 345 bp upstream of Mh3883c along with Mh3883c-Mh3880c into pMV306.hyg and then transforming the resulting plasmid, pBM263, into espBM::tn. To construct the plasmids pBM841, pBM540, and pBM810, the genes Rv3871, Rv3879c, and Rv3881c were PCR amplified from the cosmid RD1-2F9 [25] and ligated into pBM510, a derivative of pET22b+ in which the N-terminal His tag was replaced with the V5 epitope tag. To construct the plasmids pBM843 and pBM504, the genes Mh3879c and Mh3881c were PCR amplified from M. marinum M genomic DNA and ligated into pBM510. To construct the plasmids pBM332 and pBM336, a series of fragments were ligated into pLYG206 to achieve the following sequence ligated into the NotI and XbaI sites: 250 bp upstream of espBM, then the V5 epitope, then the espBM gene, and finally, in the case of the pBM336 plasmid, the His6x epitope. The plasmids pBM869, pBM870, and pBM871 were made in a manner synonymous to that of pBM332 and pBM336, where the Mh3879 promoter and gene were used. The plasmids pBM367 and pBM400v were constructed by PCR from pBM332 and re-ligation of the truncated gene fragments back into pBM332, while pBM398 was generated by quick-change mutagenesis (Stratagene, http://www.stratagene.com/). For pBM589, pBM398e, and pBM400ve, the espBM gene fragments in the plasmids pBM367, pBM398, and pBM400v were cut by restriction digest and ligated into pBM504. For pBM856, pBM550, pBM553, and pBM551, the genes Mh3879c, espBM, Rv3879c, and espBT were cut by restriction digest from the plasmids pBM843, pBM504, pBM540, and pBM810 and ligated into the GST expression vector pGex-KG. To construct the plasmid pMh3879, 250 bp upstream of the gene Mh3879c together with Mh3879c was PCR amplified from M. marinum genomic DNA and inserted into pLYG206. The plasmids pBM1010 and pBM1013 were constructed by restriction digests of pBM540 to excise portions of Rv3879c, and ligation of 5′ phosphorylated hybridized oligos that restored the frame and created the Rv3879 deletions Δ(1–166) and Δ(564–729). To construct the plasmid p(Mh3883c-Mh3880c), the 345 bp upstream of Mh3883c along with Mh3883c-Mh3880c was cut by restriction digest from the plasmid pBM263 and inserted into pLYG206. To construct the plasmid p(espBT-Rv3880c), 250 bp upstream of the operon Rv3881c-Rv3880c together with the operon were inserted into pLYG206.
M. marinum strains were grown in 40-mL cultures to 0.5 OD600 in 7H9 medium. The cultures were centrifuged and washed three times with 15 mL of PBS before re-suspension in 40 mL of Sauton's medium, supplemented with 0.015% Tween-80. When strains containing non-integrating plasmids for complementation were grown in Sauton's medium, the Sauton's medium was supplemented with Zeocin (5 μg/ml; Invitrogen, http://www.invitrogen.com/). After growth for 36 h at 30 °C, 105 rpm, in Sauton's medium, the cells were harvested by centrifugation. Supernatants were filtered through a 0.22-μm-pore-size filter with a glass pre-filter and concentrated with an Amicon Ultra-15 (5,000-molecular-weight cutoff; Millipore, http://www.millipore.com/) to 200 μL, which was saved as the culture filtrate (CF) fraction.
Pelleted cells were washed and resuspended in 1.5 mL of PBS with a protease inhibitor cocktail and 1 mM PMSF. Pellets were lysed using glass beads and the mini-bead beater (BioSpec Products, http://www.biospec.com/) with three 40-s pulses at maximum speed and incubations on ice in between each pulse, and then centrifuged at 3,000g for 2 min at 4 °C to remove unbroken cells. The resulting supernatant was collected and saved as the cell lysate (CL) fraction. M. tuberculosis (Erdman) culture filtrate and cell lysate fractions were prepared as previously described [1]. Total protein concentrations were determined by a Bradford assay.
Pellet and culture filtrate fractions were separated by SDS/PAGE on 10%–20% gradient polyacrylamide gels for detection of CFP-10; 7.5% polyacrylamide gels for detection of EspB, GroEL, or V5-tagged Mh3879; and 12.5% polyacrylamide gels for detection of FAP. Proteins were visualized by immunoblotting by using antibodies against EspB at a concentration of 1:500 (mouse polyclonal to the 100 amino acid fragment of Rv3881c [234–333 aa], Arizona State University CIM Antibody Core), and the blot was developed using ECL reagent West Dura (Pierce, http://www.piercenet.com/). Anti-CFP-10 (rabbit polyclonal; Colorado State University, http://www.cvmbs.colostate.edu/microbiology/tb/top.htm) was used at a concentration of 1:50000, blots of the culture filtrate fraction were developed using West Pico (Pierce), and blots of the cell lysates were developed using West Dura (Pierce). Anti-GroEL (rabbit polyclonal, SPA-875 / SPS-875; Stressgen, http://www.assaydesigns.com/) was used at a concentration of 1:10000, and blots were developed using West Pico (Pierce). Anti-FAP [22] for M. marinum samples was a rabbit polyclonal, used at a concentration of 1:10000 and developed using West Pico (Pierce). Anti-FAP for M. tuberculosis Erdman samples was CS-93 (Colorado State University), mouse monoclonal, used at a concentration of 1:20, and developed using West Pico (Pierce). His6x epitope was detected with a mouse monoclonal (Novagen, http://www.emdbiosciences.com/html/NVG/home.html) at a concentration of 1:1500, and V5 epitope was detected with a mouse monoclonal (R960–25, Invitrogen), at a concentration of 1:5000, and these blots were developed using West Pico (Pierce).
The genes Rv3614c, Rv3615c, and Rv3616c, which were PCR amplified from genomic DNA, and each of the genes in the region Rv3864 through Rv3883, which were PCR amplified from the cosmid RD1-2F9 [25], were cloned into the “bait” vector pBT (BacterioMatch II; Stratagene) in frame with cI. Rv3881c was cloned into the “target” vector pTRG in frame with the N-terminal subunit of RNA polymerase according to the manufacturer's instructions. The constructs were co-transformed into the E. coli two-hybrid system reporter validation strain XL1-Blue MRF′ hisB lac [F′ laqIq HIS3 aadA Kanr] and plated onto both the selective (+5 mM 3AT) and the non-selective screening medium according to the manufacturer's instructions. The non-selective screening plate is histidine-dropout M9 agar supplemented with 0.5 mM IPTG, 12.5 μg/ml tetracycline, and 25 μg/ml chloramphenicol. The selective screening plate is histidine-dropout M9 agar supplemented with 0.5 mM IPTG, 12.5 μg/ml tetracycline, 25 μg/ml chloramphenicol, and 5 mM 3-amino-1,2,4-triazole.
GST fusion proteins, GST alone, and V5-tagged proteins were expressed in the BL21-RP codon plus E. coli strain (Stratagene) by addition of 0.2 mM IPTG (3 h at 30 °C). Bacterial cultures were lysed in buffer containing 50 mM HEPES (pH 7.4), 300 mM NaCl, 1% Triton X-100, 0.5 mM EDTA, and protease inhibitor cocktail (Roche). Solubilized proteins were separated by centrifugation at 20,000g for 10 min. The GST fusion proteins and GST alone were bound to glutathione agarose beads (Amersham Biosciences, http://www.gelifesciences.com) by incubation overnight at 4 °C. The beads were then extensively washed with PBS containing 0.1% Triton X-100. Bacterial lysates containing solubilized V5-tagged proteins, in lysis buffer, were incubated with the GST protein–loaded agarose beads overnight at 4 °C. After washing three times with PBS containing 1% Triton X-100, bead-bound protein was eluted in Laemmli buffer, seperated by SDS-PAGE, and analyzed by western blot.
All macrophages used in these experiments were derived from bone marrow cells of C57BL/6 mice that were differentiated for 6 d in DMEM supplemented with 10% CMG supernatant [26] and 10% fetal bovine serum (FBS; HyClone, http://www.hyclone.com/). Immediately prior to infection, macrophage monolayers were washed once with FBS-free DMEM. M. marinum strains were each grown to OD600 of 1.0, prepared for infection, and incubated with macrophages as previously described [7]. All infections were performed at a multiplicity of infection of 1, for 2 h at 32 °C, in a 5% CO2, humidified environment, in 24-well plates. The time at which M. marinum was added to the well was designated time zero. At the end of the 2 h incubation period (T = 2 h), infected monolayers were washed twice with DMEM and further incubated in DMEM containing 0.1% FBS and 200 μg of amikacin/ml for 2 h to kill extracellular bacteria. At the end of the antibiotic treatment, monolayers were washed twice with DMEM and incubated in DMEM containing 0.1% FBS at 32 °C and 5% CO2. Intracellular bacteria were enumerated by lysing macrophage monolayers and diluting and plating bacteria exactly as described [7]. Statistical analysis was performed by calculating the one-way analysis of variance (ANOVA) with GraphPad Prism 4.0 (GraphPad Software, http://www.graphpad.com).
The GenBank (http://www.ncbi.nlm.nih.gov/Genbank/index.html) accession numbers for the gene products discussed in this paper are CFP-10 (NP_218391), ESAT-6 (YP_178023), Rv3866 (NP_218383), Rv3867 (NP_218384), Rv3868 (NP_218385), Rv3870 (NP_218387), Rv3871 (NP_218388), Rv3876 (NP_218393), Rv3877 (NP_218394), Rv3878 (NP_218395), Rv3879c (NP_218396), Rv3880c (NP_218397), and Rv3881c (NP_218398). |
10.1371/journal.ppat.1004593 | DNA Is an Antimicrobial Component of Neutrophil Extracellular Traps | Neutrophil extracellular traps (NETs) comprise an ejected lattice of chromatin enmeshed with granular and nuclear proteins that are capable of capturing and killing microbial invaders. Although widely employed to combat infection, the antimicrobial mechanism of NETs remains enigmatic. Efforts to elucidate the bactericidal component of NETs have focused on the role of NET-bound proteins including histones, calprotectin and cathepsin G protease; however, exogenous and microbial derived deoxyribonuclease (DNase) remains the most potent inhibitor of NET function. DNA possesses a rapid bactericidal activity due to its ability to sequester surface bound cations, disrupt membrane integrity and lyse bacterial cells. Here we demonstrate that direct contact and the phosphodiester backbone are required for the cation chelating, antimicrobial property of DNA. By treating NETs with excess cations or phosphatase enzyme, the antimicrobial activity of NETs is neutralized, but NET structure, including the localization and function of NET-bound proteins, is maintained. Using intravital microscopy, we visualized NET-like structures in the skin of a mouse during infection with Pseudomonas aeruginosa. Relative to other bacteria, P. aeruginosa is a weak inducer of NETosis and is more resistant to NETs. During NET exposure, we demonstrate that P. aeruginosa responds by inducing the expression of surface modifications to defend against DNA-induced membrane destabilization and NET-mediated killing. Further, we show induction of this bacterial response to NETs is largely due to the bacterial detection of DNA. Therefore, we conclude that the DNA backbone contributes both to the antibacterial nature of NETs and as a signal perceived by microbes to elicit host-resistance strategies.
| Comprising the first line of the innate immune response, neutrophils combat infectious microorganisms through the release of toxic molecules, phagocytosis of invaders and the production of the recently characterized neutrophil extracellular traps (NETs). The antimicrobial activity of NETs has been attributed to proteins bound to the DNA backbone. Our results demonstrate that the DNA lattice of each NET is potently antibacterial and elicits upregulation of protective surface modifications by the opportunistic bacterial pathogen Pseudomonas aeruginosa. These modifications, previously shown to protect bacteria from antimicrobial peptides, confer greater bacterial tolerance to DNA and NET-mediated antibacterial activity. Treatments that quench the cation chelating capacity of DNA restore bacterial viability and suppress the expression of surface modifications even in the presence of intact NETs. These observations highlight the dual function of DNA as an antibacterial component of NETs, but also a signal perceived by bacteria to induce broad host resistance strategies. Therefore, the ability of P. aeruginosa to sense and defend against the antibacterial activity of neutrophil extracellular traps may contribute to long-term survival in chronic infection sites including the Cystic Fibrosis lung.
| Neutrophils are central mediators of the innate immune defense system and perform their role by killing invading microbes through phagocytosis, degranulation, and the release of neutrophil extracellular traps (NETs) [1, 2]. The scaffold of NETs is composed of genomic DNA, which is enmeshed with antimicrobial proteins normally found in the nucleus, granules, or cytoplasm of neutrophils [1, 2]. Although largely characterized ex vivo using purified human neutrophils and the chemical inducer phorbol-12-myristate 13-acetate (PMA), the process of NETosis resulting in the generation of NETs has been observed in vivo during infection where these structures function to trap bacteria, fungi, protozoa and viruses [1–5]. The mechanism by which NETs kill microbial invaders remains controversial [5–7]. Given the detection of known antimicrobial proteins that decorate the genomic lattice structure [1, 2, 8, 9], current models describing the antimicrobial function of NETs focus on the role of NET-bound proteins. However, most of these proteins are present in low abundance and evidence of their antimicrobial function while bound to the NET structure is limited to a few proteins [1, 8–11]. NET-bound calprotectin is a zinc-chelating protein with antimicrobial activity against Candida and Klebsiella that can be neutralized with excess zinc [3, 9, 10]. Histones, the most abundant NET-bound proteins [10], possess direct membrane-acting antibacterial activity [12, 13] and were shown to contribute to killing of Staphylococcus and Shigella [1]. Cathepsin G, a granular serine protease, is required for the clearance of Neisseria by NETs [8]. To demonstrate the antibacterial contribution of the latter two NET-bound proteins, antibodies raised to histones or cathepsin G were shown to limit the bactericidal capacity of NETs towards these pathogens [1, 8].
Immunocompromised individuals, including those with Cystic Fibrosis (CF), are particularly susceptible to P. aeruginosa infection, which is a major cause of morbidity and mortality in these patients. Chronic P. aeruginosa infection of the CF lung leads to an intense inflammatory immune response, resulting in the recruitment of large numbers of neutrophils to the site of infection [14, 15]. CF sputum is highly enriched in neutrophil-derived DNA, including that of NET origin, indicating that neutrophils deploy NETs in an effort to combat infection of the lung [16, 17]. However, persistence of P. aeruginosa in the midst of sustained neutrophil presence and NETosis suggests that the pathogen is capable of evading this host immune response [18].
An important virulence strategy adopted by successful microbial pathogens is the tolerance of NETs. In a number of cases described so far, microbial invaders accomplish this goal by either avoiding or disarming neutrophil extracellular traps. Modification of the bacterial capsule and surface-localized lipoteichoic acid reduces the trapping of Streptococcus pneumonia in NETs [19]. The secretion of extracellular nucleases by Staphylococcus aureus, Streptococcus pneumonia, group A Streptococcus and Vibrio cholerae highlight a shared virulence strategy that functions to degrade the DNA-backbone of NETs, enabling evasion or liberation of the bacteria from entrapment [2, 20–22].
Although microbial or exogenous DNase is proposed to dissolve NET structures to avoid capture, it has not been considered that the DNA backbone itself may be antimicrobial. We have previously demonstrated that extracellular DNA is an efficient chelator of divalent metal cations [23]. This cation chelation has pleiotropic effects on P. aeruginosa depending on the concentration of extracellular DNA. At subinhibitory levels, sequestration of cations by DNA leads to the expression of genes controlled by the two component systems PhoPQ and PmrAB that sense Mg2+ limitation [23–25]. However, at higher concentrations, extracellular DNA causes dramatic disruptions to the bacterial envelope integrity, which leads to lysis and rapid cell death [23]. It is predicted that the phosphodiester backbone is required for cation sequestration of metal cations on the bacterial surface and the membrane-destabilizing antimicrobial activity of extracellular DNA [23]. Therefore, we hypothesize that the DNA backbone of NETs contributes to their antibacterial activity. Here we show that neutralizing the membrane destabilizing activity of extracellular DNA by quenching the capacity of the phosphodiester backbone to chelate cations protected bacteria from NETs. P. aeruginosa is capable of detecting the DNA lattice of NETs, and in response, upregulates genes required to modify the bacterial outer membrane surface to tolerate the toxic effects of DNA-mediated NET killing.
P. aeruginosa has been shown to induce the formation of neutrophil extracellular traps in purified human neutrophils [18] and NETs have been observed in CF sputum [16, 17], where P. aeruginosa is a predominant pathogen. PMA-induced NETs contained known neutrophil proteins embedded in the DNA lattice, including myeloperoxidase (MPO) and histones [2] (Fig. 1A). We also show that Gfp-tagged P. aeruginosa is efficiently trapped and aggregated in NETs (Fig. 1A). However, in vivo NETosis during P. aeruginosa infection has not been reported. Therefore intravital confocal microscopy was used to determine whether P. aeruginosa elicited NETosis in the mouse skin infection model [26]. Infection with P. aeruginosa led to the production of large NET-like structures that stained with the DNA-binding dye Sytox green and entrapped ChFP-labeled P. aeruginosa (Fig. 1B and C). In addition to the presence of NETs, neutrophils remained chemotactic and phagocytosed bacteria, suggesting that multiple neutrophil clearance mechanisms are employed in vivo to combat P. aeruginosa (Fig. 1D and S1 Movie).
Given that P. aeruginosa induces the production of NETs in vitro and in vivo (Fig. 1), we sought to compare the relative abilities of P. aeruginosa, S. aureus, E. coli and the chemical inducer PMA to elicit NETosis. In the presence of purified human neutrophils, PMA, E. coli and S. aureus induced significantly more NET formation, relative to P. aeruginosa within 1 hour of coincubation (Fig. 2A). However, at later time points (3h), P. aeruginosa elicited the formation of similar amounts of NET structures. Furthermore, quantification of the number of NETs and NET-area using the skin infection model confirmed that P. aeruginosa weakly induces NETosis relative to S. aureus in vivo (Fig. 2B and C).
One possible explanation for the reduced NET formation in the presence of P. aeruginosa is the production of a microbial secreted DNase that degrades NETs more efficiently than other organisms. Therefore, we measured the DNase activity in overnight culture supernatants, as well as in supernatants isolated from coincubating bacteria with PMA-treated neutrophils. We demonstrate that there was little, if any, DNase activity produced by P. aeruginosa, S. aureus or E. coli under these conditions (S1 Fig.). Since cations are a requirement for DNase activity, we were able to restore DNase activity in S. aureus supernatants after the addition of excess cations (S1 Fig.). However, under the cation-free conditions used to quantitate antibacterial NET function, even in the presence of PMA-stimulated neutrophils, DNase activity was not detected and thus P. aeruginosa appears to limit NETosis through an uncharacterized mechanism.
In order to characterize the bactericidal capacity of NETs, neutrophils were treated with PMA to stimulate maximal NETosis and cytochalasin D to block phagocytosis, thus restricting bacterial killing to extracellular NET function [1, 8, 21, 22]. Importantly, the addition of cytochalasin D had no effect on NETosis induced in PMA-treated neutrophils (S2 Fig.). We used the conventional method of direct bacterial counts to enumerate the number of bacteria before and after challenge with PMA-induced NETs. Direct counts of NET-exposed bacteria revealed that P. aeruginosa was most tolerant to NET killing, whereas S. aureus and E. coli were significantly more sensitive (Fig. 3A). The addition of deoxyribonuclease (DNase) restored bacterial survival of the NET-sensitive organisms E. coli and S. aureus, confirming that killing was mediated by extracellular NET function (Fig. 3A). Further, the kinetics of bacterial killing by PMA-generated NETs was determined by measuring the loss of luminescence from chromosomally-tagged luminescent P. aeruginosa strain, PAO1::p16Slux [23], and plasmid-borne luminescent E. coli / pσ70-lux [27]. This approach confirmed that P. aeruginosa was more tolerant to NET killing than E. coli, where luminescence rapidly decreased upon neutrophil challenge (Fig. 3B).
The addition of exogenous DNase I and the production of secreted DNases by bacteria are the most effective means to disable NET killing [1, 2, 20, 21]. It is thought that DNase treatment dissolves NET structures, thereby releasing and diluting the antimicrobial proteins bound to NETs. We have previously shown that extracellular DNA has a potent antibacterial activity, as purified salmon DNA (2% w/v) causes several log orders of bacterial killing within minutes and breaks the integrity of both the inner and outer membranes, leading to lysis [23]. DNA is a very efficient cation chelator and the antimicrobial activity of DNA can be blocked with addition of excess divalent metal cations [23]. It is predicted that the cation chelation is mediated by the phosphodiester backbone.
To confirm the mechanism by which extracellular DNA kills bacteria, we monitored the loss of P. aeruginosa viability in the presence of dilute extracellular DNA (0.125% w/v; Fig. 4A). Bacterial survival was restored if DNA was pretreated with DNase or excess 5mM Mg2+, which degrades DNA or saturates its cation chelating ability, respectively, thus neutralizing the antibacterial activity (Fig. 4A). Pretreatment of extracellular DNA with calf intestinal alkaline phosphatase (PTase), which cleaves 5’-phosphates, also blocked the observed antibacterial activity (Fig. 4A). The addition of decreasing amounts of DNase, PTase or Mg2+, resulted in marked, dose-dependent, decreases in bacterial survival when challenged with 0.15% DNA (Fig. 4B). The DNA-mediated damage to the outer membrane leads to the formation of ChFP-enriched outer membrane vesicle-like structures (OMVs; Fig. 4C) [23]. However, incubation of P. aeruginosa with extracellular DNA pretreated with DNase, PTase and Mg2+ greatly diminished the number of ChFP-labeled OMVs relative to control conditions (Fig. 4D), confirming that the bactericidal mechanism of extracellular DNA is through disruption of the bacterial membrane.
Extracellular DNA-mediated damage to membrane integrity was confirmed by using flow cytometry (Fig. 4E). Treatment of P. aeruginosa with DNA resulted in a new population of cells that were dual positive for SYTO9 and PI, indicative of membrane damage and increased PI uptake, and possibly dead cells. The increased staining of DNA-exposed bacteria by membrane-impermeable propidium iodide (PI) was not observed in DNase and Mg2+ pretreatments and was reduced in PTase pretreated DNA samples (Fig. 4E and F). To ensure cation chelation was responsible for the observed bacterial membrane destabilization, we assessed whether the known cation chelator EDTA could cause membrane disruption. Like DNA, EDTA caused major outer membrane disruptions and the release of OMVs, as well as a dramatic increase in PI-staining of EDTA-treated cells monitored by flow cytometry (S3 Fig.) [28]. To determine whether the antibacterial capacity of DNA requires direct bacterial contact or can be mediated through passive cation sequestration, we exposed P. aeruginosa to high concentrations of DNA spatially separated by an ion-permeable barrier. Blocking direct interaction between P. aeruginosa and DNA resulted in bacterial survival after a prolonged exposure, compared to the rapid antibacterial activity of DNA in direct contact with the bacteria (S4 Fig.). Together, these results demonstrate that the antibacterial activity of extracellular DNA requires direct contact, and the phosphate backbone for cation chelation, leading to membrane disruption and bacterial cell death.
The bactericidal activity of neutrophil extracellular traps is attributed to direct contact and exposure of bacteria to the antimicrobial proteins embedded in the DNA scaffold of NETs [1, 2]. Given the antimicrobial activity of DNA, we propose that the DNA backbone of the NET itself is antibacterial. Therefore, if DNA contributes to bacterial killing, treatments that quench the cation chelation potential of the DNA backbone will block bactericidal activity of NETs. To address this possibility, PMA-activated neutrophils were treated with the addition of DNase, PTase or excess Mg2+ and bacterial viability was monitored. The DNA-targeted treatments completely protected P. aeruginosa and E. coli from killing by neutrophil extracellular traps (Fig. 5A). To confirm these results, we monitored the luminescence of P. aeruginosa PAO1::p16Slux co-incubated with PMA-activated neutrophils and observed that the antibacterial effects of NETs were neutralized by treatment with exogenous DNase, Mg2+ cations and PTase (Fig. 5B).
To determine whether the restored bacterial viability in the presence of NETs was due to preventing damage to the bacterial envelope, we performed flow cytometry to assess membrane integrity. Increased PI staining of NET-exposed bacteria is an indicator of membrane damage, which was completely blocked by addition of DNase (Fig. 5C). The addition of excess Mg2+ or treatment with exogenous PTase also limited membrane damage (Fig. 5C). Importantly, both Mg2+ and PTase treatments neutralized the antimicrobial activity of NETs (Fig. 5A and B) did not disrupt overall NET architecture, as MPO and histones were still present within the treated NET structures (Fig. 6). To assess the function of a NET-bound protein, we measured elastase activity in PMA-treated neutrophils and showed no difference in elastase activity when NETs were treated with exogenous PTase or Mg2+ (S5 Fig.). NET structures remain intact and contain functional proteins (elastase) after treatment with PTase or excess Mg2+, but are no longer antibacterial for E. coli and P. aeruginosa (Fig. 5). Together these results suggest an antibacterial mechanism wherein the DNA backbone of NETs target and destabilize the bacterial membrane and promotes cell death.
Subinhibitory concentrations of extracellular DNA sequester Mg2+ and trigger the expression of multiple surface modifications that are known to protect the bacterial outer membrane from antimicrobial peptide (AP) damage and killing [23–25]. The arn operon (PA3552-PA3559) is required for the covalent addition of aminoarabinose to the phosphates of lipid A and the spermidine synthesis genes (PA4773-PA4774; speDE homologs) are required for production of the polycation spermidine on the outer surface [23, 24]. Both modifications substitute for divalent metal cations, mask the negative charges of the outer surface and thus contribute to AP resistance [23, 24, 29–31]. Given that these modifications stabilize the bacterial envelope, we sought to determine whether these surface modification pathways provided a more general mechanism to resist bacterial membrane damage. We noted that P. aeruginosa strains with mutations in the arn or spermidine biosynthetic pathways were significantly less capable of surviving exposure to DNA (Fig. 7A).
Given the role of these pathways for tolerance of exogenous DNA, we then investigated whether the DNA component of NETs induced expression of the arn or spermidine synthesis genes in P. aeruginosa. Expression of both pathways was strongly induced 2–6 fold following co-incubation with NETs produced by PMA-activated neutrophils (Fig. 7B). To confirm that DNA was the component of NETs that led to induction of the bacterial gene expression response, the addition of excess Mg2+ cations and enzymatic treatment with DNase (Fig. 7C) and PTase (S6 Fig.) all blocked the induction of the arn and spermidine operons response to NETs. While these treatments specifically neutralize DNA, we also considered the possibility that NET-bound antimicrobial proteins including histones or LL-37 may elicit these protective responses. We have previously shown that sub-MIC concentrations of antimicrobial peptides induce both outer surface modifications [29]. Therefore, to assess the relative capacity of each NET component to act as a bacterial signal, we compared the ability of purified histones, the well-characterized APs polymyxin B and colistin, and DNA to induce the expression of the spermidine synthesis pathway. Although all NET components induced expression of the PA4773-PA4774 spermidine synthesis pathway (S7 Fig.), DNA was the most potent inducer of this bacterial response (S7 Fig.).
The upregulation of protective outer membrane modifications (aminoarabinose-modified LPS and surface spermidine production) by NETs, and by the individual NET components of DNA and histones, suggests that these modifications are required to defend the membrane against assault from multiple innate immune components that target the bacterial membrane. Both modifications result in stable substitutions for divalent metal cations in the outer membrane, and protect P. aeruginosa from antimicrobial peptides [23, 24, 29–31] and DNA killing (Fig. 7A). P. aeruginosa mutants in the arn and spermidine synthesis genes also exhibited increased susceptibility to the disruptive effects of NETs (Fig. 7D). We next compared the susceptibilities of P. aeruginosa, E. coli and S. aureus to histone and DNA killing. Surprisingly, P. aeruginosa was the most susceptible to DNA killing, while S. aureus was the most tolerant, the opposite pattern of NET susceptibility (Figs. 3A and 8A). The bactericidal capacity of purified histones was modest, where after 2 hours exposure P. aeruginosa was the more histone tolerant (Fig. 8B). Taken together, these results highlight the ability of P. aeruginosa to quickly respond and defend against the DNA and histone mediated-antibacterial effects of NETs by stabilizing the outer membrane.
Given the observations that exogenous or secreted microbial DNases protect bacteria against NET killing and that extracellular DNA has rapid, membrane-damaging antibacterial activity, we sought to test the hypothesis that the DNA backbone of NETs contributes to their bactericidal function. Extracellular DNA possessed contact-mediated antibacterial activity that could be neutralized by enzymatic and cationic treatments that degrade or quench the capacity of DNA to chelate cations (Fig. 4 and S3 Fig.). NETs exposed to the same treatments that target the DNA scaffold were unable to cause bacterial membrane damage or to cause bacterial killing of P. aeruginosa and E. coli (Fig. 5). Therefore, we propose a novel bactericidal mechanism of NETs whereby the removal of surface-stabilizing cations by the DNA phosphodiester backbone results in bacterial lysis (Fig. 4B and C). To address the controversy surrounding the bacterial killing ability of NETs [5–7], we used multiple viability assays to measure NET killing and membrane damage, which included direct bacterial counts, luminescence viability assays and flow cytometry of PI-stained cells [32]. Taken together, these data support the general notion that NETs are directly antimicrobial.
Deciphering the specific antimicrobial mechanisms of NETs has been limited to a few candidate proteins [1, 8–11]. Most studies have focused on the NET-bound proteins as the antimicrobial components, given their important role during phagocytosis and degranulation. Although granular, cytoplasmic and nuclear proteins derived from neutrophils can be detected in NETs by immunofluorescence, the abundance of most NET-bound proteins is low (<1–6%), relative to histones, which comprise 65% of the total protein content [10]. Although classically characterized as chromatin structural proteins, NET-bound histones possess antimicrobial activity that can be neutralized through the addition of anti-histone antibodies [1, 12, 13]. However, other neutrophil proteins exhibit altered or reduced enzymatic activity when enmeshed in the NET backbone raising questions as to their antimicrobial capacity. For example, S. aureus tolerates myeloperoxidase in NETs unless supplemented with the addition of exogenous H2O2 [11]. Neutrophil elastase activity increases in DNase treated sputum from Cystic Fibrosis patients, suggesting that DNA may inhibit elastase activity, thus limiting the role of elastase as a potential NET-bound factor [17].
We observed that both the antibacterial activity of NETs and the ability to induce protective bacterial responses were blocked by treatments that target extracellular DNA, suggesting that the NET scaffold is not simply a passive structure (Figs. 5 and 7) [23–25, 31]. The most potent inducing triggers of the P. aeruginosa surface modifications are purified eDNA, followed by APs and purified histones (S7 Fig.). Since being widely induced by these components, it is not surprising that the aminoarabinose-modified LPS and spermidine synthesis pathways protect the outer membrane from DNA, NETs (Fig. 7) and antimicrobial proteins [23, 24, 29–31]. The low potency of the tested histones may be explained by the fact that our assays were performed with a mixture of full-length histones, which had modest antibacterial activity (Fig. 8). Recent evidence highlights that histones are proteolytically processed by proteases such as elastase during the process of nucleus decondensation, prior to NET release [33] It is therefore likely that potent bactericidal histone-derived peptides are present in NETs as an important antibacterial component of NETs.
P. aeruginosa appears to mount a multifunctional, outer membrane defense strategy to combat multiple antimicrobial components enmeshed in NETs. Consistent with this model, we noticed unexpected susceptibility patterns that also suggest that NET killing may be the result of the combinatorial effect of DNA and NET-bound proteins. The observation that S. aureus is susceptible to NET killing but tolerant to DNA (Figs. 3 and 8), suggests that NET killing of S. aureus is likely dependent on other anti-staphylococcal proteins enmeshed in the DNA lattice. The DNA susceptibility phenotype of P. aeruginosa may explain the potency of DNA as the strongest inducer of the protective outer surface modifications that contribute to the observed NET tolerance. It is intriguing to speculate that the NET-bound, antimicrobial proteins act in concert with the antibacterial activity of DNA to provide broad-spectrum protection against a wide range of microbial pathogens.
Modification of the bacterial cell surface and the production of secreted DNases are virulence strategies utilized by microbial pathogens to evade NET killing [19–22]. Here we report that the spermidine and the arn surface modification pathways are required to tolerate the antibacterial action of both DNA and NETs (Fig. 7). The covalent addition of aminoarabinose to the lipid A component of LPS masks the negative charges of core LPS phosphates, and the polycationic nature of spermidine (+3 charge) substitutes for surface divalent metal cations, and may also bind and neutralize DNA. In addition, spermidine possesses an antioxidant activity that protects bacterial membrane lipids from oxidative damage [24] and therefore may protect P. aeruginosa from NET-induced oxidative damage [11]. Combined, these results suggest that the spermidine and arn surface modifications possess multiple protective roles that may contribute to resisting a broad range of antimicrobial components present within NETs.
We propose that bacterial surface-bound, divalent metal cations are displaced by direct contact with extracellular DNA, and that DNA-induced surface modifications prevent outer membrane disruption and bacterial killing by NETs. Therefore, the antibacterial mechanism of cation chelation exerted by DNA is distinct from that of other previously characterized antimicrobial cation chelating proteins such as calprotectin. We have previously demonstrated that DNA chelates diverse metal cations (Mg2+, Ca2+, Zn2+, Mn2+) [23] while calprotectin chelates zinc and manganese [34]. Additionally, the antimicrobial function of calprotectin is contact-independent whereas the bactericidal function of DNA requires contact (S4 Fig.). Further, sequestration of zinc and manganese by calprotectin does not target the microbial membrane but rather sequesters cation cofactors required by bacterial enzymes such as superoxide dismutase, which protects bacteria from superoxide [34].
In summary, we have identified that the DNA backbone is a bona fide antibacterial component of neutrophil extracellular traps. The DNA scaffold structure also acts as a warning signal perceived by P. aeruginosa. Overall, these results support a model where the membrane-destabilizing activity of the DNA scaffold contributes to the bactericidal capacity of NETs, while the cation chelating activity acts as a signal perceived by P. aeruginosa that leads to upregulation of protective surface modifications. These results highlight a dynamic bacterial-host interaction between an opportunistic pathogen that causes chronic infections in the lungs of individuals with Cystic Fibrosis, an infection site known to be rich in neutrophil DNA and neutrophil extracellular traps [16, 17]. This ability to sense and defend against NETosis may help explain the long-term persistence of P. aeruginosa in CF lung infections.
All strains and plasmids used in this study are shown in S1 Table. Bacterial cultures were routinely grown at 37°C in LB or BM2 defined minimal media with 0.5 mM MgSO4, unless otherwise stated. S. aureus was grown overnight in BHI media. When necessary, the following antibiotics were used: 50 µg/mL tetracycline for P. aeruginosa mini-Tn 5-lux mutants, and 50 µg/mL kanamycin for E. coli DH5α/ pσ70-lux. Mid-log cultures were used for co-incubation experiments with neutrophils or extracellular DNA.
Neutrophils were isolated from healthy donors as previously described [35]. Whole blood was collected and mixed 5:1 in acid citrate dextrose, followed by removal of red blood cells using dextran sedimentation and hypotonic lysis with KCl. After all red blood cells were lysed, the cell pellet was subjected to Ficol-Histopaque density centrifugation. The subsequent pellet was resuspended in 2 mL of HBSS (Hank’s balanced salt solution, no cations; Invitrogen 14175–095). The viable cell concentration was determined using a haemocytometer and Trypan blue staining.
Glass cover slips were HAS-coated and placed in 6-well tissue culture plates. Neutrophils were added at 2.0×106 cells/mL per well, adhered (30 min, 5% CO2, 37°C) and treated with cytochalasin D (10 µg/well) and PMA (25 nM) to activate NETosis [35]. Mid-log bacterial cultures were diluted in HBSS (no cations) (5.0x107 CFU/mL) for an MOI of 25:1, centrifuged to the neutrophils, and coincubated for 1–4 hours (5% CO2, 37°C). Cells on cover slips were fixed with 4% paraformaldehyde, washed with 250 µL of 10% FBS (Invitrogen) in PBS and stained with either DNA dyes and/or various primary and secondary antibodies (described below).
For NET visualization with antibodies, the primary anti-human MPO antibody (DakoCytomation- A0398) was diluted into 10% FBS in PBS (1/500). 30 µL was added to adhered neutrophils, incubated (30 min, 37°C) and washed twice with sterile PBS. 40 µL of the secondary anti-rabbit Cy 5 antibody (Jackson ImmunoResearch 60354) (1/500 dilution) was added. After 15 min incubation in the dark, cover slips were washed twice with PBS, and prepared with mounting media. Anti-DNA and anti-histone antibodies where obtained from Dr. Marvin Fritzler. Either the anti-DNA (1:10) or anti-histone (1:500) antibodies [36, 37] were added as described above. The anti-human secondary antibodies with Alexa Flour 647 (Invitrogen A21445, 1/500) were added to cover slips and mounted as described above. Images of human NETs were acquired using the Leica DMI 4000B inverted microscope equipped with ORCA R2 digital camera and Metamorph software for image acquisition using the 63X or 100X objectives. The following excitation and emission filters were used for blue fluorescence (Ex 390/40; Em 455/50), red fluorescence (Ex 555/25; Em 605/52), far red fluorescence (Ex 645/30; Em 705/72) and green fluorescence (Ex 490/20; Em 525/36). Images were formatted and analyzed using the Imaris 7.0.0 imaging software. All images shown are representative of at least three experiments.
Mice were anaesthetized (10 mg/kg xylazine hydrochloride and 200 mg/kg ketamine hydrochloride) and body temperature was maintained using a rectal probe and heating pad. The mice were pretreated with intradermal MIP-2 (0.2µg/injection) diluted in sterile normal saline 30 minutes prior to imaging. The right jugular vein was cannulated to administer additional anesthetic and fluorescent dyes. The microcirculation of the dorsal skin was prepared for microscopy as previously described [26]. Briefly, after shaving the mouse’s back, a midline dorsal incision was made extending from the tail region up to the level of the occiput. The skin was separated from the underlying tissue, remaining attached laterally to ensure the blood supply remained intact. The area of skin was then extended over a viewing pedestal and secured along the edges using 5.0 sutures. The loose connective tissue lying on top of the dermal microvasculature was carefully removed by dissection under an operating microscope. The exposed dermal microvasculature was immersed in isotonic saline and covered with a coverslip held in place with vacuum grease. Alexa Fluor 649 conjugated anti-mouse GR-1 antibody (10µl per mouse i.v.; eBioscience) was used visualization of neutrophils. To visualize NETs in vivo the membrane impermeable dyes SYTOX-green or SYTOX-orange were administered (diluted 1:1000 with sterile saline, 100µl per mouse i.v.). MIP-2 injection (0.2µg/injection) was initiated 30 min prior to Pseudomonas aeruginosa or S. aureus administration. Following baseline visualization, all bacteria were directly administered into the field of view using a tuberculin needle (1×108 CFU/100µl of sterile saline, i.d.).
Spinning disk confocal intravital microscopy was performed using an Olympus BX51WI (Olympus, Center Valley, PA) upright microscope equipped with a 20×/0.95 XLUM Plan Fl water immersion objective. The microscope was equipped with a confocal light path (WaveFx, Quorum, Guelph, ON) based on a modified Yokogawa CSU-10 head (Yokogawa Electric Corporation, Tokyo, Japan). Laser excitation at 488, 561 and 649nm (Cobalt, Stockholm, Sweden), was used in rapid succession and fluorescence in green, red and blue channels was visualized with the appropriate long pass filters (Semrock, Rochester, NY). Exposure time for all wavelengths was between 500 and 600ms. Sensitivity settings were maintained at the same level for all experiments. A 512×512 pixels back-thinned EMCCD camera (C9100–13, Hamamatsu, Bridgewater, NJ) was used for fluorescence detection. Volocity Acquisition software (Improvision Inc., Lexington, MA) was used to drive the confocal microscope. Images captured using the spinning disk were processed and analyzed in Volocity 6.0.1. NET area and NET number were quantified using the Volocity software.
NET area was determined using Volocity imaging software. Briefly, in each field of view (FOV) the threshold of the corresponding fluorescent channel in which NET structures were stained was set to eliminate the background staining of the skin. The area and number of NET positive structures in the FOV was calculated and counted via the Volocity 6.0.1 software. Structures that showed no characteristic NET-like shape and resembled the staining of a nucleus of an obvious dead cell in the FOV were excluded from the quantification manually. NET image analysis was performed in at least two infected or uninfected animals and from 5 fields of view.
NETs were quantitated by measuring the amount of extracellular DNA that stains with the cell impermeant dye Styox Green [35]. Cell culture media (CCM) consisted of 48.5 mL of RPMI 1640 (Invitrogen), 0.5 mL of 1.0 M HEPES and 1.0 mL of human serum albumin (HAS; Innovative Research). Neutrophils were diluted into CCM (2.0×105 cell/well) and added to an HSA-coated, 96-well black, clear-bottom plate (Thermo Scientific). As a positive control, PMA (25 nM/well) was added to activate the neutrophils [35]. For bacterial activation of NETs, mid-log bacterial cultures were diluted in CCM (2.0×106 CFU/well) for a multiplicity of infection (MOI) of 10:1 (bacteria to neutrophils). DNase (430 kU/well; VWR 31149) was added to degrade extracellular DNA in NETs. Bacteria were gently centrifuged onto the adhered neutrophils (800xg, 10 min). Sytox green (2.5 µM; Invitrogen) was added to each well and green fluorescence (Ex 490/8; Em 535/25) was measured with Perkin Elmer 1420 Multilabel Counter Victor3 between 1 and 4 hours. All values shown are the mean from at least three individual replicates and each experiment was performed at least 3 times. We noticed variation in the background level of NET staining between individual donors, but there was a reproducible and robust 2.5 to 10-fold increase in NETosis after PMA treatment of neutrophils from various donors (S2 Fig.).
Quantification of Bacterial Viability and Gene Expression Using Plate Counts and Luminescence NET killing was examined using direct plate counting methods where a reduction in cell number indicated bacterial killing. All NET killing experiments were performed in HBSS solution lacking divalent cations. Isolated human neutrophils were mixed with mid-log bacteria in HBSS (no cations) in black, clear-bottom 96-well plates with of 2.0 × 107 CFU bacteria and 2.0 × 106 neutrophils (MOI 10:1). After a 1–4 hr incubation, 50 µL of DNase I solution (430 kU/mL) was added to every well, mixed, and incubated for 30 min at 37°C, in order to release bacteria trapped in NETs for accurate plate counts. 15 µL of suspension was serially diluted (1/10) in 0.9% NaCl solution in a sterile 96-well plate and 5 µL from each well was stamped onto LB agar plates to obtain bacterial plate count data for time zero (T0) and after 4h (T4). CFU/ml values from T4 and T0 time points were used to calculate the percentage survival by subtracting the T4—T0 plate counts and dividing the ‘bacteria and neutrophil’ conditions by the ‘bacteria alone’ conditions and multiplying by 100. For lux viability and gene expression assays [23], bacteria were centrifuged onto the adhered neutrophils and placed in the Victor3 plate reader for luminescence (CPS) measurements every 20 minutes for 3–4 hours. All values shown are the mean from at least six individual replicates and each experiment was performed at least 3 times.
SYTO9 stains the DNA in all cells and propidium iodide (PI) stains the DNA in dead cells and cells with damaged membranes [28, 32]. The sample of bacteria and neutrophils (~200 µL) was placed in 5 mL polystyrene round-bottom sample tubes and stained with SYTO9 and propidium iodide at final concentrations of 0.02 mM and 0.2 mM, respectively. The tubes were centrifuged at 300x gravity and incubated (RT, 15 min). Bacterial cells were analyzed using the BD LSRII flow cytometer (BD Bioscience, San Jose, USA) equipped with a blue laser (488nm) and a green laser (532nm). Unstained, mid-log bacterial cells were used to gate the forward scatter (FSC) and size scatter (SSC) parameters. For green and red fluorescence profiles, SYTO9 was excited by blue (488nm) laser with emission filters 525/50BP and 505LP and PI was excited using the green (532 nm) laser with emission filters 610/20BP and 600LP. All detectors were set to the logarithmic amplification with the following voltages, 500, 240, 596, and 489 and threshold was set at 200 for both FSC and SSC. For each sample, 50 000 events were acquired using the BD FACSDiva software 6.1.3. The Hierarchical gating strategy was used to determine double positive population of bacterial cells (stained with both SYTO9 and PI) where gate P1 is the total population of FSC and SSC gated events, as determined from bacteria alone control and then applied to all other samples. P2 is the population of events stained by SYTO9 and P3 is the population stained with both SYTO9 and PI. Neutrophils and mid-log bacteria controls do not contribute any autofluorescence or PI-stained events when stained with either or both of the SYTO9/PI dyes. Values displayed in each density plot represent the percentage of 50 000 cells (N value) in each quadrant gate and each experiment was performed with at least 5 times.
During the coincubation experiments of bacteria and PMA-activated neutrophils, exogenous Mg2+ was added at a final concentration of 5 mM MgSO4. For the enzyme treatments, deoxyribonuclease (DNase I, VWR) was added at a final concentration of 430 kU/well and calf intestinal alkaline phosphatase (PTase, Invitrogen) at a final concentration of 16.6 U/well. Maximum enzyme amounts were added to bacterial-neutrophil mixtures that had no effect on bacterial viability and without the addition of enzyme buffers. The killing experiments were incubated for up to 4 hours in the 5% CO2 incubator at 37°C. All % survival values shown are the mean from at least three individual replicates and each experiment was performed at least 3 times.
P. aeruginosa was grown to mid-log in LB medium (OD600 = 0.2–0.4), washed and resuspended in 10 mM Tris buffer (pH 7.4; 1.0 × 107 CFU/well). Cells were incubated with fish sperm DNA (0.125%, w/v; USB) or with DNA that had been pretreated with exogenous DNase I (150 kU/well), PTase (50 U/well) or 5 mM MgSO4. DNA was pretreated for up to 3 hrs at 37°C in order to neutralize the antimicrobial activity. To determine if DNA killing required direct cell contact, 2% w/v fish sperm DNA (USB) was resuspended in 10 mM Tris pH 7.4 was placed in sealed dialysis membranes (MW cutoff 3500 Da) and allowed to dialyze into 10 mM Tris pH 7.4 for 4 hours, exchanging the buffer every hour. Cells from mid-log P. aeruginosa PAO1 cultures (1 × 107 CFU) were washed into 10 mM Tris pH 7.4 and coincubated directly with 1% or 0.125% dialyzed DNA (final concentration), with 1% eDNA maintained inside dialysis tubing, or 10 mM Tris pH 7.4 alone as a negative control. For histone killing experiments, 1 × 107 CFU mid-log growth phase P. aeruginosa PAO1, E. coli and S. aureus were washed into 10 mM Tris pH 7.4 and subsequently coincubated directly with 1.5 µg/mL calf thymus histones (Roche). Killing experiments were performed at RT in 96-well microplates and bacterial survival was assessed by colony counts (CFU/ml) every hour. All survival values shown are the mean from 4–8 individual replicates and each experiment was performed at least 3 times. Differences in bacterial survival were statistically analyzed by two-tailed student t-test.
PAO1::OM-lipoChFP was used as an indicator for outer membrane damage. This strain of PAO1 expresses a synthetic Cherry fluorescent lipoprotein (CSFPOmlA-ChFP) anchored to the outer membrane encoded on plasmid pCHAP6656 [38]. PAO1::OM-lipoChFP was exposed to a lethal concentration of 2% w/v (20 mg/ml) extracellular sperm DNA (USB) or 2 mM EDTA and red fluorescence of untreated and DNA killed cells were monitored as described above. Fluorescent outer membrane vesicles (OMVs) were counted in 6 fields of view by ImageJ quantification using a manually controlled threshold cutoff.
Overnight cultures were grown in LB medium, diluted 1/100 (approximately 1 × 107 CFU) into 100 µl of HBSS medium lacking cations (Life Technologies) in 96-well black plates with a transparent bottom (Thermo Scientific) and overlaid with 75 µl of mineral oil (Sigma Aldrich) to prevent evaporation. Microplate planktonic cultures were incubated at 37°C in a Wallac Victor3 luminescence plate reader (Perkin-Elmer) and optical density (growth, OD600) and luminescence (gene expression, CPS) readings were taken every 20 minutes in the presence of 0.2% salmon sperm DNA, 0.125 µg/mL polymyxin B and colistin, and 0.1 µg/mL calf thymus histones (Roche). Mean gene expression was derived from triplicate samples at 180 minutes after initial dilution and error bars represent the standard deviation from 4 individual replicates. Differences were statistically assessed by two-tailed student t-test.
Overnight cultures of S. aureus and E. coli were grown in BHI medium and P. aeruginosa in BM2 medium, normalized to an OD600 = 1 and supernatants were collected by centrifugation at 8000 rpm for 3 minutes. 15 µL of supernatant was incubated with 5 µg of P. aeruginosa genomic DNA for 1 h at 37°C. Pseudomonas aeruginosa genomic DNA was purified using the Wizard Genomic DNA purification kit (Promega). DNA degradation was visualized on red safe (FroggaBio) stained 1% agarose gels. To test whether exposure to NETs induced DNase production, supernatants from S. aureus, E. coli and P. aeruginosa incubated in HBSS lacking cations with 106 PMA-stimulated human neutrophils (MOI 10:1, same method as described in the NET killing experiments section) were collected by centrifugation at 8000 rpm for 3 minutes. 100 µL of the supernatants were then coincubated at 37°C with 5 µg salmon sperm DNA stained with 2.5 µM Sytox green. 90 kU of DNase I was included as a positive control. Reactions were placed in 96-well black plates with a transparent bottom and Sytox green fluorescence quantified after 1 hour in a Wallac Victor3 luminescence plate reader.
To determine whether phosphatase treatment or the presence of excess Mg2+ altered NET-bound protein function, 2 × 105 human neutrophils were seeded in 96-well black plates with a transparent bottom and induced with 100 nM PMA. Immediately after PMA addition, 50 units of phosphatase (CIAP) and 5 mM MgSO4 were added to wells. The plate was then placed in cell culture conditions for 2 hours (37°C, 5% CO2). 300 µM elastase substrate I was added to all wells (Calbiochem), which were subsequently overlaid with 75 µL of mineral oil. The plate was then placed in a Wallac Victor3 luminescence plate reader at 37°C. Neutrophil elastase activity was monitored by measuring absorbance at OD410 nm every 20 minutes for 8 hours.
Statistical analysis was performed using GraphPad Prism v4.0 software. One-way ANOVA with Bonferroni posts tests and two-tailed students t-tests were used to calculate significant differences for plate counts, luminescence and flow cytometry analyses. Significant differences refer to P< 0.05 or less, or as otherwise denoted.
Human neutrophils were isolated from human blood samples with ethical approval by the University of Calgary Research Ethics Committee (Ethics ID# 23187), where all subjects provided written informed consent. All animal protocols were approved by the animal care committee of the University of Calgary under the protocol number AC12–0222. All protocols used were in accordance with the guidelines drafted by the University of Calgary Animal Care Committee and the Canadian Council on the Use of Laboratory Animals.
|
10.1371/journal.pntd.0006652 | The impact and cost-effectiveness of controlling cholera through the use of oral cholera vaccines in urban Bangladesh: A disease modeling and economic analysis | Cholera remains an important public health problem in major cities in Bangladesh, especially in slum areas. In response to growing interest among local policymakers to control this disease, this study estimated the impact and cost-effectiveness of preventive cholera vaccination over a ten-year period in a high-risk slum population in Dhaka to inform decisions about the use of oral cholera vaccines as a key tool in reducing cholera risk in such populations.
Assuming use of a two-dose killed whole-cell oral cholera vaccine to be produced locally, the number of cholera cases and deaths averted was estimated for three target group options (1–4 year olds, 1–14 year olds, and all persons 1+), using cholera incidence data from Dhaka, estimates of vaccination coverage rates from the literature, and a dynamic model of cholera transmission based on data from Matlab, which incorporates herd effects. Local estimates of vaccination costs minus savings in treatment costs, were used to obtain incremental cost-effectiveness ratios for one- and ten-dose vial sizes. Vaccinating 1–14 year olds every three years, combined with annual routine vaccination of children, would be the most cost-effective strategy, reducing incidence in this population by 45% (assuming 10% annual migration), and costing was $823 (2015 USD) for single dose vials and $591 (2015 USD) for ten-dose vials per disability-adjusted life year (DALY) averted. Vaccinating all ages one year and above would reduce incidence by >90%, but would be 50% less cost-effective ($894–1,234/DALY averted). Limiting vaccination to 1–4 year olds would be the least cost-effective strategy (preventing only 7% of cases and costing $1,276-$1,731/DALY averted), due to the limited herd effects of vaccinating this small population and the lower vaccine efficacy in this age group.
Providing cholera vaccine to slum populations in Dhaka through periodic vaccination campaigns would significantly reduce cholera incidence and inequities, and be especially cost-effective if all 1–14 year olds are targeted.
| While oral cholera vaccines are increasingly being used in the past few years, mainly to curtail or preempt cholera outbreaks, they have yet to be used on a large scale to control endemic cholera in a high-burden country like Bangladesh. This study examines the potential impact on disease and value of vaccinating slum dwellers in Dhaka (and by extension in other cities), who are among those at highest-risk of getting the disease. This analysis suggests that, despite the modest efficacy and limited duration of protection of existing vaccines, mass cholera vaccination can have a significant impact on reducing cholera incidence in the entire population–including those not vaccinated–as a result of herd effects–and can be a cost-effective means of controlling the disease, especially until more long-term measures, such as improved water and sanitation infrastructure, are put in place. These results should assist policymakers and potential donors in determining whether and how to use these vaccines in Bangladesh to control the disease amongst its most vulnerable populations.
| The Ganges River Delta and Bay of Bengal, including Bangladesh, are considered the birthplace of cholera and the origin of six of the seven cholera pandemics recorded in modern times[1]. While national population-based estimates of cholera incidence are lacking in Bangladesh, the perception among local experts is that cholera is increasingly becoming an urban disease. Indeed, based on long-term systematic laboratory testing of 2% of all diarrheal patients presenting at the icddr,b (International Centre for Diarrhoeal Disease Research, Bangladesh) hospital in Dhaka (locally known as the “cholera hospital”), this hospital provides care and treatment to approximately 140,000 patients of all ages in each year[2]. Dhaka has also experienced several large cholera outbreaks in the past two decades, especially during widespread floods. During major floods in 2004, 2007 and 2009, icddr,b saw an estimated 30,000 or more cholera cases annually and V. cholerae overtook rotavirus and ETEC as the main pathogen found among patients with severe diarrhea presenting at the hospital[3]. Several cholera outbreaks have also recently been documented in urban areas in other parts of the country[4, 5].
Residents of slums and low-income districts are especially vulnerable to cholera infection. A systematic sampling of every third diarrheal patient coming to the icddr,b hospital from the low-income area of Mirpur in Dhaka City found V. cholerae to be the most common pathogen isolated–accounting for 23% of cases, 70% of whom were severely dehydrated [6]. Overcrowded living conditions, inadequate sanitation, and overstressed water systems that are not keeping up with population growth are key contributors to high cholera incidence in urban slums, with tap water supplies often found to be the source. These water supplies, even if initially treated with chlorination, become contaminated during distribution due to illegal connections, leaky pipes and low or negative water pressure–resulting in the mixing of sewerage and drinking water and a dilution in chlorine levels [4, 5]. Contamination of water at the household level, due to poor hygiene, is also common.
The Government of Bangladesh has increasingly expressed interest in controlling cholera since the late 2000s. The Bangladesh delegation to the Executive Board of the World Health Organization (WHO) played a key role in the development and subsequent passage of a resolution by the World Health Assembly in 2011 calling for member states and WHO to strengthen efforts to prevent and control cholera through a series of measures, including the use of oral cholera vaccines (OCVs) “where appropriate, in conjunction with other recommended prevention and control methods and not as a substitute for such methods” [7].
That same year, the Bangladesh Ministry of Health and Family Welfare (MOHFW) played an active role in planning, implementing and monitoring a mass cholera vaccination feasibility study implemented by icddr,b in the Mirpur area of Dhaka, in which more than 123,000 persons one year and above received two doses of the bivalent, whole-cell killed oral cholera vaccine, Shanchol (produced in India), either alone or in combination with the promotion of hand washing and point-of-use safe water treatment interventions [8–10]. The vaccine–administered in two doses at least two weeks apart and licensed for use in persons one year and older–was shown in a clinical trial in Kolkata, India to have an overall efficacy of 65% lasting at least five years [11] and has been used through a global vaccine stockpile to preempt or respond to cholera outbreaks in ten countries from 2013 to October 2016. The Mirpur study found mass cholera vaccination in this impoverished, high-risk area to be feasible–with an overall estimated coverage rate for two doses of 72%, including 67% in adults 18 and older–and generally well accepted by the population [10].
The interest in cholera vaccination among local policymakers has been enhanced by the technology transfer and clinical development of a vaccine identical to Shanchol by a Bangladeshi private sector producer, Incepta Vaccine Ltd. The vaccine, to be marketed as Cholvax, is anticipated to be licensed by the end of 2018 and to cost less than Shanchol (which has a public sector price of $1.85 per dose for single-dose vials).
Analyses of the impact and cost-effectiveness of introducing a new vaccine are increasingly being conducted by countries and donors–most prominently the Gavi Alliance–to inform decisions about implementing or supporting vaccine introductions and which vaccination or targeting strategies to use. Such analyses using the TRIVAC cost-effectiveness model developed by the Pan American Health Organization (PAHO) have reportedly played an important role in decisions by a number of countries in the Latin America and Caribbean region to introduce Haemophilus influenzae type b (Hib), pneumococcal conjugate and rotavirus vaccines, as well as more recently in nine countries in Europe, Africa and the Middle East in deciding whether or not to introduce rotavirus or pneumococcal vaccines [12, 13].
Impact and cost-effectiveness analyses can especially be important in the case of a vaccine against a disease like cholera, in which the risk of getting the disease varies greatly by location (due to differing water and sanitation conditions) and by age group. Such analyses can therefore help policymakers make evidence-based decisions on whether or not to use oral cholera vaccines and which geographic areas and age groups to target in order to have the greatest impact for the lowest possible cost.
The purpose of this study was to estimate the impact, cost and cost-effectiveness of preventive cholera vaccination over a ten-year period in a high-risk population of slum dwellers in the city of Dhaka, Bangladesh in order to assist policymakers and global partners in determining whether OCVs should be used in such populations as one of the tools to reduce the cholera risk, and if so, which vaccination strategies would be most effective and efficient. To enhance the relevance and credibility of the results, this study uses local data to model the effectiveness of vaccination on cholera transmission, as well as for several key epidemiological and economic variables.
This protocol has been approved by the Research Review Committee (RRC) and Ethical Review Committee (ERC) at the icddr,b. All data used for this component was anonymised. Moreover, no active data has been collected for this component of the project.
The assumptions and data used for the major parameters for the analyses are shown in Table 1. Additional parameters for the effectiveness modeling are shown in Table A1 in the technical appendix (S1 Appendix).
To identify areas in Dhaka at high risk of cholera, we analyzed data from the laboratory surveillance of 2% of hospitalized diarrhea patients who visited the main icddr,b hospital in Dhaka and 10% of patients who came to the icddr,b treatment center in Mirpur from 2011 to 2015. The data include place of residence. Average annual cholera incidence rates by the sub-districts of Dhaka (known as thanas) were then estimated, using thana-level data from the 2011 census for the denominator. A threshold incidence rate of 1.5 cases per 1,000 per year was used to select high-incidence thanas. Fourteen of the city’s 43 thanas were found to meet this criteria, with an estimated 2015 population of around 3.5 million (supplemental S1 Table). Not all hospitalized cases of cholera in Dhaka are treated at the two icddr,b facilities, and thus this method may underestimate the number of high-incidence thanas, as well as their incidence. However, Mirpur is amongst the areas from where the highest numbers of cholera patients seek treatment at the icddr,b hospitals [6].
To further target those at the highest risk of cholera, the study selected for vaccination slum populations within the 14 high-risk thanas, under the assumption that the vast majority of hospitalized cholera cases come from slum areas. The slum population is assumed to make up around 40% of the total population of these thanas (or around 1,383,400 people), based on an estimate from the Center for Urban Studies [14].
The study assumes the use of the bivalent whole-cell oral cholera vaccine to be produced locally (Cholvax). The five-year, age-specific vaccine efficacy rates used for two doses of the vaccine are from the Kolkata clinical trial of Shanchol (42% for 1–4 year olds, 68% for 5–14 year olds, and 74% for persons 15 years and older) [11].
The analysis modeled the impact and cost-effectiveness of mass cholera vaccination campaigns for three increasingly large target age groups: 1–4 year olds, 1–14 year olds, and all persons one year and older. These targeting options were selected based on interviews conducted with MOHFW officials and other stakeholders. Although the vaccine has been shown to provide protection for five years, at least for children over five years and adults, the campaigns–to be conducted in two rounds (one for each dose)–are proposed to take place every three years over the ten-year period of the analysis. This is to account for population mobility in and out of the slum areas–which has the effect of reducing the population’s vaccination coverage over time–as well as the vaccine’s relatively low efficacy rates in the youngest (1–4 year) age group. All three targeting strategies also include the annual vaccination of new birth cohorts through the routine immunization program to protect them during non-campaign years. The first dose can be provided concurrently with the second dose of measles-containing vaccine scheduled at 15 months of age.
We assume a vaccination coverage rate for the two-dose series of 70% for children 1–14 years of age (for both the routine infant vaccination and campaigns) and 55% for persons 15 and above. These estimates are based on an average of coverage rates achieved in several OCV campaigns conducted in different countries in recent years, including the Mirpur feasibility study mentioned above (see S2 Table in supplement).
A case of cholera is defined in this study as one suffering from acute watery diarrhea requiring a visit to a treatment setting. The estimated average annual incidence rate of cholera requiring treatment in the target slum population is 2.3 per 1,000. This rate was derived by applying the proportion of diarrheal cases that were found to be confirmed cholera cases through the systematic testing of patients at the icddr,b Dhaka hospital and Mirpur treatment center to the total number of patients seeking care at the hospital for severe diarrhea[6]. Although the incidence of reported cholera varies by thana, we assume that all high-risk populations in Dhaka have the same high cholera incidence rate and therefore use the same rate for the entire target population. Under the assumption that nearly all cholera cases coming to the icddr,b Dhaka hospital are from slum areas, the denominator was the estimated size of the slum population in Dhaka, based on government population data and the Center for Urban Studies estimate of the percentage of Dhaka residents who live in slums (40%) [14]. Age-specific incidence rates were derived by applying the age distribution of cholera cases found through on-going laboratory-supported cholera surveillance in Matlab from 1997 to 2001 to the overall incidence of 2.3/1,000 per year. The estimated incidence rates are 7.86 per 1,000 for children under five years of age, 2.65/1,000 for 5–14 year olds, and 1.38/1,000 for persons fifteen and older. Thus, using these estimates, pre-school children are 5.7 times more susceptible and school-aged children nearly twice as susceptible of becoming infected with cholera requiring treatment as adults in this population. This increased susceptibility of children might actually reflect increased exposure to cholera due to age-related behavioral differences, but the actual biological mechanism behind this does not affect the model.
In the absence of data on cholera case fatality rates, an estimated rate of 1.5% was used, based on the opinion of experts at icddr,b. Estimates of the average duration of illness and duration of infection come from the literature [15, 16].
To estimate the impact of different vaccination targeting strategies on the incidence of cholera in this population over a ten-year period, we used a mathematical model that simulates the dynamics of cholera transmission. This model is based on a previously-published model of cholera transmission in Matlab that used times-series data of cholera incidence and other epidemiological data from Matlab from 1997 to 2001 [17]. Details on the model, including all parameters and differential equations used, are given in the technical appendix (S1 Appendix).
The model simulates how a person can be infected by another individual–either symptomatic or asymptomatic–or from the environment (e.g., via water) (Fig 1). It also simulates the effect of immunity on disease transmission from having been infected (natural immunity) or having been vaccinated. In brief, the model places people in one of four compartments: 1) susceptible to cholera, 2) infected and symptomatic, 3) infected but asymptomatic, and 4) recovered and immune. The concentration of V. cholerae in the environment (water) is tracked in another compartment. Ordinary differential equations are used to model the transition of people between compartments over time, which is affected by such factors as the number of infected persons in the community, the level of V. cholerae in the environment, the time it takes for an infected individual to clear the infection, and the efficacy of either natural or vaccine-induced immunity over time. The model was calibrated to simulate the seasonality of cholera in Matlab over a one-year period.
The model assumes four different levels of susceptibility to infection based on age groups (children under two years of age, 2–4 year olds, 5–14 year olds, and adults 15 and older). It also assumes that vaccine efficacy is based on the age at vaccination (1–4, 5–14 and 15+ years old), using the age-specific vaccine efficacy estimates from Bhattacharya et al. 2013 described above [11].
In the dynamic model of cholera transmission, the rate of cholera infection is proportional to the number of infected individuals and the amount of Vibrio in the environment. Therefore, vaccination not only reduces the number of cases among vaccinees, but even non-vaccinated individuals are protected indirectly because the averted cases among vaccinees reduce everyone’s risk of infection. The dynamic model was used to estimate the magnitude of this effect. Thus, the indirect (herd) protective effects of cholera vaccination are built into this dynamic model. The incorporation of herd effects, along with the simulation of the seasonal pattern of the disease, are meant to provide a more accurate picture of the impact of cholera vaccination on disease incidence over time than would a static outcomes model, in which vaccination does not reduce the incidence of cholera in the unvaccinated population so there are no herd effects.
In adjusting the model from Matlab to Dhaka, a migration factor was added to account for the high mobility of slum populations in cities like Dhaka. The model thus replaces a portion of each vaccinated population group with non-vaccinated individuals each year at a constant rate and assumes that these non-vaccinated newcomers have the same level of cholera susceptibility as the baseline (pre-vaccination) population. This has the effect of reducing vaccination coverage in the target population over time. Since little data on migration in and out of the slums of Dhaka are available, we modeled three annual migration rates: 0%, 10% and 25%, and used 10% as the base case for the main analyses. However, we assumed that the size of the at-risk population is fixed over the 10-year period of the analysis (i.e., as many people leave as enter the targeted areas).
The output of the model is the number of symptomatic cases of cholera per year, including those not treated, for each age group and each vaccination targeting strategy, varied by migration rates. We then translated the results to adjust for the age structure in Dhaka (which is somewhat different than that in Matlab), based on census data. We assume that most cholera illness is either mild or otherwise unreported, so we computed a "reporting rate" that, when multiplied by the age-adjusted number of symptomatic cases derived from 100 stochastic runs of the model using an unvaccinated population, produced 2.3 reported cases per 1,000 population per year (the estimated annual incidence of cholera requiring treatment in the target population, as described above). The analysis of cases averted is based on the "reported" number of cases produced by the model.
For each vaccination strategy modeled, we obtained the number of cases prevented each year and cumulatively over ten years by subtracting the number of cases predicted once vaccination is implemented from the expected number of cholera cases if no vaccination takes place. The number of deaths averted was calculated by multiplying the number of cases prevented by the assumed case fatality rate of 1.5%.
Measures of cost-effectiveness were obtained by dividing the net cost of vaccination over the ten-year period of the analysis by the cumulative number of cases, deaths and disability adjusted life years (DALYs) prevented as a result of vaccination for each of the targeting strategies. The net vaccination cost is the cost of the vaccination program minus the estimated savings in treatment costs resulting from a reduction in cholera incidence due to vaccination. The resulting incremental cost-effectiveness ratios (ICERs)–cost per case averted, cost per death averted, and cost per DALY averted–were calculated for two different vaccine presentations: single-dose vials and ten-dose vials.
As is standard, DALYs averted were calculated using DALY weights, a standard discount weight, and life expectancies using methods described in a paper by Fox-Rushby and Hanson [18]. No age weights were used in the analysis.
This economic analysis takes a health provider perspective, as opposed to a societal perspective. The vaccination costs are assumed to be paid by the public sector (government and/or donors) and only the cost of treating cholera paid by the health care provider are included. Thus, the costs of cholera illness borne by individuals, such as out-of-pocket expenses for medicines, transport and lodging for caregivers and the indirect costs of loss wages of patients or their caregivers from missing work–are not included in the treatment cost estimates. Nor were any private costs related to vaccination, such as the cost of transportation or of missing work to get vaccinated. The cost-of-illness from a societal perspective would include these private costs–resulting in greater treatment savings–and thus our cost-effectiveness measures will be slightly more conservative than if a societal perspective was used.
Cost-effectiveness thresholds based on a country’s per capita gross domestic product (GDP) have often been used as a measure of the cost-effectiveness of a health intervention, with a cost per DALY averted that is equal to or less than the GDP per capita indicating that the intervention is “very cost-effective” [19]. However, these thresholds have been criticized as too limited as the sole or even a major determinant in decision-making, especially since they do not take into account a country’s specific context, including its ability to afford the intervention [20]. Therefore, in addition in comparing the cost per DALY averted results to per capita GDP, we also examine the affordability of cholera vaccination, in terms of cost per vaccinees and program cost as a percentage of the routine EPI budget. A univariate deterministic sensitivity analysis was conducted to show which variables have the greatest impact on cost-effectiveness. In the Excel spreadsheet that calculated cost-effectiveness, we designated distribution functions for the variables with uncertainty and then ran the Monte Carlo simulations. The gamma distribution function was used to estimate three variables: unit vaccine cost, vaccine delivery cost, and cholera treatment cost, while the beta distribution function was used to estimate two variables that had values that were between 0 and 1: the case fatality rate and cholera incidence. This analysis varies the values of key input parameters–case fatality rate, cholera incidence rates, vaccine price, cost of treatment, and vaccine delivery cost–one by one to estimate how these affect the outcome. Monte Carlo simulations were conducted multiple times (10,000) by drawing random values from the distribution functions for the input parameters using Ersatz software (version 1.3). Two distribution functions are used to model uncertainty: 1) beta for incidence and case fatality rates, variables with values between 0 and 1; and 2) gamma for vaccine, delivery, and treatment costs. For the gamma distribution, the parameters were a shape parameter α and a mean parameter β.
For the beta distribution, the parameters are two positive shape parameters, denoted by α and β, that appear as exponents of the random variable and control the shape of the distribution.
The input variable that is the most influential on cost per DALY averted is the one with the longest confidence interval, as will be shown in the tornado graph.
Using a base case of 10% annual migration (i.e., the target population is replaced at a rate of 10% a year), Fig 2A shows the predicted number of reported cholera cases each year in the target population of around 1.4 million people once the vaccination program–consisting of mass vaccination campaigns every three years and annual vaccination of the new birth cohort–is implemented, for each vaccination strategy. 2B depicts the total number of cases for the ten-year period by vaccination strategy and age group. The model was at equilibrium when interventions were simulated, so the simulations with no vaccination also reflect the incidence of cholera before vaccination. The percent reduction in incidence for each vaccination strategy is shown in Fig 3, while the cumulative number of cases prevented over the ten-year period is shown in Fig 4. Results showing 95% confidence intervals are presented in S1 Appendix.
If no vaccination or other cholera intervention program is enacted, there would continue to be, on average, around 3,200 cases of cholera in the study population presenting at health facilities each year, or more than 32,000 cases over the ten-year period of the analysis. A strategy of vaccinating only children 1–4 years of age would reduce cholera incidence in the overall targeted population by around 7% − preventing around 2,411 cases over ten years or 241 cases per year on average (Figs 3 and 4 and Table 2). Expanding the target vaccination group to 1–14 year olds would prevent around 14,400 cases over ten years, reducing incidence in the overall population by 45%. Vaccinating all ages one and above would prevent more than 29,100 cases over this period, reducing the overall cholera burden in the target population by 91%. Viewed from another perspective, for every reported case prevented in the overall population of nearly 1.4 million, 123 children 1–4 years of age would need to be vaccinated compared to 80 children 1–14 years old and 102 persons one year and above (Table 2). The strategy of targeting 1–14 year olds is thus the most efficient.
The herd effects of cholera vaccination are clearly shown in these results. The reduction in cases overall and in each age group is modest when vaccination is limited to 1–4 year olds, who make up around 7% of the total study population. However, expanding vaccination to all 1–14 year olds not only reduces the number of cases amongst these children, it also reduces incidence in adults (who are not vaccinated) by 40% (Fig 3). Thus, while 1–14 year olds account for around 30% of the population, vaccinating them would cut cholera incidence in the entire population by 45%. When adults are also vaccinated, more than 90% of cholera cases would be prevented.
We tested these results with different levels of annual population migration, which dilutes vaccination coverage of the population over time. When the entire population one year and above is targeted for vaccination and there is no migration, cholera transmission virtually stops and the overall effectiveness of the program is nearly 100% (Figure S3 in S1 Appendix). We also simulated vaccination campaigns every 5 years instead of every 3 years. In these scenarios, we assumed that vaccine protects individuals for 5 years. Vaccinating every 5 years is somewhat less effective than vaccinating every 3 years when 10% annual migration is assumed (Figure S5A and S5B in S1 Appendix), and becomes less effective when the migration rate is high (Figures S5C and S5D). If annual migration is 25%, the effectiveness of this strategy may fall below 90%, while when migration is 10% a year, effectiveness is above 90%. The effectiveness of vaccinating 1–14 year olds on the overall cholera incidence in the population is reduced from nearly 48% at 0% migration to 45% at 10% migration and to 41% at a 25% migration rate. When only 1–4 year olds are vaccinated, the effectiveness is around 7%, regardless of the migration level.
The projected costs of vaccination for the ten-year period by vaccination strategy and vial size are shown in Table 3. Including annual vaccination of infants in all scenarios, the total costs are estimated at $1.5 - $1.9 million (depending on the vial size) for the strategy targeting 1–4 year olds, $4.4 - $5.9 million for the strategy targeting 1–14 year olds, and $10.8 - $14.3 million if all persons one year and older are targeted. The annual costs over the ten-year period therefore range from approximately $145,600 to $1.08 million if ten-dose vials are used and from $193,080 to $1.4 million if single-dose vials are used. The cost per vaccine recipient, including vaccine wastage, is estimated at $4.62 for single-dose vials and $3.48 for ten-dose vials.
Table 4 shows the net cost of vaccination once the savings in treatment costs resulting from the program are subtracted from the total vaccination program costs. The estimated savings in treatment costs for the ten-year period range from around $126,000, if the program is limited to 1–4 year olds, to $1.5 million if all persons one year and above are included.
The results of dividing the net vaccination costs by cases, deaths, and DALYs averted for each vaccination strategy and vaccine vial size are shown in Table 5. The option of vaccinating 1–14 year olds would be by far the most cost-effective, with a cost per case averted of $255 - $356 and a cost per DALY averted of between $591 and $823. The next most cost-effective strategy would be vaccinating all ages one year and above, with a cost per case averted of $318 - $439 and cost of DALY averted of $894 - $1,234–1.5 times higher than the 1–14 year old targeting strategy. On the other hand, limiting cholera vaccination to 1–4 year olds would cost $1,276 - $1,731 per DALY averted–making it 2.2 times less cost-effective than the strategy of targeting 1–14 year olds and 1.4 times less cost-effective than the option of vaccinating all persons one year and above.
Comparing the cost per DALY averted to the GDP per capita as a measure of cost-effectiveness, all vaccination strategies would cost less per DALY averted than the country’s per capita GDP threshold ($1,359) [23], except the option of vaccinating 1–4 year olds only using single-dose vials, indicating that they would be cost-effective using this definition (Fig 5). However, the 1–4 year old vaccination strategy using ten-dose vials, as well as the strategy of vaccinating all ages one and above using single-dose vials barely fall below the threshold.
Univariate deterministic sensitivity analyses were conducted on variables with the greatest uncertainty in their values: case fatality rate (CFR), vaccine price, delivery cost, treatment cost and cholera incidence. The cost per DALY averted estimates vary most with changes in CFR, cholera incidence rates, and cost of treatment, while varying vaccine price and delivery costs has less effect on incremental cost-effectiveness ratios (Table 6 and Fig 6). However, even when any of the five parameters are varied, the strategy targeting 1–14 year old children for vaccination is the most cost-effective, regardless of the vaccine vial size, followed by vaccination of all persons one year and older.
Table 7 shows the mean value and 95% confidence interval for the cost per DALY averted by age group and vial size when the uncertain parameters are varied in the Ersatz Monte Carlo simulations. The results again show that the lowest cost per DALY averted is found when the age group 1–14 is vaccinated using either vial size.
Affordability and the budgetary impact of cholera vaccination is another critical factor in assessing the value for money of this intervention. The cost per vaccinee (vaccine and service delivery) in this analysis is $3.48 if ten-dose vials are used and $4.62 if single-dose vials are used–regardless of the age group targeted for vaccination (Table 3). The annual cost of vaccination in the proposed areas of Dhaka for the most cost-effective strategy targeting 1–14 year olds − $443,924 if ten-dose vials are used, and $588,715 if single-dose vials are used–represents 0.15% to 0.2% of the 2018 routine EPI budget of nearly $300 million [24].
This analysis predicts that a program in which all children 1–14 years of age are targeted in mass cholera vaccination campaigns every three years, coupled with annual vaccination of infants 12–15 months old through the routine immunization program, would reduce cholera incidence by 45% over 10 years in a population of around 1.4 million slum dwellers in high-risk areas of Dhaka, Bangladesh, preventing 14,400 cases, including in many adults, and cost $4–6 million over ten years. This would be by far the most cost-effective of the three program options, as well as the most efficient, in terms of the number of vaccinations per case averted.
Vaccinating all ages one and above would, on the other hand, reduce cholera incidence in this population by more than 90% over 10 years, preventing more than 29,000 reported cases. However, this strategy would be considerably less cost-effective than the one targeting 1–14 year olds, and would cost, on average, $1.1–1.4 million per year or $11–14 million over ten years. The option of vaccinating only young children (1–4 year olds) would have minimum impact on the overall cholera incidence (7% reduction) and despite its lower costs ($1.5 - $1.9 million), it would be the least cost-effective of the three strategies analyzed. Moreover, if we target 1–14 years age group a total of 14,430 cases will be averted whereas 2,411 cases will be averted if we target 1–4 years age group. However, these very young children, who are most vulnerable to severe cholera, would be included in the middle option targeting 1–14 year olds.
The results for the 1–14 year old and all-ages strategies show the power of herd effects from cholera vaccination, given that the analysis assumes that only a portion of the targeted population would receive the vaccine (70% of children and 55% of adults) and since direct protection from the vaccine is quite modest (ranging from 42% in children under five to 68% in 5–14 year olds over five years). Our results are in general agreement with observations of herd immunity observed in large trials [8, 25]. A large-scale cluster randomized trial of cholera vaccination was implemented in Dhaka, and effectiveness was modest (37% over 2 years) [8]. The effectiveness could have been reduced by both the high migration rate and contamination between vaccinated and unvaccinated clusters. The large-scale vaccination campaigns represented by our modeling results are likely to result in higher effectiveness than cluster-randomized trials.
These findings reinforce those of past cost-effectiveness analyses of cholera vaccination conducted in Bangladesh and elsewhere, which also found that a strategy of vaccinating children was more cost-effective than a strategy that included vaccinating adults–a key factor being the much higher incidence rates of endemic cholera in children than in adults [24, 26, 27]. However, the results also show that limiting vaccination to children under five–those typically targeted by national immunization programs and mass vaccination campaigns–would have a minimal impact on cholera incidence and would be considerably less cost-effective than the other two targeting strategies. Key reasons are the lower efficacy of the vaccine in this youngest age group (42%), and the low level of herd protection from vaccinating such a small proportion (≈7%) of the general population. Although the efficacy of OCV among children under 5 years old has been found to be about half that of older children in endemic settings in several studies [28–30], the our model could underestimate the impact of vaccinating young children if they contribute more to transmission than older children and adults.
A strength of this study is the identification of high-risk areas in Dhaka that would be strong candidates for a targeted cholera vaccination program in Dhaka. The study also improves upon previous cost-effectiveness analyses by using local data on the cost of recently-conducted vaccination campaigns.
However, a number of limitations of the analysis need to be pointed out. The mathematical model was calibrated to the dynamics of cholera transmission in the rural area of Matlab. The transmission dynamics and epidemiological patterns of cholera (e.g., survival of V. cholerae in the environment, seasonal patterns, age distribution of cases, ratio of asymptomatic to symptomatic infections) may differ somewhat in urban slums, such as those in Dhaka. Although the population of Matlab and Dhaka differ, we believe that the attack rates of cholera in these highly vulnerable populations subject to seasonal monsoon-driven epidemics are similar. Uncertainty around these parameters, as well as those governing the dynamics of V. cholerae in the environment, is difficult to resolve and can skew projections of outbreak sizes [31]. However, the mathematical model of cholera transmission was used solely to estimate the magnitude of indirect protection from mass vaccination, which should be robust to variation in these parameters as long as the disease attack rates are calibrated. In addition, in the absence of population-based cholera incidence data, an incidence estimate of treated cases (2.3/1,000) was extrapolated from data from two hospitals run by icddr,b. While the majority of cholera cases requiring care in Dhaka are reportedly treated at these two facilities, the omission of cases treated elsewhere would have the effect of under-estimating the true cholera incidence rate, making our analyses conservative. On the other hand, the estimated incidence rate is based on the assumption that all cases coming to these hospitals are from the estimated population who live in slums, which may not be the case. There are also uncertainties with the cholera case fatality rate and level of migration in Dhaka. The uncertainty of these key parameters (CFR, incidence and migration rates) was addressed in the sensitivity analyses, which in general show that the results, when comparing the different targeting strategies, remained similar when the values of these parameters were varied.
Concerning the economic analyses, the average cost of treating hospitalized cases of cholera in Dhaka was taken from a cost study at the main icddr,b hospital, which may have higher costs of treatment (and a higher standard of care) than in other health facilities in the city. In addition, the same cost estimate was applied to all cases, regardless of age, while the cost of treating children may differ somewhat from that of treating adults, as found in past studies of the cost of cholera illness [22, 32]. However, the sensitivity analysis showed that even varying the cost of treatment estimates by a factor of two in either direction did not substantially change the incremental cost-effectiveness ratios.
It should also be noted that this analysis focuses on cholera vaccination in a highly-targeted population of around 1.4 million people considered at highest risk of cholera, in a city of 8.5 million, with more than 18 million in the Greater Dhaka Area [33]. Since there are many slums–and cases of cholera–beyond those in the 14 thanas included in this analysis, controlling or even eliminating cholera in Dhaka would likely require expanding the target population to other areas in (and possibly surrounding) the city. This would increase the costs of the program and could also reduce its cost-effectiveness, if the cholera incidence rates in other areas are lower than the rate used in this analysis.
The producer of Cholvax has recently indicated that the public sector price for single-dose vials could be reduced to as low as $1.10 (vs. $1.40 assumed in the base analysis). This lower price would significantly reduce the costs and thus increase the cost-effectiveness of cholera vaccination under the modeled scenarios.
Cost-effectiveness is not the only factor that policymakers and donors need to consider when making decisions about whether or not to implement cholera vaccination in high-risk areas and which age groups to target. Another key consideration is affordability, as the financial realities in Bangladesh make it unlikely that the country can readily implement all potentially cost-effective interventions. The most cost-effective strategy–that of vaccinating 1–14 year olds in this population of around 1.4 million people–would cost, on average, ≈$444,000 - $589,000 per year, while the most expansive strategy of targeting all ages one and above would cost around 60% more ($1.1–1.4 million). The 1–14 year old option would thus add 0.15%– 0.2% to the annual national immunization program budget (of around $300 million), while the all-ages strategy would add 0.4%-0.6%. Assuming vaccine protects for 5 years for all age groups, we found that spacing the campaigns to every 5 years led to more cases 4 or 5 years after campaigns but only decreased average effectiveness over 10 years only a modest amount. However, the 5-year campaigns were less effective when migration was higher. If protection from vaccination wanes more rapidly among young children [30], then less frequent vaccination campaigns might not adequately protect them. However, expanding the vaccination program to other areas of Dhaka or to other cities in Bangladesh would increase these costs. In terms of the estimated cost per vaccinee − $4.62 for single-dose vials and 3.48 for ten-dose vials–this compares to a cost estimate for rotavirus vaccination (using Rotarix) of $5.46 - $5.98 per child vaccinated from an analysis conducted by PATH (Clint Pecenka, personal communications).
If protection from vaccination wanes more rapidly among young children, then less frequent vaccination campaigns might not be effective. One way to reduce the costs of cholera vaccination, and thus increase its affordability and cost-effectiveness, would be to decrease the frequency of the vaccination campaigns from every three years to every five years, given that the vaccine has been shown to provide protection for at least five years, at least among persons five years and older. However, in a highly mobile area such as the slums of Dhaka, population migration would erode coverage and decrease the effectiveness of vaccination over time. If one assumes, for example, an annual migration rate of 10%, 50% of the population will have been replaced with unvaccinated people within five years, greatly reducing the level of protection in this population and potentially leading to outbreaks three or four years after a campaign. Conducting vaccination campaigns every five years may be an appropriate strategy in a less mobile population.
Government policymakers in Bangladesh have expressed interest in providing cholera vaccination in combination with other interventions to reduce cholera and water-borne diseases, as recommended by WHO [34]. A comprehensive cholera control program that combines vaccination with improvements in water distribution and water quality–such as by increasing the number of legal water connections and placing water pipes far from sewer systems—should, in fact, create synergies that result in a more rapid reduction in disease than any intervention on its own [35]. Cholera vaccination could also be a means of accelerating control of the disease before adequate water and sanitation improvements can reach the most vulnerable populations. In addition, cholera vaccination could be provided during periodic intensive routine immunization (PIRI) activities or other vaccination campaigns (e.g., measles-rubella), which would further reduce its costs.
Another key factor to consider in deciding whether or not to introduce a new vaccine is fairness and equity [20]. Because cholera predominantly strikes the most impoverished and marginalized populations, who are also those with the least access to quality health care services, cholera vaccination, especially using a strategy targeting high-risk areas, would reduce these inequities.
While cholera incidence rate estimates are not available for other urban areas of Bangladesh, it is likely that in cities where the disease is known to be endemic or where outbreaks have occurred in the recent past and where slum conditions are similar to those in Dhaka, the risk of cholera will be similar to that found in the slums of Dhaka. Thus, it is reasonable to assume that the results of this analysis can be generalized to slum populations in other cholera-affected urban areas of the country. The findings from icddr,b’s prospective cholera surveillance currently underway in twenty-two mainly urban sites throughout the country will help increase our understanding of the cholera burden in other parts of the country and thus determine the relevance of the findings of our analysis to other urban areas. This strategy of targeting urban slums for special disease control efforts would also align with a new national government priority of improving health in slums in Bangladesh.
|
10.1371/journal.pcbi.1000311 | Using Network Component Analysis to Dissect Regulatory Networks Mediated by Transcription Factors in Yeast | Understanding the relationship between genetic variation and gene expression is a central question in genetics. With the availability of data from high-throughput technologies such as ChIP-Chip, expression, and genotyping arrays, we can begin to not only identify associations but to understand how genetic variations perturb the underlying transcription regulatory networks to induce differential gene expression. In this study, we describe a simple model of transcription regulation where the expression of a gene is completely characterized by two properties: the concentrations and promoter affinities of active transcription factors. We devise a method that extends Network Component Analysis (NCA) to determine how genetic variations in the form of single nucleotide polymorphisms (SNPs) perturb these two properties. Applying our method to a segregating population of Saccharomyces cerevisiae, we found statistically significant examples of trans-acting SNPs located in regulatory hotspots that perturb transcription factor concentrations and affinities for target promoters to cause global differential expression and cis-acting genetic variations that perturb the promoter affinities of transcription factors on a single gene to cause local differential expression. Although many genetic variations linked to gene expressions have been identified, it is not clear how they perturb the underlying regulatory networks that govern gene expression. Our work begins to fill this void by showing that many genetic variations affect the concentrations of active transcription factors in a cell and their affinities for target promoters. Understanding the effects of these perturbations can help us to paint a more complete picture of the complex landscape of transcription regulation. The software package implementing the algorithms discussed in this work is available as a MATLAB package upon request.
| One of the fundamental challenges in biology in the post-genomics era is understanding the complex regulatory mechanisms that govern how genes are turned on and off. In a single organism where the functions of individual genes in a population do not differ much, many of the differences between individuals including physical phenotypes, susceptibility to disease, and response to drugs can be attributed to how genes are regulated. Previous studies have largely focused on identifying regulator and target genes whose expressions are linked to genetic variations in a population. We present work that focuses on considering a specific set of regulators called transcription factors whose targets can be verified from experiments and whose interactions with those targets have been well studied and modeled. In this setting, we can begin to understand how genetic variations perturb the concentrations and promoter affinities of active transcription factors to induce differential expression of the targets. Understanding the effects of these perturbations is important to understanding the fundamental biology of gene regulation and can help us to design and assess therapeutics and treatments for complex diseases.
| With advances in whole genome high-throughput technologies such as ChIP-Chip, expression, and genotyping arrays, it is now possible to integrate data from these sources together to decipher the complex regulatory networks that govern transcription. In addition to serving as powerful models for how basic cellular function is achieved, these regulatory networks can also help us shed light on how certain disease phenotypes are manifested. At the heart of these networks are a few regulator genes such as transcription factors (TFs), miRNAs and histones whose activity govern the behavior of many other genes. Among these regulators, transcription factors that bind the promoter regions of genes are by far the most well understood. The process of TFs activating or repressing transcription at initiation is believed to be the primary mechanism of gene regulation. A central question in genetics is how genetic variations perturb this underlying regulatory mechanism to give rise to differential gene expression and ultimately complex phenotypes.
The simplest analysis one can perform to address this question is expression quantitative trait loci (eQTL) mapping, which identifies genetic variations such as SNPs in the form of linkages and associations that are correlated with gene expression. Such studies have been carried out in a variety of organisms including yeast [1],[2] Arabidopsis [3], mouse [4],[5] and human [6]–[8]. These studies have identified many linkages between SNPs and genes in close proximity suggesting potential local regulatory mechanisms mediated by regulators such as transcription factors and miRNAs. These studies have also identified a few SNPs linked to the expressions of many genes suggesting a global regulatory mechanism mediated by master regulators such as transcription factors and histones. Unfortunately, beyond nominating candidate genes either as targets or regulators, these studies give little insight into how SNPs perturb the underlying transcription regulatory networks that control gene expression.
To gain a better understanding of the mechanisms of transcription regulation, several systems biology based methods have been proposed including clustering of co-regulated genes [9], multipoint linkage analysis [10],[11], pathway enrichment analysis [12]–[16], prediction of regulatory modules [17],[18] and the prediction of causal regulatory relationships [19]–[23]. Many of these advanced methods aim to tease out both the nodes (regulators and targets) as well as the topology (mapping of edges) in a transcription regulatory network from only considering gene expression profiles. Although these methods have predicted some interesting relationships, there are at least two aspects of transcription regulation that go unaddressed when we use them to study transcription factors and their targets. First, most previous methods rely on probabilistic models that do not provide much insight into the hidden dynamics between the activity of transcription factors and the expression of their targets. Second, the relationships inferred by these methods from the expression profiles alone can be misleading because the in vivo activity of a transcription factor does not always correlate with its expression levels [24],[25].
To overcome these problems, we adopt a framework from network component analysis (NCA) [26] that considers a simple bipartite network model of transcription regulation involving only transcription factors and their targets. In this model, the expression of a target gene is completely captured by two properties of the network, the concentrations and promoter affinities of transcription factors. In general, inferring these two quantities from the expression profiles of the target genes alone is difficult. But by leveraging protein-DNA binding data from ChIP-Chip experiments [27],[28], a partial topology of the network can be constructed and one can make the inference given certain constraints [26].
The NCA method as described by liao et al. has been successfully applied to several gene expression datasets to understand transcription regulation in a temporal setting [26] and in the context of gene knockouts [29]. In this study, we extended NCA to study transcription regulation over a population gradient by modeling three mechanisms by which genetic variations perturb the concentrations and promoter affinities of active transcription factors to induce differential expression. Figure 1 gives a simple example that illustrates the original NCA model and our extensions. Imagine we have a small experiment where we collected the gene expressions of four genes, the genotypes of three markers over three individuals. Given the topology of the bipartite network between transcription factors and their targets (Figure 1B), the NCA algorithm allows us to infer the active transcription factor concentrations (C) and the respective promoter affinities (PA) from the given gene expressions (E) in a log-linear fashion (Figure 1A, see Methods). In this example, SNP1 and SNP3 are linked to the expressions of G1 and G3 while SNP2 is linked to the expressions of G2 and G4. We propose three possible mechanisms any one SNP can perturb the regulatory network and show an instance of each using the given example.
Because the inclusion of genetic variation creates additional parameters in each of our three models compared to the original NCA model, we expected them to always fit the data better. To effectively evaluate our models, we devised a likelihood ratio statistic and a permutation scheme to assess the statistical significance of our improvements. We then applied our method to study an expression data collected over 112 segregants of Saccharomyces cerevisiae yeast and two separate ChIP-Chip datasets generated by Harbisonet al. and Lee et al..We identified several interesting global regulatory networks perturbed by SNPs located in regulatory hotspots. Some of these networks have one property perturbed (transcription factor concentration or promoter affinity) while others have both properties perturbed suggesting a complex mechanism of global regulation. We also examined linkages between SNPs and target genes located in close proximity. We found that many of these cis linked SNPs perturb the promoter affinities of transcription factors on a target gene locally confirming previous hypotheses of cis regulation.
An interesting method proposed by Sun et al. also used the NCA framework to infer the concentrations of active transcription factors from gene expression data collected over the same yeast strains. Their method was designed to detect linkages between the inferred concentrations and genetic variations and used conditional independence tests to find modules of genes controlled by the same causal regulator. Compared to this method, we expect to find similar networks of genes and transcription factors but our method does not allow us to infer additional causal relationships using statistical tests. Instead, we focus on identifying different mechanisms by which genetic variations can perturb the regulatory networks by directly modeling the effects of these perturbations into the NCA framework. We do not attempt to make rigorous causal claims but use the causal information inherent in genotyping and ChIP-Chip experiments to suggest possible mechanisms of transcription regulation.
The NCA framework is a natural model for describing how transcription factors regulate gene expression. At the heart of the model is a log linear equation that relates the expression levels of genes collected over a gradient (E) to the concentrations (C) and promoter affinities (PA) of active transcription factors. Such a model is well supported by known kinetic properties of protein-DNA interactions [30]. In linear model terms, the transcription factor concentrations are the regressors, the gene expression levels are the response variables and the promoter affinities are the coefficients that relate the two. Figure 2B shows the log-linear equations describing the graph shown in Figure 1B. The goal of NCA is to infer the matrices of concentrations and promoter affinities from the matrix of gene expressions under some restrictions in the least squares sense.
Treating genetic differences between individuals as a gradient, we applied this model to infer the matrices and from gene expressions collected from a population of yeast strains, . For the inference to have been possible, we removed a number of transcription factors and target genes to construct a network from the original ChIP-Chip data that met certain constraints [26]. After preprocessing the Lee et al. ChIP-Chip dataset, we were left with a network with 100 transcription factors and 2,294 target genes. Similarly, preprocessing the Harbison et al. ChIP-Chip dataset left 158 transcription factors and 2,779 target genes. Using a two step optimization algorithm developed by Liao et al., we inferred the concentration profile for each transcription factor over the genetic gradient and compared it to the corresponding TF expression profile by computing Pearson's correlations (). Figure S3 shows that these quantities were not well correlated with average correlation coefficients of and using the Lee et al. and Harbison et al. datasets respectively. The stability of the inferred TF concentrations were however robust when we compared results from the two ChIP-Chip datasets with a correlation coefficient of (Figure S4). The robustness was also verified by bootstrapping experiments [31] (Results not shown).
We next applied our method to study the mechanisms by which regulatory hotspots, genomic locations in yeast shown to be linked to the expression of many genes, perturb the underlying transcription regulatory networks. Although several transcription factors have been known to act as master regulators in yeast, it has been surprisingly shown in previous eQTL studies that only a few regulatory hotspots are located close to transcription factors. We hypothesized that although complex regulatory mechanisms upstream of transcription regulation such as signaling pathways exist, transcription factors ultimately mediate the global regulation of gene expressions. Using our framework, we tested our hypothesis by determining whether a regulatory hotspot is linked to the concentrations or promoter affinities of active transcription factors to achieving this regulation.
To identify the regulatory hotspots, we performed simple linkage analysis on only a subset of genes that were NCA compliant (see Methods). Similar to previous reports, only a few hotspots were located cis to any known transcription factors [1],[2]. For example, a hotspot located on chromosome 12 spanning basepairs 600,000 to 680,000 was cis to HAP1 while another hotspot located on chromosome 3 spanning basepairs 60,000 to 100,000 was cis to LEU3. Several approaches [20],[23] have identified additional putative causal regulators, many of which are not transcription factors, corresponding to these regulatory hotspots.
We first considered SNPs located in regulatory hotspots that perturbed the concentrations of active transcription factors to cause global differential expression. Extending the NCA model to incorporate SNPs as perturbations did not require changing the optimization procedure. As shown in Figure 2C, we first decomposed the inferred transcription factor concentration matrix from applying the original NCA algorithm, , into two matrices and segregated by a SNP. Next, we identified those transcription factors whose concentrations were linked to the SNP using a simple t-test, an example is shown in bold in Figure 2C, and assessed the significance of the linkage by a permutation scheme (see Methods).
Using both the Harbison et al. and Lee et al. ChIP-Chip binding data, we found many transcription factors whose concentrations were linked to at least one SNP. Table 1 lists those linkages occurring at regulatory hotspots and the corresponding transcription factors. In addition to having a strong linkage, we also required the transcription factors in the table to have at least 6 (Lee et al) or 7 (Harbison et al) downstream targets whose expression levels were significantly linked to the regulatory hotspot. A number of transcription factors known to act as global regulators were identified. Of particular note, we found HAP1 to be the mediator of hotspot 6 located on chromosome 12 spanning basepairs 600,000 to 680,000 using the Harbison et al. dataset; and YAP1 and LEU3 to be mediators of hotspot 3 located on chromosome 3 spanning basepairs 60,000 to 100,000. GCN4 was also identified as a mediator of this hotspot using the Lee et al. dataset but it was only marginally significant using the Harbison et al. dataset (Result not shown). These results are concordant with previous findings [2],[23]. In particular, LEU2 has been previously implicated to be linked to hotspot 3 where an engineered deletion of the gene occurs. Figure 4 are heatmaps showing the strong correlations between concentration levels of transcription factors, HAP1 and LEU3 respectively, and the expression levels of their downstream targets linked to the respective regulatory hotspots.
We next examined hotspot 2, a hotspot that has been previously identified by brem et al.to regulate budding and daughter cell separation through the causal regulator AMN1 [9]. We identified four transcription factors, ACE2, MBP1, SKN7 and SWI4, whose active concentrations were significantly linked to hotspot 2 in both datasets. Five other transcription factors responsible for cell cycle transitions, ABF1, FKH1, OAF1, RAP1 and SWI5 were also found to be significant in the Harbison et al. dataset. Some of these transcription factors are known to interact with each other and have similar profiles such as ACE2 and SWI5; and MBP1, SKN7 and RAP1. Figure 3A and Figure 3B are heatmaps showing the strong correlation between the concentrations of transcription factors (ACE2 and SWI4) and the expression levels of their direct targets linked to the hotspot. Our results are consistent with previous findings that suggest ACE2 as a causal transcription factor mediating the global regulation of the mitotic-exit network (MEN) by AMN1 [23] even though ACE2's direct targets were not overrepresented for any GO biological processes or functional groups. This is probably because many downstream transcripts of the MEN were not considered in this analysis because there's no direct ChIP-Chip evidence of binding between these transcripts and ACE2.
Another interesting regulatory hotspot, occurring at chromosome 12 basepairs 1,040,000 to 1,060,000, was found by Brem et al. to regulate subtelomerically encoded helicases through the causal regulator SIR3. We found two transcription factors, GAT3 and YAP5, whose concentrations were linked to this hotspot using the Harbison et al. data. YAP5 was also significant using the Lee et al. data. Figure 3D and Figure 3C show the strong correlations between GAT3 and YAP5 concentrations and the expression profiles of their targets. Unlike the previous example, the targets of YAP5 were enriched for helicases () and consisted of many genes with unknown function as represented by a significant enrichment for the GO annotation of “biological process unknown” (). These results suggest a potential novel mechanism for the regulation of subtelomerically encoded helicases mediated by YAP5 and GAT3.
We next considered SNPs located in regulatory hotspots that perturbed the promoter affinities of transcription factors to cause global differential expression. Modeling these perturbations required an extension to the NCA model. As shown in Figure 2D, in addition to decomposing the transcription factor concentration and gene expression matrices, we also decomposed the promoter affinities matrix, into and where the only difference between the two is the column corresponding to the global promoter affinities of the transcription factor of interest as shown in bold. We identified perturbed networks of genes and transcription factors by deriving a likelihood ratio statistic that compared the extended model to the original NCA model. Since the extended model included additional parameters, namely different promoter affinities between populations, we expected it to always fit the data better. Thus to assess significance, we used a permutation scheme that randomized the decomposition of individuals while preserved the topology of the bipartite graph (see Methods).
We revisited the regulatory hotspots discussed in the previous section. We speculated that transcription factors whose promoter affinities were perturbed by a regulatory hotspot must interact with other transcription factors whose concentrations were perturbed by the same hotspot to induce global differential expression of the targets. The intuition being if the in vivo concentrations of a transcription factor is relatively stable, then it could still regulate gene expression by differentially binding to other transcription factors to form a complex. A transcription factor's binding affinity for promoters is then in part determined by the concentrations of its partnering transcription factors. This is exactly what we observed in our results. For example, we found that hotspot 6 which was shown to be linked to the concentrations of HAP1 was also linked to the promoter affinities of HAP4. HAP1 and HAP4 are known to interact in a complex to regulate global respiratory gene expression. Similarly, hotspot 8 was linked to the concentrations of DIG1 and the promoter affinities of STE12. DIG1 has previously been shown to code for an inhibitor of STE12, a transcription factor involved in pheromone induction and invasive growth [32]–[34]
We next examined how two hotspots discussed in the previous section also perturbed promoter affinities of transcription factors. Figure 4 and Table 2 show that hotspot 2 was linked to the promoter affinities of ACE2, SWI4 and UME6. Hotspot 2 was also shown in the previous section to be linked to the concentrations of ACE2 and SWI4 but not UME6, see Figure S2 for the expression profiles of the downstream targets of UME6. Consistent with our speculation, UME6 has been shown to interact with SWI4 and SWI4 has been shown to interact with itself. Furthermore, we see that AMN1 is a target of ACE2 suggesting that the regulation of the mitotic-exit network might be feedback in nature.
Figure 4 also shows a similar network consisting of the two transcription factors whose concentrations linked to hotspot 7, GAT3 and YAP5. Notice that while YAP5's promoter affinities were linked to the hotspot also (thick edges), GAT3's were not (thin edges). Consistent with previous results, YAP5 has been shown to interact with itself to modulate gene expression. These results suggest that in some transcription factors, particularly those that interact with themselves, both promoter affinities and concentrations of the transcription factor could be perturbed by a regulatory hotspot. On the other hand, some transcription factors might not have their concentrations perturbed by a hotspot but because of interactions with another transcription factor, has their promoter affinities perturbed giving rise to global differential expression of their targets.
Previous eQTL analyses have shown that the most significant linkages occur cis to genes [1],[2]and often located or in LD with SNPs located in the promoter regions of genes harboring transcription factor binding sites [35]. Our model allowed us to determined if differences in expression of a single gene could be attributed to cis genetic variations perturbing the local affinities of transcription factors on the promoter.
There is a direct similarity between these perturbations and those that affect global promoter affinities. As shown in Figure 2E, SNP3 perturbs the local affinities of transcription factors for the promoter of G3. We modeled this affect by decomposing the matrix into and where the only difference between the decomposed matrices was the row corresponding to G3, as shown in bold. We used a likelihood ratio statistic to choose between two different models and assessed the significance based on permuting the genotypes of the individuals.
Of the small subset of genes examined, 2294 from using the Lee et al. dataset and 2779 from using the Harbison et al. dataset, we found ≈45% of the transcripts (972/2294 Lee, 1315/2779 Harbison) linked to at least one SNP at a FDR of with using a standard t-test. Out of these linkages, ≈30% were cis (257/972 Lee, 331/1315 Harbison). These proportions are consistent with what has been reported [10].
We postulated that many cis linked loci found by previous analyses and confirmed by our analysis are in LD with causal SNPs located in promoter regions. We further postulated that such a causal SNP corresponds to a variation in the primary sequence of a transcription factor binding site that affects the promoter affinity of a transcription factor or a complex of transcription factors. This model is consistent with the idea that a genetic variation at regulatory regions of the genome can give rise to observed subtle differences in gene expression across populations. We identified a total of 138 and 174 genes which have their local promoter affinities affected by a SNP with a FDR of .
Figure 5A shows that there is high concordance between those genes with significant cis linkages and those whose promoter affinities were perturbed. We did not expect all cis linkages to perturb promoter affinities. There are potentially other regulatory machinery that operate on intronic 3′UTRs and 5′UTRs. Next we compared the perturbed genes found using the Lee et al. dataset versus those found using the Harbison et al. dataset (Figure 5B). At a FDR of , 72 significant genes were shared between the datasets and 168 genes were not. We suspected that the different results obtained from these two datasets can be attributed to differences in network topology. The two binding datasets often reported genes with different sets of bound transcription factors and transcription factors with different sets of targets making the estimates of certain quantities inconsistent. Additional discrepancies arose from different sets of genes having been eliminated from each analysis due to the criteria placed on the network topology.
Although there is a growing wealth of literature identifying putative causal regulators in yeast and mouse using statistical approaches, some of which integrate different sources of information, it is not clear by what mechanism genetic variations perturb the underlying regulatory networks to give rise to global differential expression. We have presented an integrated framework based on network component analysis that directly models how genetic variations perturb the concentrations and promoter affinities of transcription factors to cause the differential expression of their targets. Such a model differs from current eQTL analyses in that a direct, testable mechanism of transcription regulation is specifically considered. Although these networks are limiting, both in terms of the amount of biology they explain as well as the dependence on experimental data for their inference, a substantial set of genes (≈1/3) was still considered. In our analysis, we show that many genes with cis linkages are likely to be regulated by transcription factors binding differentially to their promoter regions. We also show two representative examples of the complex mechanism of achieving global differential expression of a large number of transcripts, where the regulation of transcription factors involve two distinct processes and maybe feedback in nature.
Our approach specifically uses one variation of the NCA algorithm to infer the concentrations and promoter affinities of transcription factors. The key aspect of our approach is that we treat genetic variations as perturbations to an underlying regulatory network whose structure is already known. In theory, any NCA like approach [36]–[38] where a network is inferred from known data such as ChIP-Chip experiments, protein-protein interaction experiments or literature can be extended to take into account genetic variation.
There are also some natural extensions to the framework we have presented. First, one is not limited to considering only genetic variation as a perturbation. Other forms of perturbation such as media condition and disease pathogenesis can as well be applied in this approach to identify the corresponding effect on the networks. Second, our method considers the perturbation of only one SNP. Although several approaches have been proposed to investigate the statistical interaction of multiple SNPs on a phenotype [11],[39], it would be interesting to study the mechanistic interactions of multiple perturbations on a transcription regulatory network.
We used the expression measurements (6,216 transcripts) and genotyping data (2,956 SNPs) collected over 112 segregants of yeast derived from two parental strains BY4716 and RM11-1a originally described by Brem et al. The gene expression data is available at GEO (http://www.ncbi.nlm.nih.gov/projects/geo/) with the accession number GSE1990.
ChIP-Chip data from two datasets [27],[28] were used to generate two different transcription regulatory networks at a p-value cutoff of 0.001. Consistency was checked in each case by comparing the networks generated from using a cutoff of 0.01 and 0.001.
We next checked for NCA compliance as outlined [31]. We were left with a sub-network of 2,294 transcripts and 100 transcription factors after processing the Lee et al. dataset and 2,779 transcripts and 158 transcription factors after processing the Harbison et al. dataset.
We first performed a standard t-test to compare the population means between the segregated expression profiles of a single gene by a given SNP. We assessed the significance of our linkages by performing a permutation test as described [40].
We then identified regulatory hotspots by dividing the yeast genome into 493 20 kb bins and counted the number of significant trans linkages to unique gene expression levels each bin contained from the standard t-test. We found a total of 430 significant trans linkages using the Harbison et al. data and 290 using the Lee et al. data. Assuming a Poisson process where the rare event of a trans linkage occurs at a rate of 0.87 (430/493 Harbison) and 0.60 (290/493 Lee), the probability of observing >7 linkages in the largest bin using the harbison_transcriptional_2004 data is and the probability of observing >6 linkages in the largest bin using the Lee et al. data is . Because of the differences in the set of genes used in the different datasets, we constructed a set of 11 hotspots shared between the two.
NCA was originally developed to analyze time series based gene expression data but can be easily adapted to analyze whole genome expression data collected from different individuals in a population. In both cases, the goal is to infer the concentrations of active transcription factors and the promoter affinities from the expression levels of the target genes. This inference is made possible when the partial topology of the interaction network between transcription factors and target genes is determined from genome-wide location analysis that detects the binding of transcription factors to DNA promoter regions (ChIP-Chip).
Figure 1B shows an example of a bipartite graph where the expression levels of five genes are determined by the concentrations and promoter affinities of the three transcription factors. Formally, given a matrix of dimension where we have collected the expression levels of genes from individuals. Each column represents a separate microarray experiment that measures the expression levels of all genes in one individual. NCA approximates the relationship between the concentrations of active transcription factors and gene expression levels by a log-linear model of the type:(1)where is the gene expression level for gene in individual , is the concentration of transcription factor in individual and is the affinity of transcription factor for the promoter of gene . We can take the log of Equation 1 and transform it into a matrix representation:(2)Here, is a matrix of dimension representing the concentrations of the transcription factors in the individuals and is a matrix of dimension representing the affinities of the transcription factors for the promoters of the genes and is a matrix of dimension representing the residual. NCA analysis without incorporating genetic information seeks to iteratively find and that minimizes the quantity:(3)Finding the least squares estimates of and is equivalent to finding the maximum likelihood estimates under the assumption that the are independent identically-distributed (iid) vectors with Gaussian noise.
In our model, a genetic variation induces global differential expression either by perturbing the concentrations of a transcription factor or the promoter affinities of a transcription factor on all of its targets. Figure 1C shows the former case where the promoter affinities of TF1 on all targets remain the same but the concentration of TF1 is elevated in the group of individuals with an A allele at SNP1 while it is attenuated in the group of individuals with the C allele at SNP1. Figure 1D shows the latter case where the affinities of TF2 for the promoter region of its targets are different between two populations. Notice that in both cases, we do not make any assumptions about where the genetic variation occurs since several mechanisms can contribute to the transcription factor having different in vivo concentrations and promoter affinities. We can formally model perturbations to the promoter affinities by constructing two matrices, and that differ in the column corresponding to the transcription factor of interest.
We can also model local changes to the promoter affinities of all transcription factors on a single gene such as shown in Figure 1E where one group of individuals has the A allele and another group has the T allele (SNP3) in the binding site of the transcription factor complex. To model this change in the promoter affinities on one gene, we again construct two matrices and that differ in the row corresponding to the gene of interest.
|
10.1371/journal.pmed.1002296 | Long-term inpatient disease burden in the Adult Life after Childhood Cancer in Scandinavia (ALiCCS) study: A cohort study of 21,297 childhood cancer survivors | Survivors of childhood cancer are at increased risk for a wide range of late effects. However, no large population-based studies have included the whole range of somatic diagnoses including subgroup diagnoses and all main types of childhood cancers. Therefore, we aimed to provide the most detailed overview of the long-term risk of hospitalisation in survivors of childhood cancer.
From the national cancer registers of Denmark, Finland, Iceland, and Sweden, we identified 21,297 5-year survivors of childhood cancer diagnosed with cancer before the age of 20 years in the periods 1943–2008 in Denmark, 1971–2008 in Finland, 1955–2008 in Iceland, and 1958–2008 in Sweden. We randomly selected 152,231 population comparison individuals matched by age, sex, year, and country (or municipality in Sweden) from the national population registers. Using a cohort design, study participants were followed in the national hospital registers in Denmark, 1977–2010; Finland, 1975–2012; Iceland, 1999–2008; and Sweden, 1968–2009. Disease-specific hospitalisation rates in survivors and comparison individuals were used to calculate survivors’ standardised hospitalisation rate ratios (RRs), absolute excess risks (AERs), and standardised bed day ratios (SBDRs) based on length of stay in hospital. We adjusted for sex, age, and year by indirect standardisation. During 336,554 person-years of follow-up (mean: 16 years; range: 0–42 years), childhood cancer survivors experienced 21,325 first hospitalisations for diseases in one or more of 120 disease categories (cancer recurrence not included), when 10,999 were expected, yielding an overall RR of 1.94 (95% confidence interval [95% CI] 1.91–1.97). The AER was 3,068 (2,980–3,156) per 100,000 person-years, meaning that for each additional year of follow-up, an average of 3 of 100 survivors were hospitalised for a new excess disease beyond the background rates. Approximately 50% of the excess hospitalisations were for diseases of the nervous system (19.1% of all excess hospitalisations), endocrine system (11.1%), digestive organs (10.5%), and respiratory system (10.0%). Survivors of all types of childhood cancer were at increased, persistent risk for subsequent hospitalisation, the highest risks being those of survivors of neuroblastoma (RR: 2.6 [2.4–2.8]; n = 876), hepatic tumours (RR: 2.5 [2.0–3.1]; n = 92), central nervous system tumours (RR: 2.4 [2.3–2.5]; n = 6,175), and Hodgkin lymphoma (RR: 2.4 [2.3–2.5]; n = 2,027). Survivors spent on average five times as many days in hospital as comparison individuals (SBDR: 4.96 [4.94–4.98]; n = 422,218). The analyses of bed days in hospital included new primary cancers and recurrences. Of the total 422,218 days survivors spent in hospital, 47% (197,596 bed days) were for new primary cancers and recurrences. Our study is likely to underestimate the absolute overall disease burden experienced by survivors, as less severe late effects are missed if they are treated sufficiently in the outpatient setting or in the primary health care system.
Childhood cancer survivors were at increased long-term risk for diseases requiring inpatient treatment even decades after their initial cancer. Health care providers who do not work in the area of late effects, especially those in primary health care, should be aware of this highly challenged group of patients in order to avoid or postpone hospitalisations by prevention, early detection, and appropriate treatments.
| Today, four out of five children with cancer become long-term survivors. However, as toxic treatments, often in combination, are given at an age distinguished by growth and organ maturation, many childhood cancer survivors face significant and often uncharacterised late somatic morbidities.
No previous study has been able to examine the long-term inpatient morbidity in childhood cancer survivors including virtually all combinations of type of childhood cancer and disease-specific outcome.
We explored the lifetime somatic morbidity pattern in 21,297 5-year survivors of childhood cancer using the unique health registries of the Nordic countries, including 120 disease entities and all main diagnostic groups of childhood cancer.
We found that survivors had an overall 2-fold risk of being hospitalised and experienced longer stays in hospital than population comparisons of similar age and sex.
Major reasons for hospitalisation among cancer survivors were diseases of the nervous system (19.1% of all excess hospitalisations), endocrine system (11.1%), digestive organs (10.5%), and respiratory system (10.0%). Together, these four groups of diseases accounted for 51% of the excess cause-specific hospitalisations among cancer survivors.
The morbidity pattern was highly dependent on the type of childhood cancer, with highest risks seen for survivors of neuroblastoma, hepatic and central nervous system (CNS) tumours, and Hodgkin lymphoma.
This study provides a comprehensive overview of the lifelong, complex, and often serious disease pattern that childhood cancer patients encounter after ended treatment.
As we have used diagnostic inpatient information as measure of outcomes, our study presents the most serious somatic disease burden, which, in combination with previously described cognitive and other psychological adverse effects, may have significant impact on the quality of life and overall mortality of childhood cancer survivors.
Clinicians in primary and secondary health care should be aware of this vulnerable group of patients for better detection, prevention, and management of treatment-induced late effects.
We recommend long-term follow-up of high-risk subsets of childhood cancer survivors in a specialised setting.
| The number of childhood cancer survivors is increasing steadily in many parts of the world because of the extraordinary improvement in survival rates during the past five decades [1]. In the Nordic countries, four of five childhood cancer patients can expect to be long-term survivors [2]. These major improvements in survival come, however, at a price. Because of intense exposure to radiation and highly toxic compounds during treatment, a high proportion of survivors of childhood cancer now face somatic, mental, and cognitive late effects, many of which become clinically apparent even decades after the cancer was cured [3–7]. Only a limited number of studies have investigated the hospitalisation pattern subsequent to treatment for cancer diagnosed during childhood, adolescence, and young adulthood [8–13]. They all revealed an overall increased risk for hospitalisation. However, the record linkage studies with diagnostic inpatient information [9,10,12,13] were limited by study size and unable to give reliable risk estimates for rare types of childhood cancer or combinations of types of childhood cancer and types of subsequent disease. The larger studies were either based on self-reported information on causes for hospitalisation [7] or questionnaire data from patients’ primary health care physicians [10], with a relatively large proportion of survivors lost to follow-up.
In a population-based cohort study with virtually no loss to follow-up and exclusive use of medically verified diagnostic information from individual inpatient records, we studied the full range of somatic morbid conditions requiring hospitalisation in 21,297 5-year survivors of childhood cancer diagnosed between 1943–2008. As this is the largest long-term follow-up study of inpatient care among childhood cancer survivors conducted so far, it allowed stratifications and detailed analyses that were not possible in previous studies and provides novel information on diseases that first become symptomatic in middle age or senescence. Thus, the primary aim of our study was to present a comprehensive yet detailed overview of the long-term frequency and distribution of somatic diseases serious enough to require hospitalisations in survivors of childhood cancer combined and by type of cancer.
This retrospective, register-based cohort study is part of the collaborative study Adult Life after Childhood Cancer in Scandinavia (ALiCCS) (www.aliccs.org) [14]. The ALiCCS study was approved by the national bioethics committee, the data protection authority, or the national institute for health and welfare in the respective countries (Denmark: 2010-41-4334; Finland: THL/520/5.05.00/2016; Iceland: VSN 10–041; and Sweden: Ö 10–2010, 2011/19). Consent from study participants was not required as all data were available in national health registers.
The basic childhood cancer cohort in the present analysis is a subcohort of the Nordic ALiCCS material, comprising 33,576 individuals with cancer diagnosed in Denmark, Finland, Iceland, or Sweden in people under the age of 20 years in the periods 1943–2008 in Denmark, 1971–2008 in Finland, 1955–2008 in Iceland, and 1958–2008 in Sweden (Fig 1) [15–18]. Patients from Norway were not included, as complete hospitalisation histories with all diseases included in the present study were not available. From the cancer registries, we obtained each patient’s personal identification number, date of diagnosis, and type of cancer and assigned patients to the 12 main diagnostic groups of the International Classification Scheme for Childhood Cancer, with lymphoma divided further into Hodgkin, non-Hodgkin lymphoma, and other lymphomas [19]. For each childhood cancer patient, we randomly selected five comparison individuals from the national population registers who were alive on the date of the cancer diagnosis of the corresponding patient; of the same sex, age, and country; and without a cancer diagnosis before the age of 20 years. Fewer than five comparison individuals were available for 157 patients, leaving 167,712 participants for study. For both patients and comparison individuals, we obtained information from the population registers on vital status and emigration during follow-up.
Before linkage of study participants to the respective national hospital registers, we excluded those in whom more than one cancer was diagnosed during childhood, those who had died or emigrated before the start of the national hospital registers (Sweden, stepwise inclusion of counties in 1964–1987 and nationwide since 1987; Finland, 1975; Denmark, 1977; Iceland, 1999), and those who had died or emigrated within 5 years of the date of cancer diagnosis or an equivalent time lag for the population comparisons. These exclusions resulted in cohorts of 21,518 5-year childhood cancer survivors and 152,481 population comparison individuals (Fig 1).
The nationwide hospital registries hold information on all nonpsychiatric hospital admissions in the four countries [20,21]. Registration is mandatory, and the treating physician submits diagnostic information electronically. Each admission to a hospital initiates a record, which includes the personal identification number of the patient, the dates of admission and discharge, a primary discharge diagnosis, and supplementary diagnoses coded according to the International Classification of Diseases 7th–10th revisions (ICD-7–ICD-10).
Data on cancer survivors and comparison individuals were linked to the hospital registers, and a full hospital history with discharge diagnoses was established for each person with a previous hospital contact. We excluded cancer survivors and comparison individuals who had ever been hospitalised with a congenital malformation or chromosome abnormality (ICD-8 codes 740–759, ICD-10 codes Q00–Q99), as this could possibly confound the associations between childhood cancer and several of the outcomes, leaving 21,297 5-year survivors of childhood cancer and 152,231 population comparison individuals for the risk analysis (Fig 1).
To quantify the inpatient disease burden among study participants comprehensibly, we grouped the hospital discharge diagnoses into 120 disease categories or diagnoses, which in turn were assembled into 12 main diagnostic groups (S1 Table). The 12 main diagnostic groups were mutually exclusive; all neoplasms were grouped in the two main diagnostic groups of malignant neoplasms and benign neoplasms. Diagnoses coded according to ICD-7, ICD-9, and ICD-10 were adapted to ICD-8 to the extent possible, as shown in the table. We did not include the ICD sections of ill-defined diseases and the group of injuries and violence in the analysis, as these were regarded as too unspecific for solid conclusions. Mental disorders were the focus of a previous study [3], and childbirth and pregnancy complications require special considerations and will be investigated in separate publications.
Follow-up for diseases other than cancer was started 5 years after the date of cancer diagnosis for the cancer survivors and the corresponding date for the comparison individuals or at the beginning of the hospital registers (Denmark, 1977; Finland, 1975; Iceland, 1999; Sweden, from 1968–1987 stepwise inclusion of counties and nationwide since 1987), whichever occurred later. Follow-up for a second cancer in survivors and a first cancer in comparisons started at age 20 years at the earliest. Follow-up ended on the date of death, the date of emigration, or the end of the study (Iceland: 31 December 2008; Sweden: 31 December 2009; Denmark: 31 October 2010; Finland: 31 December 2012), whichever occurred first. As hospital registers do not reliably distinguish hospitalisations due to a relapse from those due to a primary cancer, we used the cancer registries for information on second primary cancers among childhood cancer survivors and first primary cancers among comparison individuals in analyses of hospitalisation risk. Thus, hospitalisations with childhood cancer recurrence were not included in the main analyses of hospitalisation risk.
Only the primary diagnosis, i.e., the main reason for hospitalisation at each inpatient contact, was included in the analyses. If participants had more than one hospital admission for a particular disease category, only the first record was retained. Risk was analysed for each of the 120 disease categories, and the numbers of first hospitalisations for somatic diseases in different categories were added up for the 12 main diagnostic groups. For each person, the final sum yielded the total number of first hospitalisations for diseases requiring hospitalisation in different categories. The observed number of first hospital admissions of survivors of childhood cancer for a given disease category was compared with expected numbers derived from the appropriate sex-, age-, and calendar period-specific hospitalisation rates of the comparison cohort, and the standardised hospitalisation rate ratios (RRs) were estimated. The significance and 95% confidence intervals (95% CIs) were computed using Fieller’s theorem and assuming that the observed number of first hospital contacts follows a Poisson distribution [22]. The absolute excess risk (AER), i.e., the additional risk for hospitalisation above background levels, was derived as the difference between the observed and expected first hospitalisation rates for a particular disease category per 100,000 person-years of follow-up. We did not stratify analyses by ethnicity, because this variable is not available in the Nordic population registers. However, the group of non-white individuals is historically of limited size in the Nordic countries and especially so among children.
Using the same methods as for the RR, we also added up the total number of bed days spent in hospital for cancer survivors and the number expected had they had the sex-, age-, and calendar period-specific bed day rates of the comparison individuals. We thus derived standardised bed day ratios (SBDRs) for cancer survivors. In the analyses of bed days, we included not only the first hospitalisation for a given disease category but all hospitalisations for diseases of the ICD sections of interest. Bed days for cancer recurrences were included in the analyses of SBDR.
Table 1 gives the main characteristics of the 21,297 5-year childhood cancer survivors included in the analysis. The survivors were monitored in the national hospital registers for 336,554 person-years (mean: 16 years; range: 0–42 years). Of the survivors, 27% (5,655/21,297) and 12% (2,546/21,297) were followed beyond the ages of 40 and 50 years, respectively.
Overall, 9,698 (45.5%) childhood cancer survivors were ever admitted to hospital for somatic disease, when 5,399.2 (25.4%) were expected, yielding an RR of 1.80 (1.76–1.84). The 9,698 survivors ever hospitalised experienced 21,325 first admissions to hospital for diseases in one or more of the 120 disease categories listed in S1 Table, when 10,999.0 were expected, yielding an overall RR for a new category-specific admission of 1.94 (Table 2). The underlying RRs were 1.83 (1.78–1.87) in Denmark, 1.87 (1.82–1.92) in Finland, 1.64 (1.35–1.99) in Iceland, and 2.10 (2.05–2.15) in Sweden. Based on the observed and expected hospitalisation rates of 6,336.3 and 3,268.1 per 100,000 person-years, respectively, the AER of survivors for a new category-specific admission to hospital was 3,068 per 100,000 person-years (Table 2). Thus, for each additional year of follow-up, approximately 3 of 100 survivors of childhood cancer were hospitalised for a new excess disease. Although the relative risk was significantly increased at all ages, the degree of increase diminished substantially with increasing age, i.e., from a relative risk of 3.4 in the age group 5–9 years to 1.3 in survivors aged 60 years or older. The absolute risk did not show a similar linear decline: after about 5,700 excess category-specific hospitalisations per 100,000 person-years in the age group 5–9 years, the AERs varied from 2,500 to 3,700 excess hospitalisations per 100,000 person-years for all subsequent age groups.
Fig 2 shows the relative risk for hospitalisation for somatic diseases belonging to each of the 12 main diagnostic groups; the highest risks were seen for diseases of the nervous system and sense organs (RR: 3.6 [3.4–3.7]), followed by diseases of the blood and blood-forming organs (RR: 2.8 [2.5–3.2]), endocrine diseases and nutritional deficiencies (RR: 2.8 [2.7–3.0]), and new primary cancers (RR: 2.6 [2.4–2.8]).
Table 3 shows that the pattern of excess hospitalisations among survivors is dominated by diseases of the nervous system and sense organs (AER: 603 per 100,000 person-years), followed by diseases of the endocrine system and nutritional deficiencies (AER: 349 per 100,000 person-years), diseases of the digestive organs (AER: 330 per 100,000 person-years), and diseases of the respiratory system (AER: 314 per 100,000 person-years). Together, these four main diagnostic groups constituted 50.6% (1,596/3,154) of all new excess hospitalisations, mainly for epilepsy (AER: 199 per 100,000 person-years), diseases of nerves and peripheral ganglia (AER: 157 per 100,000 person-years), pneumonia (AER: 144 per 100,000 person-years), and pituitary hypofunction (AER: 101 per 100,000 person-years). Table 3 shows the estimated relative and absolute risks of cancer survivors for each of the main diagnostic groups and a selected number of disease categories. The full table including all 120 disease categories is presented in S2 Table. Particularly high relative risks were seen for pituitary hypofunction (RR: 72.0; n = 341), testicular dysfunction (RR: 22..9; n = 19), tumours of the central nervous system (CNS) (RR: 11.8; n = 257), and herpes zoster (RR: 11.2; n = 58).
Fig 3A shows that the slight increase in overall AER seen in childhood cancer survivors over 40 years of age (Table 1) is due to marked increases in the AERs for diseases of the circulatory system, new primary cancers, and respiratory diseases with age. Moreover, the figure shows that the prevailing diagnostic groups during early life were diseases of the nervous system and sense organs and infectious and parasitic diseases. Except for a small peak among adolescent survivors for hospitalisation for diseases of the endocrine system, the AERs of the seven remaining main diagnostic groups did not show any appreciable variation by attained age (Fig 3B).
Survivors of all types of childhood cancers were at significantly increased risk for admission to hospital for subsequent disease (Fig 4). Survivors of neuroblastoma were at highest risk (RR: 2.6), followed by survivors of hepatic tumours (RR: 2.5), CNS tumours (RR: 2.4), Hodgkin lymphoma (RR: 2.4), and other lymphomas (RR: 2.3). Panels A—M in S1 Fig show the variations in the excess hospitalisation patterns among survivors of the 11 main groups of childhood cancer and of the two subgroups of lymphoma. For example, slightly more than half of all excess hospitalisations of survivors of CNS tumours were for subsequent diseases of the nervous system and sense organs and endocrine disorders and nutritional deficiencies, while infectious and parasitic diseases and diseases of the circulatory system were the primary reasons for hospitalisation of survivors of leukaemia and Hodgkin lymphoma, respectively.
The 9,698 cancer survivors who were ever hospitalised spent a total of 422,218 days in hospital whereas 85,071.7 were expected, yielding an overall SBDR of 4.96 (Table 4). The highest SBDRs were seen for survivors of hepatic tumours (SBDR: 9.9), followed by survivors of leukaemia (SBDR: 8.9), neuroblastoma (SBDR: 8.4), and CNS tumours (SBDR: 6.9). Fig 5 shows that most of the days spent by cancer survivors in hospital were for cancer recurrences and new primary cancers (SBDR: 24.2), followed by diseases of the nervous system and sense organs (SBDR: 5.8), diseases of blood and blood-forming organs (SBDR: 4.2), and benign neoplasms (SBDR: 3.4). The high overall SBDR for diseases of the nervous system and sense organs was due mainly to a particularly high SBDR of 15.9 (15.5–16.2) for epilepsy. Panels A—M in S2 Fig show the SBDR distributions for survivors of each of the 11 main groups of childhood cancer and the two subgroups of lymphoma, and S3 Fig shows the SBDR distribution for cancer recurrences and new primary cancers by main group of childhood cancer.
This population-based study of 21,297 5-year survivors of childhood cancer in the Nordic countries gives an extensive overview of the pattern of later somatic conditions that are serious enough to require inpatient care. The study shows that survivors are hospitalised because of a new somatic disease twice as often as population comparisons and that they spend five times as many days in hospital. Although cancer and its treatment may affect practically all organ systems adversely, the pattern of diseases requiring hospitalisation of survivors varied widely by type of childhood cancer and by survivors’ attained age. Despite the variations, however, the important findings are that the majority of childhood cancer survivors are at substantial risk for late effects requiring inpatient care and that the risk remains increased throughout life.
Our results concur with the mounting evidence of an excess risk for serious long-term morbidity in survivors of childhood cancer. Thus, in a clinical assessment, Geenen et al. identified chronic late effects in 74.5% of survivors treated in a single institution in The Netherlands [23]. Using self-reported outcomes, Oeffinger et al. showed that 62.3% of survivors included in the North American Childhood Cancer Survivor Study (CCSS) experienced at least one adverse, chronic late effect, equivalent to a 3.3-fold higher risk than their siblings [24]. This result is fairly consistent with the hospitalisation rate ratio for somatic disease of close to 2 seen in our study, taking into account differences in the sources of information and the maximum age at the end of follow-up (48 years in CCSS, lifelong in the Nordic study). In a later analysis of the CCSS material, Kurt et al. reported a survivor hospitalisation rate that was 1.6 times higher (95% CI 1.6–1.7) than that derived from a United States National Hospital Discharge Survey covering the period 1992–2005 [8].
In a few newer studies of the risk for late morbidity among survivors of cancer diagnosed during childhood, adolescence, and young adulthood (age range: 0–24 years), diagnostic information from hospital discharge registers was used as the outcome measure [9,10,12,13]. These studies, conducted in Utah, US; British Columbia, Canada; Scotland; and The Netherlands, included 1,499, 1,374, 5,229, and 1,382 survivors, respectively. They show a similar picture of substantial excess hospitalisation for a wide range of diseases, with increased cumulative numbers of days at hospital, as compared with population comparisons. Because of the substantially larger sample included in the present study and a follow-up period extending into middle age and beyond, we are able to provide a more detailed description of subsequent disease for a broader age range while maintaining relatively narrow 95% confidence limits. One notable difference between the four studies mentioned above and ours is that they included hospitalisations for psychiatric diseases, cancer recurrence, congenital abnormalities, injuries, suicide, and externally caused morbidities in the analysis, and the studies in British Columbia, Utah, and The Netherlands also included hospitalisations for pregnancy complications and delivery [9,12,13]. This probably explains the moderately higher overall risks for hospitalisation reported in those studies.
We used hospital-based diagnoses made by physicians as markers of disease outcome. Although this approach increased the validity of the diagnostic information, less severe late effects might have been missed because they were treated sufficiently as hospital outpatients or in the primary health care system. This implies that we almost certainly underestimate the absolute overall somatic disease burden experienced by childhood cancer survivors. As this limitation also applies to the comparison cohort, however, the validity of the relative risk estimates is acceptable, although restricted to conditions that require a hospital contact. Moreover, we cannot exclude the possibility that our results were affected by better medical surveillance of the survivors than the population comparisons, which could explain part of the longer hospital stays in the survivors. Not covered by the present study of the somatic disease burden but important to stress is the fact that many childhood cancer survivors face additional and sometimes significant challenges due to cognitive and other psychological adverse effects from cancer and its treatment [3,7].
The information on treatment currently included in the Nordic cancer registries is generally too crude or absent for meaningful analyses of type and dose of chemotherapy and radiation and specific disease outcomes. Although our study does not attribute causality, this comprehensive overview provides important clinical information on the lifelong inpatient disease burden experienced by childhood cancer survivors overall and of a number of patient characteristics, including type of childhood cancer, type of late effect, sex, and attained age. Associations, including dose-response effects, between specific treatment regimens and the risk of selected late effects are being addressed in clinical case-cohort studies within the ALiCCS cohort [14].
As our study is population based and includes a randomly selected comparison group and data from high-quality health registers, we consider our results valid for children treated for cancer in other countries with similar health care systems.
In conclusion, we found that survivors of childhood cancer have a highly increased long-term disease burden, with a broad range of late effects that require inpatient treatment and substantially longer stays in hospital as compared with the background population of similar age and sex. This will inevitably constitute a growing health care challenge for our society, affecting medical costs, and may profoundly influence the quality of life and life expectancy of childhood cancer survivors. Our findings underscore the need for continued follow-up of survivors, with particular focus on survivors of neuroblastoma, hepatic tumours, CNS tumours, Hodgkin lymphoma, and leukaemia. In particular, primary health care physicians should be aware of the risk for second primary cancers in patients who are childhood cancer survivors, as the relative risks for cancers are high and tumours may appear earlier in life than usual.
|
10.1371/journal.pntd.0002068 | Latent-Class Methods to Evaluate Diagnostics Tests for Echinococcus Infections in Dogs | The diagnosis of canine echinococcosis can be a challenge in surveillance studies because there is no perfect gold standard that can be used routinely. However, unknown test specificities and sensitivities can be overcome using latent-class analysis with appropriate data.
We utilised a set of faecal and purge samples used previously to explore the epidemiology of canine echinococcosis on the Tibetan plateau. Previously only the purge results were reported and analysed in a largely deterministic way. In the present study, additional diagnostic tests of copro-PCR and copro-antigen ELISA were undertaken on the faecal samples. This enabled a Bayesian analysis in a latent-class model to examine the diagnostic performance of a genus specific copro-antigen ELISA, species-specific copro-PCR and arecoline purgation. Potential covariates including co-infection with Taenia, age and sex of the dog were also explored. The dependence structure of these diagnostic tests could also be analysed.
The most parsimonious result, indicated by deviance-information criteria, suggested that co-infection with Taenia spp. was a significant covariate with the Echinococcus infection. The copro-PCRs had estimated sensitivities of 89% and 84% respectively for the diagnoses of Echinococcus multilocularis and E. granulosus. The specificities for the copro-PCR were estimated at 93 and 83% respectively. Copro-antigen ELISA had sensitivities of 55 and 57% for the diagnosis of E. multilocularis and E. granulosus and specificities of 71 and 69% respectively. Arecoline purgation with an assumed specificity of 100% had estimated sensitivities of 76% and 85% respectively.
This study also shows that incorporating diagnostic uncertainty, in other words assuming no perfect gold standard, and including potential covariates like sex or Taenia co-infection into the epidemiological analysis may give different results than if the diagnosis of infection status is assumed to be deterministic and this approach should therefore be used whenever possible.
| Dogs are a key definitive host of Echinococcus spp; hence, accurate diagnosis in dogs is important for the surveillance and control of echinococcosis. A perfect diagnostic test would detect every infected dog (100% sensitivity) whilst never giving a false positive reaction in non-infected dogs (100% specificity). Since no such test exists, it is important to understand the performance of available diagnostic techniques. We used the results of a study that used three diagnostic tests on dogs from the Tibetan plateau, where there is co-endemicity of E. granulosus and E. multilocularis. In this study opro-antigen ELISA and copro-PCR diagnostic tests were undertaken on faecal samples from all animals. The dogs were also purged with arecoline hydrobromide to recover adult parasites as a highly specific but relatively insensitive third diagnostic test. We used a statistical approach (Bayesian latent-class models) to estimate simultaneously the sensitivities of all three tests and the specificities of the copro-antigen and copro-PCR tests. We also analysed how some determinants of infection can affect parasite prevalence. This approach provides a robust framework to increase the accuracy of surveillance and epidemiological studies of echinococcosis by overcoming the problems of poor diagnostic test performance.
| An efficient and adequate diagnosis is at the core of effective surveillance, control and elimination programmes. For the effectiveness of such programmes, knowledge about test accuracies is indispensable, since even very accurate diagnostic tests might occasionally provide false positive and false negative test results. To diagnose canine echinococcosis, a number of tests are used including arecoline purgation, copro-antigen tests and detection of the presence of the parasite using a PCR analysis of the faeces (reviewed by [1]). Arecoline purgation, a well-established technique of high specificity, has frequently been used in the past. However, it is a laborious and potentially hazardous procedure and has been reported to show poor sensitivity [2]. Hence, alternative methods have been developed for the routine diagnosis of Echinococcus infection in dogs and other canids. These tests include copro-antigen ELISA and copro-PCR techniques, but they cannot be considered a gold standard. Only a necropsy of dogs followed by the sedimentation and parasite-counting technique can be considered close to a perfect gold standard, i.e., a gold standard with 100% sensitivity and 100% specificity. However, due to ethical reasons, this procedure cannot be used on the routine surveillance of dogs, since it would involve killing of a large number of affected as well as non-affected dogs. Even on a smaller scale, sacrificing dogs for the purpose of diagnostic test evaluation would potentially be impossible in a Buddhist country.
In the absence of a perfect gold standard in surveillance studies, the accuracy of new or alternative tests cannot be estimated robustly and without bias by comparing such test results of new or alternative results against an imperfect gold standard. For example, in the case of a well-established test with a sensitivity of less than 100%, samples which are falsely classified as negative by such a gold standard test, might be correctly detected as positive by a more sensitive alternative test, thus leading to a biased –in this case too low- estimate of the specificity of the alternative test. Given the absence of a perfect gold standard, however, test accuracies can be estimated robustly using latent-class analysis [3]. In this context, latent refers to the idea that the true disease status for each animal is unknown and needs to be estimated from the data. Hui and Walter proposed a model in which two tests with unknown test accuracies are applied to individuals from two populations to estimate sensitivities and specificities as well as prevalences. Their model can be extended to any combination of tests (R) and populations (S) as long as the condition of S≥R/(2R-1 – 1) is satisfied [3]. The Hui-Walter model relies on several assumptions, which if violated, may result in unreliable estimates [4]. The first assumption is that the tested individuals are divided into two or more populations with different prevalences. The second assumption is that sensitivities and specificities are constant across different populations. The third assumption is that test results are conditionally independent given the true disease status.
Echinococcus infections in dogs vary in parasite abundance and/or prevalence with age [5]. Diagnostic tests which are based on the detection of the parasite might be correlated if the number of parasites found affects the sensitivities of these tests [6]. Although covariance terms for conditional test dependency in the Hui-Walter latent-class model [7]–[10], have been included in a number of analyses that used a Bayesian approach, models using covariates to adjust for factors which might affect the sensitivity and specificity in different populations are scarce [11]. This contrasts with classic risk factor studies, where the outcome prevalence is routinely adjusted for covariates or confounders. In the case of infections with Echinococcus, covariates may include age or co-infection with other parasites such as Taenia spp.
Whereas the classical Hui-Walter model is based on the assumption of different populations which differ in their prevalences, in practice, it could be difficult to justify the splitting of one population into sub-populations. The separation of a population into different “prevalence populations” based on a factor which might interact with one of the tests (e.g., age or co-infection with another pathogen) is questionable [4], since sensitivities and specificities might not be constant in these different populations. Including a covariate instead offers in addition the assessment if this covariate is significantly associated with the prevalence and this association can be quantified in terms of an odds ratio.
The aims of this study were to obtain test accuracy and prevalence estimates for the diagnosis of Echinococcus granulosus and Echinococcus multilocularis in dogs in a highly endemic district of Sichuan province on the eastern Tibetan plateau. Three different tests were used for the diagnosis of E. granulosus and E. multilocularis infections in dogs, i.e. a genus-specific copro-antigen test, two species-specific copro-PCRs (one for each species of Echinococcus), and arecoline purgation. The results of these diagnostic tests on this population of dogs were used to estimate the diagnostic sensitivities and specificities of the tests using latent-class analysis. Age, sex and Taenia spp co-infection have been integrated as covariates in the latent-class models and their effects on the true prevalence have been assessed. The types of tests used and the nature of the data collected allowed for a full latent-class analysis. The use of covariates in the analysis and appropriate prior assumptions on the specificity of arecoline purgation enabled us to explore the dependence structure of these tests in this population of dogs. In addition, because we had parasite abundance data from the results of arecoline purgation, we were able to explore the hypothesis that the intensity of infection with Echinococcus spp affected the diagnostic sensitivity of copro-antigen ELISA and copro-PCR tests.
A total of 365 dogs from a highly Echinococcus-endemic region of the Eastern Tibetan Plateau in the People's Republic of China were sampled. Full details of the study animals and study area can be found in previous publications [12], [13]. Dog fecal samples were collected, and dogs subsequently received treatment, if their owners consented. Because the sampling was non-invasive, no prior ethical permission was sought. A table with data on test results classified according to Taenia co-infection is available in the supplementary online file (Table S1).
A Bayesian approach was used to obtain estimates for the test accuracies of the three tests. Initial analyses with non-informative priors as beta distributions (1, 1) were used for all parameters, except for the faecal counts of adult parasites following purge, where the specificity was set at 1. This was justified as all purge positive samples had been confirmed morphologically through microscopic examination. Conditional dependencies between tests were assessed by separately examining the impact of each of the 4 covariance terms. In the case of three tests with unknown sensitivities and specificities, three pairs of covariance terms are possible (between tests 1 and 2, tests 1 and 3 and tests 2 and 3) for both sensitivity and specificity. Fixing one test specificity to 1 results in two covariance terms becoming obsolete, since if one test has a specificity = 1, then the test specificities of the two other tests must be conditionally independent from the first test. Models allowing for age, sex or Taenia spp co-infection to be a covariate for prevalence were tested. In addition it was possible to examine the performance of the tests by fixing the specificity of PCR to 1 instead of and/or in addition to fixing the specificity of purge to 1. Model selection was performed by using the deviance- information criterion (DIC) [17]. The DIC is used as a criterion for goodness of fit of the model. Smaller DIC, with a difference of at least 2 indicate a better fit of the model.
For each model, the first 20 000 iterations were discarded as burn-in and the next 50 000 iterations were used to parameterize the model. Multiple chains were run from different initial starting points and checked for convergence. Models were fitted with the software JAGS (http://mcmc-jags.sourceforge.net/) version 2.2.0, the software R (R, 2010) and the package coda. The model code is given in the supplementary online material (Text S1). To explore the possibility that the intensity of Echinococcus infection affected the diagnostic sensitivity of other tests we also undertook the analysis after reclassifying the results of the arecoline purgation. Therefore two further analyses were undertaken. When purge results indicated that the intensity of infection was less than 20 parasites, the purge results were classified as negative and the analysis repeated. For the second analysis reclassification was undertaken when purge results indicated a parasite intensity of between 1 and 99 parasites.
The estimated test accuracies given as posterior means and their corresponding 95% credible intervals are presented in table 1 and the posterior density distributions in the figures S1, S2, S3, S4. In a Bayesian context, the results are given as posterior density or probability distributions which reflect, given the data and prior information, what would be the most probable parameter values. The reported results have the lowest DIC of a number of competing model estimates. A better model fit was obtained by including a covariance term for a conditional dependence between the sensitivities of the copro-antigen ELISA and the copro-PCR for E. multilocularis. The true prevalence of E. multilocularis infection in this population of dogs was estimated at 15.3% (95% credible intervals 10.3–21.8%) and the prevalence of E. granulosus was estimated at 11.1% (95% credible intervals 6.7–20.1%) without Taenia co-infection included as a covariate in the model. Taenia co-infection was a significant covariate with both E. multilocularis infection (odds ratio 2.06, 95% credible intervals 1.07–3.9) and E. granulosus infection (odds ratio 6.32; 95% credible intervals 2.8–15.2) (figures 1 and 2). The prevalence of E. multilocularis in Taenia test-negative dogs was estimated at 12.2% (95% credible intervals 7.6–18.9%), and in Taenia test-positive dogs was estimated at 22.3 (95% credible intervals 8.2–47.7%). The prevalence of E. granulosus in Taenia test-negative dogs was estimated at 4.1% (95% credible intervals 1.9–8%), and in Taenia test-positive dogs was estimated at 21.1% (95% credible intervals 5.1–56.9%).
Changing the cut-off for being classified as positive in the purge from at least one parasite detected to at least 20 or 100 parasites did affect the estimates of the sensitivities of the three tests differentially. The sensitivity of the purge decreased by a maximum of 13.9% (E. multilocularis) and 33.6% (E. granulosus). The sensitivity of the PCR decreased less by maximally 6.9% for E. multilocularis and 4.5% for E. granulosus. In contrast to this, the sensitivity of the copro-antigen ELISA increased by approx. 10% for both E. granulosus and E. multilocularis. The specificities for both Echinococcus species for the ELISA remained virtually the same and for the PCR decreased marginally by 6.6% for E. multilocularis and 2.9% for E. granulosus. Figures S5 and S6 show the effect of increasing the cut-off for a faecal sample to be classified as positive from at least 1 to at least 20 parasites detected. Increasing the cut-off leads to an increase in the posterior distribution of the sensitivity by approx. 10%. Results including the corresponding intervals are presented in table 2.
Fixing the specificity = 1 of the copro-PCR instead or in addition to the specificity of the purge did not affect the specificity of the copro-antigen ELISA. However, the sensitivities of the copro-antigen ELISA and the purge decreased (10 to 50%) (table 3). However, the deviance-information criteria indicated that this model was a poorer fit than allowing the specificity of the copro-PCR to vary. Sex was not a significant covariate in any analysis indicating the true prevalence of Echinococcus spp infection did not vary between male and female dogs (data not shown).
This study on the diagnosis of canine echinococcosis has used latent-class modeling to estimate the true prevalence and the diagnostic test performances of three tests for each Echinococcus spp. Arecoline purgation is a test that has been widely used in the past such as in the Echinococcus-elimination campaign in New Zealand [18] and for some transmission studies in central Asia [2], [19]. One previous study that used latent-class analysis suggested that the sensitivity of arecoline purgation was poor, perhaps as low as 38% and 21% for the diagnosis of E. granulosus and E. multilocularis infection respectively [2]. In the present study, the best fitting model (covariance with Taenia infection) suggested the sensitivity of arecoline purgation was much higher (table 1) with a sensitivity of over 75%. The specificity of the PCR test for the diagnosis of both parasitic infections converged on a lower value in the present study then in [2] where it was estimated as being 93% and 100% respectively. The two copro-PCR tests were not the same: the former study relied on egg isolation followed by PCR whereas the present studies omitted the egg isolation stage. However, it is important to reconcile these major differences. In the former study, there were only two tests used and the ability of the dog to roam as opposed to being tied all the time was a significant covariate. If the PCR test is fixed with a specificity of 100%, then the performance of the arecoline purgation drops markedly and is more similar to the values described in [2] (table 3). This indicates that the estimates of the performance of the arecoline seem to be highly dependent on the models' ability to classify PCR positives as true positives or allow for some false positives. The latter are important as there are a number of animals in both studies that are purge negative but copro-PCR positive. Indeed when the PCR tests were first developed by [15], the specificity was estimated at 100%. However this estimate was based on a sample of 10 dogs from non-endemic areas. In a naturally infected population false positive PCR results may occur due to coprophagia of faeces containing Echinococcus eggs. Thus eggs ingested in this manner might passage the intestine resulting in a positive PCR result indistinguishable from a result generated from parasite material coming from an established infection.
It should also be considered that the diagnostic performance may vary with the population of dogs. For example, the mean number of E. multilocularis parasites recovered from each dog in the present study is 131 (95% CI 62–375) [13], [14]. This is significantly higher than the mean number of 65 (95% CI 22–123) parasites recovered by purgation from the study by Ziadinov et al [2] that reports the much lower sensitivity of purgation. It is therefore possible that the dog populations from the two studies had substantially different parasite abundancies. In addition, when we reclassified the diagnostic test results as being only positive if there were at least 20 or 100 parasites, arecoline purgation was considerably less sensitive. Thus a higher mean abundance in the Tibetan population of dogs compared to the Kyrgyzstan population of dogs might also partly explain the considerable variation in the sensitivity of arecoline purgation between the two studies. When we reclassified the diagnostic test results as being only positive if there were at least 20 or 100 parasites, the sensitivity of the copro-antigen ELISA increased by approx.10% for both Echinococcus species. This might be explained by the ELISA performing better with higher parasite abundance in faecal samples. Previous studies have suggested that the sensitivity of copro-antigen increases as the intensity of infection increased [20]. There is little variation in the sensitivity of the PCR test regardless of which scenario is studied, indicating a sensitivity of approximately 89% for the diagnosis of E. multilocularis and 84% for the diagnosis of E. granulosus. This appears to be somewhat more sensitive than the test described in [2]. However, the previous test could only diagnose patent infections as it relied on prior egg isolation from the faeces. For patent infections the two tests are more comparable with the previous test able to detect an estimated over 87% and 72% for E. granulosus and E. multilocularis respectively. In another study with three tests based on antigen detection, DIC was also used as model selection criteria and indicated the same “best” models as likelihood ratio tests [21].
This analysis failed to find age of dog as a significant covariate and hence concluding that prevalence of Echinococcus infection does not vary significantly with age. This is consistent with the previous analysis [12], although age may affect abundance of E. granulosus in this group of dogs [13]. Relatively lower parasite abundances in older canids has been suggested to be a result of exposure to the parasite and immunity to reinfection or by variations in infection pressure with age [5] and changes in abundance may not accompanied by changes in prevalence.
The significance of Taenia as a covariate is to be expected as dogs are infected with both Taenia and Echinococcus spp through predation and local prey species are infected with metacestodes from both genera of parasites. In the previous analysis of this set of purge and faecal samples, significant correlations of abundance of Taenia and Echinococcus spp were found [12]. However, the present analysis failed to identify dog sex as a significant covariate. This is in contrast to a previous analysis of the same data using logistic regression and assuming that the results of arecoline purgation were definitive [12] which suggested that the prevalence in male dogs was higher than in females. The present study examined dog sex as a covariate in the latent-class analysis of diagnostic test performance and hence included the sensitivity of the arecoline purgation in the analysis. Thus a number of dogs in the previous study would have been misclassified and would have affected the results of the regression analysis. Techniques are now becoming available to incorporate the latent but unknown infection status in regression analysis [22] and these should be used where possible to avoid reaching inappropriate conclusions about the possible significance of covariates in epidemiological studies.
In conclusion the results of this study demonstrate how the unknown true prevalence of Echinococcus spp in dogs can be estimated if a number of diagnostic tests are used in parallel with a suitable covariate structure. It also demonstrates that an identical diagnostic test may have a considerable difference in performance between different study populations. Sensitivity and specificity are population-dependent [23] and the terminology of “intrinsic diagnostic test characteristics” implying that these are “constant and universally applicable” across populations should be discouraged [24]. Thus multiple tests should ideally be used routinely in the population of interest if no perfect gold standard is available. In contrast to (formerly) used approaches like Kappa tests to assess agreement of test results beyond chance, Bayesian latent-class approaches are more suitable to model the prevalence and associated influencing factors in a robust way. Finally using the true prevalence rather than the test prevalence may give different results with regard to the importance of determinants (such as Taenia in the case of this data set) which are associated with infection. This is due to misclassification errors following false positive or false negative test results when using test results in a deterministic manner.
|
10.1371/journal.ppat.1003425 | Identification of a Novel Antimicrobial Peptide from Human Hepatitis B Virus Core Protein Arginine-Rich Domain (ARD) | The rise of multidrug-resistant (MDR) pathogens causes an increasing challenge to public health. Antimicrobial peptides are considered a possible solution to this problem. HBV core protein (HBc) contains an arginine-rich domain (ARD) at its C-terminus, which consists of 16 arginine residues separated into four clusters (ARD I to IV). In this study, we demonstrated that the peptide containing the full-length ARD I–IV (HBc147-183) has a broad-spectrum antimicrobial activity at micro-molar concentrations, including some MDR and colistin (polymyxin E)-resistant Acinetobacter baumannii. Furthermore, confocal fluorescence microscopy and SYTOX Green uptake assay indicated that this peptide killed Gram-negative and Gram-positive bacteria by membrane permeabilization or DNA binding. In addition, peptide ARD II–IV (HBc153-176) and ARD I–III (HBc147-167) were found to be necessary and sufficient for the activity against P. aeruginosa and K. peumoniae. The antimicrobial activity of HBc ARD peptides can be attenuated by the addition of LPS. HBc ARD peptide was shown to be capable of direct binding to the Lipid A of lipopolysaccharide (LPS) in several in vitro binding assays. Peptide ARD I–IV (HBc147-183) had no detectable cytotoxicity in various tissue culture systems and a mouse animal model. In the mouse model by intraperitoneal (i.p.) inoculation with Staphylococcus aureus, timely treatment by i.p. injection with ARD peptide resulted in 100-fold reduction of bacteria load in blood, liver and spleen, as well as 100% protection of inoculated animals from death. If peptide was injected when bacterial load in the blood reached its peak, the protection rate dropped to 40%. Similar results were observed in K. peumoniae using an IVIS imaging system. The finding of anti-microbial HBc ARD is discussed in the context of commensal gut microbiota, development of intrahepatic anti-viral immunity and establishment of chronic infection with HBV. Our current results suggested that HBc ARD could be a new promising antimicrobial peptide.
| Antibiotics-resistant pathogens have been a major problem to our public health. Recently, in our studies of human hepatitis B virus (HBV), we accidentally discovered potent and broad spectrum antimicrobial peptides from HBV core protein (HBc) arginine-rich domain (ARD). The peptides are mainly composed of SPRRR repeats and are effective against both Gram-positive and Gram-negative bacteria, as well as fungi. We found different bactericidal mechanisms of the ARD peptides, which involved LPS binding, DNA binding and membrane permeabilization in various tested bacteria, such as P. aeruginosa, K. pneumoniae, E. coli and S. aureus. We also found that this ARD peptide was effective for colistin-resistant A. baumannii. The peptides exhibited no hemolysis activity to human red blood cells and no cytotoxicity to human hepatoma cells and kidney cells. Furthermore, the ARD peptide was shown to be safe and protective in the animal model. Recently, intestinal flora was found to influence the development of immunity. We discussed here the potential involvement of the antimicrobial activity of HBc ARD in the establishment of HBV chronic infection in the newborns. We proposed here that the HBc ARD peptides could serve as an alternative to the conventional antibiotics in clinical medicine.
| The increase of drug-resistant pathogens caused by the extensive use of traditional antibiotics is a serious concern worldwide. There is an urgent need to develop more effective treatment to overcome the drug-resistance problem. Antimicrobial peptides (AMP) are a new class of antibiotics with a new mode of action and remarkable therapeutic effects [1]. In general, they contain 10–50 amino acids, with an overall positive charge and an amphipathic structure. Under hydrophobic environment, AMPs can fold into four classes of structures, including α-helix, β-sheets, extended structures, and loops [1], [2], [3]. It is well known that most AMPs can directly bind to bacteria membrane and kill them by disrupting membrane or targeting intracellular components [3], [4], [5]. Most importantly, they are effective to antibiotics-resistant pathogens [6], [7]. This unique feature has encouraged the development of AMPs as novel antibiotics in the last few decades. To date, more than one thousand AMPs have been identified in various species including plants, insects, fish, frogs, and mammals [8], [9], [10], [11], [12], [13]. Although their sequences vary greatly, certain amino acids such as cysteine, lysine, proline or arginine are key compositions of AMPs [12], [14], [15], [16], [17].
Hepatitis B virus (HBV) remains a major human pathogen, and there are new challenges for the treatment of viral hepatitis B [18], [19]. HBV encodes a 21 KDa core (HBc) protein, which is essential for viral replication [20], [21], [22]. It contains a capsid assembly domain at N-terminus (residue 1 to 149) and an arginine-rich domain (ARD) at C-terminus (residues 150 to 183) [23], [24]. ARD contains 16 arginines separated into four arginine-rich clusters (ARD I, II, III, IV) and has a function of binding to nucleic acids. When it binds to HBV pre-genomic RNA or polyanions, HBc can assemble into a stable capsid [25], [26], [27]. In addition, ARD contains important signals for nuclear export and import of HBc core protein and particles [28]. We have found that the growth of E. coli expressing HBc1-183 was much slower than that of E. coli expressing HBc1-149 (unpublished results). It appears that it is HBc 150-183 that somehow retarded the growth of E. coli, and dramatically reduced the yield of HBc 1-183 protein. When we examined the sequences of HBc 150-183 in further detail, we noted that it shares a certain degree of sequence similarity with known antimicrobial peptides [29], [30]. This finding suggests the possibility that ARD may have the antimicrobial activity. In this study, we determined the in vitro antimicrobial activities of HBc147-183 against a wide variety of bacteria, including multidrug resistant (MDR) and colistin (polymyxin E)-resistant A baumannii. Using a peritoneal sepsis mouse model, we demonstrated further that ARD peptides can effectively protect all the mice challenged with a lethal dose of Staphylococcus aureus. Treatment of ARD peptide also caused significant reduction of bacterial load of S. aureus and K. pneumoniae in infected mice. Potential mechanisms for the bactericidal activity were investigated. The ARD peptides appeared to be capable of direct binding to the Lipid A moiety of lipopolysaccharide (LPS) in several different binding assays. We discussed further the potential significance of the anti-microbial activity of the HBc ARD peptide in the commensal microbiota and the development of the intrahepatic antiviral immunity in HBV infected newborns. In summary, with high antimicrobial activity and very low toxicity against human cells and animal models, these HBc ARD peptides may have a therapeutic potential in the future.
All animal experiments were conducted under protocols approved by Academia Sinica Institutional Animal Care & Utilization Committee (ASIACUC permit number 12-02-322). Research was conducted in compliance with the principles stated in the Guide for the Care and Use of Laboratory Animals, National Research Council, 1996.
The antimicrobial activities of HBc ARD peptides were tested using a number of bacterial strains from ATCC, including Pseudomonas aeruginosa Migula strain (ATCC 27853, ampicillin-resistant), Pseudomonas aeruginosa Migula strain (ATCC 9027, ampicillin-resistant), Klebsiella pneumoniae strain (ATCC 17593), Escherichia coli strain (ATCC 25922), Staphylococcus aureus subsp. strain (ATCC 25923, methicillin-resistant), Staphylococcus aureus subsp. strain (ATCC 29213, methicillin-resistant), Staphylococcus aureus subsp. strain (ATCC 19636, methicillinresistant), and Candida albicans strain (ATCC 10231).
Clinical isolates of Pseudomonas aeruginosa (NHRI-01, NHRI-02 and NHRI-04) were obtained through the program of Taiwan Surveillance of Antimicrobial Resistance (TSAR), National Health Research Institutes (NHRI), Taiwan. Acinetobacter baumannii (ATCC 17989, ATCC 17978 CR, ATCC19606, ATCC 19606 CR, TCGH 45530 and TCGH 46709) were obtained from Tzu-Chi Buddhist General Hospital (TCGH) in Taiwan, and clinical isolates (TCGH 45530 and TCGH 46709) were identified using the Vitek system (Biomerieux Vitek, Inc., Hazelwood, MO, USA) [31]. A. baumannii is defined as multidrug-resistant, when the organism is resistant to piperacillin, piperacillin-tazobactam, ampicillin/sulbactam, imipenem, ceftazidime, gentamicin, amikacin, tetracycline, chloramphenicol, ciprofloxacin, and cotrimoxazole [32]. Susceptibility to colistin was determined using the broth-dilution method, in accordance with the guidelines of the Clinical and Laboratory Standards Institute (CLSI) [33].
All peptides were purchased from Yao-Hong Biotechnology Inc. (Taipei, Taiwan). Vendors provided data of peptide characterizations, including HPLC and Mass (data not shown). Antimicrobial activity was determined as described [30] with some modifications as detailed below. Bacteria were grown overnight in Mueller–Hinton broth (Difco) at 37°C, and during the mid-logarithmic phase, bacteria were diluted to 106 CFU (colony formation unit)/ml in phosphate buffer (10 mM sodium phosphate and 50 mM sodium chloride, pH 7.2). Peptides were serially diluted in the same buffer. Fifty microliter (µl) of bacteria was mixed with fifty µl of peptides at varying concentrations followed by incubation at 37°C for 3 hours without shaking. At the end of incubation, bacteria were placed on Mueller-Hinton broth agar plates, and allowed growth at 37°C overnight for measurement of minimal bactericidal concentration (MBC). The lowest peptide concentration on the agar plate, which displayed no bacterial growth (zero colony), is defined as MBC. All peptides were tested in triplicate.
For measurement of killing kinetics, bacteria and peptides were prepared as described above. Fifty µl of bacteria were mixed with fifty µl of peptides at the concentrations corresponding to MBC and were incubated at 37°C. At the indicated time, bacteria were serially diluted and placed on Mueller-Hinton broth agar plates for the viability measurement.
The localization of peptide was monitored by confocal fluorescence microscopy. Bacteria were grown to mid-logarithmic phase and were collected by centrifugation. Approximate 107 CFU were resuspended in phosphate buffer containing FITC-labeled HBc147-183 at a concentration corresponding to 0.5×MBC. Following incubation for 1 hour at 37°C, cells were washed, fixed, and immobilized on poly-L-lysine coated glass slides. ProLong Gold antifade reagent with DAPI (Invitrogen) was added to the slides prior to mounting. Localization of labeled-peptide was observed using an Olympus Ultraview confocal microscopy equipped with a 100× oil immersion lens.
Briefly, bacteria (107 CFU) were prepared and mixed with 1 µM SYTOX Green (Invitrogen) for 15 minutes in the dark. After the addition of peptides to the final concentrations corresponding to their respective MBC, fluorescence intensity was measured at 37°C using wavelengths 485 nm and 520 nm filters for excitation and emission. Melittin (Sigma), the major toxin of bee venom, was used as a positive control to provide maximal permeabilization [34].
The proportion between amino nitrogen (NH3+) of HBc147-183 and phosphate (PO4−) of DNA was defined as N/P ratio [35]. Briefly, HBc147-183 was incubated with pSUPER plasmid DNA at different N/P ratio (0, 0.2, 0.4, 0.6, 0.8, 1, 2, 3 and 4) for 30 minutes at 37°C. The mobility of pSUPER plasmid DNA was analyzed by electrophoresis on 1% agarose gel.
Several kinds of peptide-LPS or peptide-Lipid A binding assays were performed in this study: 1) Streptavidine-conjugated beads (Dynabeads MyOne Streptavidin T1, Invitrogen) were blocked by P. aeruginosa LPS (Sigma) at 37°C for 1.5 hour. After washing with PBST (PBS, pH 7.4 containing 0.1% (w/v) Tween-20), aliquots containing 250 pmol streptavidine-conjugated beads were incubated with a reaction mixture overnight at 4°C. The reaction mixture was prepared by mixing increasing amounts of biotinylated peptide HBc147-183 (0, 0.004, 0.02, 0.1, 0.5 and 2.5 µM) and 5 µg/ml P. aeruginosa LPS (Sigma) or 200 µg/ml E. coli lipid A (Sigma), at 37°C for 3 hour. After incubation overnight at 4°C, the reduction of LPS (or Lipid A) in the supernatants were measured by the Limulus Amebocyte Lysate (LAL) test (Charles River Endosafe) with an ELISA reader (Molecular Devices). The amount (EU/ml) of LPS was calculated according to the standard curve prepared with Endosafe Control Srandard Endotoxin. 2) To directly measure the increasing amounts of LPS bound to the increasing amounts of peptide HBc147-183 on the peptide-coated beads, the beads were then washed with PBST three times and incubated with 100 µl of PBS containing 0.15 units of trypsin agarose (Sigma) for overnight digestion at 37°C. After trypsinization, the trypsin-released LPS in the supernatants were collected and measured by the Limulus Amebocyte Lysate (LAL) test. Similar results were obtained by another LPS testing method: Endosafe- PST Cartridges (Charles River Laboratories). 3) To perform the LPS/Lipid A competition assay, one µg of LPS was coated on High Binding ELISA plates (Corning) overnight at 4°C. The LPS-coated plates were washed by PBST and then blocked with PBST containing 5% BSA for 1 hour at 37°C. After washing, HBc147-183 (10 nM), mixed with varying concentrations of E. coli Lipid A (0 to 10 µg/ml), were added into each well and incubated for 1 hour at 37°C. Plate-bound HBc147-183 was measured by streptavidin conjugated with HRP (1∶10000 dilution) for 1 hour at 37°C. TMB substrates were added into each well for color development. The absorption was measured at 450 nm with a reference wavelength at 655 nm.
The hemolytic activities of peptides were determined by hemolysis against human red blood cells (hRBCs). Human blood was obtained in EDTA-containing tube and was centrifuged at 450 g for 10 min. The pellet was washed three times with PBS buffer, and a solution of 10% hRBCs was prepared. hRBCs solution was mixed with serial dilutions of peptides in PBS buffer, and the reaction mixtures were incubated for 1 h at 37°C. After centrifugation at 450 g for 10 min, the percentage of hemolysis was determined by measuring the absorbance at the wavelength of 405 nm of the supernatant. Blank and 100% hemolysis were determined in PBS buffer and in the presence of 1% Triton X-100, respectively.
Cytotoxicity was measured for HepG2, Huh7, HEK293, and Vero cells by MTT assay. Cells were seeded at 104 cells/well in a 96-well plate and serial dilutions of peptides were added into each well. PBS was used as a negative control and melittin was used as a positive control. After 1 hour of incubation, the medium were replaced by fresh medium containing 10% MTT solution (Promega), and the plate was incubated for 4 hours in 5% CO2 at 37°C. The absorbance at the wavelength of 595 nm was measured by an ELISA reader (Bio-Rad model 680).
To set up CFSE cell proliferation assay, 293 cells (human kidney origin) and Vero cells (monkey kidney origin) were resuspended in PBS to a final concentration of 106 cells/ml, before incubation with 10 µM CFSE dye (CellTrace CFSE cell proliferation kit, Invitrogen) at 37°C for 10 min. To quench the staining, ice-old culture media were then added and incubated on ice for 5 min. Labeled cells were then pelleted and washed three times by fresh medium containing 10% FBS before seeding into six well plates at a density of 3.3×105cells/well. After 20 h, the medium was removed and incubated with fresh medium containing 5, 25 and 100 µM HBc 147-183 for one hour (FITC-labeled ARD peptide had been largely internalized in 10 minutes after the addition of ARD peptides to the medium of HepG2 cells). Forty-eight hours later, cells were harvested and analyzed by flow cytometry (FACSCanto, BD Bioscience).
Three-week old male ICR mice (19 to 21 g) were purchased from BioLASCO (Taiwan). Overnight culture of bacteria in BHI broth (Difco) was subcultured in fresh BHI broth to log phase. Inoculums were diluted in BHI broth to indicated densities. To test the acute toxicity of ARD peptide in vivo, ICR male mice were inoculated intraperitoneally (i.p.) with 10 and 20 mg/kg HBc147-183 in PBS, respectively. Each group contained 5 mice. After peptide injection, the number of dead mice was recorded daily for 7 days post-injection. To test the antimicrobial activity of the ARD peptide in vivo, all mice were inoculated i.p. with Staphylococcus aureus ATCC 19636 (4×106 CFU/mouse) in BHI broth. Peptide HBc147-183 (10 mg/kg) was administered i.p. at 1, 1.5 and 2 hours post-inoculation. PBS (10 ml/kg) control was administered at 1 hour post-inoculation. Each group contained 10 mice. Mortality was monitored daily for 7 days post-inoculation. In a separate experiment to measure the bacterial load, mice were inoculated i.p. with Staphylococcus aureus ATCC 19636 (106 CFU/mouse) in BHI broth. All mice were administered at 1 hour post-inoculation with peptide HBc147-183 (10 mg/kg) or PBS (10 ml/kg) control, and sacrificed at 4 hours post-inoculation. Blood samples (200 µl) were mixed with 100 mM EDTA (10 µl) and were diluted 20-fold in PBS (calcium and magnesium free). Liver and spleen samples (0.1 g) were homogenized in sterile PBS (500 µl). Samples were diluted approximately 100-fold and plated on BHI agar for scoring the colony numbers. To test the in vivo antimicrobial activity of the ARD peptide against Gram-negative bacteria, mice were inoculated with Klebsiella pneumoniae Xen39 (107 cfu/mouse) (Caliper LifeSciences), an engineered strain containing a modified Photorhabdus luminescens luxABCDE operon. One hour post-inoculation, mice received either 10 ml/kg PBS (n = 5) or 10 mg/kg ARD peptide (n = 5), respectively. In vivo imaging was carried out at 4 hours post-inoculation. The mice were anesthetized first before transferring to the IVIS imaging system (IVIS spectrum), and luminescence was measured with an exposure time of 1 minutes or less. The image system measured the number of photons and translated the data to false color images that depicted the region of strong luminescence with red, moderate luminescence with yellow and green, and mild luminescence with blue. Decreasing bioluminescence indicated reduction of bacteria. The images were overlay of photographic images and bioluminescence using a computer-generated color scale. Total flux (RLU) of region of interest (ROI) was quantified by the IVIS imaging software.
As shown in Figure 1 and Table 1, HBc147-183 displayed a broad-spectrum activity against Gram-negative bacteria (P. aeruginosa, K. pneumoniae and E. coli), Gram-positive bacteria (S. aureus), and fungi (C. albicans). Among these tested strains, P. aeruginosa and K. pneumonia were the most sensitive to this peptide. The MBCs of HBc147-183 were lower than 4 µM for P. aeruginosa and K. pneumonia, and around 4 µM for E. coli, and S. aureus. C. albicans was the least sensitive to this peptide (MBC ∼8 µM).
To further map the active sequences of the antimicrobial activity, various peptides (Figure 1) in different length were synthesized and tested as before. Peptide HBc147-175, with the deletion of the last eight amino acids at the C-terminus, maintained strong activity against Gram-negative bacteria, albeit it lost the activity against S. aureus and C. albicans. We detected no activity against all of the tested bacteria and fungi from peptides ARD I–II (HBc147-159) or ARD III–IV (HBc164-176 and HBc162-175). In contrast, all peptides containing ARD II–IV (HBc153-176, HBc157-176, HBc153-175, HBc155-175, and HBc157-175) and ARD I–III (HBc147-167) exhibited strong activity against P. aeruginosa and K. pneumonia, respectively, albeit they were weak against E. coli (Table 1). Therefore, peptide ARD II–IV and ARD I–III appeared to be necessary and sufficient for the bactericidal activity against P. aeruginosa and K. pneumonia.
Our phophorylation studies on serine residues S155, S162, S170, S176 and S181 revealed that serine phosphorylation in general weakened the potency of antimicrobial activity. For example, as shown in Table 1, we found that all HBc peptides, once phosphorylated, lost their activities against C. albicans. For bacteria, the phosphorylation on S181 showed no effects, whereas phosphorylations on S155, S162, S170, and S176 reduced the antimicrobial activity. The MBCs dropped to 8 µM for HBc155p and HBc176p, and 32 µM for HBc162p and HBc170p, respectively. When S155, S162 and S170 were simultaneously phosphorylated (HBc155p162p170p), the antimicrobial activity was completely lost (>32 µM). The results here suggested that, except for S181, serine phosphorylation is generally detrimental to the antimicrobial activity of HBc ARD peptide. To confirm the importance of arginine residues for bactericidal activity, we synthesized and tested peptide HBc147-183-III–IV AA, which has two R-to-A substitution mutations in each of ARD III and ARD IV. Similar to phosphorylated HBc ARD peptides, the MBC of HBc147-183-III–IV AA was significantly increased compared to HBc147-183. The result here indicated that arginine residues are required for the antimicrobial activity.
Cationic peptides, such as polymyxin B and E (colistin), have become one of the last options for multidrug resistant bacteria these days [36]. We therefore tested the antimicrobial activity of HBc147-183 against colistin-resistant P. aeruginosa and A. baumannii. As shown in Table 2, while HBc147-183 killed colistin-sensitive P. aeruginosa at 4 µM, colistin-resistant P. aeruginosa were cross-resistant to HBc147-183 (MBC>16 µM). In contrast to P. aeruginosa, the MBCs of HBc147-183 against colistin-sensitive and colistin-resistant A. baumannii are in a similar range of 0.5–1 µM. This result indicates that, for colistin-resistant A. baumannii, there is no cross-resistance to our ARD peptide HBc147-183.
Time course of bacterial viability was determined after the tested bacteria (P. aeruginosa, K. pneumonia, E. coli and S. aureus) were treated with HBc147-183 at the concentrations corresponding to the MBC (Figure 2). The results showed that P. aeruginosa was immediately killed within 20 minutes upon the addition of HBc147-183 (2 µM). Although K. pneumonia and E. coli were members of Gram-negative bacteria, they were killed by 4 µM HBc147-183 in 180 minutes. For S. aureus, complete killing by 4 µM HBc147-183 was observed in 120 minutes.
As shown in Figure 3, P. aeruginosa, E. coli and S. aureus were treated with FITC-labeled HBc147-183 corresponding to 0.5×MBC, and the localization of HBc147-183 was visualized using confocal fluorescence microscopy. The results showed that, upon peptide treatment, P. aeruginosa, K. pneumonia and E. coli appeared as hollow rods with fluorescence clearly defined bacteria surface, suggesting that HBc147-183 was accumulated on the membrane (Figure 3A–D). To understand better the effect of HBc peptides on the membranes, SYTOX Green uptake assay was performed. As shown in Figure 4A, a significant degree of membrane permeabilization was induced on P. aeruginosa upon the addition of 2 µM HBc147-183. Although it was also accumulated on the membrane of K. pneumonia and E. coli, 4 µM HBc147-183 was not able to induce membrane permeabilization as observed on S. aureus. Consistent with the bactericidal activity of HBc147-183 against P. aeruginosa, HBc153-176 caused the same membrane permeabilization within 10 minutes in a dose-dependent manner (Figure 4B). The results indicated that the bactericidal effect of HBc peptides on P. aeruginosa is directly through the membrane permeabilization with a fast kinetics similar to that of killing kinetics (Figure 2). On the other hand, HBc147-183 was found to penetrate through the membrane of S. aureus and localized in the cytoplasm (Figure 3E–G). In order to investigate the potential interaction between HBc 147-183 and DNA, HBc147-183 was mixed with pSUPER plasmid DNA at different N/P ratio (Materials and Methods) and analyzed by gel electrophoresis (Figure 4C). The results showed that the mobility of DNA was decreased when the ratio of peptide/DNA increased and the plasmid DNA was completely retarded at the ratio of 1, suggesting that HBc147-183 has a strong binding activity to plasmid DNA. Overall, it suggests that the bactericidal mechanisms of HBc147-183 on Gram-positive and Gram-negative bacteria may be completely different (see Discussion for further detail).
To determine whether LPS of Gram-negative bacteria could serve as a potential target of HBc147-183, LPS (0.05 to 50 µg/ml) from either P. aeruginosa or E. coli (Sigma) were incubated with both P. aeruginosa and 2 µM HBc147-183 for three hours, respectively. The results showed that the bactericidal activity of HBc147-183 was significantly reduced by the addition of either LPS at the concentration of 50 µg/ml (Figure 5). In addition, HBc147-183 preferentially bound to the LPS from P. aeruginosa, rather than that from E. coli. However, the addition of anti-LPS antibody (Genetex Co.) cannot sufficiently neutralize the bactericidal activity of HBc147-183 (Figure 5). Taken together, it suggests that HBc147-183 could bind to not only LPS but also other target molecules on the membrane. Alternatively, HBc147-183 and the anti-LPS polyclonal antibody used here could bind predominantly to two different epitopes on the LPS.
As shown in the cartoon illustration of Figure 6A, the potential interaction between HBc147-183 and LPS (or Lipid A moiety) in vitro was investigated using several different binding assays (Materials and Methods). In Figure 6B, when increasing amount of HBc147-183 was bound to the strepavidine-conjugated Dynabeads and allowed incubation with constant amount of LPS, gradually increasing amount of LPS appeared to be depleted from the supernatant. HBc147-183 8p, which has eight ser/thr phosphorylations, was used in parallel as a control peptide. Similar results were obtained by another LPS testing method: Endosafe-PTS Cartridges (Charles River Laboratories). In Figure 6C, the beads-captured LPS were dissociated from the beads by trypsin agarose digestion of the ARD peptide HBc147-183. The amount of released LPS was measured by the LAL test (Materials and Methods). LPS contains mainly the polysaccharide and Lipid A moieties. To determine whether the ARD peptide can bind to Lipid A directly, we tested in Figure 6D the binding between Lipid A and the ARD peptide in a manner similar to Figure 6B. As expected, increasing amounts of HBc147-183 on the beads led to decreasing amounts of Lipid A remaining in the supernatant. The inverse correlation between the ARD peptide on the beads and the Lipid A in the supernatant (Figure 6D) is strikingly similar to what was observed previously between the ARD peptide HBc147-183 on the beads and LPS in the supernatant (Figure 6B). To directly demonstrate that the ARD peptide can bind to the Lipid A moiety of LPS, we performed a competition experiment between Lipid A and LPS (Figure 6E). The LPS-coated ELISA plates were incubated with constant amount of HBc147-183 (10 nM), which was premixed with varying concentrations of E. coli Lipid A (0 to 10 µg/ml). After extensive washing, plate-bound (i.e., LPS-bound) biotinylated peptide HBc147-183 was measured by streptavidin conjugated with HRP, followed by adding TMB substrates and color development (Materials and Methods). Indeed, the binding of HBc147-183 to LPS was significantly decreased by the increasing concentrations of Lipid A (Figure 6E). The result here lends support for the notion that Lipid A moiety of LPS can serve as a direct target for ARD peptide HBc147-183.
To determine the cytotoxicity of HBc peptides, we measured the hemolytic activity of HBc147-183. Compared to the melittin control, no detectable hemolysis by HBc147-183 was observed after one hour of incubation (Figure 7A). In addition, MTT assay was performed to determine the cytotoxicity of HBc147-183 to human hepatoma (Huh 7 and HepG2 cells) and kidney cells (Vero and HEK293 cells). The viability of cells treated with melittin at low dose (3.125 µM) was significantly decreased. In contrast, HBc147-183 caused only a low level of cytotoxicity at the concentration of 100 µM (Figure 7B). The CFSE cell proliferation assay was also performed to determine the effect of HBc147-183 on the proliferation of Vero and HEK293 kidney cells. In comparison to day 1, CFSE intensity of cells treated with HBc147-183 (5, 25 and 100 µM) decreased to the same level as the mock control on day 3 (Figure 7C), suggesting that ARD peptide HBc147-183 has no significant effect on cell proliferation.
To conduct the experimental infection with bacteria, we i.p. inoculated mice with Staphylococcus aureus ATCC 19636 (4×106 cfu/mouse). Bacterial load in blood at 1, 2, 4 and 6 hours post-inoculation was determined. As shown in Figure 8A, bacteria rapidly transferred to the blood compartment from peritoneal cavity. Within 2 hours, the number of bacteria in the blood achieved the maximum (106 cfu/ml). Thereafter, the number of bacteria in the blood gradually decreased spontaneously. To distinguish the ARD peptide-mediated from the spontaneous clearance, we therefore tested the in vivo protection activity of the ARD peptide within 2 hours post-inoculation. Briefly, mice were i.p. inoculated with Staphylococcus aureus ATCC 19636 and received a single dose of 10 ml/kg PBS or a single dose of 10 mg/kg ARD peptide at 1, 1.5, 2 hours post-inoculation, respectively. Mice (n = 10) treated with PBS died within 24 hours post-inoculation (Figure 8B). In contrast, administration of ARD peptide (10 mg/kg) at 1 hour post-inoculation can effectively protect all mice (n = 10) from death at day 7. When we administered ARD peptide at 1.5 (n = 10) and 2 (n = 10) hours post- inoculation, survival rates were decreased to 70% and 40%, respectively. Instead of using death as a surrogate indicator of the antimicrobial activity of ARD peptide, we also determined directly the in vivo effect of ARD peptide on bacterial load of infected mice (Figure 8C). Mice were inoculated with Staphylococcus aureus as before and treated with 10 ml/kg PBS (n = 5) or 10 mg/kg ARD peptide (n = 5) at 1 hour post-inoculation. Four hours post-inoculation, bacterial load in blood, liver and spleen samples of control mice were in the range of 106 cfu/ml (Figure 8C and 8D). Administration of ARD peptide significantly reduced the bacterial load (∼104 cfu/ml) by 100-fold in blood, liver and spleen than the PBS control mice (P<0.01). In addition to Staphylococcus aureus, we also examined the in vivo antimicrobial activity of ARD peptide on K. pneumoniae using an IVIS imaging system. Similar to the change in bacterial load of S. aureus, bioluminescence of mice inoculated with K. pneumoniae Xen39 peaked at 2 hour post-inoculation (data not shown). We then treated K. pneumoniae Xen39-infected mice with PBS or ARD peptide at 1 hour post-inoculation, respectively. The results showed that the bioluminescence of ARD peptide-treated mice was very weak, whereas PBS control showed a more extensive bioluminescence (Figure 9A). There was a significant difference in the overall RLU values of mice treated with PBS versus ARD peptide (P<0.01) (Figure 9B). Taken together, the results indicated that HBc147-183 exhibited significant antimicrobial activity in vivo.
In this study, we present a novel antimicrobial peptide (HBc147-183) isolated from the C-termial domain of HBc. The computer program, based on the antimicrobial peptide database [37], predicted unfavorably that HBc147-183 could serve as an antibacterial peptide, due to its very low content of hydrophobic amino acids. In contrast to the computer prediction, surprisingly, HBc147-183 exhibited a broad-spectrum antimicrobial activity.
Many clinical isolates of P. aeruginosa, K. pneumonia, A. baumannii and S. aureus are highly pathogenic and are resistant to aminoglycosides, beta-lactams, and fluoroquinolones [38], [39], posing a serious threat to human health. In the past decades, polymyxins, such as polymyxin B and colistin (polymyxin E), were considered the “Last Hope” antibiotics, and were increasingly used in clinical settings to treat multidrug resistant bacteria [40], [41]. However, polymyxin-resistant bacteria have also emerged most recently, and the studies of adaptive resistance to polymyxin have been reported [36], [42]. In Table 2, we tested the bactericidal activity of HBc147-183 to colistin-resistant P. aeruginosa and A. baumannii. While colistin-resistant P. aeruginosa exhibited cross-resistance to ARD peptide HBC147-183, we found a strong activity of HBc147-183 (MBC = 0.5–1 µM) against all tested colistin-resistant A. baumannii. Therefore, it appears that A. baumannii and P. aeruginosa may have adopted different strategies to acquire resistance to colistin. It would be interesting to investigate further in the future whether the so-called two component regulatory systems, such as ParR-ParS and CprR-CprS, could contribute to the colistin-induced cross-resistance to the HBc ARD peptide [41], [42], [43]. It has been proposed that Lipid A modification could be responsible, at least in part, for the resistance to polymyxin in P. aeruginosa [43]. In this regard, it is noteworthy that our ARD peptide could bind to Lipid A of E. coli and LPS of P. aeruginosa (Figure 6). It might be relevant to compare the lipid A structures between colistin-resistant P. aeruginosa and A. baumannii in the future (Table 2). Most importantly, our studies here open up the possibility that ARD peptide could be used for treatment of colistin- resistant A. baumannii in the future.
As shown in Table 1, the bactericidal activity of phosphorylated peptides and mutant peptide (Arg to Ala) were significantly reduced. In addition, a highly phosphorylated peptide (HBc147-183 8p) showed significantly reduced binding activity to LPS compared to non-phosphorylated HBc147-183 (Figure 6B and 6D). It suggested that arginine residues and positive charge are very important for the activity. While only the full-length HBc147-183 (ARD I–IV) was effective against the tested Gram-positive bacteria, S. aureus, ARD II–IV (HBc153-176) and ARD I–III (HBc147-167), in a less than full-length context, exhibited strong activity against Gram-negative P. aeruginosa and K. pneumoniae, respectively (but not E. coli). Consistently, phosphorylation of S162 and S170 resulted in much weaker activity against P. aeruginosa and K. pneumoniae, while phosphorylation at several other positions showed no apparent attenuation effects (Table 1). In summary, a minimal amount of both arginines and positive charge of HBc ARD peptides appeared to be important for effective bactericidal activity against these different Gram-positive and Gram-negative bacteria.
It is surprising that the ARD domain of HBc protein (HBc147-183) exhibits novel and broad spectrum antimicrobial activity. This potent peptide shares some degree of similarity with several antimicrobial peptides in literature, such as protamine (PRRRRSSSRPVRRRRRPRVSRRRRRRGGRRRR) [29] and Drosocin (GKPRPYSPRPTSHPRPIRV) [44]. Protamine is a polycationic peptide found in the nuclei of sperm of different animal species [29]. It consists of four arginine clusters. Radial diffusion assay has shown that a single arginine-rich domain (RRRR) is sufficient for antimicrobial activity, especially against Gram-negative bacteria [16]. Unlike protamine, the arginine-rich domain of HBc147-183, such as ARD I–II and ARD III–IV, were not sufficient for the antimicrobial activity. In addition, sequence alignment by anitimicrobial peptide database revealed that HBc153-176 shares 44% amino acid sequence homology with Drosocin, which is a proline-rich peptide isolated from Drosophila. However, except for P. aeruginosa, Drosocin is predominately active against most Gram-negative bacteria. Drosocin kills bacteria via an apparently non-membranolytic mechanism [44], [45]. Taken together, HBc ARD is a novel peptide with a broad spectrum bactericidal activity quite distinct from the other known arginine-rich antimicrobial peptides.
AMPs can act by several mechanisms, including permeabilization of the membrane, as well as inhibition of the synthesis of protein, DNA, or cell wall [12]. Gram-negative bacteria have been shown to contain receptors for antimicrobial peptides, such as lipopolysaccharide [46], and membrane proteins, such as OprI and Lpp [47], [48]. Our results here showed the membrane localization of HBc147-183 on Gram-negative bacteria (Figure 3) and the neutralization activity of LPS from either P. aeruginosa or E. coli (Figure 5). It suggests that HBc147-183 could have a strong binding activity to LPS. Indeed, based on several in vitro binding assays, our results revealed a direct binding of HBc147-183 to LPS and Lipid A (Figure 6B, 6C and 6D). Furthermore, Lipid A moiety of LPS was shown to be one major direct target of HBc147-183 (Figure 6E). However, incubation of LPS antibody with P. aeruginosa and HBc147-183 failed to neutralize the bactericidal activity of HBc ARD peptide (Figure 5). One interpretation for this negative result, among several others, is that HBc147-183 could bind not only LPS but also some other molecules on the bacterial membrane.
Previous studies have shown that the LPS of P. aeruginosa and E. coli have diverse structures of lipid A, a core component of LPS [49], [50]. Indeed, we found a better neutralization effect of the LPS from P. aeruginosa than that from E. coli (Figure 5). The preference of binding by HBc147-183 for the LPS of P. aeruginosa is correlated with its stronger bactericidal activity against P. aeruginosa.
Previously, it has been shown that AMPs causing membrane permeabilization exhibited fast killing kinetics, while AMPs and antibiotics targeting intracellular components exhibited slow killing kinetics [51]. The mode of action of HBc147-183 on P. aeruginosa could be related to membrane permeabilization based on the fast killing kinetics (Figure 2) and its membrane localization (Figure 3A and 3B). This speculation is also supported by the results of SYTOX Green uptake experiment (Figure 4A). Like P. aeruginosa, HBc147-183 was also accumulated on the membrane of K. pneumonia and E. coli. However, the killing kinetics and SYTOX Green uptake experiments of K. pneumonia and E. coli did not support for a mechanism of membrane permeabilization. Bactericidal mechanisms other than membrane permeabilization can be cited. For example, mammalian peptidoglycan recognition protein (PGRP) has been reported to kill bacteria by activating protein sensing two-component systems [52]. It remains to be further investigated how bacteria can be killed by the ARD peptides using a mechanism other than membrane permeabilization.
In the case of Gram-positive bacteria, we found that HBc147-183 was not accumulated on the membrane (Figure 3). Instead, it can enter the cytoplasm of S. aureus without any apparent development of membrane permeabilization (Figure 4A). In addition to LPS, HBc147-183 can also bind strongly to plasmid DNA (Figure 4C). Taken together, the bactericidal mechanism of HBc147-183 against Gram-positive bacteria appeared to be more similar to Buforin II, which was reported to kill bacteria by binding to DNA and RNA after penetrating bacterial membrane [53].
Other arginine-rich peptides, such as Penetratin [54], Tat peptide [55], and oligoarginine [56], have been reported to be able to enter the mammalian cells. Although HBc147-183 can penetrate through the cell membrane of Huh 7 and HepG2 cells (data not shown), we observed no significant cytotoxic effect on human hepatoma cells Huh 7 and HepG2, as well as kidney cells Vero and HEK293, even at high peptide concentration (100 µM) by MTT assay (Figure 7B) and proliferation assay (Figure 7C). Taken together with the results from the hemolytic assay (Figure 7A), HBc147-183 appears to be much safer relative to melittin in cell culture [57]. Indeed, in our animal model study, we observed no apparent in vivo toxicity of ARD peptide at 20 mg/kg dose in the ICR mice by i.p. injection (Figure 7D). In fact, at as low as 10 mg/kg level, treatment of ARD peptide can protect mice from death (Figure 8B). In contrast, all mice receiving the PBS control were dead soon after bacterial inoculation. In addition to the sepsis survival model, treatment of ARD peptide (10 mg/kg) also resulted in a significant reduction of bacterial load of S. aureus and K. pneumoniae, whereas PBS control mice showed high levels of bacterial load (Figure 8C–D and Figure 9). The results demonstrated the in vivo antimicrobial potency of HBc ARD peptide.
|
10.1371/journal.pcbi.1000903 | Pavlovian-Instrumental Interaction in ‘Observing Behavior’ | Subjects typically choose to be presented with stimuli that predict the existence of future reinforcements. This so-called ‘observing behavior’ is evident in many species under various experimental conditions, including if the choice is expensive, or if there is nothing that subjects can do to improve their lot with the information gained. A recent study showed that the activities of putative midbrain dopamine neurons reflect this preference for observation in a way that appears to challenge the common prediction-error interpretation of these neurons. In this paper, we provide an alternative account according to which observing behavior arises from a small, possibly Pavlovian, bias associated with the operation of working memory.
| The theory of Reinforcement Learning (RL) has been influential in explaining basic learning and behavior in humans and other animals, and in accounting for key features of the activity of dopamine neurons. However, perhaps due to this very success, paradigms that challenge RL are at a premium. One case concerns so-called ‘observing behavior’, in which, at least in some versions, animals elect to observe cues that are predictive of future rewarding outcomes, although the observations themselves have no direct behavioral relevance. In a recent experiment on observing, the activity of monkey dopaminergic neurons was also found to be incompatible with classic RL. However, as is often the case, this was a task that allowed for potential interactions from a secondary behavioral system in which responses are directly triggered by values. In this paper we show that a model incorporating a next order of refinement associated with such Pavlovian interactions can explain this type of observing behavior.
| Animal behavior all too rarely follows the precepts of simple theories such as normatively optimal choice. Prominent examples of this arise in the florid fancies of Breland & Breland's animal actors [1], or in the complexities of negative automaintenance or omission schedules [2]–[4]. Such failures and irrationalities have been important sources of theory revision and refinement, for instance leading to suggestions about the competition and cooperation of multiple systems of control [5]–[7], some instrumental and adaptive; others Pavlovian and hard-wired.
In this paper, we study one apparent departure from optimality, namely a type of ‘observing behavior’ [8], [9], which has been the subject of a recent important electrophysiological study [10]. In brief, subjects are programmed to receive either a large or small reward, with its size being determined stochastically. When faced with the choice of finding out (by being presented with a suitably distinctive cue) sooner rather than later which of the two rewards they will ultimately receive, subjects prefer to know sooner. A lack of indifference despite the equality of the outcomes has been found to be widely true even if the knowledge cannot influence the outcome, and, at least in other experiments, even if this choice is expensive [8], [9], [11]–[13]. In economics, the same anomaly is referred to in terms of “temporal resolution of uncertainty” [14], explained by such notions as savoring [15]–[17], with subjects enjoying the anticipation of good things to come.
The correct interpretation of this form of observing behavior has been the subject of substantial debate (see, e.g. [9]). Superficially attractive theories, such as a desire to gain Shannon information [18] have been dealt fatal blows, for instance with animals preferring to observe more even when the number of bits they receive by doing so is less (e.g., as the probability of getting the large reward becomes smaller than , [12]).
A recent study on observing behavior in macaques [10] has offered a new perspective on the problem. These authors recorded from putative dopamine neurons in the midbrain whilst monkeys chose to observe. According to a common theory, these neurons report a temporal difference error in predictions of future reward [19], [20] as in reinforcement learning accounts of optimal instrumental choice [21]. Bromberg-Martin and Hikosaka [10] showed that: (a) the macaques did observe; and furthermore (b) the activity of dopamine neurons was associated with the choice they make. However, although the behavior and activity are mutually consistent, observing behavior offers no instrumental benefit and therefore it should also not be associated with any prediction errors. Bromberg-Martin and Hikosaka suggested that this means that the dopamine cells are reporting on some aspects of the benefit of information gathering in addition to aspects of reward.
In this paper, we examine the extent to which this form of observing behavior can be explained by temporal difference learning, coupled with the same mechanism that provides an account of a wide range of departures from normative choice, namely a Pavlovian influence over instrumental actions [4]. In particular, we assume that subjects only make associative predictions when they are appropriately engaged in the task. If the level of this engagement is influenced by the size of the predictions (the putatively Pavlovian effect), then stimuli predicting certain or deterministic large future rewards (one outcome of an observing choice) will lead to more engagement than stimuli that leave uncertain the magnitude of the future rewards. This idea can be seen as a realization of the suggestion made by Dinsmoor [9] that the predictions of future reward associated with stimuli influence the attention paid to them. We show that occasional failures of engagement, modeled as a breakdown in the working memory for the representational state, can lead directly to both the preference for observing and the apparently anomalous dopamine activity, without need for any reference to ‘information’. We also examine the various factors that control the strength of observing in this model.
Bromberg-Martin and Hikosaka's experiment (see Methods and Figure 1) involved the most precise conditions for establishing observing behavior. On each trial, thirsty subjects had a 50% chance of receiving a small or large volume of water directly into their mouths. There were three sorts of trials: forced-information, forced-random and free choice. On forced-information trials, the subjects were presented with a single target (C; just an orange square in the figure) and, after looking at it, would receive one of two cues (S; an orange ‘+’, or S; an orange ‘’) according to the volume they were to receive in a couple of seconds. On forced-random trials, looking at the single target (C; green square) led again to one of two cues (S; green ‘*’, or S; green ‘o’). However, either of these could be followed by either small or large rewards; and thus they provided no discriminative information about the forthcoming reward. Finally, on free choice trials, both orange and green targets were provided, and the subjects could choose whether to receive the discriminative (orange) or non-discriminative (green) cues.
Figures 2a;b show primary behavioral results from the study for two subjects – both gradually expressed a bias towards the discriminative (orange) option in the free-choice trials. As Bromberg-Martin and Hikosaka stressed, under a standard associative learning or temporal difference scheme, there is no difference between the expected reward for the discriminating and non-discriminating option, and so no reason to expect this strong and enduring preference.
We built a model of this which, with one critical exception that we discuss below, involves a standard temporal difference learning algorithm [21], [22]. Forced-choice and free-choice trials permit learning about the future expected rewards associated with the various targets and stimuli, training the values of the states. Then, on free-choice trials, the selection depends on the relative values, via a softmax function (see methods). Figure 2c;d shows the results from simulations of our model, with parameters chosen to match Bromberg-Martin and Hikosaka's two subjects. The model closely matches qualitative features of the monkeys' performances.
In standard models such as this, in which there is a delay between the presentation of cues and the rewards that they predict, an assumption has to be made about the way that the subjects maintain knowledge about their state in the task, and indeed keep time. Many different possibilities have been explored, from delay lines to complex patterns of activity evolving in dynamical recurrent networks (e.g., [23]–[28]). All of these amount to forms of working memory – and so present the minimal requirement that the subjects continue to be engaged in the task throughout the delay in sufficiently intense a manner as to maintain this ongoing memory. Thus the critical exception to conventional temporal difference learning in our model is to assume that this maintained engagement is influenced by the current predicted value. That is, if the value is high, then engagement is readily maintained; if the value is low, then engagement can be weakened or lost.
Losing engagement is detrimental to the subject in the context of the present task; by analogy with a similarly detrimental effect in negative automaintenance, we consider it a form of Pavlovian misbehavior [4]. Pavlovian responses are typically elicited in an automatic manner based on appetitive or aversive predictions, and can exert benign or malign influences over the achievement of subjects' apparent goals. Normally, such responses are overt behaviors; here, along with several recent studies [29], [30], we consider internal responses, associated with the operation of working memory. Mechanistically, these could come, for instance, from the influence dopamine itself exerts on the processes concerned [31].
In the model, we consider engagement to be lost completely on some trials as a stochastic function of the evolving predicted value. Such losses have the effect of decreasing the subjective value of cues and states associated with lower values below their objective worth; in particular exerting a negative bias on the non-discriminative cues (S; S) compared with the discriminative cue associated with the large reward (S), which will more rarely experience such losses. Figure 3 shows the effective probability of disengagement at different timepoints as well as showing the effect this has on the expected reward. Disengagement associated with S is benign, since the outcome on those trials is modelled as being close to in any case. Altogether, this creates a bias towards choosing the discriminative option on free-choice trials, as is evident in Figure 2c;d.
The difference between the parameters for Figures 2c;d is in the parameter governing the strength of the competition in the softmax ( and for Figure 2c;d respectively). Monkey V's results are consistent with a larger value of than monkey Z; smaller leads to more stochasticity and a lower overall degree of preference. The asymptotic preference for observing is monotonic in .
Bromberg-Martin and Hikosaka [10] also recorded the activity of putative midbrain dopaminergic cells during the performance of the task. Figure 4a shows the activity of an example neuron in the various conditions. The population response is similar (Figure 4 of [10]) albeit, as has often been seen, with an initial brief activation to the forced choice non-discriminative case, likely because of generalization [32]. Firing at the time of the discriminative or non-discriminative cues (marked ‘cue’) and the delivery or non-delivery of reward (‘reward’) is just as expected from the standard interpretation of these neurons, i.e., that they report the temporal difference prediction error in the delivery of future reward [19], [20].
However, it is their activity at the time of the targets indicating the forced-informative or forced-random trials (marked ‘target’) that is revealing about observing. The target indicating a forced-informative trial was associated with a small but significant phasic increase in activity; whereas that indicating the random cues was followed by a small decrease in the firing rate. Under the temporal difference interpretation of the neurons, this is consistent with the preference exhibited by the monkeys, but not with the objective value of the options.
Figure 4b shows modelled dopamine activity in the variable engagement temporal difference model (here, negative prediction errors have been compressed compared with positive ones, see methods; [33], [34]). This shows exactly the same pattern shown in the monkey data. Note that, once the subject has learned the associations and learned the preference for choosing the discriminative option in the free choice trials, these trials will overall be more frequent than the forced-random trials, and so the negative prediction error associated with the latter will be larger than the positive prediction error associated with the former.
Figure 5 decomposes the modelled responses in the cases that there is successful and failed engagement between cues and reward or non-reward. The most significant effect of the complete failure to engage given an non-discriminative cue, is that if the large reward is provided, then there is a greater response than expected from a 50% prediction. The possibility of using this to test the theory is discussed below.
In a version of the task that involved choice between immediate or delayed information about upcoming rewards, Bromberg-Martin and Hikosaka [10] further showed that switching the colors of the cues without warning led to a slow reversal of the observing choice (Figure 6a;b). Figure 6c;d shows the same for the model using identical softmax parameters to those in Figure 2c;d. The switch in preference evolves at a similarly glacial pace.
Various other features of observing can be examined through the medium of the model. Figure 7a;b show the consequence of the reinforcing outcome being aversive (e.g., an electric shock) rather than appetitive. One key question in this case is whether failure to engage is controlled more by salience or valence. Figure 7a shows the former case, for which a prediction of a large punishment also protects engagement (symmetrically with reward; inset plot). In this case, subjects prefer the random to the discriminative cues, since disengagement leads to subjective preference. Such preference for random cues might also come from adding a fixed value to all the potential rewards, thus allowing the moderately large disengagement in S to have a subtractive value on its expected values (Bromberg-Martin, personal communication, 2010). However such an effect would likely be small.
Figure 7b shows the case in which valence (from appetitive to aversive) determines disengagement, with predictions of punishments leading to more failures of engagement than small rewards. This again supports observing behavior. Unfortunately, experimental tests of the case involving punishment [35] have not enjoyed the precision of the paradigm adopted by Bromberg-Martin and Hikosaka, leaving open the question as to which of these patterns arises.
Another important experimental manipulation has been to vary the probability of the larger versus the smaller reward. As decreases from 1 towards 0.5 there is an increase in the observing bias (i.e., a greater tendency to choose the discriminative option). Below this, the nature of the bias depends on the assumption about how the choices are generated. A choice rule that depends on the difference in expected values () leads to a bias that ultimately decreases towards as these values themselves decrease towards . However, the bias is asymmetric about (black curve in Figure 7c). If, instead, the choices are based on the ratio of the values (), the choice bias can continue to increase as approaches (red curve). Just such an increase in observing was shown by Roper and Zentall [12] as reward schedules thinned. While some studies have also manipulated the size of the reward [36]–[38], our model does not make any direct predictions about this. It is possible that adaptation would scale the response to the overall sizes of available rewards (as indeed found for phasic dopamine activity in [39]), and the metrics of this would have to be known in order to make predictions about disengagement.
One extra factor that is important for analysing behavior is that the biases inherent in disengagement are small and develop over a long time-scale, consistent with the stately progress evident in Figure 2. However, this means that the initial course of learning can be subject to significant influence from the initial values ascribed to the different options, leading to biases that are incommensurate with the final, long term, state. Figure 7d shows an example. For the blue curve, the initial values of all states are low (), but the probability of a reward is high (); for the red curve, the initial values are high (), but the probability of a reward is low (). In the former case, there is substantial initial over-observation; in the latter, initial under-observation.
We have provided an account of ‘observing behavior’ that shows how it can arise from a small Pavlovian bias over instrumental behavior associated with disengagement from a task, rather than any aspect of information seeking. Pavlovian biases are rife in decision-making; and accommodating them does not necessitate any further change to the standard underlying theory of the activity of dopaminergic neurons that has not already been suggested to accommodate other data. What we have done here is specify the shape of such an interaction based on disengagement in the task. We intended specifically to capture [10] experiment on macaques. However our results do touch upon other, but emphatically not all, instances of observing in the literature.
Experiments such as [10] into observing are designed to maximize the effects of what is a relatively small anomaly in decision making (compared, for instance, with the more extreme misbehavior evident in negative automaintenance [2] or the schedule task [40]). Indeed, in this case, the subjects did not have to pay a penalty for observing. Thus, under standard decision-making conditions, we may expect the net effect of disengagement to be modest, leaving near-optimal behavior within the scope of the model.
Dinsmoor [9] suggested an account of the phenomenon based on his observation of ‘selective observing’, i.e., that the subjects would preferentially focus on stimuli associated with higher probabilities of reward. This idea met some resistance (some of which is contained in the commentary to [9]), partly based on experimental tests in which the subjects were not able to avoid the low value predictive cues. Our account can be seen as a form of selective observing, but involving internal actions associated with the allocation of engagement and attention, rather than external actions involving preferential looking. It might seem that these accounts are close to Mackintosh's [41] suggestion that attention is preferentially paid to stimuli that are strong predictors of affectively important outcomes. However, in Mackintosh's account, attention particularly influences the speed of learning (the associability of the stimulus) rather than the fact of it (at least in the absence of competing predictors), and so would not have the asymptotic effect that is apparent in the experiments we have discussed.
Another interesting account of observing is Daly and Daly's DMOD [42], which learns predictions associated with frustration (when reward is expected, but does not arrive), and courage (when reward is actually delivered during a state of frustration). These extra predictions warp the net expected values associated with the different cases in observing, favoring observing responses. The theory underlying DMOD is the original Rescorla-Wagner [43] version of the delta rule [44], whose substantial modification by Sutton and Barto [45] to account for secondary conditioning led to the original prediction error treatment of the activity of dopamine neurons in appetitive conditioning [19]. It would be necessary to extend DMOD in a similar way, and to make an assumption about which of its three prediction errors (or other quantities) are reflected in the activity of dopamine neurons, in order to determine its match to the neurophysiological data. The failure of TD models to capture behavioral aspects of frustration is, however, notable.
To some tastes, the most theoretically appealing accounts of observing start from the notion that animals seek to acquire information about the world [46]. However, formal informational theories have difficulty with the results of reducing the probability of reward (Figure 7c; [12]), which reduce the uncertainty and the information gained, but increase observing. More informal theories, such as that suggested by [10] require more precise specification to be tested against accounts such as the one here. The sloth of initial learning and reversal apparent in Figure 6 (taking 1200–2400 choice trials, 3000–7000 trials overall) might be considered suggestive evidence against an informational account, since it implies at the very least a nugatory value for the information.
In terms of our account, there are various routes by which predicted values could influence persistent engagement. Failure to engage can be seen as the same sort of malign Pavlovian influence over behavior that is implicated in the poor performance of monkeys in tasks in which they know themselves to be several steps away from reward [40], [47]. In that paradigm, it is an explicitly informative cue that the reward is disappointingly far away that leads to disengagement; this parallels the disappointment associated with the non-discriminative cue in observing. The most obvious mechanism associated with engagement is the influence of dopamine itself over working memory [31]; however, whether this is the phasic dopamine signal associated with prediction errors for reward [19] or a more tonic dopamine signal associated with a longer term average reward rate [48], [49] is not clear. Alternatively, some theories suggest that working memory is controlled by a gating process [29], [30] associated with the basal ganglia, treating internally- and externally directed action in a uniform manner. Dopamine certainly influences the vigor associated with external actions [48]–[50]; it is therefore reasonable to assume that it might also influence internal engagement.
We specialized our description of the model to the particulars of the experiment conducted by Bromberg-Martin and Hikosaka [10]. The most important question for other cases concerns the conditions under which re-engagement occurs. Since disengagement is seemingly rather rare, it is hard to get many hints from this experiment, and we might assume that it is reward delivery itself that causes re-engagement. However in a more general setup (e.g. without reward delivery at fixed time points), a mechanism for re-engagement is necessary. One possible way to do that would be by stochastically re-engaging based on either the reward prediction error or expected value. Such a mechanism of re-engagement could happen at any time point but would be extremely likely to happen at the delivery of reward, as well as for the initiation of a new trial. To be fully generalizable we also need to specify the case for disengagement at the time of an action selection. While in a disengaged state we envision the animal not performing an explicit choice, thus potentially not responding within an allocated time. If a choice is required to progress in the behavioral setup it would happen after an eventual re-engagement.
The model raises some further questions. First, we assumed that the probability of disengagement is a function of the actual prediction. However, it is possible that this function scales with the overall magnitude or scale of possible rewards, making the degree of observing relative rather than absolute. There is a report that phasic dopamine itself scales in an adaptive manner [39], , and this would be a natural substrate.
A second issue is whether disengagement is occasioned by the change in predictions associated with the phasic dopamine activity, or the level of the prediction itself. If the former, then in tasks such as the one studied by Bromberg-Martin and Hikosaka [10], where substantial prediction errors only happen with phasic targets and cues, the state could, for instance, just be poorly established in working memory at the outset, because of a weak dopamine signal, and this could lead to a subsequent chance of disengagement. We adopted the simpler scheme in which it is the ongoing predictive value that controls the chance of disengagement. One experiment that hints in the direction of change is that of Spetch et al. [52] (for a more recent study see [53]). In this, pigeons were given the choice between a certain (100%) or uncertain (50%, but observed) reward. Surprisingly, the level of engagement to the latter (measured by the number of pecks to the illuminated key) was many times to that of the former, and the pigeons duly made the suboptimal choice. The model presented in this paper does tie engagement to choice in a similar way, but we would be unable to explain such a strong effect. A variant of the model for which engagement is governed by prediction errors rather than predictions would show some contrast effect that could favor the uncertain, but observed, reward. However, it would be hard to explain such a stark contrast.
A third issue is whether disengagement is complete (and stochastic), or partial (and, at least possibly, deterministic). We considered the former case, and indeed, this leads to a straightforward prediction that the histogram of the dopamine response at the time of a delivered reward in the non-discriminative case might have two peaks; one associated with continuing engagement to the point of reward; the other, which would be roughly twice as high, associated with prior disengagement. However, it is also possible that less dramatic changes in engagement occur during the interval between cues and reward. If many individual neural elements are involved in the engagement (for instance in working memory circuits devoted to timing), then some could disengage before others. This might even lead to a non-uniform behavior among different dopamine cells. Unfortunately, the low firing rates of these cells make it hard to discriminate between these various possibilities.
Finally, the question arises as to the computational rationale for value-dependent disengagement. Other instances of Pavlovian misbehavior, such as withdrawal from cues associated with predictions of low values, can find plausible justifications in terms of evolutionary optimality. Disengagement might be seen in the same way, as a Pavlovian spur to exploration [54] in the face of poor expected returns.
From the perspective of conditioned reinforcement, our account suggests that the issue that is often studied is not really the one that is critical. Various investigators (see, for instance, the ample discussion in Lieberman et al. 1997 [55] about the differences between their findings and those of Fantino and Case 1983 [56]) have considered whether stimuli like S are conditioned reinforcers because of their association with the reward. For us, S and S and S are all conditioned reinforcers. The key question for observing behavior is instead an apparent concavity: the average worth of two different stimuli associated deterministically with small and large rewards is greater than the worth of a single stimulus associated stochastically with the same outcome statistics (see [57]). It is this non-linearity that demands explanation, and not merely the fact, for instance, of savoring or anticipation of the future reward, which could quite reasonably also be purely linear. Some accounts put the weight of the non-linearity onto the stimulus associated surely with the large reward. By comparison, our account places this emphasis onto the non-discriminative stimuli, suggesting that they are more likely to lead to disengagement. The same is true of other sources of non-linearity, for instance a mechanism that accumulates distress from the prolonged variance/uncertainty in the non-discriminative pathway.
Various versions of the ‘observing task’ have also been tested on humans [55], [56], [58]. These studies have shown consistent observing behavior, but, partly because of the different reading of the issue of conditioned reinforcement to the one discussed above, have often focused on different questions and methods from those in Bromberg-Martin and Hikosaka [10]. For instance, one question has been whether subjects would observe if they only ever found out S and never S – the idea being that conditioned reinforcement could support observing of the latter but not the former. Unfortunately, the answers have been confusing [55], perhaps partly because of issues about how cognitive effects (e.g., expectations of controllability) influence the results. Note, in particular, that we have only modeled observing behavior associated with repeated experience and learning, and not the sort of single-instance decisions that are often used in human cases.
In conclusion we have shown that the often observed effect of ‘observing’, preferring a behaviorally irrelevant discriminating stimulus cue, can readily be explained by a bias caused by Pavlovian misbehavior, putting it in the same category as a range of other suboptimalities. Informational accounts, however seductive, are not necessary.
We model value learning using a modified version of a standard temporal difference model [21], [22]. We assume the task can be specified as a Markov process, where the participant estimates the expected long run future reward (value) of each state as , updating it according to(1)where is the learning rate, and is the change in expected value given by:(2)where is the delivered reward, and is the state that follows . Learning proceeds for all three sorts of trials (forced disc., forced non-disc. and choice trials). The modelled dopamine signal for Figures 4 and 5 is .
The only deviation from the standard TD model is in assuming that the correct updating of this system is dependent on maintaining engagement, for instance in working memory. We assume the probability of disengagement of the course of state to be(3)per unit of time (in seconds). Hence, for a given state the probability of a correct updating is given by , where is the amount of time spent in the state (see Figure 1). and are fixed parameters. We assume the consequence of disengagement to be the transition to a specific fixed (non-updating) state of value and hence the updating signal for is(4)
The system stays in this state, until a reward is delivered at the end of the trial. At this point the system is ‘re-engaged’ creating a TD error relative to the fixed state (see Figure 5). We assume that any potential disengagement in the intertrial interval is negated by the initiation of a new trial.
Choice is only possible at one state , between progressing to either state and state . Given the learned values, we assume the subject performs choice based on the Softmax or Luce choice rule [59](5)
Note that it is straightforward to see that this version of softmax is dependent on the difference in values (), whereas using the logarithm of the value (as in Figure 7c) causes the function to be dependent on the ratio of values ().
In the limit without any failures in updating the learned values would approach the true value , where the expectation is taken over states . However with a chance of failure dependent on the value, the iterative solution in Figure 7c can be given by solving(6)numerically.
For all figures we assumed and . For Figs. 2 and 6 we used parameters, and . For the aversive stimuli in Figure 7a–b we assumed negative reward values. For Figure 7a the parameters were . For Figure 7b the parameters were . For Figure 7d the parameters were . To mimic the fact that dopamine neurons have less dynamic range for increases than decreases in firing rate, for Figure 4 we truncated the negative responses at −25 percent of the maximal positive response of the neuron.
|
10.1371/journal.pntd.0004150 | Mass Administration of Ivermectin for the Elimination of Onchocerciasis Significantly Reduced and Maintained Low the Prevalence of Strongyloides stercoralis in Esmeraldas, Ecuador | To evaluate the effect of ivermectin mass drug administration on strongyloidiasis and other soil transmitted helminthiases.
We conducted a retrospective analysis of data collected in Esmeraldas (Ecuador) during surveys conducted in areas where ivermectin was annually administered to the entire population for the control of onchocerciasis.
Data from 5 surveys, conducted between 1990 (before the start of the distribution of ivermectin) and 2013 (six years after the interruption of the intervention) were analyzed. The surveys also comprised areas where ivermectin was not distributed because onchocerciasis was not endemic.
Different laboratory techniques were used in the different surveys (direct fecal smear, formol-ether concentration, IFAT and IVD ELISA for Strongyloides stercoralis).
In the areas where ivermectin was distributed the strongyloidiasis prevalence fell from 6.8% in 1990 to zero in 1996 and 1999. In 2013 prevalence in children was zero with stool examination and 1.3% with serology, in adult 0.7% and 2.7%.
In areas not covered by ivermectin distribution the prevalence was 23.5% and 16.1% in 1996 and 1999, respectively. In 2013 the prevalence was 0.6% with fecal exam and 9.3% with serology in children and 2.3% and 17.9% in adults.
Regarding other soil transmitted helminthiases: in areas where ivermectin was distributed the prevalence of T. trichiura was significantly reduced, while A. lumbricoides and hookworms were seemingly unaffected.
Periodic mass distribution of ivermectin had a significant impact on the prevalence of strongyloidiasis, less on trichuriasis and apparently no effect on ascariasis and hookworm infections.
| Strongyloides stercoralis (Ss) is a soil-transmitted helminth (STH) that is not yet targeted by control programs, although it is highly prevalent in many areas of the world and may cause severe consequences, in particular to immunosuppressed patients, with a high fatality rate. Unfortunately, albendazole, the drug most commonly used for the control of the other STH (hookworm, Ascaris lumbricoides and Trichuris trichiura) has little effect on Ss. The drug of choice, ivermectin, has been extensively used in mass drug administration (MDA) for the filarial worms Onchocerca volvulus and Wuchereria bancrofti. In the province of Esmeraldas, in Ecuador, we studied Ss (and other STH) prevalence from 1990 (prior to MDA initiation) to 2013 (6 years after MDA cessation) in rural communities where MDA was regularly executed for onchocerciasis compared with neighboring communities where ivermectin was not distributed because onchocerciasis was not present. Ss prevalence remained high over the years in the areas with no MDA, while in those with MDA prevalence fell to zero, and remained very low 6 years after MDA cessation. A less important effect was observed for T. trichiura. Adding ivermectin to MDA programs for STH would importantly contribute to the control of Ss infection.
| The Onchocerciasis Elimination Programme of the Americas (OEPA) is a regional initiative with the goal of eliminating morbidity and interrupting transmission of river blindness in six endemic countries in the Americas [1]. In Ecuador, the main endemic focus of onchocerciasis was in the province of Esmeraldas, North of the country, where the elimination of onchocerciasis was achieved thanks to a community-based, sustained effort, based on the mass drug administration (MDA) of ivermectin once a year from 1991 to 2000, then twice, from 2001 to 2007 in all communities along Santiago river where onchocerciasis was endemic (while in the other endemic areas the treatment was carried on until 2009). Ivermectin was administered to adults of any age and to chidren over 5 years. A high level of coverage (ranging from 81.9% to 98.0% at each treatment instance) was obtained until 1996 [2], to increase in the following years to >90% in most treatment sessions and until the end of the program [3].
There is no doubt that ivermectin is also effective on Strongyloides stercoralis [4] and community-wide preventive chemotherapy has been proposed [5]. This paper is the first published evidence to our knowledge that such a strategy would be successful. Objective of this work is to estimate the effect of ivermectin MDA on the prevalence of S. stercoralis as well as of other soil transmitted helminths (STH) (Ascaris lumbricoides, Trichuris trichiura and hookworm) through: a) comparison of results from fecal surveys carried out before, during and after the distribution program; b) comparison of results from fecal and serological surveys carried out in areas targeted versus areas not targeted to ivermectin treatment, during and after the onchocerciasis elimination program.
The study area (Fig 1) is Borbon district, a rural tropical forest area located in the basin of the Cayapas, Santiago and Onzole rivers, bordering at a longitude of 78°30’ and 79°05’ West and a latitude 1°12’ and 0°35’, North, in its extreme limits, and at an altitude at sea level under 40 m.
The zone is inhabited by Afro-descendant people and the indigenous Chachi people, although there is a segment of mestizo population linked to a process of land colonization from other provinces of Ecuador.
The main economic activities are agriculture, fishing, and exploitation of forest resources. According to the data of the System of Social Indicators of Ecuador, in this area, 94.5% of the population are under the poverty line and 52.7% is in extreme poverty, currently only 16.36% of families have access to piped water at homes, 59.9% have access to latrines but only 3.6% are connected to sewage.
Five, cross-sectional surveys were carried out by CECOMET—Esmeraldas, Universidad Nacional of Quito and CTD—Negrar, Italy, with the field collaboration of the local health district Director and personnel, from 1990 to 2013, in areas along Santiago river where ivermectin was distributed for onchocerciasis elimination (Oncho-areas) and in areas around the village of Borbon where ivermectin was not distributed as they were not affected by onchocerciasis (Non-oncho-areas) (Fig 1).
The laboratory method used was microscopic examination of formol-ether concentrated stool at CTD; serology for S. stercoralis with Immune Fluorescence Antibody Test (IFAT) was also performed at CTD (in-house method described in detail elsewhere) [6]. IFAT cutoff titer for certain cases was >80 (the test is virtually 100% specific at this titer) [6]. Final result was defined as certain in case of positive stools AND/OR serology above the predetermined cutoff (1/80), possible in case of negative stools and positive serology (but below the cutoff), and negative in case of negative stools and serology. For logistical reasons, serology was only performed in children of the untreated area.
Laboratory methods used were: microscopic examination of formol-ether concentrated stool at CTD; serology for S. stercoralis with IVD-ELISA (S-stercoralis serology Microwell ELISA Kit, IVD Research Carlsbad, CA) was carried out at the reference laboratory for parasitology of Universidad Nacional of Quito. The method is described in detail elsewhere [6]. Cutoff for certain cases was ≥0.5 Optical Density (the test is virtually 100% specific at this OD) [6]. Final results were defined as certain, possible or negative according to fecal and serologic results, similarly to 1996.
The timeline of the different surveys and of ivermectin MDA is reported in Fig 2.
In all the surveys the laboratory personnel received pre coded, anonymous samples and were unaware of any characteristics of the subjects, including their origin.
Data was colleted in Excel and analyzed using the software STATA IC 14 (StataCorp, 4905 Lakeway, College Station, TX, 77845, USA—www.stata.com).
All variables considered were categorical. The absolute, relative and (when indicated) cumulative frequencies were calculated with the respective 95% Confidence Intervals (CI).
Informed consent was obtained from members of the communities and from the children’s parents or guardians, before the collection of any biological samples. Consent was obtained verbally for surveys carried out from 1990 to 1999, that were not submitted to the Ethics Committee, as they were not primarily intended as research, but rather as screening and treatment programs promoted by the Ministry of Health, including School Health program. All exams were carried out free of charge, results were made available to all study subjects and treatment was also offered free of charge, when indicated, and according to the best local practice guidelines. The Ethics Committee of the Central University of Ecuador in Quito (“Comité de Bioetica”—COBI) approved the study protocol in December, 2012, including the last survey (for which a written informed consent was obtained from all the study subjects and/or their guardians) and the retrospective analysis of the previous ones. A minor amendment was approved in November, 2013.
Longitudinal data in adults in Oncho-areas from 1990 to 1996 showed that prevalence fell from 6.8%; (C.I. 3.5–12.8) in 1990 (just before the start of MDA with ivermectin) to 0 (C.I. 0.0–3.3) in 1996 (Table 1).
In children in 1996, the prevalence evaluated with formol-ether concentration was 0 (CI 0.0–3.3) in Oncho-areas and 23.5% (C.I. 18.2–29.8) in Non-oncho-areas (Table 2).
In the same year, serology (IFAT) for S. stercoralis was carried out in Non-oncho-areas only. Results at different cutoff are reported in Table 3.
The results were further classified according to the combined result of fecal examination and IFAT as defined in Materials and Methods. Prevalence was estimated between 35.5% (C:I: 29.2–42.4) considering certain cases and 67.5% (C.I: 60.7–73.6) if we included also possible cases (Table 4).
Results from cross-sectional studies in children, year 1999, Oncho-areas versus Non-oncho-areas are shown in Table 5. In Oncho-areas, S. stercoralis prevalence in children evaluated with formol-ether concentration was 0 (C.I. 0.0–3.6), while in Non-oncho-areas the prevalence was 16.1% (C.I. 10.8–22.8).
In 2013, six years after the conclusion of mass treatment program, and 14 years after the last survey, the prevalence of S. stercoralis measured using stool examination in children in Oncho-areas was 0% (C.I. 0.0–1.4) (Table 6).
By serology, the prevalence was 1.1% (C.I. 0.4–3.2) and 4.0% (C.I. 2.3–7.0) for certain and certain plus possible cases, respectively (Table 7).
In Non-oncho-areas, the prevalence in children resulted 0.6% (CI 0.1–3.0) at stool examination (Table 6), while combining results of fecal examination and serology the prevalence was 9.3% (C.I. 5.9–14.4) and 16.4% (C.I. 11.8–22.4) for certain and certain plus possible cases, respectively (Table 7). In adults, S. stercoralis prevalence measured with stool examination was 0.7% (CI 0.1–3.7) in Oncho-areas and 2.3% (CI 1.0–6.4) in Non-oncho-areas (Table 6). By serology, in Oncho-areas the prevalence of certain and certain plus possible cases was 2.7% (C.I. 0.7–6.7) and 9.3% (C.I. 5.6–15.1), respectively, while in Non-oncho-areas the prevalence was 17.9% (C.I. 12.7–24.7) and 37.8% (C.I. 30.6–45.6), respectively (Table 7).
The prevalence of A. lumbricoides, T. trichiura and hookworm is summarized in Tables 1, 2, 5 and 6. In adults the prevalence of the three STH in 1990, before the start of ivermectin mass treatment, and in 1996 was similar (66.1 vs 60.7%, 16.9 vs 26.8% and 18.6 vs 21.4%, respectively).
In children, comparing the two areas during the mass administration program in 1996 and in 1999 (Tables 2 and 5), the prevalence of A. lumbricoides was similar in both areas (68.1 vs 73% and 55.9 vs 51.6%, NS), while that of hookworm was significantly higher in Oncho-areas: 38.9% (CI 30.5–48.2) vs 19.5% (CI 14.6–25.5) in 1996 and 45.1% (CI 35.8–54.8) vs 16.8% (CI 11.4–23.5) in 1999. Conversely, prevalence of T. trichiura was significantly lower in Oncho-areas: in 1996, 28.3% (CI 20.8–37.2) vs. 86% (CI 80.5–90.1), and in 1999, 45.1% (CI 35.8–54.8) vs 70.8% (CI 63.1–77.7).
In 2013 the prevalence of the other STH declined in both areas and age groups, with the partial exception of A. lumbricoides. Prevalence of the latter in children was 47.3% (CI 41.5–53.2) in Oncho-areas and 53.0% (CI 45.8–60.1) in Non-oncho-areas, in adults it was 37.3 (CI 30.0–45.3) and 40.4 (CI 33.0–48.2), respectively. Hookworm prevalence declined in both areas and age groups, but remained higher in Oncho-areas: 8.7 (CI 5.9–12.7) vs. 5.5 (CI 3.0–9.8) in children, and 4.7 (CI 2.3–9.3) vs. 1.9 (CI 0.7–5.5) in adults, though the difference was not statistically significant. Conversely, T. trichiura prevalence remained lower in Oncho-areas: 9.5% (CI 6.5–13.5)vs 38.3% (CI 31.5–45.5) in children and 3.3% (CI 1.4–7.6) vs 18.0% (CI 12.7–24.7) in adults.
The results of several surveys reported in this unique study document the evolution of the prevalence of selected STH over a quarter of a century, starting in 1990, prior to the onset of the MDA with ivermectin, and up to 2013, six years after the conclusion of the mass administration program.
The mass administration (once a year from 1991 to 2000, then twice a year until 2007) of this drug in the context of onchocerciasis control has caused a significant and sustained impact on the prevalence of S. stercoralis infection. The study was conducted in areas with very high prevalence of strongyloidiasis. The benefits of the MDA were still present in 2013, six years after the conclusion of ivermectin distribution. Over time the prevalence of S. stercoralis measured by microscopy declined in both the age groups considered in Non-oncho-areas (remaining virtually absent in Oncho-areas), while serology still indicated a higher prevalence in Non-oncho-areas than in Oncho-areas. The difference between the results of serology and those of microscopy was more marked than in 1996 survey.
We hypothesized that this might be due to the higher sensitivity of serology because: i) we recently showed that serology is virtually 100% specific above a given cutoff [6] ii) the serologic titer sharply declines in a matter of months after effective treatment [7–11], and ivermectin has not been available up to now in Non-oncho-areas. Moreover, the striking difference in prevalence between the two areas would not to be explained by problems inherent to the methods used.
The higher sensitivity of serology is therefore the most plausible explanation as documented by many studies [9,12–18]. It is possible that in 2013 the average larval load in feces was lower, partly because of the widespread (albeit irregular) use of albendazole since the year 2000 (Mariella Anselmi, personal communication). This, along with an improvement in hygienic conditions and housing in the area of Borbon in the past few years, has seemingly had an important effect on other STH (see below). Albendazole, however, is poorly effective on S. stercoralis [19], and the repeated administration of this drug could have reduced the parasitic burden in most infected subjects below a detection threshold for the (far from ideal) fecal methods used, but not for the much more sensitive serology [20].
It is noteworthy that, six years after the administration of ivermectin in Oncho-areas, the prevalence of S. stercoralis, measured with serological methods, remained much lower than elsewhere. However, there are some adults who resulted positive despite several courses of ivermectin. This might be due to the persistence of the transmission in the area (hypothesis that might also be confirmed by the fact that there are younger children infected, although they might also have acquired strongyloidiasis moving to other areas). HTLV1, that is present in Esmeraldas province [21], may have played a role in the lack of response to repeated ivermectin courses in a few subjects, posing obstacles to the definitive eradication of S. stercoralis. Ad hoc surveys would be necessary to confirm this hypothesis. This issue also raises the relevant question of which markers of cure should be used for S. stercoralis infection in endemic areas.
Regarding other STH, the prevalence of A. lumbricoides, hookworm and T. trichiura was high or very high before and during the MDA program. However, while in 1996 and in 1999 the prevalence of A. lumbricoides and hookworm appeared to be unaffected by ivermectin administration, T. trichiura prevalence was, conversely, significantly lower in Oncho-areas, suggesting an effect of ivermectin MDA on the prevalence of this parasite. This coincides with similar results found by other researchers [22], although in other studies ivermectin has been shown to have a good effect on ascariasis [23–26]. A general improvement in hygiene and sanitation conditions over the years has certainly played a role in the reduction of STH prevalence.
This study suffers from important limitations: 1) the difference in the diagnostic methods used, some of them far from ideal for the diagnosis of S. stercoralis, the main target of our study; 2) the sampling methodology that was not on a random basis but rather based on convenience criteria, although the number of the subjects tested in each community represented an important proportion of the whole population; 3) the very long time interval elapsed between the surveys carried out during and after ivermectin intervention, thus making it difficult to closely follow the evolution of prevalence over time; 4) the use of S. stercoralis serology, the most sensitive diagnostic method for this parasite, was only possible in 1996 during the ivermectin program, and only in Non-oncho-areas. Only in the last survey (2013), 6 years after the conclusion of the intervention, was serology performed in both areas; 5) the diagnostic methods used did not allow to quantify the intensity of infection for the other STH, therefore the impact of ivermectin MDA on the latter might be underestimated. While duly acknowledging these limitations, mostly inherent to the retrospective study design, we believe that the effect of ivermectin on the prevalence of S. stercoralis is apparent, and that the difference between the two areas is so huge that it is unlikely to be affected by any bias.
In an area with high prevalence of strongyloidiasis, regular ivermectin mass administration has showed an important impact on the prevalence of this parasite. This finding, obtained through a retrospective study, needs to be confirmed by appropriate longitudinal studies, including randomized community trials, however it clearly indicate the possibility of controlling strongyloidiasis with large scale distribution of ivermectin.
In addition the data from our study also support the possibility of routinely include ivermectin during preventive chemotherapy campaigns targeting soil transmitted helminthiasis because of the beneficial effect on trichuriasis, a parasite that is more difficult to treat with albendazole and mebendazole only [27,28], and the possible advantage of the albendazole—ivermectin combination therapy to prevent the insurgence of drug resistance [29]. Moreover it could be also considered to replace with the latter combination the commonly-used regimen of albendazole and diethylcarbamazine for preventative community treatment (PCT) of lymphatic filariasis where S. stercoralis and LF are co-endemic. And finally, PCT with ivermectin has also been proposed to control scabies [30], which would be an important added value.
|
10.1371/journal.pntd.0001368 | Proteomic Identification of IPSE/alpha-1 as a Major Hepatotoxin Secreted by Schistosoma mansoni Eggs | Eggs deposited in the liver of the mammalian host by the blood fluke parasite, Schistosoma mansoni, normally drive a T-helper-2 (Th2)-mediated granulomatous response in immune-competent mice. By contrast, in mice deprived of T-cells and incapable of producing granulomata, egg-secreted proteins (ESP) induce acute hepatic injury and death. Previous work has shown that one such ESP, the T2 ribonuclease known as omega-1, is hepatotoxic in vivo in that specific antisera to omega-1 prevent hepatocyte damage.
Using an in vitro culture system employing mouse primary hepatocytes and alanine transaminase (ALT) activity as a marker of heptocyte injury, we demonstrated that S. mansoni eggs, egg-secreted proteins (ESP), soluble-egg antigen (SEA), and omega-1 are directly hepatotoxic and in a dose-dependent manner. Depletion of omega-1 using a monoclonal antibody abolished the toxicity of pure omega-1 and diminished the toxicity in ESP and SEA by 47 and 33%, respectively. Anion exchange chromatography of ESP yielded one predominant hepatotoxic fraction. Proteomics of that fraction identified the presence of IPSE/alpha-1 (IL-4 inducing principle from S. mansoni eggs), a known activator of basophils and inducer of Th2-type responses. Pure recombinant IPSE/alpha-1 also displayed a dose-dependent hepatotoxicity in vitro. Monoclonal antibody depletion of IPSE/alpha-1 abolished the latter's toxicity and diminished the total toxicity of ESP and SEA by 32 and 35%, respectively. Combined depletion of omega-1 and IPSE/alpha-1 diminished hepatotoxicity of ESP and SEA by 60 and 58% respectively.
We identified IPSE/alpha-1 as a novel hepatotoxin and conclude that both IPSE/alpha-1 and omega-1 account for the majority of the hepatotoxicity secreted by S. mansoni eggs.
| The flatworm disease, schistosomiasis, is a major public health problem in sub-Saharan Africa, South America and East Asia. A hallmark of infection with Schistosoma mansoni is the immune response to parasite eggs trapped in the liver and other organs. This response involves an infiltration of cells that surround the parasite egg forming a “granuloma.” In mice deprived of T-cells, this granulomatous response is lacking, and toxic products released by eggs quickly cause liver damage and death. Thus the granulomata protect the host from toxic egg products. Only one hepatotoxic molecule, omega-1, has been described to date. We set out to identify other S. mansoni egg hepatotoxins using liver cells grown in culture. We first showed that live eggs, their secretions, and pure omega-1 are toxic. Using a physical separation technique to prepare fractions from whole egg secretions, we identified the presence of IPSE/alpha-1, a protein that is known to strongly influence the immune system. We showed that IPSE/alpha-1 is also hepatotoxic, and that toxicity of both omega-1 and IPSE/alpha-1 can be prevented by first mixing the proteins with specific neutralizing antibodies. Both proteins constitute the majority of hepatotoxicity released by eggs.
| Schistosomiasis is a chronic parasitic disease that affects more than 200 million people worldwide [1]. The central pathological characteristic during chronic infection is a granulomatous reaction around trapped parasite eggs in the host's liver, bladder, or intestine [2]. Granulomatous inflammation in the liver may result in fibrosis, scarring, portal hypertension, and in the worst cases hemorrhaging and death [3].
Schistosoma mansoni infection in mice is the most common experimental model employed. Approximately five-to-six weeks post-infection, parasite eggs deposited by adult worms induce a T-helper-2 (Th2)-type-polarized immune response [4]. A number of the immunodominant molecules in eggs have been described [5], [6], [7], [8], in addition to the two molecules central to this report (see below). The ability of S. mansoni eggs to induce Th2-type differentiation during infection is underscored by the observation that eggs alone, or soluble egg antigen (SEA) released by the eggs through pores in the shell, are sufficient to drive Th2 polarization in naïve uninfected mice [9], [10], [11].
Research from the late 1960s and 1970s has documented that mice lacking T-cells due to genetic loss or surgical removal of the thymus [12], [13], [14], or administration of specific immunosuppressives [15], [16], [17], do not develop a typical granulomatous response to trapped parasite eggs. In these T-cell depleted mice, infection was associated with extensive hepatic parenchymal damage suggestive of a cytotoxic egg product(s) diffusing into hepatic tissue [18]. Histopathology of livers from schistosome-infected immunocompromised mice displayed microvesicular hepatocyte steatosis [18], [19], [20], nuclear degeneration, and hepatocyte apoptosis [21]. Coincident with hepatocyte injury, there is an increase in liver cell transaminases in the plasma [19]. Immunosuppressed mice also have higher mortality once egg deposition in the liver begins [17], [18], [19], [22]. In immunologically intact mice, circumoval granulomata, and antibody responses to released S. mansoni egg components, likely protect against hepatocyte damage. Also, hepatotoxicity is prevented in infected T cell-deprived mice by transfer of serum from intact mice immunized with S. mansoni eggs or egg homogenate, whilst antisera against other lifecycle stages do not prolong survival [19]. Egg-induced hepatotoxicity appears to be specific to S. mansoni; it is not observed during infection of T-cell deprived mice with S. haematobium or S. bovis [23]. Finally, transfer of sera from S. japonicum-infected mice failed to alleviate hepatotoxicity induced by S. mansoni eggs in T-cell deprived mice [24].
Research in the 1980s identified several proteins in S. mansoni egg antigen preparations based on their electrophoretic mobility and their recognition by sera from mice with chronic infection [18], [25]. Two of these proteins, termed omega-1 and alpha-1, were isolated from S. mansoni egg homogenates (SmAE) by cation exchange chromatography in a single salt-eluted fraction that was termed cationic egg fraction 6 (CEF6) [26].
Omega-1 is a 31 kDa monomeric glycoprotein [26] released from S. mansoni eggs [27] that has previously been reported to be hepatotoxic [18]. Monospecific antisera against omega-1, which is a highly immunoreactive egg antigen, were protective against hepatocyte damage in T-cell deprived mice [18], [23]. Immunochemical characterization of omega-1 using sera from humans and mice infected with different schistosome species suggested that the antigen is specific to S. mansoni [26]. Omega-1 is a functional T2 ribonuclease (RNase) [28]. Cytotoxic RNases (which includes T2 family RNase members) have been found in a wide range of species from bacteria to mammals. A range of biological roles has been suggested, including serving as extra or intracellular cytotoxins and modulating host immune responses [29]. Omega-1 was also reported to drive Th2 polarization in human monocyte-derived dendritic cells (DCs), whereas SEA depleted of omega-1 loses this ability [30]. Omega-1 directly affects both DC morphology and the ability of these antigen-presenting cells to interact physically with CD4 T-lymphocytes [31]. Furthermore, when injected into IL-4 dual reporter mice, omega-1 is a potent inducer of the Th2 response in vivo [30].
A second major protein in S. mansoni egg homogenates, originally termed alpha-1 [25], was recently reported to be identical to IPSE (IL-4 inducing principle from S. mansoni eggs) [32]. IPSE/alpa-1 is a glycoprotein [33] that has been crystallized [34], occurs naturally as a dimer (33–35 kDa), and is enriched in the sub-shell area of S. mansoni eggs from where it is secreted into the surrounding tissue [32]. It is not found in the miracidium residing within the egg [35]. Various proteomics analyses have identified IPSE/alpha-1 as an abundant protein in egg secretions [5], [36], [37], [38]. IPSE/alpha-1 binds immunoglobulin and activates naïve basophils, leading to histamine release and facilitating the production of Th2-type cytokines [32]. In vivo, IPSE/alpha-1 induces interleukin (IL)-4 secretion from murine basophils in an IgE-dependent but antigen-independent manner [39]. Most recently, IPSE/alpha-1 has been shown to contain a functional C-terminal, monopartite, nuclear localization sequence that binds DNA such that it may alter gene expression in the host cell [40].
We employed a primary hepatocyte in vitro culture system to identify and measure direct toxicity of S. mansoni eggs and their derived fractions, including pure proteins. Hepatotoxicity of omega-1 was confirmed, and IPSE/alpha-1 was identified as a novel hepatotoxin. Both proteins together account for more than half of the egg-derived toxicity measured.
These studies were performed in accordance with the recommendations by the University of California San Francisco Institutional Animal Care and Use Committee. The protocol was approved by the Committee on the Ethics of Animal Experiments of the University of California San Francisco (Permit Number: AN080237-03). All surgery was performed under sodium pentobarbital anesthesia, and all efforts were made to minimize suffering. The protocol followed these guidelines in the study: All U.S. Federal Policy and Guidelines governing the use of laboratory animals, Public Health Service Policy on Humane Care and Use of Laboratory Animals, Guide for the Care and Use of Laboratory Animals, National Academy Press, USDA Animal Welfare Act and Regulations, and NIH Office of Laboratory Animal Welfare Guidelines.
S. mansoni eggs were isolated from the livers of female golden hamsters six weeks following infection with 500 cercariae, as previously described [41]. Approximately 0.5 million eggs were washed twice in serum-free RPMI-1640 supplemented with 100 mg/ml streptomycin. Eggs were then resuspended in 2 ml RPMI-1640, and 500 µl aliquots were placed in 12-well culture plates (Costar). Cultures were checked daily by microscopy to ensure sterility. Medium was harvested after 72 h, and centrifuged for 10 min at 200×g to remove eggs. ESP, usually containing approximately 0.5 mg/ml protein by Bradford assay [42], was stored at −80°C. After collection of ESP, egg viability was confirmed by hatching of miracidia; normally >85% of the eggs hatched. Hatching of eggs during the collection period was <1%. SEA was prepared, as described previously [43].
Purified natural omega-1 [30] and recombinant (r)IPSE/alpha-1 proteins [32], and anti-omega-1 (140-3E11) and anti-IPSE/alpha-1 (74-1G2) monoclonal antibodies [32] were kindly supplied by Drs. Gabriele Schramm and Helmut Haas of the Research Center Borstel, Germany. Experiments to deplete ESP and SEA (each 10 µg/ml) of omega-1 and IPSE/alpha-1 involved incubation for 1 h with 5 µg/ml of the respective monoclonal antibodies bound to Protein G Sepharose (GE healthcare Biosciences Pittsburgh, PA). Protein G Sepharose was then removed by centrifugation for 30 min at 100×g.
ESP (2 mg) were added to 2 ml 30 mM Tris-HCl, pH 8.0, and centrifuged at 5000×g for 10 min at 4 C. The supernatant was loaded onto an Hr 5/5 Mono Q column (GE Healthcare), and equilibrated with the same buffer and elute by a 0 to 1 M linear NaCl gradient in six column volumes of Tris-NaCl buffer. Flow-through and eluted fractions (1 ml) were stored at −80°C prior to testing for toxicity with cultured primary hepatocytes (applied volume 100 µl of each fraction to 2×105 hepatocytes/0.5 ml).
Hepatocytes were isolated from C57/BL6 mice by in situ perfusion of liver with collagenase, as described previously [44]. The portal vein was severed to permit outflow followed by cannulation of the inferior vena cava with a 22-guage catheter. The liver was then flushed with a calcium-chelating buffer (liver perfusion medium) for 3 to 5 min, followed by perfusion with collagenase (liver digest medium) for an additional 6–8 min. At the end of the digestion, the liver was removed to a sterile dish and minced thoroughly with a scissors. This crude liver cell isolate was suspended in 25 ml of Dulbecco's modified Eagle's medium/Ham's F-12, filtered through sterile gauze, centrifuged at 70×g for 2 min, and resuspended in Dulbecco's modified Eagle's medium/Ham's F-12. After an additional round of centrifugation and resuspension, hepatocytes were isolated by centrifugation using a 50% Percoll gradient. Hepatocytes were cultured at a density of 2×105/0.5 ml per well in a 12-well culture plate (Costar), previously coated with a 5 mm layer of matrigel (BD Bioscience) [45], which is a tumor biomatrix prepared from the Engelbroth-Holm-Swarm mouse sarcoma [46]. Hepatocytes were allowed to attach for 1 hour at 37°C. Culture plates were gently swirled and the medium containing unattached cells and debris was aspirated. Cultures were then incubated for 72 h in a final volume of 0.5 ml RPMI medium containing 50, 100, or 200 eggs, unfractionated ESP (10 µg/ml) or 100 µl of chromatography fractions. Supernatants were collected and stored at −20°C and analyzed for alanine transaminase (ALT), a serum marker of hepatoxicity [47]) using a Beckman Chemical Analyzer in the Clinical Chemistry Laboratory of the San Francisco Veterans Affairs Medical Center (VAMC).
ESP was fractionated by SDS-PAGE then silver stained [48], [49], and 40 evenly spaced protein bands were sliced out of the gel (Fig S1). The gel slices were then diced into small cubes, reduced and alkylated with dithiothreitol and iodoacetamide, and in-gel digested with trypsin [50], [51]. The resulting peptides were extracted and analyzed by on-line liquid chromatography/mass spectrometry, using an Eksigent nanoflow pump and a Famos autosampler, which were coupled to quadrupole-orthogonal-acceleration-time-of-flight hybrid mass spectrometer (QStar XL or Pulsar, Applied Biosystems). Peptides were fractionated on a reversed-phase column (C18, 0.75×150 mm), and a “5–50% B gradient-in-gradient” was developed in 35 min at a 350-nl/min flow-rate. Solvent A was 0.1% formic acid in water and solvent B was 0.1% formic acid in acetonitrile. Data were acquired in information-dependent acquisition mode: 1 sec MS surveys were followed by 3 sec CID experiments on computer-selected multiply charged precursor ions. Peak lists were generated using Analyst 2.0 software (Applied Biosystems) with the Mascot script 1.6b20 (Matrix Science). Database searches were performed using ProteinProspector v. 5.1.7 (http://prospector2.ucsf.edu) [52]. Searches were performed first on the SwissProt databank (December 16, 2008, 405,506 entries) to evaluate sample purity, followed by searching in the S. mansoni database SchistoDB v. 4.0 (www.schistodb.net; 13,174 entries downloaded July 2009). Batch-Tag settings were selected for samples prepared with trypsin allowing a maximum of one missed cleavage and no non-specific cleavages. Peptide modifications searched for included carbamidomethyl (Cys) as the only fixed modification, and up to two variable modifications from among the following: oxidation (Met), acetyl (N-term), oxidized acetyl (N-term), pyroglutamate (Gln) and Met-loss (N-term). The mass accuracy considered was 200 ppm, and 300 ppm for the precursor and fragment ions, respectively. The following acceptance scores for database matches were required: a minimum protein score of 22, a minimum peptide score of 15, and a maximum expectation value of 0.02 required for both peptide and protein identification. These criteria resulted in an approximate 2% false determination rate. Protein identifications are reported with a minimum of two peptide matches per protein. For the analysis of ESP anion exchange fraction #11, the maximum expectation value was changed to 0.05, and no decoy proteins were identified using these acceptance criteria.
To measure hepatotoxicity caused by parasite eggs ESP, and their chromatographic fractions, we employed an in vitro system involving murine primary hepatocytes cultured on matrigel. ALT was employed as a hepatoxicity biomarker. With parasite eggs, measurements were taken 24, 48 and 72 h. No alteration in ALT levels was seen at 24 or 48 h (not shown); however, by 72 h, ALT had increased markedly and in a dose-dependent manner (Figure 1A). Dose-dependent hepatocellular injury elicited by ESP was also measured after 72 h (Figure 1B).
To aid identification of those ESP constituents responsible for hepatotoxicity, ESP was fractionated by Mono Q anion exchange chromatography. The flow-through (component not bound to the column), and each of the eluted fractions (100 µL), was co-incubated with cultured hepatocytes. At 72 h, fraction #11 induced an approximate two-fold greater release of ALT relative to the other eluted fractions and flow-through, such that it was the equivalent of 10 µg/ml unfractionated ESP (Figure 2).
SDS-PAGE and tryptic digestion followed by mass spectrometry of ESP and hepatotoxic fraction #11 identified 99 and nine proteins, respectively (Tables 1 and 2). Previously, total proteomic analysis of ESP identified 188 proteins [36], and many are common between this and the present dataset (Table 1). Among the nine proteins identified in fraction #11 were metabolic enzymes involved in glucose metabolism, glycogen storage, in addition to chaperones. IPSE/alpha-1 was also identified; and, given its potent immunomodulatory properties [32], [34], was of immediate interest in discovering of whether or not it was hepatotoxic.
Omega-1 is an egg-secreted glycoprotein with RNase activity and in vivo-demonstrated hepatotoxicity [18], [28]. To measure direct toxicity to primary cultured hepatocytes, purified native omega-1 was co-incubated with primary hepatocytes. At 72 h, a dose-dependent release of ALT was measured (Figure 3A). Likewise, Aspergillus oryzae T2 RNase (25 U/µl) (Invitrogen, # 18031-013 Carlsbad, CA) was toxic. Importantly, pre-incubation of pure omega-1 with a monoclonal anti-omega-1 antibody bound to Protein G Sepharose abolished cytotoxicity (Figure 3B). Depleting ESP and SEA with the same antibody decreased toxicity by 47 and 33%, respectively. All reductions in hepatoxicity were statistically significant (Figure 3B).
IPSE/alpha-1 was an abundant protein in the hepatotoxic fraction #11 from anion exchange chromatography (Table 2). Recombinant IPSE/alpha-1 was added to hepatocyte cultures, and a dose-dependent toxicity was measured at 72 h ALT levels that was significantly elevated relative to negative controls (Figure 4A). Similar to that found for omega-1, specific neutralization of rIPSE/alpha-1 with an anti-rIPSE/alpha-1 monoclonal antibody abolished activity and decreased the cytotoxicity of ESP and SEA by 32 and 35%, respectively (Figure 4B). All reductions in hepatoxicity were statistically significant.
To measure the combined contributions of omega-1 and IPSE/alpha-1 to the hepatotoxicity of ESP and SEA in vitro, both egg-derived preparations were depleted of both omega-1 and IPSE/alpha-1 with specific monoclonal antibodies prior to incubation with hepatocytes. The combination of both antibodies diminished hepatotoxicity of ESP and SEA by 60 and 58%, respectively (Figure 5).
The pathogenesis of hepatic schistosomiasis is due to the host's granulomatous response to eggs deposited in the liver [2]. The initial cellular granuloma is characterized by the presence of activated macrophages, lymphocytes, and eosinophils, as reviewed in both Agnew and Pearce [23], [53]. Over time, granulomata become fibrotic, and their accumulation in periportal areas, as is the case in chronic S.mansoni infection, can lead to portal hypertension, hemorrhaging, and death [3]. Ironically, in the absence of a granulomatous response, experimental hepatic schistosomiasis in mice leads to a more acute and lethal disease [17], [18], [19], [20], [22], [54]. The understanding from such observations is that schistosome eggs release hepatotoxins, toxins that are normally prevented from diffusing by circumoval granulomata. To date, the only hepatotoxin characterized in S. mansoni eggs is omega-1 [18] which is RNaseT2 [28], that also induces a Th2 response [30].
We established an in vitro primary hepatocyte culture system using ALT as the metric for cell injury to identify egg components with direct hepatotoxicity. We first confirmed the toxicity of S. mansoni eggs and their derivatives, ESP and SEA, and then showed that pure native omega-1 is hepatotoxic in vitro, consistent with previous in vivo observations. Based on the present system, omega-1 is a major toxin released by S. mansoni eggs, as depletion of ESP or SEA with a specific monoclonal antibody decreased ALT levels by 47 and 33%, respectively.
To search for additional hepatotoxins, we combined anion exchange chromatography of ESP with proteomics. A single hepatotoxic fraction (#11) was identified which contained a short list of nine proteins. IPSE/alpha-1 stood out as a molecule of interest given its potent immunomodulatory properties [39], [55]. Subsequent characterization of pure rIPSE/alpha-1 demonstrated that the molecule is indeed directly hepatotoxic. The finding was confirmed using a specific monoclonal antibody that essentially neutralized IPSE/alpha-1 toxicity while decreasing the cell injury produced by both ESP and SEA by approximately one-third. Further depletion of ESP and SEA with a combination of monoclonal antibodies targeting both omega-1 and IPSE/alpha-1 indicated that approximately 60% of the toxicity of the egg-derived material is due to these two proteins. This leaves room for additional hepatotoxins to be identified, perhaps by different chemical and physical separation approaches. We also note that although both omega-1 and IPSE/alpha-1 were identified in the total ESP proteome, only IPSE/alpha-1 was subsequently found in the single hepatotoxic fraction #11. This suggests that omega-1 was below the mass spectrometry detection limits used to identify proteins.
Recently, IPSE/alpha-1 was reported to be internalized by Chinese hamster ovary cells (CHO) and primary monocyte-derived dendritic cells, but not by peripheral blood basophils [40]; and in each case without apparent toxicity. This suggests that host cell-specific factors determine how cells interact with and respond to IPSE/alpha-1. Such factors might explain why IPSE/alpha-1 is directly toxic to hepatocytes. Studies to understand the mechanism of hepatoxicity induced by IPSE/alpha-1, and other hepatotoxins such as omega-1, can now be undertaken with the present in vitro system. Ribonuclease activity is often associated with cytotoxicity, and Steinfelder et al noted that omega-1 was initially characterized as a hepatotoxic agent from S. mansoni [28]. Nevertheless, the Th2-promoting activity of omega-1 cannot be explained by a cytotoxic effect, as the molecule failed to induce a detectable reduction in dendritic cell viability. The exact mechanism(s) by which the ribonuclease activity of omega-1 may promote Th2 responses is currently under investigation [31].
The results presented here underscore the paradox of the granulomatous response in hepatic schistosomiasis. Though detrimental to the host in the longer term due to its contribution to disease sequelae such as portal hypertension, it nevertheless protects against more acute hepatocyte injury resulting from toxins released by the schistosome egg.
|
10.1371/journal.pgen.1007776 | C. elegans ZHP-4 is required at multiple distinct steps in the formation of crossovers and their transition to segregation competent chiasmata | Correct segregation of meiotic chromosomes depends on DNA crossovers (COs) between homologs that culminate into visible physical linkages called chiasmata. COs emerge from a larger population of joint molecules (JM), the remainder of which are repaired as noncrossovers (NCOs) to restore genomic integrity. We present evidence that the RNF212-like C. elegans protein ZHP-4 cooperates with its paralog ZHP-3 to enforce crossover formation at distinct steps during meiotic prophase: in the formation of early JMs and in transition of late CO intermediates into chiasmata. ZHP-3/4 localize to the synaptonemal complex (SC) co-dependently followed by their restriction to sites of designated COs. RING domain mutants revealed a critical function for ZHP-4 in localization of both proteins to the SC and for CO formation. While recombination initiates in zhp-4 mutants, they fail to appropriately acquire pro-crossover factors at abundant early JMs, indicating a function for ZHP-4 in an early step of the CO/NCO decision. At late pachytene stages, hypomorphic mutants exhibit significant levels of crossing over that are accompanied by defects in localization of pro-crossover RMH-1, MSH-5 and COSA-1 to designated crossover sites, and by the appearance of bivalents defective in chromosome remodelling required for segregation. These results reveal a ZHP-4 function at designated CO sites where it is required to stabilize pro-crossover factors at the late crossover intermediate, which in turn are required for the transition to a chiasma that is required for bivalent remodelling. Our study reveals an essential requirement for ZHP-4 in negotiating both the formation of COs and their ability to transition to structures capable of directing accurate chromosome segregation. We propose that ZHP-4 acts in concert with ZHP-3 to propel interhomolog JMs along the crossover pathway by stabilizing pro-CO factors that associate with early and late intermediates, thereby protecting designated crossovers as they transition into the chiasmata required for disjunction.
| The creation of a viable individual from the fusion of egg and sperm requires that they each contain the correct number of chromosomes. This is ensured through the meiotic divisions, which initially fasten identical chromosomes through DNA linkages that hold them together until the cell is ready to separate them. To make these linkages, called crossovers, the cell breaks the DNA in many places, and must repair them to create a crossover, or a noncrossover. We investigate here the role of ZHP-4, and its partner ZHP-3 which form a complex that associates along paired chromosomes and finally with crossover sites. ZHP-3/4 are conserved proteins found in many organisms that function in recruiting proteins required to decide which DNA event will become a crossover and how this DNA event is coordinated with changes in chromosome structure. Using mutations that reduce the function of ZHP-4, we show that the complex cannot localize normally to meiotic chromosome and that crossing over fails. Our results suggest that ZHP-3/4 work at early and late steps in the process to stabilize other factors required for crossover formation.
| During the two specialized divisions of meiosis, a single round of DNA replication is followed by two rounds of segregation that ultimately produce gametes with half the parental number of chromosomes. Central to chromosome segregation accuracy is the formation of chiasmata between paired homologous chromosomes that are the visible product of genetic crossing over. The critical series of events leading to the formation of these linkages occurs during meiotic prophase when programmed meiotic DNA double-strand breaks (DSBs) are repaired using homologous recombination (HR). An early step in this process is resection of a DSB end to form a single-stranded stretch of DNA that can recruit Rad51, an event that initiates invasion of the homologous chromosome and the formation of a joint molecule (JM) intermediate to link the homologs (reviewed in [1]). Resolution of these JMs can proceed through a crossover (CO) or noncrossover (NCO) pathway and the route chosen at any given site is carefully monitored. To ensure crossover formation, the number of induced DSBs is in excess of the final number of COs (reviewed in [2]), however, the number of crossovers is in turn strictly regulated in any given organism (e.g. [3,4]). Consequently, a decision must be made to stabilize certain JM intermediates for entry into the CO pathway, while the remaining events are repaired as NCOs [5]. These events are particularly tightly regulated in Caenorhabditis elegans, where an estimated 5–12 DSBs along a chromosome pair must be processed to yield a single exchange event [6–8] that serves to both physically link the homologous chromosomes and asymmetrically reconfigure the bivalent in preparation for interaction with the segregation machinery (reviewed in [9]). C. elegans meiotic chromosomes exhibit robust interference that effectively limits each homolog pair to a single crossover [10–11]. As in other organisms, CO formation in the nematode is promoted by conserved players that act to stabilize and protect JM intermediates in the CO pathway, including the scaffolding protein RMH-1 (RM1), MSH-4/5 (MutSγ), and cyclin-like COSA-1(CNTD) [11–14]. In addition to these factors, a family of proteins resembling SUMO E3-like ligases has emerged as pivotal regulators of the decision to transform JM recombination intermediates into crossovers [15–17]. The canonical budding yeast Zip3p [18] exhibits E3 SUMO activity in vitro [19] and orthologs have subsequently been identified in mammals, plants, nematodes, other fungi, and Drosophila [20–29]. Members of the Zip3 E3-ligase related family are required for CO formation and share similar protein structures: an N-terminal RING finger domain, followed by a coiled-coil domain and a C-terminal domain enriched in serine residues [25]. In these organisms, the Zip3-like proteins diverge into two possible clades, one defined by Zip3, the vertebrate RNF212 and nematode ZHP-3/ZHP-4, and the other represented by HEI10 and its orthologs [21,23,29]. Budding yeast possesses a single member of the Zip3/RNF212 group [18], while plants and the filamentous fungus Sordaria appear to carry a single ortholog of the HEI10 subgroup [23,24,26], and C. elegans and mammals possess members from both subgroups [20,21,25,27,29]. However, all members of both groups share a similar pattern of localization by appearing as numerous foci or stretches along the synaptonemal complex (SC) and eventually persisting at the few obligate CO sites. In C. elegans, for example, a predicted ZHP-3/ZHP-4 heterodimer is required for crossing over, localizes to the SC and finally restricts to the six late CO intermediates typically observed in each meiotic nucleus at late pachytene stages [22,29, this study]. The pattern of ZHP-3/4 localization is reminiscent of other pro-crossover factors (RMH-1, MSH-5, and COSA-1), which similarly begin with abundant early localization that is then confined to the sites of the obligate crossovers at late pachytene stages, and finally disappears as chromosomes desynapse and chiasmata emerge [11,12].
An elusive question in the study of meiosis is how the well-studied molecular events of DNA strand exchange that lead to CO formation transform into the microscopically evident chiasmata required for chromosome segregation. Early microscopy studies of these events revealed a physical connection (chiasma) between chromatids [30] and their correlation in number with the frequency of genetic exchange [31]. Consequently, while it is widely accepted that chiasmata originate with the formation of crossovers, the question of how HR at the DNA level becomes a cytologically evident chiasma capable of supporting chromosome segregation remains largely unexplored. In the case of C. elegans, the emergence of chiasmata is coupled to remodelling of the bivalent in preparation for interaction with the spindle machinery and regulated cohesion loss [32,33]. In this study, we have shown that zhp-3/4 are required at distinct stages in this transition. First, we show that at early pachytene stages, zhp-3/4 are required to promote the formation of an RMH-1-mediated JM competent intermediate to recruit pro-CO factors. Second, we show that zhp-3/4 are required at late pachytene exit stages for the transition from the late crossover intermediate (likely the double Holliday Junction, dHJ [34]) to chiasmata. In fact, zhp-4 is required to stabilize RMH-1 at early JMs and is necessary for recruitment/stabilization of MSH-5 and signalling the end to meiotic DSB induction. Furthermore, genetic crossovers that occur in zhp-4 mutants and are not marked by the pro-crossover factors (RMH-1, MSH-5, COSA-1) are unable to form chiasmata capable of triggering the bivalent remodelling required for accurate chromosome segregation at meiosis I. Together, our data suggest that the ZHP-3/4 complex is recruited to the SC as it forms [22] to convene the complex in proximity to early recombination intermediates where it stabilizes pro-crossover factors that first promote JM resolution along the crossover pathway and finally resolve crossover-designated sites into chiasmata.
An EMS screen for recessive nondisjunction mutants (Materials and Methods) isolated a mutation (vv96) in Y39B6A.16 (ZHP-4), a gene with significant predicted protein sequence similarity to ZHP-3 (13% identity and 23% similarity). Since ZHP-4 and ZHP-3 share the structural features of the C3HC4-type RING finger domain characteristic of known SUMO E3 ligases (Fig 1A; reviewed by [35]), we investigated its role in meiosis and its relationship to ZHP-3.
The zhp-4(vv96) mutation results in a premature translation termination codon at amino acid (a.a.) 160 that is predicted to produce a truncated protein with an intact RING finger domain (Fig 1A) and is compatible with other data presented below that it represents a severely hypomorphic allele. To determine the null phenotype, CRISPR-Cas9 mutagenesis (Materials and Methods) was used to generate the deletion allele zhp-4(vv103), a frameshift mutation that creates a premature stop codon before the last two cysteines in the predicted RING-finger domain (a.a. 56; Fig 1A). The germlines of both zhp-4 mutants displayed no overt defects in nuclear morphology as assessed by DAPI staining (S1A Fig). The broods of zhp-4(vv96) and zhp-4(vv103) mutant homozygotes were marked by statistically similar levels of high embryonic lethality and incidence of XO males amongst the surviving progeny, phenotypic hallmarks of autosomal and X-chromosome nondisjunction ([36] Fig 1B and 1C).
Previous studies observed that ZHP-3 first localizes to synapsed chromosomes in an SC-dependent manner, and is then restricted at late pachytene stages to foci that correspond to sites of crossing over [21,22]. To investigate if ZHP-4 functions with its paralog ZHP-3 in CO formation, we first examined the localization of the proteins during meiotic prophase and tested their codependency in recruitment to chromosomes (Fig 2). In the case of ZHP-4, antibodies raised against the C-terminal 123 a.a. and a CRISPR-generated HA tag (Materials and Methods) both revealed that ZHP-4 localizes to the SC from earliest pachytene and is similarly restricted to the 5–7 late pachytene foci reported for ZHP-3 (Figs 2A and S2A; [21,22]). Like ZHP-3, ZHP-4 recruitment to chromosomes is SC dependent (S2C Fig), and the protein colocalizes with ZHP-3 at the SC and at the CO sites that emerge at late pachytene (Fig 2A). ZHP-3 localization was reduced to weak background levels throughout meiotic prophase in the absence of ZHP-4 (Fig 2B), and conversely ZHP-4 localization was similarly abrogated in zhp-3(jf61) mutant germlines (Fig 2B), indicating that the two paralogs are co-dependent in their recruitment to meiotic chromosomes. Our observations that ZHP-3/4 colocalize in a co-dependent manner to the same chromosome features and the results of a recent characterization of the ZHP-3 family [29] are most simply reconciled with a model in which the two paralogs physically cooperate to mediate the formation of crossovers.
Consistent with the crossover-specific function of ZHP-3 [21], ZHP-4 is similarly not required for pairing, synapsis (S1B and S1C Fig), or for meiotic DSB induction [29] as evidenced by the formation of foci of the strand exchange RAD-51 protein [37,38]. While wild-type diakinesis oocytes stained with DAPI invariably contained 6 bivalents (representing the 12 chromosomes linked by chiasmata), zhp-4(vv103) and zhp-4(vv96) mutants respectively exhibited an average of 11.4 and 8.2 DAPI-stained figures (p < 0.001 compared to WT, Fig 3A and 3B), instead of the 12 predicted in the event of complete loss of crossover potential [39]. To directly assess the effect of loss of ZHP-4 function on crossing over, we genetically measured the frequency of genetic exchange between visible markers (Fig 3D; Materials and Methods) in large genetic intervals comprising ~ 1/3 of the chromosome III (12 m.u.) and ~3/4 of the X chromosome (38 m.u.). We were unable to measure crossing over in zhp-4(vv103) null mutants, as the mutation in combination with several visible markers tested was near inviable (could not be maintained as a strain) and the embryonic lethality and aneuploidy phenotypes made it impossible to attain data sets for statistical comparisons; however, rare recombinants (<1/100 wild-type progeny) that segregated progeny of the recombinant phenotype were recovered in both intervals. The frequency of isolation of the rare recombinants in vv103 mutants is similar to the rare events previously reported for zhp-3 null mutants (2/93) [21]. In contrast, zhp-4(vv96) mutants attained 33% of wild-type crossover levels on the X chromosome (p < 0.001) and 77% of wild-type crossover levels on chromosome III (p > 0.05), indicating that vv96 mutants remained competent for significant levels of genetic exchange.
ZHP-3/4 contain a conserved C3HC4-type RING finger domain required for the catalytic activities of E3 ubiquitin and SUMO ligases (reviewed by [35,40]). We investigated its contribution to ZHP-3/ZHP-4 function by targeting conserved histidine residues at positions known to be essential for RING finger function during meiosis in other organisms (S3 Fig); for example, S. cerevisiae zip3H74A and zip3H80A mutants exhibit defects in SC assembly and sporulation efficiency [19], while Sordaria hei10H30A mutants are defective in crossing over and chiasma formation [26]. While zhp-4(H26A) mutants exhibited high embryonic lethality and males amongst surviving progeny (p < 0.01 compared to WT), zhp-3(H25A) mutants produced only 3% dead embryos and 3% male progeny (p > 0.05 in comparison to WT, Fig 1B and 1C), indicating that the RING domain of ZHP-3 is largely expendable for its function. The severity of the embryonic lethality defects observed in the two mutants was also reflected at diakinesis where only 5% of zhp-4(H26A) nuclei exhibited the 6 DAPI bodies observed in wild-type nuclei (average of 8.0, p < 0.001), while 95% of zhp-3(H25A) diakinesis nuclei (average of 5.9, p > 0.05) did so (Fig 3C). Since neither zhp-3(H25A) nor zhp-4(H26A) single mutants replicated the phenotypes of the respective null mutant, we addressed the possibility of RING domain redundancy by examining the consequence of loss of both RING domains. zhp-3(H25A); zhp-4(H26A) double mutants exhibited phenotypes that were more severe than those of zhp-4(H26A) single mutants and not different from those observed for the null mutant; homozygotes segregated 95% dead embryos and 43% male progeny (Fig 1B and 1C, p > 0.05) and 80% of diakinesis nuclei showed 12 univalents (average of 11.7 p < 0.001; Fig 3C).
In zhp-3(H25A); zhp-4(H26A) double mutants, ZHP-4 was not detectably recruited to synapsed chromosomes at any stage, and no enriched nuclear localization could be detected (Fig 4B). We could not detect ZHP-3H25A localization in the double RING mutant using α-ZHP-3 antibodies (the aliquot did not provide a reliable signal even in wild-type controls); however, its localization is likely to be equally abrogated given the co-dependent co-localization of the proteins and the relative fertility of zhp-3(H25A). In the case of zhp-3(H25A) single mutants, ZHP-4 adopted the wild-type pattern of SC localization throughout pachytene (Fig 4B) to finally restrict to ~ 6 foci marking putative crossover/chiasma sites that correlate with the lack of meiotic defects in this mutant. In zhp-4(H26A) single mutants, however, ZHP-4H26A localization to the SC was reduced, discontinuous, and evident only as punctate foci of varying intensities (Fig 4B), indicating that an intact ZHP-4 RING domain is required for the contiguous pachytene pattern of ZHP-3/4 association with the SC that is observed in wild-type. Despite this disrupted localization during early/mid-pachytene stages, 1–3 bright ZHP-4H26A foci/nucleus emerged at late pachytene (Fig 4C); this in combination with the detection of significant levels genetic crossing over (57% and 40% of wild-type chromosome X and III frequencies; Fig 3C) and the presence of bivalents at diakinesis (Fig 3C) indicates that zhp-4(H26A) mutants can support reduced levels of crossover formation and localization of the protein to those sites. Our results are collectively consistent with a model in which 1) the RING domain of ZHP-4 is critical for the localization of the heterodimer to the SC where it cooperates with ZHP-3 to promote crossover formation and 2) the ZHP-3 RING domain can partially compensate for loss of ZHP-4 RING domain activity to foster the formation of a severely reduced number of crossovers/chiasmata.
In crossover-defective mutant backgrounds, RAD-51-marked recombination intermediates typically appear on time and at wild-type levels in early pachytene, however, they accumulate and persist into late pachytene stages [38]. We measured RAD-51 foci in nuclei of the mitotic (zone 1 and 2), leptotene/zygotene (referred to as the transition zone)/pachytene entry (zone 3), early pachytene (zone 4), mid-pachytene (zone 5), and late pachytene stages (zone 6) (Fig 5). In both zhp-4(vv103) and zhp-4(vv96) mutants, RAD-51 foci appeared, peaked in number, and disappeared with wild-type like timing as previously reported for zhp-3 mutants [21], suggesting appropriate recombination initiation and timely DSB processing and repair. However, the levels of RAD-51 foci in both zhp-4 mutants were dramatically elevated at every stage until their disappearance at late pachytene (Fig 5). In particular, meiotic RAD-51 foci first emerge in zone 3 where 70% of the wild-type nuclei have no foci (30% have 1–6 foci) while only 4% of vv96 and 14% of vv103 mutant nuclei lack any foci; at this same stage, 34% of vv96 and 17% of vv103 mutant nuclei have >7 RAD-51 foci, a category that does not appear in wild-type germlines until zone 4. This initial increase of RAD-51 foci in zhp-4 mutants persists into later stages, but with wild-type-like dynamics; their average number per nucleus peaks in zone 4 (4.2 in wild-type, 10.5 and 11.1 in vv96 and vv103 mutants, respectively. p < 0.001), and appropriately disappears in the last zone (1.0 in wild-type, 2.0 and 1.4 in vv96 and vv103 mutants, respectively. p < 0.01 for vv96 and p > 0.05 for vv103). While the elevated numbers of RAD-51 foci observed in zhp-4 mutants could originate in their impaired turnover, RAD-51 foci kinetics followed the wild-type pattern and did not exhibit the accumulation observed in other CO-defective mutants [38]. This suggests that the initial elevation that appears in concert with meiotic DSB formation is not a reflection of an impairment in processing RAD-51-marked early recombination intermediates but reflects a requirement for ZHP-4 in down regulating DSB formation. To test this possibility, we examined the levels of RAD-51 foci in a rad-54(RNAi) background where DSB repair is blocked and RAD-51 foci are not removed, thereby permitting quantitation of the total number of DSBs formed [41]. Hermaphrodites injected with rad-54 RNA did not show the sterility previously reported for rad-54 deletion mutants or rad-54(RNAi) [41]. However, cytological analysis at 2 days post-injection (3 days post L4 stage) revealed an increase in RAD-51 foci until early/mid-pachytene stages (zone 4) at levels comparable with those previously reported [41]. We observed that the levels of RAD-51 foci decreased in mid-pachytene (zone 5) and disappeared in late pachytene stages (zone 6), in contrast to rad-54 mutants in which the levels remain high until the end of pachytene. This suggests a partial effect of the RNAi treatment, and consequently we measured RAD-51 foci levels up to zone 4 in zhp-4(vv103); rad-54(RNAi) germlines and compared them to the levels observed in the same zone in rad-54(RNAi) controls (S4 Fig). We observed a significant increase of RAD-51 foci in zone 4 in the double mutant in comparison to rad-54(RNAi) alone (average of 13.5 versus 9.6 for rad-54(RNAi), p < 0.00001). This result is consistent with the interpretation that the higher levels of RAD-51 foci observed in vv103 germlines are the consequence of increased DSB formation rather than defective DSB repair. This suggests a role for ZHP-4 in negatively regulating recombination initiation, possibly by being required for the formation of a crossover intermediate that can feedback to negatively regulate DSB formation (reviewed in [42]).
Since zhp-4 mutants robustly initiated recombination but failed to resolve these events into crossovers, we used known markers of crossing over to probe the origin of this defect. We first characterized the dynamics of RMH-1, a conserved scaffolding component thought to co-operate with Bloom’s helicase (BLM; nematode HIM-6) during an early step in crossover designation [43] and the resolution of recombination intermediates into crossovers or noncrossovers [12]. In C. elegans, GFP::RMH-1 is first recruited to synapsed chromosomes at early pachytene in numbers exceeding the number of obligate COs, suggesting that it marks both COs and non COs at this stage; by late pachytene, the number of RMH-1 foci decreased to ~6 per nucleus putatively marking the obligate crossover sites [12]. We observed a similar kinetics of appearance and disappearance of RMH-1 foci in wild-type germlines (Fig 6A–6F) as previously reported [12]. Both vv103 null mutants and vv96 hypomorphs acquired low levels of RMH-1 foci in very early pachytene stages in numbers not significantly different from wild-type (p > 0.05, Fig 6A), indicating that ZHP-4 mutants are competent for formation of these early RMH-1-marked recombination intermediates with appropriate timing and levels. In zhp-4(vv103) mutants, the number of RMH-1 foci did not significantly increase throughout pachytene (p < 0.001), and the 2–3 foci that did form disappeared at the approach of pachytene exit (Fig 6E and 6F), similar to the timing and kinetics observed for RMH-1 foci in zhp-3(jf61) mutants [12].
In contrast to vv103 null mutants, hypomorphic zhp-4(vv96) mutants steadily accumulated RMH-1 foci as pachytene progressed, reaching levels not different than those observed in wild types at mid-late pachytene stages (median of 10, p > 0.05; Fig 6D, 6E and 6G). However, the accumulation of these foci was delayed with respect to the kinetics observed in wild types germlines (Fig 6B and 6C); for example, while the number of RMH-1 foci in wild-type peaked at mid-pachytene with a median of 16 RMH-1 foci, vv96 mutants exhibited 8 foci and displayed a peak later at mid-late pachytene with 13 foci (Fig 6D). Although zhp-4(vv96) mutants appeared competent in the formation of early RMH-1 marked recombination, they proved to be defective in presenting the bright RMH-1 foci at very late pachytene stages in which RMH-1 marks the sites of the obligate crossovers (Fig 6F and 6H). While in wild types the excess early RMH-1 foci disappeared to a final median population of 6, zhp-4(vv96) mutant nuclei exhibited no detectable foci at the stage approaching pachytene exit (Fig 6F and 6H; p < 0.001 in comparison to wild types), despite their abundant presence earlier (Fig 6B–6E). Collectively, these data support dual roles for ZHP-4 in RMH-1 dynamics. Our data suggest that wild-type levels of early RMH-1 foci can form in the absence of ZHP-4, however, ZHP-4 is then required to stabilize or localize RMH-1 at early JMs that will become either COs or NCOs, and later at designated crossover sites where the final intermediate will be resolved as a crossover.
In addition to RMH-1, the crossover-designated sites at mid-late pachytene stages also colocalize with ZHP-3, MSH-5, and COSA-1 [11,12]. zhp-4(vv96) mutants displayed punctate staining of ZHP-3 on chromosome tracks at late pachytene (Fig 4A), and a zhp-4(ha::vv96)-tagged variant that phenocopies the genetic mutant (Materials and Methods) adopted a pattern similar to the ZHP-3 localization in the zhp-4(vv96) background (S5 Fig). Given that zhp-4(vv96) mutants exhibit evidence of significant levels of bivalent formation and crossing over (Fig 3), these data suggest that the ZHP-4vv96 mutant protein can associate with crossover pathway intermediates in the absence of its contiguous recruitment to the SC to facilitate crossing over.
We next examined the localization of MSH-5, a component of the pro-crossover MutSγ complex, that is required for formation of the obligate crossovers from a population of earlier recombination intermediates. Abundant early MSH-5 foci form during mid-pachytene where they colocalize with a subset of RMH-1 foci and finally mark the six crossover sites in late pachytene stages [11,12]. In the absence of ZHP-4, only a few sporadic small MSH-5 foci that did not colocalize with chromosome tracks could be detected (Fig 7A). Given the correlated deficit of RMH-1-marked recombination intermediates in zhp-4(vv103) mutant germlines, these data suggest that MSH-5 recruitment to a potential crossover intermediate may require an RMH-1 processed JM and/or that the two proteins also co-operate in one anothers’ localization/stabilization at these sites. In zhp-4(vv96) hypomorphs, abundant MSH-5 foci formed in mid-late pachytene, however, these foci were of varying sizes and intensities that precluded accurate scoring and ultimately did not reduce to the ~6 bright foci observed in wild types. Given that RMH-1-marked recombination intermediates are fewer or more unstable in vv96 mutants, a likely consequence is that their loss translates to impaired MSH-5 recruitment/stabilization [12], in turn leading to the reduced crossing over and chiasma defects observed in these mutants. Furthermore, no bright wild-type-like COSA-1 foci were detected on chromosome tracks in the germlines of vv96 and vv103 mutants (Fig 7B). Its localization was instead reduced to a faint and diffuse signal punctuated by numerous foci of variable sizes and intensities that occasionally overlapped with the synapsed chromosomes, but were often extranuclear. This localization differed from the few faint foci observed in spo-11 mutants that do not initiate meiotic recombination and may reflect protein interactions outside of the context of crossover formation. Based on these results we conclude that COSA-1-mediated steps in crossover designation require ZHP-4 function.
The pro-crossover factors that showed disrupted localization to late crossover intermediates in vv96 mutants participate in the resolution of recombination intermediates into class I crossovers that show interference (the reduced probability that a second crossover forms in the vicinity of the first). An alternative class II pathway that includes the MUS-81 structure-specific endonuclease is required for a small subset of meiotic crossovers that are noninterfering [44] and we investigated whether the bivalents that formed in vv96 mutants depended on the activity of MUS-81 or RMH-1 (Fig 8). The diakinesis nuclei of zhp-4(vv96); mus-81(tm1937) double mutants did not show significantly different distributions of DAPI figures in comparison to vv96 single mutants (p > 0.05), indicating that the residual chiasmata observed in zhp-4(vv96) mutants do not required mus-81. In contrast, an average of 12 DAPI figures and no bivalents were observed at diakinesis in zhp-4(vv96); rmh-1(jf92) double mutants (Fig 8B and 8C), indicating that bivalent formation in vv96 mutants is dependent on RMH-1. Furthermore, the residual chiasmata observed in rmh-1 single mutants was in turn dependent on zhp-4, suggesting possible co-dependent functions in crossover formation. In summary, our collective results suggest that ZHP-4 is first required for the stabilization of RMH-1 at early recombination intermediates; this event either generates a crossover intermediate recognized by pro-crossover factors or stabilizes those factors at the site (or both), leading to resolution of the intermediate into a crossover at pachytene exit and its transition into a chiasma.
Characterization of the severe hypomorph zhp-4(vv96) revealed a functional paradox: although the levels of embryonic lethality and frequency of males in the self progeny did not differ from the null allele (p > 0.05, Fig 1B and 1C), zhp-4(vv96) mutants exhibited surprisingly substantial levels of both bivalent structures and of crossovers as measured by genetic exchange (Fig 3A, 3B and 3D) despite the defects in acquiring late-pachytene stage RMH-1, MSH-5 and COSA-1 foci that mark sites of the obligate crossovers in wild-type.
Close examination of chromosomes in the diakinesis nuclei of zhp-4(vv96) mutants revealed an unexpected phenotype; the occasional appearance of well-condensed bivalent-like structures that had separated chromosomes and were tethered to one another by a chromatin mass, with the axial element HTP-3 often congregated within (Fig 9A). While abnormal bivalent morphology in which chromosomes linked by chromatin bridges has been observed in mutants that disrupt Holliday junction resolution and crossover intermediate processing, the chromatin linkages in these cases appear more thread-like and do not contain axis components [12,45]. The anomalous diakinesis structures observed in zhp-4(vv96) mutants appeared in a SPO-11-dependent manner, indicating that they were the outcome of a meiotically-programmed DSB intermediate (Fig 3A and 3B), and are for simplicity referred to as “tethered bivalents”. To characterize this disruption, we probed the bivalents observed in vv96 mutants for evidence of the remodelling associated with chiasma formation (Fig 9), including restriction of the AIR-2 kinase (aurora B kinase; [46,47]) and SC component SYP-1 to the short arms of the bivalent [32], and the meiotic sister chromatid cohesin regulator HTP-1 to the long arms [33]. In vv96 mutant oocytes at diakinesis, bivalents (as assessed by size and cruciform staining of the axis marker HTP-3) failed to appropriately remodel (Fig 9A–9C) and instead exhibited SYP-1 (13/13 bivalents) and HTP-1 (11/12 bivalents) along the axes of both the long and short arm. zhp-4(vv96) mutants similarly displayed disruptions to AIR-2 localization that interfered with quantitation; these included failure to localize or its appearance in chromatin masses between the tethered bivalents. Taken together, these results indicate that the crossovers forming in vv96 mutants are defective in triggering the associated remodelling of factors implicated in chromosome segregation, or that this triggering is not executed in vv96 mutants.
To further investigate the functional implications of this localization defect, we next examined the localization of the single C. elegans SUMO ortholog whose conjugation to target proteins promotes chromosome alignment at metaphase I [48]. SMO-1 localizes to the chromatin of germline nuclei and then to the axes of the bivalent short arms in late diakinesis stage oocytes ([48]; S6 Fig); this dynamic localization is recombination dependent, since SMO-1 remains diffusely associated with chromatin rather than restricted to HTP-3-marked chromosome axes in spo-11(ok79) mutants (Fig 9D). In both zhp-4(vv103) null and zhp-4(vv96) hypomorphic mutants, SMO-1 localizes appropriately to chromatin of mitotic and prophase chromosomes and remains associated with the chromatin of univalents present in late-stage diakinesis nuclei, similar to the localization observed in spo-11 mutants (Figs 9D and S6). In contrast to wild types, however, SMO-1 remained associated with the chromatin of diakinesis bivalents in vv96 mutant nuclei and with the chromatin of a rare bivalent observed in a vv103 mutant nuclei (Fig 9D.2-3). In the case of vv96 mutants, SMO-1 also occasionally appeared enriched in the chromatin at the ends of tethered bivalents (Fig 9D.4). Consequently, we conclude that the crossovers that form in vv96 or vv103 mutants can give rise to bivalent-like figures, however, these exhibit structural anomalies and are defective in the chromosome remodelling that normally accompanies crossover formation and chiasma emergence. In contrast, the bivalents that form in the diakinesis nuclei of zhp-4(H26A) mutants show appropriate localization of SYP-1, HTP-1, and SMO-1 and no tethered bivalent phenotypes (Fig 9B–9D; no tethered bivalents observed in 86 diakinesis nuclei), suggesting that the reduced number of crossovers that form are competent to trigger chromosome remodelling, correlated with improved embryonic survival (Fig 1B).
To investigate the possible functional consequences of the formation of aberrant bivalent structures in zhp-4(vv96) mutants, we examined chromosome congression and segregation at the metaphase plate in meiosis I. In wild-type, bivalents align at the metaphase plate I between overlapping microtubule bundles that form channels through which the chromosomes move during segregation [49,50]. At metaphase I, zhp-4(vv96) oocytes displayed a spectrum of phenotypes from occasional wild-type spindle organization to stray chromosomes and abnormal spindle morphology as assessed by disorganized microtubule channels (S7 Fig). These structural defects in spindle assembly correlated with aberrant segregation behaviour at anaphase I; while in wild-type oocytes two distinct masses of chromatin were separated by the microtubule channels, the chromosomes of some zhp-4(vv96) mutant oocytes appeared unresolved and tangled while being pulled to the poles (S7 Fig). Given the presence of bivalents in zhp-4(vv96) mutants, we favour the interpretation that the crossovers that form in the absence of full ZHP-4 function do not trigger the bivalent remodelling associated with preparation for segregation and consequently lead to the same level of embryonic lethality and X chromosome nondisjunction as observed in zhp-4(vv103) null mutants. These results are consistent with a model in which crossover and chiasmata formation are genetically separable events that require ZHP-4 for the physical transformation of crossovers into the chiasmata capable of directing chromosome segregation at the meiotic spindle.
zhp-4(H26A) mutants shared several similar phenotypic features with zhp-4(vv96) mutants, including faint and punctate SC localization and significant levels of genetic crossing over. In contrast to zhp-4(H26A) mutants, however, zhp-4(vv96) mutants 1) did not form ZHP-3/4 late pachytene foci, despite the presence of genetic crossovers, 2) contained aberrant tethered bivalents in diakinesis nuclei which were not observed in zhp-4(H26A) mutants, and 3) exhibited significantly higher levels of embryonic lethality (Fig 1B). Given that the zhp-4(vv96) mutant defects correlated with disrupted RMH-1 stabilization at recombination intermediates and failure to retain/recruit markers of designated crossovers/chiasmata, we investigated these processes in ZHP-4 RING domain mutants. zhp-4(H26A) mutants exhibited early RMH-1 foci dynamics that resembled those observed in zhp-4(vv96) mutants (Fig 6A–6E, p > 0.05 for all except p < 0.05 for C), including: 1) their appropriate appearance at very early pachytene, 2) a delay in their accumulation, and 3) the appearance of wild-type levels of RMH-1-marked recombination intermediates at mid-late pachytene stages (Fig 6D, p > 0.05 vs. WT). However, in nuclei entering the pachytene exit stage in which designated crossover/chiasma markers emerge, the majority of zhp-4(H26A) nuclei exhibited 1–3 RMH-1 foci/nucleus (consistent with the appearance of 1–3 ZHP-4 foci in the RING mutant), while none were detected in zhp-4(vv96) mutants (Fig 6F–6H, p < 0.05 compared to vv96 mutants). The presence of these RMH-1 foci at this late stage suggests that the RING domain mutant is competent to form a reduced number of the crossover intermediates that are observed in wild types. To investigate this possibility, we next examined zhp-4(H26A) pachytene exit stage nuclei for the appearance of COSA-1 marked-crossover intermediates, which could not be detected on chromosomes above background levels in zhp-4(vv96) mutants (Fig 7B). zhp-4(H26A) mutants exhibited levels of COSA-1 foci formation similar to that observed for RMH-1 focus formation at late pachytene (median of 2; Fig 7C, p < 0.001 in comparison to wild types), and in both cases the bright COSA-1/RMH-1 foci appeared on HTP-3-marked synapsed axes as observed in wild types. Furthermore, well-formed bivalents with respect to DNA condensation and the localization of SYP-1/SMO-1 to the short arm and HTP-1 to the long arm appeared in zhp-4(H26A) mutants at late diakinesis, indicating that the genetic crossovers detected (Fig 3D) correlate with the presence of designated crossover markers and appropriate remodelling of the bivalent (Fig 9B–9D, white arrow). The association of late pro-crossover factors with the designated crossover site correlated not only with bivalent remodelling, but also with the ability of the chiasmata to direct segregation as evidenced by the lower embryonic lethality observed in zhp-4(H26A) mutants in comparison to zhp-4(vv96) mutants (Fig 1B). These results strongly suggest that the crossover intermediates that do acquire RMH-1, ZHP-4 and COSA-1 at very late pachytene stages in zhp-4(H26A) mutants define crossovers competent to form chiasmata and trigger the bivalent remodelling that is required to ensure accurate chromosome segregation. Consistent with this interpretation, zhp-4(vv96) mutants are competent for the formation of RMH-1/MSH-5 foci in mid-pachytene stages and reduced levels of crossing over; however, the crossovers that do form are not cytologically visible as foci containing the chiasmata markers, and correlate with defects in chromosome remodelling and spindle function. We propose that ZHP-4 acts in concert with ZHP-3 to stabilize RMH-1 at early recombination intermediates to foster the formation of an early crossover intermediate competent for negatively regulating meiotic DSB induction and association with other pro-crossover factors like MSH-5 and COSA-1. Our analysis suggests that the ZHP-4-mediated stabilization/recruitment of the pro-crossover complex at designated crossover sites by late pachytene is required to convert the DNA exchange events into the chiasmata solely capable of triggering bivalent remodelling in preparation for meiotic spindle assembly and chromosome segregation.
A further evolution in our understanding of E3 ligases has been the observation that some can function as heterodimers, including the Ub ligases BRCA1-BARD1 [51–53] and SUMO-directed Ub ligases Slx5-Slx8, [54–57]; however, no such examples have yet emerged for SUMO E3 ligases. Here, we have shown that ZHP-3/4 colocalize throughout meiotic prophase and that their localization to the SC and to designated CO sites is interdependent, suggesting a cooperative activity. Although the ZHP-3 RING finger is competent to support the formation of rare COs, the RING finger domain of ZHP-4 is the preferred contestant in localizing the complex to the SC since ZHP-3 RING activity can only be detected in its absence. This difference in terms of requirement between the two RING finger domains mirrors other examples of heterodimeric Ub E3 ligase complexes: in the case of BRCA1 and BARD1, BRCA1 is the ‘active’ partner while BARD1, the ‘inactive’ partner, stabilizes the complex in vivo (reviewed by [40]). Although ZHP-3/4 has been suggested to act as a heterodimeric E3 SUMO ligase [29], such a function is not supported by the phenotype of mutants in the single nematode SUMO gene, which form bivalents rather than the predicted univalents [22,58]. An outstanding question is whether ZHP-3/4 is a ubiquitin ligase, and if so, if it could perform a structural role at crossover sites that becomes a catalytic role in restructuring the resulting bivalent.
The family of RNF212-like orthologs have differences and similarities in terms of localization and their relationship with DSB formation and SC assembly that largely reflects the relationship between recombination initiation and SC initiation in each organism. Recent studies have revealed that S. cerevisiae Zip3 and mouse RNF212 recruitment to meiotic chromosomes occurs in two distinct modes, one being DSB-dependent foci, and the other requiring SC formation for localization [18,25,59], indicating a DSB-independent mechanism of a CO-promoting factor. In the divergent world of plants and filamentous fungi, the single Zip3-like protein (HEI10) similarly localizes in pachytene to the SC central regions and is then restricted to detectable foci at late stages that correspond to sites of both COs and chiasmata [23,24,26].
Although SC initiation is independent of DSB formation in C. elegans, ZHP-3/4 exhibit two patterns of localization that grossly reflect the localization of RNF212 in mice: 1) ZHP-3/4 initially localize along the SC in an SC-dependent (and SPO-11-independent) manner, and 2) are restricted to a few sites of crossovers (~ 6 foci) that at late pachytene stages are dependent on both the SC and SPO-11 (S2C Fig; [21]). While in Drosophila, the RNF212-like ortholog Vilya is required for DSB formation to occur [28], our study indicates that ZHP-3/4 are not required for the initiation of meiotic DSB formation (DSBs form in the absence of the proteins) and SC assembly (Figs 5 and S1C; [21]). Instead, ZHP-3/4 are required to foster the transition of a limited number of crossover intermediates into bona fide crossover entities, a function which includes negatively regulating meiotic DSB induction once crossover intermediates have been formed. Overall, a common and recurring feature of all RNF212-like orthologs is their localization to the SC either as continuous linear stretches, or as a population of small foci that later emerge as larger discrete foci that mark crossing over sites. Given these localization dynamics across species with differential requirements for synapsis initiation (DSB dependent or independent), our results are consistent with an intimate relationship between RNF212 family members and the SC from the earliest stages of recombination initiation that supports relatively rare events to go forward as the crossovers that will support chromosome segregation.
In this study, we have shown that zhp-3/4 are required at distinct stages in the transition of JMs to chiasmata; at mid-pachytene stages where they promote the formation of an RMH-1-mediated JM competent to recruit pro-CO factors and at pachytene exit stages where they are required for the transition from crossover-designated sites to chiasmata. Since DSB formation, as visualized by RAD-51, loading initiates on time and at robust levels in zhp-4 mutants (see below), the deficit of chiasmata observed in the diakinesis oocytes indicates a defect in post-initiation/strand exchange process. At mid-pachytene stages RMH-1 cooperates with BLM (nematode HIM-6) to promote crossover outcomes at JMs and repair the remaining recombination intermediates as noncrossovers [12,34]. We observed that the first population of RMH-1 appears on time and at appropriate levels in zhp-4 mutants, indicating that recruitment of RMH-1 to early recombination intermediates does not require ZHP-4; at early/mid-pachytene, however, both null and hypomorphic mutants exhibited severe defects in presenting wild-type numbers of RMH-1 foci, collectively consistent with the interpretation that ZHP-4 is required for stabilization of RMH-1 at JM rather than recruitment per se. The failure to stabilize RMH-1 at the early-mid-pachytene stage in zhp-4 mutants correlates with altered dynamics of other markers of recombination progression. First, RAD-51-marked recombination intermediates showed dramatically increased levels from the transition zone/earliest pachytene stages that followed wild-type-like kinetics of appearance and disappearance, suggesting an overall increase in meiotic recombination initiation rather than a defect in RAD-51 turnover. Second, the mid-pachytene MSH-5 foci that colocalize with RMH-1 at JMs in wild-type fail to form in the absence of zhp-4 and in vv96 hypomorphs appear with delayed timing and altered morphologies. Both phenomena are most parsimoniously explained as outcomes of a single event; zhp-3/4 is required for stabilization of RMH-1 at early/mid-pachytene stages to produce a JM intermediate that can signal the end to meiotic DSB initiation, leading to stabilization of interhomolog intermediates, CO designation, and crossover formation. Consequently, we favour the interpretation that the loss or reduction of crossing over observed in zhp-4 mutants originates in the failure to form a stable RMH-1-associated JM intermediate that can progress into the crossover pathway, and is instead repaired as an NCO.
In addition to the early role of zhp-3/4 in the crossover pathway, our analysis of the zhp-4 mutants revealed a genetically separable role for ZHP-4 at pachytene exit stages in the designation of crossover intermediates destined to become chiasmata. zhp-4(vv96) and zhp-4(H26A) RING mutants share a similar early/mid-pachytene phenotypic profile: in both cases, ZHP-3/4 do not show localization along the SC, RMH-1 appears with delayed kinetics that ultimately reaches wild-type levels, detectable genetic crossing over occurs, and similar distributions of bivalent structures appear at diakinesis. At late pachytene stages when pro-crossover markers are restricted to CO sites, zhp-4(H26A) RING mutants exhibit 1–3 bright foci appropriately marked by RMH-1, ZHP-4, and COSA-1, suggesting that the earlier problems in RMH-1 dynamics are overcome to generate a reduced number of wild-type crossovers. In zhp-3(H25A); zhp-4(H26A) double RING mutants, these ZHP-4-marked foci fail to form, a phenotype which is accompanied by loss of bivalent formation (Figs 3C and 4B). This suggests that crossing over in ZHP-4 RING mutants is dependent on ZHP-3 and likely reflects a scenario in which the ZHP-3 RING domain is sufficient to support highly reduced ZHP-3/4 localization to chromosomes where its function is unaltered. In the case of vv96 mutants, however, crossing over and bivalent formation are not accompanied by the appearance of crossover-designated sites as defined by RMH-1/COSA-1 focus formation, indicating that CO designation and formation are separable events. The failure to form late RMH-1/COSA-1 foci in vv96 mutants correlates with the appearance of anomalous bivalent structures unique to zhp-4(vv96) mutants and well-formed bivalents that fail to exhibit the CO-directed remodelling associated with preparation for segregation. These chromosomes often show gross defects in alignment and congression at the metaphase I spindle and remain entangled at anaphase I, consistent with the chromosome segregation defects and high embryonic lethality observed in zhp-4(vv96) mutants. Similarly, zhp-3::gfp mutants (the gfp construct does not fully rescue the null mutant phenotype at standard culture temperatures [22]) display a competency for crossover formation that is nevertheless accompanied by unexpectedly high levels of embryonic lethality and X-chromosome nondisjunction, suggesting that the significant levels of crossing over in the presence of altered ZHP-3 function does not always guarantee chiasmata formation and bivalent remodelling [22]. In many organisms, the correct placement of crossovers on the chromosomes has been proven to be pivotal for promoting segregation; COs at the centromeres or ends of chromosomes are less effective at ensuring disjunction (reviewed by [3]). In the case of zhp-4 mutants, an argument can be made that the inability of the COs to ensure accurate segregation is a consequence of their displacement to disjunction-ineffective regions. However, the nematode rec-1 mutant redistributes a wild-type number of COs in a pattern reflecting the physical map without compromising chromosome segregation [60], indicating that CO redistribution per se is not sufficient to provoke nondisjunction.
The early prophase localization of the ZHP-3/4 complex presents an elegant solution to the requirement for zhp-3/4; they function in promoting CO intermediate formation, while being dispensable for initial pairing or meiotic DSB induction [39,61], The SC is required for crossing over and recruitment of the complex to the structure and concentrates its activities in proximity to nascent HR intermediates from the earliest time point that JMs can enter the crossover pathway. An outstanding question that remains is the function of ZHP-3/4 at the crossover-designated sites that appear at late pachytene stages. The behaviour of vv96 mutants suggests that lost or disrupted retention of late pro-crossover markers at designated sites does not necessarily abolish crossing over, but does disrupt some aspect of CO formation that has functional consequences for the resulting bivalent during chromosome segregation. Recent work on the architecture of nematode recombination complexes and their relationship to the SC during meiotic prophase has observed that CO/NCO outcomes are visibly manifested at late pachytene stages [34]. HR repair proteins are lost from NCO sites (presumably indicative of completed repair) and pro-crossover MSH-5, COSA-1, and BLM (RMH-1-interacting nematode protein HIM-6; [12]) appear at CO-designated sites in the context of central region components of the SC that envelop them in a bubble-like structure. Although the function of this structure is not known, an intriguing possibility is that it reflects an enzymatic caging which can concentrate the pro-CO activities within and protect the CO intermediate from the NCO activities taking place outside. Since ZHP-3/4 are dependent on central region SC components for their localization, it is possible that their function in this compartment is to stabilize the pro-crossover factors until desynapsis at diplotene frees the double Holliday junction (dHJ) precursor for resolution into a crossover. In this context, the consequences of the inability of vv96 mutants to form these late pro-crossover factor-enriched sites may result in premature exposure of the dHJ to resolvases that temporally uncouple crossover formation from chiasma emergence and regulated swapping or remodelling of the axes to which the involved DNA is tethered [62]. Such a function may explain the fact that zhp-4(vv96) mutants are in part marked by the appearance of diakinesis bivalents that are tethered by chromatin masses engaged with axis components, in addition to other bivalent structural anomalies that are suggestive of perturbed coordination between dHJ resolution and CO-triggered chromosome morphogenesis. We speculate that the failure to coordinate these events may distinguish a genetic crossover at the DNA level from a chiasma competent to direct chromosome segregation by disrupting axis exchange and/or patterning of sister chromatid cohesion.
C. elegans strains were cultured under the conditions described by Brenner [63] and all experiments were conducted at 20°C. The N2 var. Bristol strain was used as a wild-type reference and the following mutations and rearrangements were used: zhp-3 (jf61::unc-119+) /hT2 I. meIs8 [Ppie-1::gfp::cosa-1 + unc-119(+)] II; cosa-1(tm3298) III. jfsi38 [gfp::rmh-1 cb-unc-119+] II. dpy-18(e364) unc-25(e156) III. spo-11(ok79)/nT1 IV. spo-11(me44)/nT1 IV. dpy-3(e27) unc-3(e151) X. rmh-1(jf92[M01E11.3::unc-119+]) I. mus-81(tm1937) I. syp-2(ok307) V. msh-5::gfp IV.
Hermaphrodites were singled at L4 stage and transferred daily to fresh plates for three consecutive days. The number of eggs of each hermaphrodites was recorded immediately after each transfer; in the last plate, the number of eggs was recorded 24 hours after transfer. The number of hermaphroditic and male progeny were scored three days later. Embryonic lethality rate was calculated as the total number of surviving progeny divided by the total number of eggs. Incidence of males was calculated as the number of males divided by the total number of surviving progeny.
Recombination was assayed using visible markers by crossing zhp-4(vv96)/+ males with zhp-4(vv96) V; dpy-3(e27) unc-3(e151) X and zhp-4(vv96) V; dpy-18(e364) unc-25 (e156) III hermaphrodites. Similarly, zhp-4(H26A) males were crossed with zhp-4(H26A) V; dpy-3(e27) unc-3(e151) hermaphrodites. NonUnc, nonDpy F1 cross progeny were picked and allowed for self-fertilize. F1s that were homozygous for zhp-4(vv96) and zhp-4(H26A) respectively were identified by embryonic lethality (Emb) and high incidences of males (Him) in the F2 progeny. For wild-type, 1973 (1334 wild-type and 639 recombinants) F2 progeny were scored from 10 dpy-3 unc-3/+ + and 3451 (2966 wild-type and 485 recombinants) F2 progeny were scored from 15 dpy-18 unc-25/+ + heterozygotes. For zhp-4(vv96) mutants, 507 (437 wild-type and 70 recombinants) F2 progeny were scored from 37 dpy-3 unc-3/+ +; zhp-4(vv96)/zhp-4(vv96) and 815 (710 wild-type and 105 recombinants) F2 progeny were scored from 67 dpy-18 unc-25/+ +; zhp-4(vv96)/zhp-4(vv96). For zhp-4(H26A) mutants, 1165 (913 wild-type and 252 recombinants) F2 progeny were scored from 57 dpy-3 unc-3/+ +; zhp-4(H26A)/zhp-4(H26A) heterozygotes and 558 (524 wild-type and 34 recombinants) F2 progeny were scored from 35 dpy-18 unc-25/+ +; zhp-4(H26A)/zhp-4(H26A) heterozygotes. Recombination frequencies were calculated as previously described [61], where the frequency (p) between two markers was calculated using the formula p = 1 - (1 - 2R)1/2, where R is the number of visible recombinant individuals divided by the number of total progeny. The number of total progeny for the hermaphrodite was calculated as 4/3 X (number of Wts + one recombinant class) to compensate for the inviability of the double homozygote class. Both classes of recombinants were used in the calculations.
The zhp-4(vv96) allele was recovered from a “Green Egg” mutagenesis screen (50 mM EMS) that isolated mutants with X-chromosome segregation defects [14]. Cloning of vv96 revealed a C to T substitution at the 160th codon, which changes the glutamine residue (Q) into a premature stop codon in the coding sequence Y39B6A.16, predicted to be a paralog of ZHP-3 and named ZHP-4 [29]. The wild-type tagged line of zhp-4(vv117[zhp-4::ha]) was generated by Shaolin Li (Gene Editing Services). All the other alleles were generated by directed mutagenesis using CRISPR-Cas9 protocol previously described [64] with the only difference that Cas9 protein was purchased from PNA Bio (CP01-200). For a list of sgRNAs and repair templates refer to S1 Table. In the case of zhp-4(vv103) an indel mutation was introduced in the RING-finger domain in front of the last two cysteine residues, which resulted in a frameshift and eventually a premature stop codon. The wild-type sequence is ATTATGTCATCCACCGGAAG-AAG while the mutant sequence is ATTATGTC——CGGAAGAAAG. The mutant zhp-4(vv96::ha) has been created by the positioning of the tag right before the stop codon introduced by vv96 mutation. This strain perfectly mimics the zhp-4(vv96) untagged worms allowing us to use either of them according to necessity. The ring mutants zhp-3(vv137[H25A]) and zhp-4(vv138[H26A]) harbour the following mutations respectively: CAC-to-GCC and CAT-to-GCC, both substituting a highly conserved His to an Ala.
To raise antibodies against ZHP-4 and avoid cross-reactivity with other RING domain containing proteins, a fragment of 372 base pairs corresponding to the C-terminus of ZHP-4 was cloned into two bacterial expression vectors: pGEX-6p-2, containing the GST tag at the N-terminus (GE Healthcare) and pET28a (Qiagen), to generate an N-terminal 6xHis-fusion protein. Recombinant proteins were purified under native conditions using anti-GST beads (GE Healthcare) and Ni-NTA matrix (Qiagen) respectively following the manufacturer’s instructions. GST::ZHP-4 was used for antibody production in rat and 6xHis::ZHP-4 was used for sera purification (Medimab). ZHP-4 antibody was purified using activated supports according to the manufacturers’ protocols (Affi-Gel 10, BioRad).
For whole embryo staining, thirty-forty of 24-26h post-L4-staged adults were dissected in 1xPBS, followed by freeze crack using liquid nitrogen and fixation in methanol -20°C for 30 minutes. For whole germline staining, gonads of 24-26-h post-L4 staged adults were dissected in 1XPBS and fixed by 1% paraformaldehyde for 5 minutes, followed by freeze crack and fixation in 100% methanol for 5 minutes at -20°C. After fixation in methanol, slides were washed with PBS-T (0.1% Tween-20) for 5 minutes 3 times. Gonads were then blocked with 1% BSA in PBS-T for an hour and incubated with primary antibodies overnight at 4°C. The following day, slides were washed for 3 times 15-minutes each and then incubated with secondary antibodies for two hours. Afterwards, they were washed 3 times, 15 minutes each. 1μg/μL of DAPI in anti-fading agent (Vectashield) was added onto the slides. Images consisting of 15–20 stacks (of 0.2μm increments), were acquired and processed using a Delta Vision Deconvolution system equipped with an Olympus 1X70 microscope or a Spinning-disc confocal microscope (Leica DMI 6000B inverted microscope equipped with a Quorum WaveFX spinning Disc and EM CCD camera). The following antibodies were used in this study: guinea pig and rabbit α-HTP-3 (1:500–1:750), goat α-SYP-1 (1:1000, gift from M. Colaiacovo), rabbit α-HIM-8 (1:200, Novus Biological, 41980002), mouse α-GFP (1:200, AbCAM ab290), rabbit α-RAD-51 (1:1125), guinea pig α-ZHP-3 (1:750), rabbit α-AIR-2 (1:200), rabbit α-HTP-1 (1:400) rat α-ZHP-4 (1:200), mouse α-HA (1:100, BioLegend, 901513), mouse monoclonal α-SMO-1 6F2 (1:10, DSHB), tubulin-FITC conjugate (1:500, Sigma F-2168), guinea pig α-SUN-1 S8Pi (1:700). For specificity of anti-HA staining see S2B Fig. Secondary antibodies used in this study were: AlexaFLuor 555 goat α-guinea pig (Molecular Probes, A21435) and α-rabbit (Invitrogen, A21429), AlexaFluor 488 goat α-rabbit (Molecular Probes, A11034), AlexaFLuor 488 donkey α-guinea pig, AlexaFluor 555 donkey α-goat (Abcam 150130), AlexaFluor 488 goat α-rat and AlexaFluor 488 goat α-mouse (Jackson ImmunoResearch, 106498), all of which were used in 1:1000 dilution.
RMH-1 and COSA-1 are both fused with GFP and, following anti-GFP staining, all the foci were scored in each nucleus in the entire pachytene region of each gonad. For representation of the data, the pachytene region was divided into six equal zones and labelled as: early, early-mid, mid, mid-late, late pachytene, and pachytene exit; three gonads were scored for each genotype. For RAD-51 an anti RAD-51 antibody was used for the staining, and foci were scored along the whole gonad of each genotype untill the end of pachytene and then divided into six equal zones (three gonads per genotype were scored).
RNA interference experiments were performed as described previously [65]. In brief, dsRNA was generated using PCR amplification of 946 bp of rad-54 gene using the primers TTCAGGACGAACGGAGGAAC and TTCCACTGTCCACTGGCATC, followed by in vitro transcription with T7 RNA polymerase (Ambion). At 6–8 hours post-L4, very young hermaphrodites were injected and after 2 days processed for cytological analyses. The efficacy of the RNAi was never complete since the injected animals never showed sterility as expected by complete knockdown of rad-54 [41]. All the eggs laid by injected animals hatched, and their progeny were not sterile; however, cytological analysis of two day post injection animals showed an increase of RAD-51 foci in their germlines until mid-pachytene stage (S4B Fig). Scoring of these foci revealed levels of RAD-51 foci that were comparable to those previously reported [41]. Therefore, our analysis was based on scoring RAD-51 foci in zones 1 through 4 which were affected by rad-54(RNAi).
Distributions of DAPI-stained bodies, RMH-1 foci and COSA-1 foci were statistically tested by Kruskal-Wallis test followed by Dunn’s multiple pairwise coparison tests. The significance of RAD-51 foci scoring was tested by Mann-Whitney U test while embryonic lethality and incidence of males by ANOVA followed by multiple pairwise comparison tests. All calculations were performed with Prism 5 (GraphPad) and p < 0.05 was considered significant.
|
10.1371/journal.ppat.1007791 | A fungal ABC transporter FgAtm1 regulates iron homeostasis via the transcription factor cascade FgAreA-HapX | Iron homeostasis is important for growth, reproduction and other metabolic processes in all eukaryotes. However, the functions of ATP-binding cassette (ABC) transporters in iron homeostasis are largely unknown. Here, we found that one ABC transporter (named FgAtm1) is involved in regulating iron homeostasis, by screening sensitivity to iron stress for 60 ABC transporter mutants of Fusarium graminearum, a devastating fungal pathogen of small grain cereal crops worldwide. The lack of FgAtm1 reduces the activity of cytosolic Fe-S proteins nitrite reductase and xanthine dehydrogenase, which causes high expression of FgHapX via activating transcription factor FgAreA. FgHapX represses transcription of genes for iron-consuming proteins directly but activates genes for iron acquisition proteins by suppressing another iron regulator FgSreA. In addition, the transcriptional activity of FgHapX is regulated by the monothiol glutaredoxin FgGrx4. Furthermore, the phosphorylation of FgHapX, mediated by the Ser/Thr kinase FgYak1, is required for its functions in iron homeostasis. Taken together, this study uncovers a novel regulatory mechanism of iron homeostasis mediated by an ABC transporter in an important pathogenic fungus.
| Essential element iron plays important roles in many cellular processes in all organisms. The function of an ATP-binding cassette (ABC) transporter Atm1 in iron homeostasis has been characterized in Saccharomyces cerevisiae. Our study found that FgAtm1 regulates iron homeostasis via the transcription factor cascade FgAreA-HapX in F. graminearum and the function of FgHapX is dependent on its interaction with FgGrx4 and phosphorylation by the Ser/Thr kinase FgYak1. This study reveals a novel regulatory mechanism of iron homeostasis in an important plant pathogenic fungus, and advances our understanding in iron homeostasis and functions of ABC transporters in eukaryotes.
| Iron is an essential element for growth, and can be present in various forms such as iron ions, heme and iron sulfur clusters that play critical roles in respiration, DNA synthesis and repair, ribosome biogenesis, metabolism and other cellular processes in all organisms [1–3]. In mammals, iron deficiency anemia is the most extended and common nutritional disorder in the world [4, 5]. In pathogenic fungi, the defects in iron uptake lead to decreased virulence [6, 7]. However, excess iron has the ability to generate toxic reactive oxygen species (ROS) through Fenton's reaction resulting in damage to cellular components [8]. Iron overload in liver and other organs from hepcidin regulation disorder is associated with hereditary hemochromatosis [9, 10]. Consequently, all organisms have developed tightly homeostatic regulatory mechanisms to balance uptake, consumption and storage of iron.
ATP-binding cassette (ABC) transporters that contain transmembrane domains (TMDs) and structurally conserved nucleotide-binding domains (NBDs) actively transport a wide variety of compounds across biological membranes [11]. ABC transporters play important roles in transporting compounds and regulating various physiological processes, including fatty acid metabolism, ribosome biogenesis, and mRNA translation [12, 13]. Recently, ABC transporters have been implicated in endocytosis and hyphal formation in Candida albicans [14], autophagy in human [15, 16], lifespan regulation in Drosophila [17], and the establishment of terrestrial lifestyle in plants [18]. However, our understanding of ABC transporters involved in iron homeostasis is limited. In S. cerevisiae, two ABC transporters Atm1 and Mdl1 have been found to be associated with iron homeostasis. Atm1 regulates the assembly of cytoplasmic and nucleic iron-sulfur (Fe-S) proteins might via transporting glutathione (GSH)-linked [2Fe-2S] clusters ((GS)4-[2Fe-2S]) from mitochondria to cytosol [19–20]. Mdl1 exports the proteolytic products generated by the m-AAA protease, and the over-expression of Mdl1 partially restores the defects in Atm1 mutant [21, 22]. In addition, an ABC transporter in Mycosphaerella graminicola (MgAtr7) harboring a dityrosine/pyoverdine biosynthetic domain is required for siderophore production and subsequently modulates iron homeostasis [23].
To date, Atm1 homologs have been found to control assembly of cytoplasmic and nuclear Fe-S proteins in S. cerevisiae, Arabidopsis thaliana and Homo sapiens [24–26]. But the regulatory mechanism of iron homeostasis modulated by Atm1 is only characterized in S. cerevisiae [27–30]. In the budding yeast, the Fe-S proteins monothiol glutaredoxins Grx3/4 sense GSH-linked [2Fe-2S] clusters exported by Atm1 from mitochondria to cytoplasm [31]. Depletion of ATM1 impairs the loading of GSH-linked [2Fe-2S] clusters onto monothiol glutaredoxins Grx3/4, thus hindering the formation of the complex containing Grx3/4 and the cytosolic proteins Fra1/2. This subsequently enhances retention of the transcription factor Aft1/2 at the promoter of iron acquisition genes, therefore leading to constitutive gene activation [27–31]. Except for the budding yeast, the functions and regulatory mechanisms of Atm1 orthologs in iron homeostasis have not been documented in other organisms.
F. graminearum is an economically important plant pathogen that causes cereal scab disease worldwide [32]. In addition to yield reduction, mycotoxins such as deoxynivalenol (DON) and zearalenone (ZEA) produced by the causal agent constitute a serious threat to food security and human health [33]. F. graminearum contains many more ABC transporters than most other representative fungi from major evolutionary lineages within the fungal kingdom [34, 35]. After the screening of 60 ABC knockout mutants for sensitivity to iron stress, we found that only the FgAtm1 (Atm1 ortholog) mutant was highly sensitive, whereas the mutants of Mdl1 (FGSG_01885) and MgAtr7 (FGSG_03735) orthologs were not involved in iron regulation in F. graminearum. We therefore focused on exploring the functions of FgAtm1 in regulating iron homeostasis in F. graminearum.
In this study, we revealed that the deletion of FgATM1 impedes the activity of cytosolic Fe-S proteins nitrite reductase and xanthine dehydrogenase, which in turn induces transcription factor FgAreA, and subsequently activates the transcription factor FgHapX. The phosphorylation of FgHapX is mediated by the Ser/Thr kinase FgYak1 and is further required for the transcriptional regulation of iron-related genes. It is worth to note that this interaction between FgHapX and the monothiol glutaredoxin FgGrx4 is also required for the transcriptional activity of FgHapX, which is dramatically different from what is known in the budding yeast. Overall, results from this study reveal a regulatory mechanism of iron homeostasis mediated by FgAtm1 and the transcription factor cascade FgAreA-HapX in F. graminearum, which will help us improve the understanding of iron-homeostatic regulation in eukaryotes.
F. graminearum contains 62 putative ABC transporters. In order to explore functions of ABC transporters, we deleted each of them using a homology recombination strategy. Among 62 ABC transporter genes, 60 were deleted successfully, and two genes FGSG_07101 and FGSG_04181 are essential for F. graminearum growth [35]. To explore functions of ABC transporters in iron homeostasis, we screened these 60 deletion mutants for the sensitivity to iron stress and found that the mutant of FGSG_10911 was supersensitive to iron stress (S1 Fig), indicating that this ABC transporter may play important roles in iron homeostasis regulation. The BLAST analysis showed that FGSG_10911 is homologous to S. cerevisiae Atm1 (Fig 1A), and thus we named the gene FgAtm1. We complemented ΔFgAtm1 with FgAtm1-GFP and N-terminal mitochondrion-targeting sequence of FgAtm1 (FgAtm1N1-111)-GFP, respectively. Subcellular localization observation revealed that FgAtm1-GFP co-localized with the mitochondrial dye MitoTracker, and the N-terminal mitochondrion-targeting sequence (http://www.cbs.dtu.dk/services/TargetP/) is thought to be responsible for its mitochondrial localization (Fig 1B and S2 Fig).
Phenotypic characterization showed that ΔFgAtm1 displayed hypersensitivity to 0.5 mM Fe2+ and 2 mM Fe2+ supplemented into minimal medium (MM) and potato dextrose agar (PDA), respectively (Fig 1C). In contrast, after treatment with iron-specific chelating agent bathophenanthroline disulfonate (BPS) at 0.3 mM, ΔFgAtm1 grew better than untreated cultures on MM and PDA (Fig 1C). Determination of intracellular iron by using fluorescent iron-binding dye FeRhoNox-1 and a colorimetric ferrozine-based assay revealed a high level of iron accumulation in both mitochondria and the whole cell of ΔFgAtm1 (Fig 1D). The iron content in mitochondria was clearly higher than that in whole cell (Fig 1D). In addition, the effect of iron stress on conidial germination was also determined. As shown in Fig 1E, 98% of the wild type conidia germinated after incubation at 28°C for 24 h in the trichothecene biosynthesis induction (TBI) medium that contains a trace amount of Fe2+ [36], while, ΔFgAtm1 conidia did not germinate even after 48 h under the same conditions. When 0.3 mM BPS was added into TBI to chelate Fe2+, 44% and 72% of ΔFgAtm1 conidia germinated after incubation for 24 and 48 h, respectively. Conidial germination in the wild type was not affected by BPS treatment (Fig 1E). To further confirm that the supersensitivity of ΔFgAtm1 to iron stress is due to the deletion of the FgATM1 gene, the mutant was complemented with a full-length wild-type FgATM1 gene amplified with the primers listed in S1 Table. The complemented strain ΔFgAtm1-C contained a single copy of FgATM1, which was inserted into the genome of ΔFgAtm1 (S3 Fig). The defects of mycelial growth, conidial germination and accumulation of iron in ΔFgAtm1 were restored to the wild-type phenotypes in ΔFgAtm1-C (Fig 1C–1E). These results strongly indicate that the lack of FgAtm1 leads to accumulation of intracellular iron in F. graminearum.
To determine whether FgAtm1 regulates the assembly of cytosolic Fe-S proteins, we studied the activity of Fe-S proteins isopropyl malate isomerase (FgLeu1, FGSG_ 09589), aconitase (FgAco1, FGSG_07953), fumarase (FgFum1, FGSG_08712). S. cerevisiae Leu1 is a cytosolic Fe-S protein and catalyzes the second step in leucine biosynthesis [26]. Cytosolic and mitochondrial Fe-S proteins Aco1 and Fum1 both participate in glyoxylate shunt in cytosol and TCA cycle in mitochondria in the budding yeast, respectively [37, 38]. Enzyme activity assays showed that the activities of FgLeu1, FgAco1 and FgFum1 in the cytosol of ΔFgAtm1 were attenuated by 37, 44 and 28% respectively, when compared to those in the wild type. In contrast, the activities of FgAco1 and FgFum1 in ΔFgAtm1 mitochondria were not significantly changed (Fig 2A). Further, feeding ΔFgAtm1 with leucine (the final catalytic product of FgLeu1) or the final catalytic product of other cytosolic Fe-S proteins nitrite reductase, glutamate dehydrogenase or xanthine dehydrogenase [39, 40] also accelerated the growth of ΔFgAtm1 on MM (Fig 2B). In S. cerevisiae, GSH-linked [2Fe-2S] clusters were reported to be the substrate of Atm1, and deletion of ATM1 caused increased GSH content in the whole cell [19, 20, 41]. We therefore tested the content of GSH, and found that it was increased by 59% and 50% in mitochondria and the whole cell of ΔFgAtm1, respectively (Fig 2C). Similar to these reported in the budding yeast [19–20], the results of this study indicate that FgAtm1 also modulates the assembly of cytosolic Fe-S proteins likely via transporting GSH-linked [2Fe-2S] clusters from mitochondria into F. graminearum cytoplasm.
To further explore the regulatory mechanism of FgAtm1 in iron homeostasis, we determined iron stress sensitivity for nine cytoplasmic Fe-S protein mutants constructed in our laboratory and found that nitrite reductase (FgNiiA, FGSG_08402) and xanthine dehydrogenase (FgXdh, FGSG_01561) mutants showed increased sensitivity to Fe2+ (Fig 3A and S5 Fig). Similar to what were reported in the yeasts [37, 38, 42, 43], the remaining proteins that we tested were not involved in iron stress responses. Previous studies have shown that nitrite reductase and xanthine dehydrogenase are key enzymes for non-preferred nitrogen source utilization, and are regulated by nitrogen metabolism regulator AreA in Aspergillus nidulans, F. oxysporum and F. graminearum [40, 44–47]. Nitrite reductase is responsible for nitrate utilization [40] and xanthine dehydrogenase is required for oxidizing hypoxanthine to xanthine [48]. Quantitative reverse transcription PCR (qRT-PCR) assays showed that transcription of FgAREA was induced by the deletion of FgATM1, FgNIIA (encoding nitrite reductase) or FgXDH (encoding xanthine dehydrogenase), as well as by the non-preferred nitrogen sources, NaNO3 or hypoxanthine (Fig 3B). Surprisingly, we found that ΔFgAreA also exhibited elevated sensitivity to Fe2+ (Fig 3A). Therefore, we hypothesized that the deletion of FgAtm1 leads to reduced activities of FgNiiA and FgXdh, which induces overexpression of FgAreA.
To explore the role of FgAreA in regulating iron homeostasis, we first performed serial analysis of gene expression (SAGE) assay for the mutant ΔFgAtm1, and found that 56 iron-related genes were differentially expressed (>2-fold) in ΔFgAtm1 (S2 Table). Further, qRT-PCR assay confirmed that the transcription level of the transcription factor FgHAPX was dramatically increased in ΔFgAtm1 as compared to that of the wild type (S6 Fig). To understand the mechanism by which FgAreA regulates FgHAPX expression, we studied the binding ability of FgAreA to the promoter of FgHAPX in the wild type bearing FgAreA-GFP (PH-1::FgAreA-GFP) and in ΔFgAtm1 bearing FgAreA-GFP (ΔFgAtm1::FgAreA-GFP) using chromatin immunoprecipitation and quantitative PCR (ChIP-qPCR) assay. A strain transformed with GFP alone was used as a negative control. ChIP-qPCR analyses showed that enrichment of FgAreA at the FgHAPX promoter was induced by the deletion of FgATM1 as well as the treatment by NaNO3 or hypoxanthine (Fig 3C). GFP enrichment at the FgHAPX promoter was undetectable in the negative control strain (Fig 3C). Additionally, qRT-PCR assays revealed that FgHAPX transcription was also induced with NaNO3 or hypoxanthine treatment. Moreover the induced expression of FgHAPX upon non-preferred nitrogen source treatment was dependent on FgAreA (Fig 3D). These results indicated that FgAreA binds to the promoter of FgHAPX and regulates its transcription.
To explore the function of FgHapX in iron homeostasis, we first constructed a FgHAPX deletion mutant ΔFgHapX, and tested the sensitivity of ΔFgHapX to iron stress. As shown in Fig 4A and 4B, ΔFgHapX became more sensitive to iron stress in comparison with the wild type, although ΔFgHapX did not show an obvious change in total iron content. Furthermore, we determined the content of extra- and intracellular siderophores secreted by ΔFgHapX with a chrome azurol S (CAS) assay, and found that the lack of FgHAPX caused reduced extracellular siderophore but not intracellular siderophore (Fig 4C). Similarly, qRT-PCR assays revealed that iron acquisition genes were down-regulated and iron-consuming genes were up-regulated in ΔFgHapX (Fig 4D). We knocked out FgHAPX in ΔFgAtm1, and checked whether the defects of ΔFgAtm1 were partially recovered by deletion of FgHAPX. As we expected, the double mutant ΔFgAtm1-HapX grew better and displayed decreased sensitivity to iron stress than ΔFgAtm1 (Fig 4A). Determination of iron and siderophore revealed that lack of FgHAPX in ΔFgAtm1 led to a reduced iron concentration, and decreased extra- and intracellular siderophores in comparison with those in ΔFgAtm1 (Fig 4B and 4C). Expression levels of iron acquisition genes in ΔFgAtm1-HapX were reduced and the transcription of iron-consuming genes were elevated compared to those in ΔFgAtm1 (Fig 4D).
HapX homologs in A. nidulans and A. fumigatus have been found to repress the transcription of iron-consuming genes by binding to CCAAT motif [49, 50]. The multiple EM for motif elicitation (MEME) analyses showed that the genes FgCYCA, FgHEMA, FgLYSF and FgACOA involved in the iron-consuming have the CCAAT motif (Fig 5A). Electrophoretic mobility shift assay (EMSA) further confirmed that FgHapX bound the promoters of iron-consuming genes (Fig 5B). Iron acquisition genes contained the GATA, but not CCAAT, motif in their promoters (Fig 5A), indicating that other regulator(s) modulates the transcription of iron uptake genes directly in F. graminearum. In A. fumigatus, HapX activates the expression of siderophore-mediated iron uptake genes via transcriptional repression of SreA that suppresses the transcription of iron acquisition genes via binding to the GATA motif in their promoter [50, 51]. The MEME analysis and EMSA assay showed that FgHapX could bind the FgSREA (SreA homolog) promoter (Fig 5A and 5B). Moreover, the qRT-PCR assay revealed that deletion of FgHAPX led to elevated transcription of FgSREA (Fig 5C). We further obtained a FgSREA deletion mutant ΔFgSreA, and found that ΔFgSreA displayed increased sensitivity to iron stress (Fig 5D). The deletion of FgSREA caused iron accumulation, and increased extra- and intracellular siderophores (Fig 5E and 5F), and qRT-PCR assays showed that deletion of FgSREA caused elevated expression of iron acquisition genes (Fig 5G). These results indicated that FgHapX represses the transcription of FgSreA, and subsequently activates transcription of iron acquisition genes.
To explore the regulatory mechanism of FgHapX in iron homeostasis, we performed a yeast two-hybrid (Y2H) screen of F. graminearum cDNA library, and found 50 potential FgHapX-interacting proteins (S3 Table), including the monothiol glutaredoxin FgGrx4 that is homologous to S. cerevisiae Grx3/4. Furthermore, Y2H, Co-IP and BiFC assays revealed that FgGrx4 interacted with FgHapX in the nucleus and the interaction was independent of FgAtm1 (Fig 6A–6D). Moreover, Y2H and BiFC assays showed that FgGrx4 interacted with FgHapX through its GRX domain but not its TRX domain (Fig 6B and 6D). In S. cerevisiae, lack of Grx3/4 leads to constitutive expression of iron acquisition genes, which contributes to iron accumulation in cells [52, 53]. We therefore generated a FgGRX4 deletion mutant, ΔFgGrx4, and found that ΔFgGrx4 displayed increased sensitivity to iron stress although it did not exhibit an obvious alteration in the total iron content (Fig 6E and S7A Fig). To explore the effect of FgGrx4 on FgHapX functions, we determined the quantity and localization of FgHapX in ΔFgGrx4, and found that FgGRX4 deletion did not cause an obvious change in the localization and quantity of FgHapX (S7B and S7C Fig). However, similar to the FgHapX deletion, the deletion of FgGRX4 led to significantly decreased expression of iron acquisition genes, and increased transcription of iron-consuming genes (Fig 6G). These results indicated that FgGrx4 is required for the transcriptional activity of FgHapX in F. graminearum.
In eukaryotes, phosphorylation of transcription factors frequently has been found to regulate their activities. Phosphoproteome assay showed that FgHapX contains two predicted Ser residues at 245 and 338 sites that may be subject to phosphorylation (S8 Fig). To confirm the function of these two residues, we constructed a strain containing a constitutive dephosphorylated FgHapX isoform. Briefly, the two phosphorylated Ser residues were replaced by alanine, the mutated FgHapXS245A/S338A was transformed into ΔFgHapX and the resulting strain was designated as ΔFgHapX-CS245A/S338A. Next, we performed a phos-tag assay to detect the phosphorylation level of FgHapX in ΔFgHapX-C and in ΔFgHapX-CS245A/S338A. As shown in Fig 7A, the dephosphorylated level of FgHapX in ΔFgHapX-CS245A/S338A was significantly higher than that in ΔFgHapX-C. To further explore the function of FgHapX phosphorylation, we determined the sensitivity of ΔFgHapX-CS245A/S338A to iron stress. As shown in Fig 7B and 7C, similar to ΔFgHapX, ΔFgHapX-CS245A/S338A still remained highly sensitivity to iron stress. Consistently, the qRT-PCR assays showed the expression levels of iron acquisition genes were reduced and the transcription of iron-consuming genes was elevated in ΔFgHapX-CS245A/S338A (Fig 7D). In addition, these mutations did not change the quantity and localization of FgHapX (S9A and S9B Fig). Collectively, these results indicate that phosphorylation of FgHapX is required for regulating expression of iron-related genes.
To identify the potential kinase that phosphorylates FgHapX, we screened 96 kinase mutants and found that the mutant of Ser/Thr protein kinase FgYak1 (FGSG_05418) showed dramatically increased sensitivity to iron stress (Fig 7B and S10 Fig). Furthermore, co-immunoprecipitation (Co-IP) confirmed that FgYak1 interacted with FgHapX (Fig 7E). Immunofluorescence assay also revealed that FgYak1 interacted with FgHapX in the nucleus (Fig 7F). The qRT-PCR assays showed that, similar to those in ΔFgHapX and ΔFgHapX-CS245A/S338A, the transcription levels of iron acquisition genes were reduced and those of iron-consuming genes were elevated in ΔFgYak1 (Fig 7D). Importantly, the phos-tag assay revealed that the dephosphorylated level of FgHapX in ΔFgYak1 was higher than that in ΔFgHapX-C (Fig 7A). Meanwhile, the phos-tag assays showed that the phosphorylated levels of FgHapX in ΔFGSG_13318, ΔFGSG_00408, ΔFGSG_10381, ΔFGSG_06832, ΔFGSG_05734 or ΔFGSG_11812 were similar with that in ΔFgHapX-C, although these kinase mutants also showed elevated sensitivity to iron stress (S10A–S10D Fig). In addition, deletion of FgYak1 did not alter the quantity and localization of FgHapX (S9A and S9B Fig). Collectively, these results indicated that FgYak1 phosphorylates FgHapX in F. graminearum.
In S. cerevisiae, the Atm1-mediated iron regulation has been well characterized. The depletion of Atm1 impedes the loading of GSH-linked [2Fe-2S] clusters onto monothiol glutaredoxins, subsequently disrupting formation of the Grx3/4-Fra1/2 complex, which results in the failure of Aft1/2 dissociation from the promoters of iron acquisition genes [27–30]. The iron acquisition genes are therefore activated constitutively in the Atm1-depleted cells. In this study, we found that lack of FgAtm1 also leads to an overload of intracellular iron. However, the regulation mechanism of iron homeostasis mediated by FgAtm1 in F. graminearum is different from what is known in S. cerevisiae. We uncovered that deletion of FgAtm1 impedes the activity of cytosolic Fe-S proteins nitrite reductase and xanthine dehydrogenase, which conversely activates the nitrogen metabolism regulator FgAreA (Fig 8). Subsequently, FgAreA activates the transcription of repressor FgHapX via binding the FgHAPX promoter (Fig 8). Moreover, we found that FgHapX directly represses the transcription of iron-consuming genes, but also activates the expression of iron acquisition genes indirectly via suppressing the transcription of another repressor FgSREA (Fig 8). It is worthy to note that S. cerevisiae does not contain a HapX ortholog, and F. graminearum and other filamentous fungi do not have the yeast Aft1/2 orthologs. These results indicate that the regulatory networks of iron homeostasis can be distinct in different fungi.
AreA belongs to the GATA nitrogen regulator and is required for the transcription of genes responsible for the utilization of non-preferred nitrogen sources in several fungi [54]. Previous studies have found that the transcription of AREA is induced by nitrogen starvation or the treatment with nitrate in A. nidulans [55], F. graminearum [56, 57] and Fusarium fujikuroi [58]. In this study, the lack of FgAtm1 compromised the activity of nitrite reductase FgNiiA and xanthine dehydrogenase FgXdh resulting in utilization defects of the non-preferred nitrogen sources nitrite and hypoxanthine, and subsequently induced the transcription of FgAREA. This finding indicates that iron metabolism is able to affect nitrogen utilization via the Fe-S proteins in the filamentous fungus F. graminearum.
HapX is an important transcriptional repressor of intracellular iron homeostasis in filamentous fungi [59]. In A. fumigatus, HapX not only represses genes involved in iron-consuming pathways to spare iron, but also activates iron acquisition genes to acquire iron via suppressing another repressor SreA during iron starvation [50], which is consistent with our finding. However deletion of HapX homologs does not change the transcription of iron acquisition genes in A. nidulans, F. oxysporum and C. albicans [49, 60, 61]. Previous studies have found that the transcription factor SreA may also represses the expression of HapX during iron overload [49, 62]. In Cryptococcus neoformans, carbon metabolism regulator Mig1 promotes the transcription of HAPX under low-iron conditions [63]. In this study, we found however that FgHapX is regulated by the GATA transcription factor FgAreA. To our knowledge, this is the first observation that the iron regulator HapX is regulated by a nitrogen metabolism regulator AreA in filamentous fungi.
In the current study, we also found that the functions of Grx4 orthologs vary dramatically in F. graminearum and yeasts. First, monothiol glutaredoxin FgGrx4 is required for the transcriptional activity of FgHapX via its interaction with FgHapX in F. graminearum. In yeasts, however, Grx4 homologs combine with the GSH-linked [2Fe-2S] clusters and then interact with transcription factors S. cerevisiae Aft1/Aft2 or Schizosaccharomyces pombe Php4/ Fep1 to disassociate these factors from the promoters of iron regulation genes [28–30, 64–66]. Second, the interaction of FgGrx4 and FgHapX is independent of the presence of FgAtm1, and the TRX domain of FgGrx4 doesn’t interact with FgHapX. In S. cerevisiae, the interaction of Grx4 and Aft1/2 is dependent on Atm1 [30]. In S. pombe, the interaction of GRX domain of Grx4 with Php4 or Fep1 is dependent on iron conditions, whereas the TRX domain continuously binds to Php4 or Fep1 [65, 66]. Third, similar to FgHAPX deletion, deletion of FgGRX4 did not change intracellular iron content (S7A Fig). However, lack of Grx3/4 leads to iron accumulation in S. cerevisiae cells [52]. Fourth, deletion of FgGRX4 caused the down-regulation of iron acquisition genes in F. graminearum. However, deletion of GRX3 or GRX4 causes constitutive expression of iron acquisition genes in S. cerevisiae [52, 53]. In S. pombe, GRX4 disruption causes constitutive transcription of iron acquisition genes regulated by Fep1, or constitutive transcription of iron-consuming genes regulated by Php4 [64, 66]. Taken together, these results indicate that the function of FgGrx4 in F. graminearum is dramatically different from that of Grx4 in the yeasts.
Phosphorylation of transcription factors mediated by various kinases has been found to modulate their localization, protein accumulation and DNA binding ability in eukaryotic organisms [67]. In mammals, phosphorylation of the organismal lifespan-related transcription factor FOXO by serum and glucocorticoid-induced kinase (SGK) results in the exclusion from the nucleus and repression of transcriptional activity [68]. In A. thaliana, multisite light-induced phosphorylation of phy-interacting basic Helix Loop Helix (bHLH) transcription factor PIF3 causes its degradation [69]. In the fission yeast, phosphorylation of sterol biosynthesis regulator SpSre1 mediated by casein kinase Hhp2 reduces its protein quantity by accelerating its degradation [70]. Whereas in Lotus japonicus, phosphorylation of a root nodule development-associated transcriptional activator CYCLOPS by calcium- and calmodulin-dependent kinase (CCaMK) increases its DNA binding activity at the target gene promoters [71]. In this study, we discovered that phosphorylation of FgHapX is required for its transcription activity, but not for its quantity and localization (Fig 6D, S9A and S9B Fig). Furthermore, we identified that FgHapX is subject to phosphorylation mediated by the kinase FgYak1. Previous studies have reported that the Ser/Thr protein kinase Yak1 controls cell growth in response to glucose depletion by negatively regulating the cAMP-PKA pathway in S. cerevisiae, and regulates the emergence and maintenance of hyphal growth of C. albicans [72, 73]. To our knowledge, it is the first report on involvement of a kinase (Yak1) in regulating iron homeostasis by phosphorylation of a transcription factor in fungi.
In A. thaliana, Atm1 ortholog Atm3 was found to regulate the assembly of cytoplasmic molybdenum cofactor (Moco) proteins, besides Fe-S proteins [Bernard et al., 2009]. The precursor of Moco cyclic pyranopterin monophosphate (cPMP) that is synthesized in mitochondria was reported as another substrate of Atm3 [25, 74]. In the current study, we also found that F. graminearum FgAtm1 modulates the assembly of Moco protein nitrate reductase (FgNiaD) (S11 Fig). Further, we also found that F. graminearum FgAtm1 modulates mitochondrial function and redox balance besides iron homeostasis, which are in agreement with the studies in these reports of S. cerevisiae and C. neoformans [41, 75, 76]. The ΔFgAtm1 mutant displayed decreased sensitivity to the mitochondrial respiratory inhibitors diphenylene iodonium (DPI) and rotenone (complex I) and antimycin A (complex III) (S12A Fig). The mutant also exhibited increased reactive oxygen species (ROS) content and elevated sensitivity to hydrogen peroxide (H2O2) (S12B–S12D Fig), since mitochondrial respiratory complexes I and III are known to be major generators of ROS in eukaryotic cells [77]. In addition, phenotypic determination showed that deletion of FgATM1 led to the defects in asexual and sexual development, virulence and secondary metabolite production (S13 Fig). The phenotypic defects might result from the imbalance of nutrient, iron and redox and impaired mitochondrial functions in ΔFgAtm1.
F. graminearum wild-type strain PH-1 was used as a parental strain for transformation experiments in this study. Mycelial growth of the wild type and the resulting transformants were assayed on potato dextrose agar (PDA) or minimal medium (MM) as described previously [78, 79]. To determine sensitivity to iron stress, 5-mm mycelial plugs of each strain taken from a 3-day-old colony edge were inoculated on PDA or MM supplemented without/with Fe2+, H2O2, or catalytic product of each Fe-S protein, and then incubated at 25°C for 3 days in the dark. Three biological replicates were used for each strain and each experiment was repeated three times independently.
The double-joint PCR approach [80] was used to generate the gene replacement construct for each target gene. Briefly, for each gene, 5’ and 3’ flanking regions were amplified with the primer pairs listed in S1 Table and the resulting amplified sequences were then fused with the hygromycin resistance gene cassette (HPH) driven by the constitutive trpC promoter which was amplified from the pBS-HPH1 vector [81]. Protoplast transformation of F. graminearum was carried out using the protocol described previously [82]. Putative gene deletion mutants were identified by PCR assays with relevant primers (S1 Table) and the FgATM1 deletion mutation was further confirmed by Southern hybridization assays (S3 Fig).
To construct the FgAtm1-GFP cassette, FgATM1 containing the promoter region and open-reading frame (without the stop codon) was amplified with the relevant primers (S1 Table). The resulting PCR products were co-transformed with XhoI-digested pYF11 containing a geneticin resistance gene (NEO) [83] into the yeast strain XK1-25 [84] using the Alkali-Cation Yeast Transformation Kit (MP Biomedicals, Solon, USA) to generate the recombined FgAtm1-GFP fusion vector. Subsequently, the FgAtm1-GFP fusion vector was recovered from the yeast transformant using the Yeast Plasmid Kit (Solarbio, Beijing, China) and then transferred into Escherichia coli strain DH5α for amplification. Using the similar strategy, FgYak1 (FGSG_05418)—and FgTri1 (FGSG_00071) -GFP fusion cassettes were constructed. Using the similar strategy, FgGrx4 (FGSG_01317)—and FgYak1-Flag fusion cassettes were constructed by co-transformation with XhoI-digested PHZ126 vector. Similarly, FgHapX-CYFP and FgGrx4-NYFP fusion cassettes were constructed by co-transformation with XhoI-digested PHZ68 and PHZ65 vectors, respectively.
The double-joint PCR approach [80] was also used to construct FgHapX-, FgHapXS245A/S338A- and FgLeu1 (FGSG_09589)-mCherry cassettes. Briefly, the target gene containing the promoter region and open-reading fragment and the geneticin resistance gene (NEO) were amplified, and then fused with mCherry fragment. Before protoplast transformation, each fusion construct was verified by DNA sequencing. The transformation of F. graminearum was carried out using the previously described protocol [82]. All the mutants generated in this study were preserved in 15% glycerol at −80°C.
For determination of mitochondrial and total iron content, About 50 mg of fresh mycelia were lysed by 2% cellulase (Ryon Biological Technology CO, Ltd, Shanghai, China), 2% lysozyme (Ryon Biological Technology CO, Ltd, Shanghai, China) and 0.2% driselase from Basidiomycetes sp. (Sigma, St. Louis, MO, USA) for 4–6 hours. After filtration with funnel and filter paper, the filtrate was centrifuged at 5000 g at 4°C for 10 min. The protoplast was used for total iron determination and mitochondrial extraction with a Cell Mitochondria Isolation Kit (Beyotime Industrial Co., Ltd., Shanghai, China). Iron content was determined using the colorimetric ferrozine-based assay previously reported, with ferrozine as chelator and ferric chloride as a standard [28, 29, 85]. Briefly, aliquots (100 μl) of cell lysates were mixed with 100 μl of 10 mM HCl (the solvent of the iron standard FeCl3), and 100 μl of the iron-releasing reagent (a freshly mixed solution of equal volumes of 1.4 M HCl and 4.5% (w/v) KMnO4 in H2O). The mixtures were incubated within a fume hood at 60°C for 2 h. After the mixtures had cooled to room temperature, 30 μl of the iron-detection reagent (6.5 mM ferrozine, 6.5 mM neocuproine, and 2.5 M ammonium acetate and 1 M ascorbic acid) was added to each tube. After 30 min, 200 μl of the solution in each tube was transferred into a well of a 96-well plate and the absorbance was measured at 550 nm on a microplate reader. The linear range of the ferrozine assay is from 0.2 to 30 nmol.
To determine the content of extra- and intracellular siderophores, each strain was cultured in CM for 36 hours, then transferred to MM lack of FeSO4 and amended 0.3 mM iron-specific chelating agent bathophenanthroline disulfonate (BPS) at 25°C for 8 hours. After filtration with funnel and filter paper, the supernatant was used for determining the content of extracellular siderophore, and the mycelia was used for determining the content of intracellular siderophore. Finely ground mycelia (50 mg) were resuspended in 50 mM potassium phosphate buffer. After vigorous vortexing, the cellular debris was pelleted, and the supernatant was determined for the content of intracellular siderophore. An aliquot of supernatant (1 ml) was mixed with 1 ml chrome azurol S (CAS) assay solution that was prepared. After incubation in the dark for 1 hour at room temperature, the absorbance of each sample was measured at 630 nm and the relative content of siderophore was calculated according to previous study [86].
For the conidiation assay, fresh mycelia (50 mg) of each strain were inoculated in a 50-ml flask containing 20 ml of carboxymethyl cellulose (CMC) liquid medium. The flasks were incubated at 25°C for 4 days in a shaker (180 rpm). Subsequently, the number of conidia in each flask was determined using a hemacytometer. Three biological replicates were used for each strain and each experiment was repeated three times independently.
Virulence assays on wheat spikelets, corn silks, and wheat seedling leaves were conducted as described previously [35]. For virulence on wheat spikelets, a 10 μl aliquot of conidial suspension (105 conidia/ml) was injected into a floret in the central section spikelet of a single flowering wheat head of susceptible cultivar Jimai 22. Fifteen days after inoculation, the infected spikelets in each inoculated wheat head were recorded. For virulence on corn silks and wheat seedling leaves, a 5-mm mycelial plug of each strain was inoculated on the middle of corn silks and wheat seedling leaves, and then cultured for 4 days. Ten replicates were used for each strain and each experiment was repeated three times independently. To determine DON production, each strain was grown in liquid trichothecene biosynthesis induction (TBI) medium at 28°C for 3 d in a shaker (150 rpm) in the dark. A DON Quantification Kit (Wise Science, Zhenjiang, China) was used to quantify the DON production for each sample. The experiment was repeated three times.
The colony edge of each strain cultured on PDA plates at 25°C for 7 days was examined with a Leica TCS SP5 imaging system (Leica Microsystems, Wetzlar, Germany). Morphology of conidia cultured in CMC liquid medium for 4 days was observed with a Leica TCS SP5 imaging system after staining with the cell wall-damaging agent calcofluor white (CFW) at 10 μg/ml. Germinating conidia in TBI liquid medium or iron-depleted TBI liquid medium were observed with a Leica TCS SP5 imaging system. Fluorescence signals were examined with a Zeiss LSM780 confocal microscope (Gottingen, Niedersachsen, Germany). For the observation of proteins tagged with GFP, mCherry or YFP, each strain was cultured in CM at 25°C for 24 h in a shaker (180 rpm) before staining with MitoTracker or 4’,6-diamidino-2-phenylindole (DAPI) [87]. For examination of iron content, each strain was stained with 5 μM FeRhoNox-1 (Goryo Chemical, Inc., Sapporo, Japan) [88–91] for 1 hours after culture in complete medium (CM) for 36 hours. For the observation of FgTri1-GFP, each strain was cultured on wheat seedling leaves at 25°C with 100% humidity for two days.
The wild type, ΔFgAtm1 and the ΔFgAtm1-C were grown in CM for 36 hours, and then were harvested after rinsing 3 times with sterile water. To determine FgLeu1 activity in cytosol, 50 mg of finely ground mycelia was resuspended in 1 ml of lysis buffer (1 M Tris-HCl pH 7.4, 1 M NaCl, 0.5 M EDTA, 1% Triton 100) for 10 min. The lysate was centrifuged at 14,000 g for 10 min at 4°C. Each protein sample (50 μg) was used for FgLeu1 activity determination. DL-threo-3-isopropylmalic acid (Wako Pure Chemical Industries, Ltd, Japan) was used as the substrate and β-isopropylmalate formation was measured by monitoring absorbance at 235 nm for 10 minutes [92, 93]. To determine FgAco1 and FgFum1 activity in cytosol and mitochondrion, the separation of mitochondrion and cytosol (without nuclei) was conducted according to the above method described for iron content determination. Each protein sample (50 μg) was used for FgAco1 activity determination. Cis-aconitic acid (Sigma-Aldrich, St. Louis, MO, USA) was used as the substrate and monitor absorbance at 340 nm for 2–4 minutes [92]. FgFum1 activity of each protein sample (50 μg) was determined with a Fumarase Specific Activity Assay Kit ab110043 (Abcam, Cambridge, UK) according to the manufacturer’s instruction. These experiments were repeated three times independently.
To examine ROS content, fresh mycelia of each strain grown in CM liquid medium for 36 hours were harvested after rinsing 3 times with sterile water. 50 mg of finely ground mycelia were resuspended in 1 ml of extraction buffer. After homogenization with a vortex shaker, the lysate was centrifuged at 14,000 g in a microcentrifuge for 10 min at 4°C. An aliquot of 10 μl supernatant was used for ROS determination with ROS ELISA Kit (Tong Wei Industrial Co., Ltd., Shanghai, China). Meantime, to examine ROS, each strain cultured on PDA plates at 25°C for 3 days was stained with 0.05% (wt/vol) nitroblue tetrazolium (NBT) for 2 h. These experiments were all repeated three times.
Total RNA isolation from mycelia of each sample with the TaKaRa RNAiso Reagent (TaKaRa Biotechnology, Dalian, China), and reverse transcription was performed with a HiScript II 1st Strand cDNA Synthesis Kit (Vazyme Biotech, Nanjing, China). The expression level of each gene was determined by qRT-PCR with HiScript II Q RT SuperMix (Vazyme Biotech, Nanjing, China). The expression of the FgACTIN gene was performed as a reference. The experiment was repeated three times.
ChIP was performed as previously described [94, 95] with additional modifications. Briefly, fresh mycelia were cross-linked with 1% formaldehyde for 15 min and then stopped with 125 mM glycine. The cultures were ground with liquid nitrogen and re-suspended in the lysis buffer (250 mM, HEPES pH 7.5, 150 mM NaCl, 1 mM EDTA, 1% Triton, 0.1% DeoxyCholate, 10 mM DTT) and protease inhibitor (Sangon Co., Shanghai, China). The DNA was sheared into ~300 bp fragments with twenty pulses of 10 s and 20 s of resting at 35% amplitude (Qsonica*sonicator, Q125, Branson, USA). After centrifugation, the supernatant was diluted with 10×ChIP dilution buffer (1.1% Triton X-100, 1.2 mM EDTA, 16.7 mM Tris–HCl, pH 8.0 and 167 mM NaCl). Immunoprecipitation was performed using monoclonal anti-GFP ab290 (Abcam, Cambridge, UK) antibody together with the protein A agarose (Santa Cruz, CA, USA) respectively. DNA was immunoprecipitated by ethanol after washing, eluting, reversing the cross-linking and digesting with proteinase K. Further, ChIP-enriched DNA was used for quantitative PCR analysis using SYBR green I fluorescent dye detection with the relative primers (S1 Table). Relative enrichment of each gene was determined by quantitative PCR and calculated first by normalizing the amount of a target cDNA fragment against that of FgACTIN as an internal control, and then by normalizing the value for the immunoprecipitated sample against that for the input. The ChIP-qPCR was independently repeated three times.
The cDNA encoding the N terminal 1–230 amino acids of FgHapX (FgHapXN1–230) was amplified and cloned into pGEX-4T-3 vector to generate GST-tagged protein. The resulting construct was transformed into the E. coli strain BL21 (DE3) after verifying the cDNA sequence. The recombinant GST-FgHapXN1–230 was purified by Ni sepharose beads and eluted by reduced glutathione. Promter DNAs were amplified using the primers in Table S1. For EMSA, Reaction mixtures containing purified GST -FgHapXN1–230, promter DNAs and 10×Binding buffer (100 mM Tris-HCl (PH 7.5), 0.5 M NaCl, 10 mM DTT, 10 mM EDTA, 50% glycerol) were incubated for 20 min at 25°C. The purified GST was used as negative controls. The reactions were electrophoresed on 1.2% agarose gel in 0.5×TAE for 45 min in 80 V under low temperature. Signals were detected in J3-3000 imaging system after dying DNA dye ethidium bromide (EB) for 15 min. The experiment was conducted independently three times.
To construct plasmids for Y2H analyses, the coding sequence of each tested gene was amplified from the cDNA of PH-1 with primer pairs indicated in S1 Table. The cDNA of each gene was inserted into the yeast GAL4-binding domain vector pGBKT7 (Clontech, Mountain View, CA, USA) and GAL4 activation domain vector pGADT7 (Clontech, Mountain View, CA, USA), respectively. The pairs of Y2H plasmids were cotransformed into S. cerevisiae strain Y2H Gold lithium acetate/single-stranded DNA/polyethylene glycol transformation protocol. In addition, a pair of plasmids, pGBKT7-53 and pGADT7 and another pair of plasmids, pGBKT7-Lam and pGADT7, served as a positive control and negative controls, respectively. Transformants were grown at 30°C for 3 d on synthetic medium (SD) lacking Leu and Trp, and then transferred to SD stripped of Ade, His, Trp and Leu to assess binding activity. Three independent experiments were performed to confirm yeast two-hybrid assay results.
To search for FgHapX-interacting proteins, we performed Y2H screens. FgHapX was cloned into the yeast vector pGBKT7. A F. graminearum cDNA library was constructed in the Y2H vector pGADT7 using total RNA extracted from mycelia and conidia. The Y2HGold that was co-transformed with the cDNA library as well as FgHapX-pGBKT7 were directly selected using SD-Trp-Leu-His. Approximately 150 potential yeast transformants containing cDNA clones interacting with FgHapX were further confirmed in selection medium SD-Trp-Leu-His-Ade.
For western blotting assay, protein samples of strains were prepared and extracted as described previously [96]. Proteins separated on the SDS-PAGE gel were transferred onto a polyvinylidene fluoride membrane with a Bio-Rad electroblotting apparatus. The polyclonal anti-Flag A9044 (Sigma, St. Louis, MO, USA), monoclonal anti-GFP ab32146 (Abcam, Cambridge, MA, USA) and monoclonal anti-mCherry ab125096 (Abcam, Cambridge, UK) antibodies were used at a 1:2000 to 1:10000 dilution for immunoblot assays. The samples were also detected with the monoclonal anti-GAPDH antibody EM1101 (Hangzhou HuaAn Biotechnology co., Ltd.) as a reference. Incubation with a secondary antibody and chemiluminescent detection were performed as described previously [97]. The experiment was repeated three times independently.
The mCherry and Flag fusion constructs were verified by DNA sequencing and transformed in pairs into PH-1. Transformants expressing the fusion constructs were confirmed by western blot analysis. In addition, the transformants bearing a single fusion construct were used as references. For Co-IP assays, total proteins were extracted and incubated with the anti-Flag (Abmart, Shanghai, China) agarose overnight at 4°C as described previously [97]. Proteins eluted from agarose were analyzed by western blotting detection with the monoclonal anti-mCherry ab125096 (Abcam, Cambridge, UK) antibody. The protein samples were also detected with monoclonal anti-GAPDH antibody EM1101 (Hangzhou HuaAn Biotechnology co., Ltd.) as a reference.
The FgHapX-CYFP and FgGrx4-NYFP fusion constructs were generated by cloning the related fragments into pHZ68 and pHZ65 vectors, respectively. FgHapX-CYFP and FgGrx4-NYFP constructs were co-transformed into PH-1 and ΔFgAtm1. In addition, a pair of constructs, FgHapX-CYFP and NYFP and another pair of constructs, FgGrx4-NYFP and CYFP were used as negative controls. Transformants resistant to both hygromycin and zeocin were isolated and confirmed by PCR. YFP signals were examined with a Zeiss LSM780 confocal microscope (Gottingen, Niedersachsen, Germany).
Proteins of the wild-type PH-1 were extracted in the lysis buffer (8 M urea, 50 mM Tris 8.0, 1% NP40, 1% sodium deoxycholate, 5 mM dithiothreitol, 2 mM EDTA, 30 mM nicotinamide, 3 μm trichostatin A, 1% protease inhibitor Cocktail and 1% phosphatase inhibitor cocktail). For trypsin digestion, the protein sample was diluted in 0.1 M TEAB (triethylammonium bicarbonate), and digested with 1/25 trypsin (Promega, Madison, USA) for 12 h at 37°C. The digestion was terminated with 1% TFA (trifluoroacetic acid), and the resulting peptides were cleaned with a Strata X C18 SPE column (Phenomenex, Torrance, USA) and vacuum-dried in a scanvac maxi-beta (Labogene, Lynge, Denmark). Then, the resulting peptides (2 mg per sample) were reconstituted in 120 μl 0.5 M TEAB, and treated with a TMTsixplex label reagent kit (Pierce, Idaho, USA). Both fractionations were performed with an XBridge Shield C18 RP column (Waters, Milford, USA) in a LC20AD HPLC system (Shimadzu, Kyoto, Japan). For phosphorylation enrichment, peptides were dissolved in 80% ACN/6% TFA and then incubated with IMAC-Ti4+ beads (Sachtopore, Sachtleben Chemie, Germany) at room temperature. The beads were washed once with 50% ACN/0.5%TFA/200 mM NaCl and once with 50% ACN/0.1% TFA. The bound peptides were then eluted with 10% NH4OH and 80% ACN/2% FA (formic acid). All of the eluted fractions were combined, vacuum-dried and cleaned with C18 ZipTips (Millipore, Billerica, USA) according to the manufacturer’s instructions, followed by LC-MS/MS analysis according to a previous study [98]. The database search and bioinformatics analyses were performed as described previously [99].
For Phos-tag assay, the FgHapX-mCherry fusion construct was transferred into the ΔFgHapX and ΔFgYak1 mutants and the FgHapXS245A/S338A-mCherry fusion construct was transferred into the ΔFgHapX mutant. Protein samples were prepared and extracted as described above. Each resulting protein sample was loaded on 8% SDS-polyacrylamide gels prepared with 25 μM Phos binding reagent acrylamide (APE×BIO, F4002) and 100 μM ZnCl2. Gels were electrophoresed at 20 mA/gel for 3–5 h. Prior to protein transfer, gels were first equilibrated three times in transfer buffer containing 5 mM EDTA for 5 min and further equilibrated in transfer buffer without EDTA for 5 min for two times. Protein transfer from the Zn2+-phos-tag acrylamide gel to the PVDF membrane was performed for 4–5 h at 100 V on ice, and finally the membrane was analyzed by western blotting with the monoclonal anti-mCherry ab125096 (Abcam, Cambridge, UK) antibody.
FgYak1-Flag and FgHapX-mCherry were co-transformed into the wild type. The nuclei of corresponding strains were extracted as described previously [100] and then fixed on the polylysine slides. Slides were incubated with the polyclonal anti-Flag antibody A9044 (Sigma, St. Louis, MO) and monoclonal anti-mCherry ab125096 (Abcam, Cambridge, UK) antibody at 1:500 dilution, respectively, followed by the secondary antibodies Andy Fluor 594 goat anti-mouse L119A (red fluorescence) (GeneCopoeia, Maryland, US) and Andy Fluor 488 goat anti-rabbit L110A (green fluorescence) (GeneCopoeia, Maryland, US) at 1:300 dilution. Nuclei were stained with DAPI for 15 minutes before fluorescence observation.
|
10.1371/journal.pmed.1002488 | Increased risk of ischemic heart disease, hypertension, and type 2 diabetes in women with previous gestational diabetes mellitus, a target group in general practice for preventive interventions: A population-based cohort study | Gestational diabetes mellitus (GDM) is associated with developing type 2 diabetes, but very few studies have examined its effect on developing cardiovascular disease.
We conducted a retrospective cohort study utilizing a large primary care database in the United Kingdom. From 1 February 1990 to 15 May 2016, 9,118 women diagnosed with GDM were identified and randomly matched with 37,281 control women by age and timing of pregnancy (up to 3 months). Adjusted incidence rate ratios (IRRs) with 95% confidence intervals (CIs) were calculated for cardiovascular risk factors and cardiovascular disease. Women with GDM were more likely to develop type 2 diabetes (IRR = 21.96; 95% CI 18.31–26.34) and hypertension (IRR = 1.85; 95% CI 1.59–2.16) after adjusting for age, Townsend (deprivation) quintile, body mass index, and smoking. For ischemic heart disease (IHD), the IRR was 2.78 (95% CI 1.37–5.66), and for cerebrovascular disease 0.95 (95% CI 0.51–1.77; p-value = 0.87), after adjusting for the above covariates and lipid-lowering medication and hypertension at baseline. Follow-up screening for type 2 diabetes and cardiovascular risk factors was poor. Limitations include potential selective documentation of severe GDM for women in primary care, higher surveillance for outcomes in women diagnosed with GDM than control women, and a short median follow-up postpartum period, with a small number of outcomes for IHD and cerebrovascular disease.
Women diagnosed with GDM were at very high risk of developing type 2 diabetes and had a significantly increased incidence of hypertension and IHD. Identifying this group of women in general practice and targeting cardiovascular risk factors could improve long-term outcomes.
| The prevalence of gestational diabetes mellitus (GDM) is increasing rapidly in most developed countries.
Although it is well documented that women diagnosed with GDM have a greatly increased lifetime risk of developing type 2 diabetes, there is a paucity of reports on its association with cardiovascular disease.
Only 3 previous population-based cohort studies have reported on GDM and the long-term risk of cardiovascular events (including 1 utilizing a primary care database), and no studies have been reported from the United Kingdom (UK).
This population-based retrospective cohort study utilized a large UK primary care database that included more than 9,000 women diagnosed with GDM between 1 February 1990 and 15 May 2016.
The association between GDM and the development of type 2 diabetes, hypertension, and cardiovascular disease postpartum was examined.
Women diagnosed with GDM were over 20 times more likely to develop type 2 diabetes, 2.8 times more likely to develop ischemic heart disease, and twice as likely to develop hypertension, although no association was found for cerebrovascular disease.
Less than 60% of women diagnosed with GDM were screened in primary care for type 2 diabetes in the first year postpartum, and the proportion screened rapidly declined after this period.
Screening rates were also low for cardiovascular risk factors such as smoking, high body mass index, hypertension, and dyslipidemia, and were similar to the screening rates in control women.
GDM in women increases the risk for hypertension and ischemic heart disease in addition to type 2 diabetes.
Although the National Institute for Health and Care Excellence guidelines recommend annual screening for type 2 diabetes in women diagnosed with GDM, this study found that follow-up screening was poor for type 2 diabetes and other cardiovascular risk factors such as hypertension.
Women diagnosed with GDM are an identifiable group of at-risk women and ideal for targeting preventative metabolic and cardiovascular interventions.
Clinical guidelines should include postpartum screening and management for all cardiovascular risk factors in women diagnosed with GDM and not restrict it to diabetes.
| Gestational diabetes mellitus (GDM) is increasing, largely due to the obesity epidemic [1] and increasing maternal age [2]. Although inconsistencies exist across countries for screening for GDM [1] and diagnostic cutoff points for the oral glucose tolerance test [3], reported prevalences are 2%–6% for Europe [1], 7% for North America [4], and 1%–9% and 4%–24% for white British and South Asian (SA) women, respectively, in England and Southern Ireland [5], reflecting the higher 10%–20% prevalence in high-risk populations [2]. It is well accepted that the early identification and treatment of women with GDM reduces pregnancy and perinatal complications [1,6,7] and improves infant birth weights [8]. Women with GDM are also more likely to have markers for insulin resistance and beta cell dysfunction [9–13], particularly if overweight [14], and GDM is a well-established predictor for progression to type 2 diabetes and carries up to a 70% lifetime risk [15].
Although the association between GDM and type 2 diabetes is well established [15,16], onset of the latter following delivery is less well documented and understood in terms of underlying genetic and lifestyle factors [16]. Only 3 previous large population-based studies quantifying the increased risk of cardiovascular disease following delivery in women diagnosed with GDM were identified [17–19]. One Canadian retrospective study identified 8,191 women with GDM and age-matched them with 81,262 control women without GDM utilizing primary care records in Ontario [17]. One French study utilizing hospital records identified over 1.5 million women who delivered infants during 2007 and 2008 and included 62,958 women with GDM who were compared with all women without GDM who delivered a healthy infant during that period [18]. More recently, the North American Nurses’ Health Study II group reported on 5,992 nurses who self-reported a history of GDM from a total cohort of 89,479 [19].
All 3 studies reported an increase risk of cardiovascular events for women with GDM compared with women without GDM during the 12, 7, and 26 years of postpartum follow-up for the 3 studies, respectively. The increased risk for major cardiovascular events in people with type 2 diabetes in addition to other traditional risk factors [15,20] and at an early age [21] is well documented. Despite this, and the recommendation for annual screening for type 2 diabetes in women diagnosed with GDM [22] and evidence that lifestyle changes can improve outcomes [23], there is a paucity of reports on screening, and low rates have been reported [24,25]. The current National Institute for Health and Care Excellence (NICE) guidelines recommend screening for type 2 diabetes (between 6 and 13 weeks postpartum and an annual glycated hemoglobin [HbA1c] test) and lifestyle changes (weight control, diet, and exercise) for women diagnosed with GDM [26]. There is no recommendation to screen, identify, and actively manage cardiovascular risk factors (including hypertension, dyslipidemia, and smoking) in women diagnosed with GDM in the postpartum period in the current 2015 NICE guidelines [26].
The aim of this current study is to examine the risk of cardiovascular disease in women previously diagnosed with GDM in a population that is representative of all women diagnosed with GDM in the United Kingdom (UK). In addition, the proportion of women assessed for cardiovascular risk factors in the first 3 years postpartum in primary care will be documented. The results are expected to assist general practice in identifying and targeting cardiovascular risk factors in a group of relatively young women at high risk of long-term metabolic and cardiovascular disorders.
The study protocol was approved by the Scientific Review Committee (SRC Reference Number: 17THIN001) of the data provider, IQVIA.
A retrospective cohort study design was used to compare long-term cardiometabolic outcomes in women diagnosed with GDM and randomly matched pregnant control women not diagnosed with GDM, utilizing The Health Improvement Network (THIN) database—following the pre-analysis study plan (S1 Text). This database captures electronically recorded medical records in primary care and is designed to encourage research and improve healthcare delivery in the UK [27]. Over 675 general practices contribute to the THIN database, which captures about 6% (3.6 million) of the total registered population [27], is representative of the age structure of the UK population, and is made up predominantly of a white British, Welch, and Irish population (94% in 1991 although decreasing to 86% by 2011) [28]. Patient information is entered into the Vision patient record software, which uses Read code data (version 2) [29], rather than the World Health Organization International Classification of Diseases designed for hospital records. The Vision software also captures all British National Formulary drug prescription records [30]. General practices that had electronic medical record software for at least 1 year and had an acceptable mortality recording for at least 12 months were included in the analyses to ensure data quality and that all important covariates were recorded.
All records for women who became pregnant between 1 February 1990 and 15 May 2016 and were aged less than 50 years were accessed for possible inclusion in the study. Women diagnosed with GDM prior to delivery were identified and randomly matched with up to 4 pregnant control women without GDM by age and timing of entry of a code for pregnancy (up to 3 months).
The primary outcomes were the clinical diagnosis of coronary artery disease (ischemic heart disease [IHD]) and cerebrovascular disease (stroke or transient ischemic attack [TIA]). Secondary outcomes were cases of incident hypertension and type 2 diabetes. Outcomes were identified through clinical codes (S1 Data). Recording of diabetes, hypertension, and cardiovascular disease is considered accurate in UK primary care because there is a mandatory requirement for maintaining a register for these conditions, and incentive payments are made for identification and management of these cardiometabolic outcomes [31].
Women with a diagnosis of the outcome of interest prior to baseline (i.e., index date: at diagnosis of GDM for cases and confirmation of pregnancy for control women) were not included in the analysis for that outcome. All potential and available risk factors for all outcomes of interest (Townsend quintile, smoking, and body mass index [BMI]) and confounding variables (hypertension and lipid-lowering medication at baseline for IHD and stroke or TIA) were extracted for each of the women included in the study to adjust for confounding.
Patients were not involved in the design of this research project, in conducting the study, or in preparing results or reports. In acknowledgment of the patients and general practices that contributed information to the THIN dataset, the published paper will be circulated via IQVIA to all current practices that contribute to the dataset.
Characteristics of women with GDM and matched control women in the cohort were reported using appropriate descriptive statistics (mean and median for continuous variables and proportions for categorical variables). Incidence rate ratios (IRRs) and 95% confidence intervals (CIs) were calculated using the Poisson regression model, offsetting the exposure for person-years of follow-up. Adjusted IRRs were constructed by including age, BMI, Townsend quintile (a measure of deprivation), and smoking status in the Poisson models for hypertension and diabetes outcomes. In addition to these covariates, baseline hypertension and lipid-lowering medication prescription were included in the models for cardiovascular disease outcomes. BMI (in kg/m2) was treated as a categorical variable and grouped into <25, 25 to 30, and >30 kg/m2, based on the World Health Organization BMI categories [32].
Missing data for BMI, Townsend quintile, and smoking were included in the regression model as a missing categorical variable. We did not include ethnicity in our primary analysis because of poor recording in the primary care setting (<50%). However, in a sensitivity analysis, we included the available recording of ethnicity along with a missing category in the model to assess its impact on findings. Statistical significance was set at 0.05. Cumulative incidence curves were generated utilizing the cumulative incidence function of the survival curves. In addition, we report on the proportions of women with GDM and control women who were screened in the subsequent 3 years postpartum for smoking, BMI, diabetes, hypertension, and dyslipidemia. All analyses were conducted using STATA 14.0 [33].
A total of 9,118 women with GDM (based on an electronic code entry for GDM) were identified in the dataset. Table 1 outlines the demographic characteristics of the women diagnosed with GDM compared with pregnant control women (matched by age and timing of pregnancy) at baseline. Mean age at the time of delivery was 33 years and ranged from 14 to 47 years. A significantly greater proportion of women with GDM compared with controls were from economically deprived areas (Townsend quintile 4 or 5), were overweight or obese (BMI ≥ 25 kg/m2, 63% compared with 35%), and had been diagnosed with hypertension (including 2.5% prior to pregnancy), but women with GDM were less likely to be current smokers (16% versus 19%). The follow-up period varied from less than 1 to 25 years (median 2.9 years). There was a high proportion of SA women (17.1%) among those with a recording for ethnicity in the GDM cohort. Fig 1 outlines the sampling frame for the total number of pregnant women, those diagnosed with GDM, and control women matched by age and time of pregnancy.
Table 2 shows that women diagnosed with GDM were over 20 times more likely to develop type 2 diabetes (IRR = 21.96; 95% CI 18.31–26.34; p-value < 0.001) and had almost a 2-fold higher risk of developing hypertension (IRR = 1.85; 95% CI 1.59–2.16; p-value < 0.001) after adjusting for age, Townsend quintile, BMI, and smoking compared with control women. Further, after controlling for baseline lipid-lowering medication and hypertension in addition to the above covariates, women with GDM were more than 2.5 times more likely to develop IHD (IRR = 2.78; 95% CI 1.37–5.66; p-value = 0.005), but no increase in risk was found for cerebrovascular disease (IRR = 0.95; 95% CI 0.51–1.77; p-value = 0.87). Of the 14 women with GDM who developed IHD, only 5 also developed type 2 diabetes in the postpartum period, suggesting that the risk of cardiovascular disease is not always mediated through type 2 diabetes.
Fig 2 shows that the cumulative incidence of type 2 diabetes, hypertension, and IHD was higher for women with GDM compared with control women and that this difference persisted throughout the 25-year study period. The increased risk was specific for type 2 diabetes, hypertension, and IHD but not for stroke or TIA.
The sensitivity analysis that included ethnicity in the model did not significantly alter the effect sizes for any of the outcomes: IRR for type 2 diabetes = 21.08 (95% CI 17.57–25.30), IRR for hypertension = 1.82 (95% CI 1.56–2.13), IRR for IHD = 2.72 (95% CI 1.33–5.60), and IRR for stroke or TIA = 0.91 (95% CI 0.48–1.70). In an analysis restricted to women with GDM, SA women were twice as likely (IRR = 2.09; 95% CI 1.52–2.85) as white women to develop type 2 diabetes, and Afro-Caribbean (AC) women were 1.6 times more likely (IRR = 1.65; 95% CI 1.05–2.62). There was no increased risk for hypertension in AC women (IRR = 1.35; 95% CI 0.60–3.02) or SA women (IRR = 0.85; 95% CI 0.41–1.77) with GDM compared with white women with GDM. Ethnic subgroup analysis showed that white, AC, and SA women with GDM were at higher risk of developing type 2 diabetes than women without GDM, with IRR (95% CI) values of 35.2 (20.0–58.5), 22.15 (6.42–76.4) and 15.40 (6.54–36.25), respectively. Similar analyses for other outcomes were not possible due to the small number of outcomes among the relatively small number of women from the 2 minority ethnic groups.
Medical records for women with GDM showed that only 58% had some form of glycemic measurement in the first year following delivery (Table 3). Although 62% of women with GDM were tested after 1 January 2010, per the original 2008 NICE guideline publication recommending screening for type 2 diabetes [34], this proportion was not markedly different from the 53% tested prior to 2010. In the second and third year following delivery, the proportion with glycemic measurement decreased to less than 40%, and 24% of women did not have any glycemic measurement in the first 3 years postpartum. No difference was noted in the proportion of women with GDM who had blood pressure measured and recorded prior to and from 2010, with about 80% in the first year following delivery and declining to around 50% in the subsequent 2 years. BMI was recorded for almost half of women with GDM in the first year following delivery, decreasing to about a third in the subsequent 2 years, and did not change following the original guideline publication. Smoking status was recorded for 46% of women in the first year following delivery, and although this proportion declined in the second and third year, 73% had at least 1 record over this 3-year period. Lipid profiles were the most poorly recorded for women with GDM in the 3 years following delivery. Only 28% and 23% of women had any serum cholesterol or triglyceride level, respectively, recorded in this 3-year period.
In the control population, surveillance for risk factors in the postpartum period was comparatively lower than for women diagnosed with GDM across all assessments in year 1, and this difference persisted for type 2 diabetes and lipid measurements in years 2 and 3. In contrast, the proportion assessed for hypertension and smoking in years 2 and 3 did not differ between control women and women diagnosed with GDM. In a sensitivity analysis limited to patients who did not develop hypertension in the first year and were followed-up for more than 1 year, the increased risk for hypertension persisted (hazard ratio [HR] = 1.87; 95% CI 1.57–2.24), suggesting that surveillance bias did not affect this outcome.
This is, to our knowledge, the first large population-based study in the UK that reports on the increased risk of cardiovascular disease in women diagnosed with GDM, and quantifies the high incidence of type 2 diabetes and hypertension for these women in the postpartum period. Women diagnosed with GDM were over 20 times more likely to develop type 2 diabetes, had almost twice the risk of developing hypertension, and were 2.8 times more likely to develop IHD in the postpartum period compared with control women. The increased risk persisted throughout the 25-year follow-up period. Despite the high risk of developing type 2 diabetes and cardiovascular disease, postpartum screening was poor, with less than 60% of women undergoing any type of screening test for diabetes in the 12 months following delivery and with the proportion declining to below 40% in the second year. Further, only half of the women with GDM had their blood pressure recorded in the second year following delivery. About a third had smoking status recorded, and very few women had lipids recorded, in the third year postpartum. The only improvement noted following publication of the 2008 NICE guidelines [34] was in any form of measurement for glycemia, and this improvement was only moderate.
Our findings are broadly consistent with the French study utilizing hospital records, the Canadian study utilizing primary care records, and the Nurses’ Health Study II using self-reported diagnosis of GDM. The French study reported a higher adjusted odds ratio (OR) for hypertension (2.72; 95% CI 2.58–2.88) than our study’s IRR, but a lower adjusted OR for cardiovascular outcomes that included angina pectoris (1.68; 95% CI 1.29–2.20) and myocardial infarction (1.92; 95% CI 1.36–2.71). Similar to our finding, the French study reported no effect for stroke [18]. The Canadian study reported an attenuated effect (HR = 2.09; 95% CI 1.19–3.67) for the development of coronary artery disease [17]. A significant but lower effect size was reported for the Nurses’ Health Study II for myocardial infarction (HR = 1.56; 95% CI 1.09–2.23) compared with our study, and no effect for stroke (HR = 1.22; 95% CI 0.80–1.86). In addition, a large Swedish case–control study also utilizing hospital records reported an increased risk of cardiovascular events (OR = 1.51; 95% CI 1.07–2.14) and hypertension (OR = 5.10; 95% CI 3.18–8.18) for women previously diagnosed with GDM compared with control women who had not had a cardiovascular event prior to the matched pregnancy and without a history of GDM [35]. Smaller studies have also shown associations between GDM and hypertension and cardiovascular disease in women with a family history of type 2 diabetes [36], and a higher risk of hypertension for Hispanic compared with white North American women diagnosed with GDM [37].
The increased incidence of type 2 diabetes for women with GDM found in our study is consistent with previous studies. Despite this, the incidence in this study was far higher than that reported in a review of observational studies (relative risk = 7.43, 95% CI 4.79–11.51) that also reported an increased risk after 5 years compared with the first 5 years postpartum [16]. However, the high incidence of type 2 diabetes in the first few years postpartum found in this current study was similar to findings in an older review that reported a higher cumulative incidence in the first 5 years postpartum than in subsequent years, after adjusting for cohort retention [15].
Historically, follow-up screening for type 2 diabetes in women diagnosed with GDM is poor in the postpartum period [38]. This study shows that fewer than 60% of women with diabetes in pregnancy were screened for type 2 diabetes, and far fewer had smoking status or lipids recorded, in the first year following delivery. Blood pressure recordings also decreased from 80% in the first year to 50% in the second year following delivery. The current 2015 NICE guidelines recommend annual screening for diabetes and lifestyle advice on “weight control, diet and exercise” for all women diagnosed with GDM [26]. However, there is no recommendation for screening and management of cardiovascular risk factors such as hypertension, dyslipidemia, or tobacco use following delivery in this group of high-risk women. In addition to enhancing early identification of type 2 diabetes, targeting this group also ensures that women know of their increased risk, which is important to them [39], and presents an opportunity to provide education and support for the lifestyle changes required to improve long-term outcomes [9,11], although there is currently a lack of evidence on exactly how to achieve this [40].
The study has several limitations. Our study captured women with GDM only if the condition was documented in the primary care medical records. Our estimates suggest we may have only captured around 49% of women with GDM in the THIN database (S1 Table). Selective documentation of women with more severe GDM may have resulted in an overestimation of the effect size, while any women with GDM misclassified in the control population may have resulted in an underestimation of the effect size. Though overall our findings are consistent with previous studies, the effect estimates for diabetes and IHD were higher than the estimates observed in other studies, while effect estimates were lower for hypertension and similar for stroke. Moreover, women with GDM may have underreported the use of tobacco during pregnancy as they were more likely to be overweight and to live in economically deprived areas, both of which are strongly associated with tobacco use in pregnancy [41]. A previous review of observational studies also found no association between smoking and GDM [42].
The timing and frequency of diabetes-related screening tests during and after pregnancy varied, potentially leading to a nondifferential error, with more women with GDM being diagnosed with diabetes and hypertension. Higher frequency of blood pressure measurements in the GDM population was noted only in year 1 of follow-up. When limiting our analysis to women without hypertension in the first year postpartum and with follow-up recordings beyond 1 year, the effect sizes remained the same for hypertension. Surveillance bias is less likely to affect outcomes that are symptomatic such as IHD and stroke. Further, there is limited information on baseline characteristics such as ethnicity and the number and order of pregnancies, limiting further in-depth analyses on factors that modify each of the outcomes. In particular, not sufficiently controlling for ethnicity might have resulted in an overestimation of the risk of type 2 diabetes in women with previous GDM, but this overestimation is likely to be small considering that the majority of women in the UK over the study period were white [28].
Additional limitations include the short median follow-up period, resulting in few women diagnosed with cardiovascular disease, and possible misclassification of outcome data related to the Read codes, which, although are ideally suited for general practice, are not always accurate. However, the outcomes are expected to be reported as part of the Quality and Outcomes Framework and have been shown to be reliable [43].
Despite these limitations, this study is, to our knowledge, the first UK and the largest population-based study of women with GDM utilizing primary care records to report on incidence of cardiovascular disease not requiring a hospital admission [17]. The findings add an important insight into the trajectory of the development of type 2 diabetes, hypertension, and cardiovascular disease in the early and later postpartum periods. Findings are consistent with previous reports on the risk of developing type 2 diabetes and cardiovascular disease. Furthermore, the findings report on a large population and identify an at-risk group of relatively young women ideally suited for targeting of risk factor management to improve long-term metabolic and cardiovascular outcomes. Targeting these high-risk women may also provide better value for money for prevention programs, as they are already known to general practice. While the value of preventing cardiovascular outcomes requires further studies, there is some evidence that targeting this subgroup of women may yield benefits in reducing conversion to type 2 diabetes [44].
Results showed that women diagnosed with GDM were significantly more likely to develop type 2 diabetes, hypertension, and IHD at a relatively young age compared with women without a previous diagnosis of GDM. The risk was greatest for type 2 diabetes in the first year following delivery and persisted for 25 years. Follow-up screening for type 2 diabetes was poor, with less than 60% of women with GDM undergoing screening in the first year following delivery, and the proportion decreased to less than 40% by the second year. Guideline recommendations for screening and management of hypertension, lipids, and smoking cessation are lacking and need to be reviewed.
|
10.1371/journal.ppat.1006155 | RNA-Seq analysis of chikungunya virus infection and identification of granzyme A as a major promoter of arthritic inflammation | Chikungunya virus (CHIKV) is an arthritogenic alphavirus causing epidemics of acute and chronic arthritic disease. Herein we describe a comprehensive RNA-Seq analysis of feet and lymph nodes at peak viraemia (day 2 post infection), acute arthritis (day 7) and chronic disease (day 30) in the CHIKV adult wild-type mouse model. Genes previously shown to be up-regulated in CHIKV patients were also up-regulated in the mouse model. CHIKV sequence information was also obtained with up to ≈8% of the reads mapping to the viral genome; however, no adaptive viral genome changes were apparent. Although day 2, 7 and 30 represent distinct stages of infection and disease, there was a pronounced overlap in up-regulated host genes and pathways. Type I interferon response genes (IRGs) represented up to ≈50% of up-regulated genes, even after loss of type I interferon induction on days 7 and 30. Bioinformatic analyses suggested a number of interferon response factors were primarily responsible for maintaining type I IRG induction. A group of genes prominent in the RNA-Seq analysis and hitherto unexplored in viral arthropathies were granzymes A, B and K. Granzyme A-/- and to a lesser extent granzyme K-/-, but not granzyme B-/-, mice showed a pronounced reduction in foot swelling and arthritis, with analysis of granzyme A-/- mice showing no reductions in viral loads but reduced NK and T cell infiltrates post CHIKV infection. Treatment with Serpinb6b, a granzyme A inhibitor, also reduced arthritic inflammation in wild-type mice. In non-human primates circulating granzyme A levels were elevated after CHIKV infection, with the increase correlating with viral load. Elevated granzyme A levels were also seen in a small cohort of human CHIKV patients. Taken together these results suggest granzyme A is an important driver of arthritic inflammation and a potential target for therapy.
Trial Registration: ClinicalTrials.gov NCT00281294
| The largest chikungunya virus (CHIKV) epidemic ever recorded began in 2004 in Africa and spread across Asia reaching Europe and recently the Americas, with millions of cases reported. We undertook a detailed analysis of the mRNA expression profile during acute and chronic arthritis in an adult wild-type mouse model of CHIKV infection and disease. Gene induction profiles showed a high concordance with published human data, providing some validation of the mouse model. The host response was overwhelmingly dominated by type I interferon response genes, even after type I interferon induction was lost. The analysis also provided information on CHIKV RNA, with no adaptive viral genome changes identified. An important goal of the analysis was to identify new players in arthritic inflammation. Granzyme A was prominent in the RNA-Seq data and granzyme A deficient mice showed reduced arthritis, with no effects on viral loads. Arthritic disease could also be ameliorated in wild-type mice with a granzyme A inhibitor. Elevated circulating granzyme A levels were seen in non-human primates infected with CHIKV and in human CHIKV patients. Granzyme A thus emerges to be a major driver of CHIKV-mediated arthritic inflammation and a potential target for anti-inflammatory interventions.
| Chikungunya virus (CHIKV) belongs to a group of mosquito-borne arthritogenic alphaviruses that include the primarily Australian Ross River and Barmah Forest viruses, the African o’nyong-nyong virus, the Sindbis group of viruses and the South American Mayaro virus [1]. The largest documented outbreak of CHIKV disease ever recorded began in 2004 in Africa and spread across the Indian Ocean to Asia, east to Papua New Guinea and several pacific islands, with small outbreaks also seen in Europe. In late 2013 the epidemic reached the Americas, spreading through the Caribbean, Central and South America, with autochthonous transmission also reported in the USA [2,3]. Millions of cases have been reported. Symptomatic infection of adults with CHIKV is nearly always associated with acute and often chronic polyarthralgia and/or polyarthritis, which can be debilitating and usually lasts weeks to months, occasionally longer [1,4]. At present, no particularly effective drug or licensed vaccine is available for human use for any of these alphaviruses; although paracetamol/acetaminophen and non-steroidal anti-inflammatory drugs can provide relief from rheumatic symptoms [1,5] and CHIKV vaccines are in development [6,7].
CHIKV infection usually results in a 5–7 day viraemia, which is primarily controlled by a rapid type I IFN response [8–11] and subsequently by anti-viral antibodies [12–15]. Infection also drives a pro-inflammatory response with the up-regulation of multiple inflammatory mediators [16–24]. CHIKV arthropathy is generally viewed as an immunopathology [13,25,26], with the pro-inflammatory arthritogenic response sharing similarities with rheumatoid arthritis [27]. The arthritogenic response is triggered by viral infection of joint tissues and is associated with a robust mononuclear cell infiltrate comprised primarily of monocytes, macrophages, NK cells and some T cells [28,29]. An important role for CD4 T cells in driving CHIKV arthritis has been established [27,30], although the role of IFNγ is less clear [27,30,31].
A major burden of CHIKV disease is chronic or persistent polyarthralgia/polyarthritis [4,32], with the evidence currently suggesting that such ongoing arthritic disease is due to persistence of virus and/or viral material in joint tissues [13,20,33]. Whether such viral material (i) represents replicating virus or replicating viral RNA [13] with mutations that promote persistence [34,35] or (ii) simply represents delayed clearance of non-replicating viral material [2], remains unclear. Whether chronic rheumatic disease is associated with the development of new inflammatory processes (distinct from those prominent during the acute phase) is also unclear.
We have developed an adult C57BL/6J (wild-type) mouse model of acute and chronic CHIKV infection and arthritis that recapitulates many aspects of human disease [13,28]. The model has been widely adopted for testing new interventions [25,36–43], although how well the mouse recapitulates the full spectrum of inflammatory responses seen in humans remains unclear.
A key goal of CHIKV arthritis research is to identify potential new targets for anti-inflammatory drug interventions to improve treatment options for CHIKV arthritis [25,26] and perhaps related diseases [44]. Such interventions should clearly neither compromise anti-viral immunity [25,45] nor trigger other immunopathologies [46]. Herein we describe an RNA-Seq study of lymph nodes and feet in the adult wild-type mouse model of CHIKV infection. The study was undertaken to explore in depth the anti-viral and pro-inflammatory responses in acute and chronic infection, and to identify new players in arthritic inflammation.
We undertook transcriptional profiling of whole hind feet and inguinal lymph nodes using the previously described adult wild-type mice model of acute and chronic CHIKV infection and arthritic disease [13,28]. Poly-adenylated RNA from whole hind feet (days 2, 7 and 30 post infection) and lymph nodes (days 2 and 7 post infection) from infected mice, and from feet and lymph nodes of mock infected mice were analyzed by RNA-Seq. Day 2 represents the day of peak viraemia, day 7 acute arthritis [28], with day 30 representing chronic arthritic disease [13]. Three biological replicates, each comprising pooled RNA from 4 mice, were sequenced using 3 lanes of the Illumina HiSeq 2000 platform. Quality control analyses and read alignment data are shown in S1 Fig. The Tuxedo pipeline was used to identify differentially expressed genes (DEGs) in the infected tissues at the different times post infection compared to mock infected controls. The DEG lists (where q<0.01 and fold change >2) and the up and down-regulated genes (with the additional filter of FPKM > 1 in at least one sample) are provided in S1 Table.
The genes and/or proteins reported to be up-regulated in previous studies on CHIKV patients were all identified in this RNA-Seq analysis of mouse tissues (Table 1). Most of these genes/proteins are associated with inflammation (Table 1), suggesting a good concordance in pro-inflammatory gene expression in this mouse model and in human patients following CHIKV infection.
Many of the up-regulated genes identified in this RNA-Seq analysis have also been shown to be up-regulated (at the gene and/or protein level) in previously published mouse and monkey studies (S2 Table).
Venn diagram presentation of the up-regulated genes in feet illustrated that many up-regulated genes were shared between days 2, 7 and 30, with these shared genes also showing the highest mean fold change (Fig 1A). These 247 shared genes (fold change >2, FPKM>1, q<0.01, S3 Table) were overwhelmingly type I IRGs (as defined by Interferome [53]) and contained many anti-viral effectors, some of which have previously been described in alphavirus studies, such as Mx1 [54], viperin [55], ISG15 [56] and Ifit1 [57] (Table 2). Sensing and signaling proteins were also prominent and included IRF7 [9], Usp18 [58], Stat1, IRF1, IRF5 and IRF8. Tmem731 (STING) [59] (Table 2) and Trex1 [60] (S1 Table) were up-regulated, although the mechanisms and implications remain to be established [61–63]. CXCL10 was the most up-regulated chemokine (Table 2), with only some chemokines, such as CCL2, well studied in alphavirus infections [33,46]. As might be expected, complement [64], immunoproteasome genes and T cell response associated genes [27,30] were present (Table 2). Although granzyme B up-regulation has been noted previously [65], its prominence (Table 2, Gzmb) was perhaps unexpected given the limited role played by cytotoxic T cells and NK cells in protection against alphavirus infections [13,30,65,66]. Also prominent were interferon-inducible guanylate binding proteins, immunity related GTPases [67], C-type lectins and membrane-spanning 4-domains subfamily A (Ms4a) genes (Table 2), which have not been extensively studied in alphavirus infections. Most of the genes in the latter two groups and Cd300a, recently reported as a virus attachment factor [68], are expressed by monocytes/macrophages [69], which dominate the CHIKV inflammatory infiltrate.
Ingenuity pathway analysis (IPA) of up-regulated genes illustrated a high degree of similarity in the upstream regulators (direct and indirect) identified at the 3 time points in feet (Fig 1B). IPA canonical pathway analysis also showed considerable overlap (with many pathways associated with T cells) (S2 Fig). Gene induction profiles and inflammatory pathways were therefore surprisingly similar despite the different stages of infection and disease; day 2 (peak viraemia), day 7 (acute arthritis, no viraemia) and day 30 (chronic disease, persistent viral RNA) [13,28]. IPA analysis of genes uniquely up-regulated on day 30 (i) identified pathways already identified for days 2 and 7 and (ii) showed that many of the genes were associated with tissue repair (using the IPA Diseases & Functions feature).
For the down-regulated genes in feet, a large number of genes were uniquely down-regulated on day 7 (Fig 1C). IPA canonical pathway analysis also showed minimal overlap in pathways between the 3 time points (Fig 1D, heat map; S3 Table). These analyses and the low mean fold change (-2.5) suggest that a major influence on this data set is the pronounced cellular infiltration seen on day 7 [28], which would effectively dilute (and down-regulate) the mRNA of resident cells. This contention is supported by the observation that the top 10 terms identified by DAVID (v6.7) functional gene annotation analysis were associated with keratinocytes (Fig 1D, bar chart; S3 Table), cells that are not a major target of infection in C57BL/6 mice [9]. The top IPA canonical pathways for day 7 were associated with muscle (Fig 1D, heat map; S3 Table), with both the dilution effect and viral infection [9,71] likely responsible.
In contrast to feet, DEGs up-regulated in lymph nodes showed only minimal overlap between days 2 and 7 (Fig 1E). However, up-regulated genes in lymph nodes on day 2 showed a considerable overlap with genes up-regulated in feet on day 2 (Fig 1F, Venn diagram). IPA upstream regulator analysis also showed a high degree of concordance between pathways in lymph nodes and feet on day 2 (Fig 1F, heat map). This likely reflects the systemic nature of the infection and argues that early innate responses are not overly tissue specific.
Up-regulated genes on day 7 in lymph nodes, as might be expected, were dominated by immunogobulin genes, which represent 60% of the top 150 genes (S1 Table). The top terms from a DAVID functional gene annotation analysis were associated with cell division, consistent with the expected proliferation of B and T cells. In contrast with day 2 (Fig 1F), there was minimal overlap between up-regulated genes in lymph nodes and feet on day 7 (Fig 1G). By the time adaptive immune responses are developing and arthritis is peaking, the infiltrates in lymph nodes and arthritic feet thus appear to share relatively few genes.
The down-regulated genes from lymph nodes on days 2 and 7 showed some overlap (Fig 1D). IPA upstream regulator analysis also showed some overlap in pathways, with top pathways (as expected) generally indicative of immune activation.
Reads that did not map to the mouse genome were mapped to the CHIKV genome (S1D Fig). For feet on day 2, >8% of all sequencing reads aligned to the CHIKV genome, with 84% of reads aligning to the mouse genome (Fig 2A, S1D Fig). The number of reads aligning to the CHIKV genome dropped to 0.003% by day 30 (Fig 2A, S1D Fig), a reduction consistent with previous qRT-PCR data [13]. Examples of read alignments to the CHIKV genome are shown in Fig 2B and S3A Fig. The higher sequence coverage for the structural genes (evident at all 3 time points), reflects the known higher levels of subgenomic 26S RNA (encoding C to E1) compared to genomic RNA present in alphavirus infected cells [34]. The 3′ bias in sequence coverage, clearly evident on day 30 (Fig 2B), may represent an artifact of the Illumina HiSeq sequencing platform [72].
Although the low fidelity RNA replication of CHIKV [73] might predict the rapid emergence of sequence variants, we were unable to identify any consistent or high frequency changes (S3B Fig). Although some changes were identified (i) for each nucleotide position the percentage of reads showing a different nucleotide to the reference sequence rarely exceeded 10%, (ii) nucleotide sites with >2% of reads showing changes from the reference sequence were associated with areas of low read coverage (S3B Fig) and (iii) some consistent deletions/insertions (present in up to 10% of reads) were associated with runs of identical polynucleotides. Changes above a background sequencing error rate of ≈2% thus appear largely to represent sequencing artifacts. The ratios of synonymous to non-synonymous mutations were also consistent with random changes (S3C Fig).
The importance of the type I interferon (IFN) response for protection against lethal CHIKV infection is well established [9–11]. The upstream regulator analysis also showed that many of the top upstream regulators were associated with the type I IFN response (Fig 1B). qRT-PCR analysis illustrated a good correlation between IFNβ or IFNα6 mRNA levels and the tissue CHIKV titers, with feet and lymph nodes (the tissues analyzed by RNA-Seq in this study) showing the highest levels of both CHIKV and IFNβ/IFNα6 mRNA levels (Fig 3A). Type I IFN induction was thus more virus titer-dependent than tissue-dependent.
Interferome (v2.01) analysis of the DEGs identified by RNA-Seq (S1 Table) illustrated that about half of up-regulated genes (in all samples except day 7 lymph nodes), and 10–20% of down-regulated genes, were type I IFN regulated genes (IRGs) (Fig 3B). In addition, 10–30% of up-regulated genes in all samples were identified as genes regulated by IFNγ (type II IRGs) (Fig 3B). This analysis provides the first quantitative assessment of the very considerable dominance of IFN responses, particularly the type I IFN response, during both acute and chronic CHIKV infection.
The RNA-Seq analysis provided the first detailed picture of all the type I IFN genes induced after CHIKV infection, with IFNβ and α4 dominating (Fig 2C and S4 Fig). The surprising observation (given Fig 3B) was the overall low abundance of type I IFN transcripts, which did not exceed an FPKM = 7 and was often close to FPKM = 1 (Fig 3C, S4 Fig), a frequently used cut-off for expression analyses [74,75]. Low abundance of type I IFN mRNAs may also explain why reporter mice expressing GFP from IFNα or IFNβ promoters [76] express undetectable levels of GFP after CHIKV infection [77]. These results suggest high bioactivity for type I IFN proteins and/or highly efficient translation of type I IFN mRNAs [9–11]. Despite persistence of viral RNA (Fig 2A), by day 7 and 30 type I IFN mRNA levels had dropped to background levels (Fig 3C, S4 Fig).
The continued dominance of type I IRGs on days 7 and 30 (Fig 2C) despite the loss of significant type I IFN mRNA induction (Fig 2B, S4 Fig), argues that type I IFN-independent induction of type I IRGs (although well described [63,78,79]) essentially takes over after the brief period of type I IFN production. To better understand this process, an examination of transcription factor usage was undertaken. The direct upstream regulator function of IPA identified IRF7, STAT1, IRF3, IRF1 and IRF5 [80] in the top 10 upstream regulators for each time point (ranked by activation Z scores), with these transcription factors also showing (i) high fold change (with the exception of IRF3) and (ii) high FPKM (mRNA expression) values (Fig 3D). Other transcription factors identified by this analysis were IRF8 [81], Stat3 [82], IRF1 [83,84], IRF2 [85], and Stat2/IRF9 (with unphosphorylated ISGF3 able to signal [86]) (Fig 3D, S5A Fig). RELA was also identified in the top 10 (ranked by activation Z scores), but was only marginally up-regulated (Fig 3D, S5A Fig). These results were supported by a transcription factor site analysis using a new program (CiiiDER, Gearing et al, in prep) (S5B Fig). Although identification of IRF7 and IRF3 would be expected [9,78,87]; the role(s) of the other transcription factors identified herein remain to be fully explored in alphaviral infections [79,81,83,84,86].
The accumulated data might suggest IFNγ plays an important role in CHIKV infections [28] (Figs 1B and 3B–3D, S4 and S5B Figs.), both to promote inflammation [27,31] and to mediate anti-viral activity [88–90]. However, CHIKV infection of IFNγ-/- mice led to only a slightly elevated/extended RNAemia [30] or viraemia (S6A Fig), and only a marginal decrease in arthritic disease (S6B Fig), which was largely due to a reduction in edema (S6C and S6D Fig).
The limited effects of IFNγ deficiency prompted an analysis of putative transcription factor sites in the promoters of the type II IRGs up-regulated in feet (white bars, Fig 3B) using the CiiiDER program. Contrary to expectations, putative IRF7, ISGF3, IRF8 and consensus IRF sites were significantly over-represented in these genes (Fig 3E; formulas for calculating x and y values and the analysis inputs and outputs are provided in S7 Fig). Putative Stat1:Stat1 sites, although present in ≈45% of the type II IRGs, were also present in ≈34% of background genes thereby reducing significance scores (Fig 3E, S7 Fig). The mild phenotype in IFNγ-/- mice might thus be explained by redundancy in the induction of type II IRGs. The reverse, a compensatory role for IFNγ in the absence of IFNα/β has also previously been suggested [91].
An important objective of the RNA-Seq analysis was to identify new players in arthritic inflammation that may present new targets for intervention. Interrogation of the data revealed that granzyme A, B and K often show highly significant induction, high fold change, and for granzyme A and B, high FKPM values (Fig 4, Table 2). These granzymes are classically associated with cytolytic activities and their expression and secretion by cytotoxic T cells and NK cells is well described [92–95]. However, granzymes (particularly A and K) have also been associated with promoting inflammation in a number of settings [92–94,96,97].
To assess the role of granzymes A, B and K in CHIKV infection and disease, mice deficient in these proteases (GzmA-/-, GzmB-/- and GzmK-/- mice) were infected with CHIKV. Strikingly, GzmA-/- mice showed a dramatic reduction in foot swelling (Fig 4B). An independent repeat experiment with similar results is shown in (S8A Fig). No significant effect on foot swelling was evident in GzmB-/- mice, but GzmK-/- mice showed a significant, but less dramatic, reduction in foot swelling (Fig 4B).
None of the granzyme deficient mice showed significant changes in the viraemia (Fig 4C), consistent with the general view that controlling the viraemia of cytopathic viruses (such as alphaviruses) is not overly reliant on T cell- or NK cell-dependent cytolytic activities [65,66,98]. Granzyme A deficiency has been associated with a failure to clear certain viral infections [99,100]; however, feet tissue titers were not significantly affected in GzmA-/- mice (Fig 4D, Feet). In addition, the level of persistent CHIKV RNA in feet on day 30 post infection was not increased in GzmA-/- mice when compared with C57BL/6 mice (S8B Fig). (In both humans and C57BL/6 mice, viral RNA persists for extended periods and is associated with chronic arthritic disease [13,20]). Cytotoxic T cells have been reported to be important for clearing alphavirus from muscle tissues in certain settings [101]; however, muscle tissue titers were also not significantly different in GzmA-/- mice (Fig 4D, Muscle). The reduced inflammation in GzmA-/- mice (Fig 4B and Fig 5) was thus not due to an effect of granzyme A deficiency on virus levels in inflamed tissues.
GzmA-/- mice did not show any significant differences from C57BL/6 mice in their CHIKV-specific IgG2c and IgG1 responses (S8C Fig), indicating that anti-CHIKV antibody responses and the Th1/Th2 balance [102] were not significantly affected by granzyme A deficiency.
Histological examination of feet from GzmA-/- mice showed that the densities of cellular infiltrates (a prominent feature of CHIKV arthritis [28]) were significantly reduced when compared with C57BL/6 mice (Fig 5A and 5B). This result is consistent with the reduction in foot swelling and supports the contention that granzyme A has a role in promoting arthritic inflammation.
Immunohistochemical analyses of whole foot sections from CHIKV-infected mice during peak arthritis illustrated that the densities of NK (Fig 5C and 5D) and T cells (Fig 5E and 5F), but not monocytes/macrophages (Fig 5G and 5H), was significantly reduced in GzmA-/- when compared with C57BL/6 mice.
The pro-inflammatory activity of granzyme A is believed to be due to its proteolytic activity [92,96,103,104], with extracellular or circulating granzyme A remaining proteolytically active [105,106]. Furthermore, a potent and specific endogenous inhibitor of mouse granzyme A has been identified, Serpinb6b [107]. To determine whether Serpinb6b might show therapeutic activity, C57BL/6 mice were injected i.v. with purified recombinant Serpinb6b [107] from day 2 to 6 post CHIKV infection. Treatment with this granzyme A inhibitor significantly reduced foot swelling (Fig 6A) without impacting the viraemia (Fig 6B). (Treatment was not associated with any noticeable side-effects during daily monitoring of mice). Proteolytic inactivation of Serpinb6b with trypsin reversed the anti-inflammatory activity back to that seen in untreated mice (Fig 6A). H&E staining also showed a reduction in the arthritic infiltrates in the feet of Serpinb6b treated mice (Fig 6C). These results support the view that granzyme A has an extracellular pro-inflammatory role in this setting, and that granzyme A represents a potential target for anti-inflammatory drugs.
Elevated levels of circulating granzyme A have been detected in humans with a number of viral infections [105,108,109] or suffering from rheumatoid arthritis [110]. We have previously reported a non-human primate (NHP) model of CHIKV infection [33]. Using commercial ELISA kits, granzyme A and K levels were determined in plasma samples from such CHIKV-infected NHPs. In 9 out of 11 NHPs, plasma granzyme A levels increased relative to levels on day -1 (prior to infection) and usually peaked on day 4–8 post infection (Fig 6D). (The initial drop in granzyme A levels in NHPs 3, 5, 6 and 7 on day 2 coincides with the transient lymphopenia often seen at this time [33]). Taken as a group, the peak granzyme A levels for the 9 animals were significantly elevated when compared with levels prior to infection (day -1) (Fig 6E). Using data from all 11 animals, significance was retained (S9A Fig). In addition, when mean levels of granzyme A for all NHPs were plotted over time, a significant elevation was again evident (S9B Fig). A similar treatment of granzyme K data showed no significant elevation in mean circulating granzyme K levels (S9C Fig). This may in part reflect a sensitivity issue, as increases in circulating granzyme K levels after viral infections can be substantially more modest than those seen for granzyme A [111].
For most plasma samples in which granzyme levels were tested, viral loads were also measured; viral loads for all nine animals over time are shown in S9D Fig. No correlation between granzyme A level and viral load was apparent when using data from individual samples (S9E Fig). However, when peak viral loads (which occurred on day 2 post infection, S9D Fig) were plotted against the increase in granzyme A levels (i.e. the increase from day -1 to the peak) for each of the 9 NHPs in Fig 6D, a clear and significant positive correlation was observed (Fig 6F). Plotting peak viral loads against peak granzyme A levels also showed a significant correlation (S9F Fig). Thus the higher the viral load (the higher the disease severity [33] and) the higher the subsequent increase in circulating granzyme A levels.
Serum levels of granzyme A were also measured in serum from control patients and a small cohort of deidentified symptomic CHIKV patients who had tested IgM positive for CHIKV. (IgM usually remains detectable by serology for 1–3 months [1]). The CHIKV patients showed significantly higher levels of serum granzyme A than controls (Fig 6F), suggesting that CHIK-infected humans, like NHPs, show elevated granzyme A levels after CHIKV infection.
Herein we describe the first detailed RNA-Seq analysis of CHIKV infection, covering the time of peak viraemia, and acute and chronic arthritis, in a widely adopted adult wild-type mouse model of CHIKV infection and disease [77]. The inflammatory mediators identified previously in CHIKV infected humans were also identified by this RNA-Seq analysis (Table 1), illustrating that the mouse model recapitulates known aspects of the human inflammatory response to CHIKV infection. This analysis also highlights the potential for using RNA-Seq data to provide a level of validation of mouse models in general.
The RNA-Seq analysis provided information on CHIKV genome expression and sequence. In feet up to 8% of all the reads from poly adenylated RNA mapped to the CHIKV genome (S1D Fig), attesting to the remarkably high replicative capacity of this virus [112]. We are unaware of any study suggesting such a high proportion of viral RNA to host mRNA in vivo, although it should be noted in this model CHIKV infection is via s.c. inoculation into the feet [28]. The persistence of CHIKV RNA in joint tissues seen herein is also consistent with previous reports of persistent CHIKV RNA in mice, monkeys and humans [13,20,33]. The notion that the viral genome might undergo adaptive changes to promote persistence [34,35,113] was not supported by the RNA-Seq analysis, with no consistent or abundant genomic changes identified. The lack of changes in persisting CHIKV RNA thus suggests the CHIKV RNA is either not replicating [2] or is replicating, but not adapting over time. Continuous viraemia in Rag1-/- mice for 100 days also resulted in surprisingly few changes in the CHIKV genome [13].
Global expression profiles for feet and lymph nodes on day 2 (peak viraemia), feet day 7 (acute arthritis, no viraemia) and feet day 30 (chronic arthritis, persistent viral RNA) showed a surprisingly high level of overlap in both up-regulated genes and pathways, despite the differences in levels of infection, disease manifestations, immunity and tissues types [13,28]. This might in part be explained by the dominance of the IFN and inflammatory responses, which are largely independent of time and tissue in this robust systemic infection [58]. The high degree of overlap between feet on day 7 and 30 in both genes and pathways also argues that chronic arthritic disease represents a tailing off or extension of acute disease, rather than the activation of some fundamentally new inflammatory immunopathology [44]. This notion is further supported by the observation that only two of the genes, that were shared between days 2, 7 and 30 in feet, showed significantly higher fold change on day 30 than day 7. These were Tnip3 and Clec4d (S1 Table), which are genes associated with inflammation resolution [114,115].
Interferome analysis of up-regulated genes provided a quantitative assessment of the dominance of the type I IFN response after CHIKV infection, with ≈50% of genes identified as type I IRGs at all time points and tissues tested except day 7 lymph nodes. This dominance of type I IRGs was retained (in feet) on day 7 and 30, despite the loss of type I IFN gene induction. The loss of type I IFN gene induction also occurred despite the persistence of viral RNA. Type I IFN-independent induction of type I IRGs (although well described [63,78,79]) thus entirely and seamlessly takes over after the brief period of type I IFN-dependent induction of IRGs. A number of transcription factors potentially responsible were identified; some were perhaps expected (e.g. IRF7, IRF3 [78,79,116]), whereas others (e.g. IRF1, IRF2, IRF5, IRF8) have not been extensively studied in alphavirus infections.
Therapeutic targeting of IFNγ would appear to have limited utility for treating CHIKV arthropathy, as CHIKV infection in IFNγ-/- mice produces a relatively mild phenotype (reduced edema), despite abundant type II IRG induction and a robust IFNγ signature. The discrepancy may, at least in part, be explained by the transcription factor analysis, which suggested that genes induced via Stat:Stat1 can also be induced via other transcription factors (with several interferon response factors implicated). One might also speculate that the large volume of data sets on type II IRG induction may result in some pro-IFNγ bias in bioinformatics programs. Perhaps of note (given the similarities in the expression profiles of CHIKV and rheumatoid arthritis [27]), a phase II study of the anti-IFNγ agent, fontolizumab, in rheumatoid arthritis patients failed to show efficacy (ClinicalTrials.gov Identifier: NCT00281294).
Herein we describe the first phenotype of the recently generated GzmK-/- mouse, which showed no evidence of anti-viral or cytolytic functions in lymphocytic choriomeningitis virus or ectromelia virus infections (manuscript in preparation). The reduction in foot swelling in CHIKV-infected GzmK-/- mice reported here is consistent with previous suggestions of a non-cytolytic, pro-inflammatory role for granzyme K [117–119]. Granzyme A and granzyme K are related tryptic proteases that have arisen by gene duplication [120], and have overlapping substrate specificities [121].
Herein we illustrate for the first time that granzyme A is a key proinflammatory mediator during CHIKV arthritis, and (to our knowledge) are the first to show (in any setting) that inactivating granzyme A may have therapeutic benefit. The observation is consistent with the view that granzyme A’s proinflammatory role is associated with its proteolytic activity [92,96,103,104]. Our evidence argues against granzyme A having a significant role in suppressing CHIKV replication or in promoting viral clearance. Instead, our data is consistent with a number of studies in various systems that granzyme A promotes inflammation in both mice and humans [94,96,97,102–104,108,110,122–124]. Although a role for granzyme A in driving IL-1β-mediated inflammation in macrophages was recently reported [97], treatment with anakinra (a licensed IL-1 receptor antagonist) [125] provided only marginal amelioration of foot swelling in the CHIKV mouse model (manuscript in preparation).
The reduced arthritic disease in CHIKV-infected GzmA-/- mice was associated with the reduction of NK cell and T cell infiltrates, with NK cells and CD4 T cells previously shown to promote arthritis in this model [27,30,65]. NK cells and CD4 T cells are also well described in human alphaviral arthritides [126,127], and both NK cells and cytotoxic CD4 T cells express and secrete granzyme A (as well as granzyme K) [128,129]. Although cytotoxic CD4 T cells are prominent in dengue infections [130], they have yet to be formally demonstrated in alphavirus infections. However, CRTAM was recently identified as critical for differentiation of cytotoxic CD4 T cells at sites of inflammation [93], and CRTAM was up-regulated in feet (but not lymph nodes) on days 2 and 7 with a fold change of 3.3 and 3.8, respectively (S1 Table). How loss of granzyme A expression by NK cells and T cells might lead to reduced numbers of these cells in arthritic lesions remains unclear. A role for granzyme A in type IV collagen degradation and lymphocyte migration has been reported [131], although migration through matrigel was not impaired in GzmA-/- lymphocytes [132]. Perhaps of note, both granzyme A and K activate the protease activated receptor 1 (PAR1) [118,133], with PAR1 previously shown to be involved in inflammation [134,135], chemotaxis [136,137], and NK and T cell recruitment [138,139]. Granzyme A/K and thrombin (also known to activate PAR1) are trypsin-like proteases that cleave behind positively charged amino acids, with canonical proinflammatory signaling by PAR1 induced by cleavage at arginine 41 (Arg41) [140]. Granzyme B specifically cleaves behind aspartic acid residues; cleavage of PAR1 at Arg41 by granzyme B would thus be unlikely, perhaps explaining the lack of a CHIKV arthritis phenotype in GzmB-/- mice.
Granzyme A (rather than IFNγ) from differentiated NK cells [126,141] and CD4 T cells [27,65] thus appears to be an important driver of CHIKV arthritis, with granzyme A dispensable for control of CHIKV infection [25]. We also show that granzyme A is elevated in CHIKV-infected NHPs and in CHIKV patients, with circulating granzyme A levels in NHPs peaking day 4–8 post infection, which coincides with the peak of circulating IFNγ levels [33]. Circulating granzyme A levels have previously been shown to be elevated in patients suffering from infections with dengue [105], EBV, HIV-1 [108], primary CMV [109], malaria [142], and bacteria [106,143], and also in rheumatoid arthritis patients [110]. Taken together these observations argue that granzyme A represents a potential target for anti-inflammatory interventions not only in alphaviral arthritides, but perhaps also in other inflammatory diseases.
All mouse work was conducted in accordance with the “Australian code for the care and use of animals for scientific purposes” as defined by the National Health and Medical Research Council of Australia. Mouse work was approved by the QIMR Berghofer Medical Research Institute animal ethics committee (P1060 A705603M) and was conducted in biosafety level-3 facility at the QIMR Berghofer. Mice were euthanized using carbon dioxide.
NHP plasma samples were available from previous CHIKV studies [33,144]; a full ethics statement is provided in [144]. No additional NHPs were used for this study.
Human CHIKV serum samples were provided by the Centre for Infectious Diseases and Microbiology Laboratory Services (CIDMLS), Westmead Hospital (Sydney, Australia). Samples had been obtained from symptomatic patients who had returned to Australia from overseas and were collected for diagnostic purposes. All samples were IgM positive for CHIKV. Serum samples from healthy individuals were provided by the Australian Red Cross. Written and oral informed patient consent was obtained from all patients. No new human samples were collected as part of this study. Serum samples were deidentified before being provided for the research project and no patient data was provided or accessed. The study was approved by Griffith University Human Research Ethics Committee (BDD/01/12/HREC).
C57BL/6J mice (6–8 weeks) were purchased from Animal Resources Center (Canning Vale, WA, Australia). Interferon-γ deficient mice (IFNγ-/-) mice (JR3288 B6.129S7-Ifng/J) were obtained from the Jackson Laboratory. Granzyme A deficient (GzmA-/-) and granzyme B deficient (GzmB-/-) mice were generated as described [145] and were backcrossed onto C57BL/6J mice a total of 12 times and were provided by the Peter MacCallum Cancer Centre, Melbourne, Victoria, Australia [146]. GzmK-/- mice on a C57BL/6J background were provided by Prof Phillip Bird (manuscript submitted). Female mice were inoculated with 102 or 104 CCID50 of the Reunion Island isolate (LR2006-OPY1) in 40 μl of medium (RPMI1640 supplemented with 2% fetal calf serum), s.c. into both hind feet as described previously [13,28]. The virus (GenBank KT449801) was grown in C6/36 cells [13]. Serum viraemia was determined as described [9,13]. Foot swelling was measured using digital calipers and is presented as a group average of the percentage increase in foot height times width for each foot compared with the same foot on day 0 [13].
C57BL/6 mice were infected with 104 CCID50 CHIKV as described above and whole feet (cut above the ankle) and inguinal lymph nodes harvested on days 2, 7 and (for feet) 30 post infection. Mock infected mice were injected s.c. in the feet (i) with medium (and harvested 2 days later) or (ii) with heat inactivated (60°C, 30 mins) viral inocula (and harvested on day 30). Tissues were placed in RNAlater (Life Technologies) overnight at 4°C and then homogenized in TRIzol (Invitrogen) using 4 x 2.8 mm ceramic beads (MO BIO Inc., Carlsbad, USA) and a Precellys24 Tissue Homogeniser (Bertin Technologies, Montigny-le-Bretonneux, France) (3 x 30 s, 6000 rpm on ice). Homogenates were centrifuged (12,000 g x 10 min) and RNA extracted from the supernatants as per manufacturer’s instructions. RNA concentration and purity was determined by Nanodrop ND 1000 (NanoDrop Technologies Inc.). Eight RNA pools were generated in triplicate with each of the 24 samples containing equal amounts of RNA from four different mice; (i) feet day 2, (ii) feet day 7, (iii) mock feet day 2, (iv) feet day 30, (v) mock feet day 30, (vi) lymph node day 2, (vii) lymph node day 7 and (viii) mock lymph node day 2. The 24 samples were DNase treated using RNAse-Free DNAse Set (Qiagen), purified using an RNeasy MinElute Kit, and sent to the Australian Genome Research Facility (Melbourne, Australia).
Library preparation and sequencing were conducted by the Australian Genome Research Facility (Melbourne, Australia). cDNA libraries were prepared using a TruSeq RNA Sample Prep Kit (v2) (Illumina Inc. San Diego, USA), which includes isolation of poly-adenylated RNA using oligo-dt beads. cDNA libraries were mixed and sequenced from both ends (100 bp) using Illumina HiSeq 2000 Sequencer (Illumina Inc.). To obtain a high sequencing depth (total ≈55,000,000 paired end reads per sample) each library was sequenced three times using independent lanes. The CASAVA v1.8.2 pipeline was used to separate the bar-coded sequences and extract 100 base pair, paired end reads into FASTQ files.
Bowtie v2.0.2 and Tophat v2.0.6 [147,148] were used to align paired end read sequences to the UCSC mus musculus full genome build (mm10, Dec. 2011) to generate bam files (default parameters). The Cufflinks suite v2.1.1 (default parameters) [148,149] was then used to assemble transcripts (MapQ > 20) and calculate relative abundance and generate differentially expressed gene (DEG) lists. Differential gene expression for day 2 and 7 post CHIKV infection was determined relative to mock inoculated mice that had received medium 2 days previously, and differential gene expression for day 30 post CHIKV infection was determined relative to mice that had received heat inactivated viral inocula 30 days previously. For further analysis, DEGs were selected where (i) the q-value (false discovery rate adjusted p-value) was < 0.01, (ii) the fold change was > 2 relative to mock and (iii) FPKM was > 1 in the mock or the infected sample [74,75]. DEGs were analyzed by the Database for Annotation, Visualization and Integrated Discovery (DAVID) v6.7 [150], Ingenuity Pathway Analysis (IPA; Ingenuity Systems) and Interferome v2.01 [53].
Reads that did not map to the mouse genome where aligned to the CHIKV genome (LR2006-OPY1; GenBank KT449801) (excluding the polyA tail) using Bowtie v2.0.2. The frequency allele threshold was set to 5% with a mapQ>20. Reads alignments were visualized using the Integrative Genomics Viewer (IGV) version 2.3.34 [151]. Single nucleotide polymorphism analyses of the CHIKV sequences was undertaken using Geneious v. 7.1.5 [152] using minimum coverage of 20 reads per position and a minimum variant frequency of 0.5%.
Gene lists were analyzed by the recently developed software, CiiiDER (Gearing et al., in prep), which predicts key transcription factors regulating co-expressed genes. Using motifs released by TRANSFAC (2011), the software used a Java-based implementation of the Match algorithm [153] to identify putative tissue factor binding sites in sets of up-regulated gene, as well as in sets of background genes (for each time point), whose expression was not significantly changed by viral infection. A Fisher's exact test was used to identify sites significantly over-represented (enriched) in the up-regulated genes compared to the background genes and to provide p values as described [154].
Histology, immunohistochemistry and quantitation was performed as described previously [13,28,46]. Briefly, feet were fixed in paraformadehyde, decalcified and embedded in paraffin, and sections stained with hematoxylin and eosin (H&E). For immunohistochemistry, sections were stained with anti-NKp46 (rabbit polyclonal; Biorybt, Berkeley, CA) or anti-CD3 (A0452; Dako, North Sydney, Australia), with detection using MACH 2 (Biocare, Concord, CA) and Nova Red. F4/80 staining was undertaken as described [28]. Sections were scanned using Aperio AT Turbo (Aperio, Vista, CA) and analyzed using Aperio ImageScope software (v10) and the Positive Pixel Count v9 algorithm.
His-tagged recombinant Serpinb6b was produced at the Monash University Protein Production Unit using Pichia pastoris and purified using a nickel column followed by HiTrap Q column anion exchange chromatography (GE Healthcare Life Sciences) [107,155]. As a negative control the recombinant Serpinb6b was digested with tissue culture grade trypsin (Sigma) (1:1 molar ratio) for 30 mins at 37°C prior to injection. Serpinb6b (0.6 mg/ml) was diluted in RPMI 1640 and injected i.v. daily, 10 μg in 100 μl.
Plasma samples were available from previous experiments in which Macaca fascicularis NHPs had been infected with a range of doses of CHIKV as described [33,144]. Granzyme levels were determined using Monkey Granzyme A and K ELISA Kits (MyBioSource, San Diego, CA) according to manufacturer’s instructions. Viral loads were measured by quantative RT PCR as described [33].
Human serum samples were tested for granzyme A using the Human Granzyme A Flex Set (BD Cytometric Bead Array) and the LSRFortessa Cell Analyser (BD Biosciences, San Diego, CA, USA) according to manufacturer’s protocols.
Statistical was performed using IBM SPSS Statistics (version19). The t test was used if the difference in the variances was <4, skewness was >-2, and kurtosis was <2; where the data was nonparametric and difference in variances was <4, the Mann Whitney U test was used, if >4 the Kolmogorov-Smirnov test was used [13]. A 2 way ANOVA was used for some Aperio data and included a term for section. For NHP data paired t tests, Pearson and Spearman correlations were also used. The Kruskal-Wallis test was used for human serum granzyme A levels.
|
10.1371/journal.pcbi.1002715 | The Mechanistic Basis of Myxococcus xanthus Rippling Behavior and Its Physiological Role during Predation | Myxococcus xanthus cells self-organize into periodic bands of traveling waves, termed ripples, during multicellular fruiting body development and predation on other bacteria. To investigate the mechanistic basis of rippling behavior and its physiological role during predation by this Gram-negative soil bacterium, we have used an approach that combines mathematical modeling with experimental observations. Specifically, we developed an agent-based model (ABM) to simulate rippling behavior that employs a new signaling mechanism to trigger cellular reversals. The ABM has demonstrated that three ingredients are sufficient to generate rippling behavior: (i) side-to-side signaling between two cells that causes one of the cells to reverse, (ii) a minimal refractory time period after each reversal during which cells cannot reverse again, and (iii) physical interactions that cause the cells to locally align. To explain why rippling behavior appears as a consequence of the presence of prey, we postulate that prey-associated macromolecules indirectly induce ripples by stimulating side-to-side contact-mediated signaling. In parallel to the simulations, M. xanthus predatory rippling behavior was experimentally observed and analyzed using time-lapse microscopy. A formalized relationship between the wavelength, reversal time, and cell velocity has been predicted by the simulations and confirmed by the experimental data. Furthermore, the results suggest that the physiological role of rippling behavior during M. xanthus predation is to increase the rate of spreading over prey cells due to increased side-to-side contact-mediated signaling and to allow predatory cells to remain on the prey longer as a result of more periodic cell motility.
| Myxococcus xanthus cells collectively move on solid surfaces and reorganize their colonies in response to environmental cues. Under some conditions, cells exhibit an intriguing form of collective motility by self-organizing into bands of travelling alternating-density waves termed ripples. These waves are distinct from the waves originating from Turing instability in diffusion-reaction systems, as these counter-traveling waves do not annihilate but appear to pass through each other. Here we developed a new mathematical model of rippling behavior based on a recently observed contact signaling mechanism – cells that make side-to-side contacts can signal one another to reverse. We hypothesize that this signaling is enhanced by the presence of prey-associated macromolecules and compare modeling predictions with experimentally observed waves generated on E. coli prey cells. The model predicts a modified relationship between the wavelength and individual predatory cell motility parameters and provides a physiological role for rippling during predation. We show that ripples allow predatory cells to increase the rate of their spreading to quickly envelope the prey, and subsequently to decrease their random drift to remain in the prey region for longer. These and other predictions are confirmed by the experimental observations.
| Spatial self-organization of developing cells, which results the formation of complex dynamic structures, remains one of the most intriguing phenomena in modern biology [1]–[4]. Analogous developmental behaviors are observed as bacterial cells form biofilms, which are populations of surface-associated cells enclosed in a self-produced matrix [5], [6]. The dynamic self-organization in biofilms formed by the soil bacterium Myxococcus xanthus is dependent on the ability of the cells to move on solid surfaces [7], [8], while sensing, integrating and responding to a variety of intercellular and environmental cues [9]–[12].
M. xanthus is the preeminent model system for bacterial social development. At high density and under nutrient stress M. xanthus cells execute a complex multicellular developmental program by aggregating into multicellular mounds, termed fruiting bodies, and differentiating into dormant, environmentally resistant myxospores [11]. In addition, these bacteria exhibit complex behaviors when they cooperatively prey on other microorganisms by collectively spreading over the prey cells, producing antibiotics and lytic compounds that kill and decompose their prey [13], [14]. One of the most intriguing forms of collective dynamics exhibited by M. xanthus is their ability to self-organize into ripples – travelling bands of high-density wave crests [15]–[18]. Although the M. xanthus counter-traveling waves appear to pass through each another, they actually reflect off of one another and are termed “accordion waves” [16], [18]–[21]. These waves are distinct from the waves originating from Turing instability diffusion-reaction patterns, such as those in chemical systems or observed during development of the other well-studied model social microorganism, the amoeba Dictyostelium discoideum [22]–[24].
The initial studies of the mechanisms underlying M. xanthus rippling motility focused on this behavior during starvation-induced multicellular fruiting body development [16]–[20], [25]–[27]. The application of mathematical modeling to developmental rippling revealed that the wave properties are consistent with contact-induced reversal signaling [18]–[21], [28]. This signaling was hypothesized to originate from ‘head-to-head’ collisions of cells moving in opposite directions and to result in an exchange of C-signal that accelerates the reversal clock [16], [19], [20]. C-signal is an extracellular protein that controls aggregation and sporulation via contact-dependent pole-to-pole transmission [12]. Developmental aggregation and motility coordination are induced through the C-signal-dependent stimulation of the frz chemotaxis-like system, which includes an unconventional soluble cytoplasmic chemoreceptor homologue FrzCD [12], [29], [30].
An opportunity to reevaluate and replace the pole-to-pole collision-mediated model was prompted by a new report of FrzCD protein clusters that appear to transiently align and stimulate reversals in cells making side-to-side contact [8] and by the recent discovery that more robust rippling occurs during predation [13], [15]. In this paper we have investigated predatory rippling behavior with a combination of mathematical modeling and experimentation. We have constructed a mathematical model that faithfully reproduces the travelling wave behavior by adapting the recently proposed reversal-inducing side-to-side contact-mediated signaling model [8] and incorporating the properties of the patterns resulting from these interactions.
To model collective cell behavior we needed a modeling formalism that would allow us to connect the motility of individual cells, intercellular interactions, and the resulting population patterns. To this end, we employed an agent-based model (ABM) approach [19], [31]–[33]. Individual cells are represented as agents that move and interact according to the rules and equations that correspond to experimental observations. Unlike continuous, cell-density-based approaches, the ABM approach allows cell variability and modular implementation of interactions to be easily incorporated. The details and equations describing our ABM are summarized in the Materials and Methods Section. Here we qualitatively describe the main model ingredients that result in predatory rippling behavior.
Each agent is simulated as a self-propelled rod on a 2-D surface. The agents move continuously along their long axis and periodically reverse by switching the polarity of their two ends simultaneously. As in the previous models [19]–[21], we expected the ripples to emerge as a result of intercellular signaling, which leads to synchronized cellular reversals among the cell population. The side-to-side contact-induced signaling mechanism used here is based on the recent observations by Mauriello et al. [8], which demonstrated that when cells make transient side-to-side contact, their FrzCD clusters align causing one or both of the cells to reverse. The reversals stimulated by this intercellular signaling would be somewhat similar to the reversals induced by pole-to-pole collisions that were hypothesized to occur due to C-signal exchange during M. xanthus development [18]–[21]. Based on this and other experimental observations, our model incorporates four rules to guide the agents' interactions (see below). These rules are converted to mathematical equations that describe rippling motility (see the Materials and Methods Section).
The first three rules are sufficient for the model to produce rippling behavior (Figure 1, top row; Video S1). Starting from a uniform aligned population of agents (0 hrs), the model results in their self-organization into periodic traveling bands (ripples) within about 3 hrs. As in previous models [18]–[21], [25]–[27], the ripples emerge from the synchronized reversals. However, this model, which is based on a side-to-side contact-mediated signaling mechanism, appears to be more robust than the previous models that utilized pole-to-pole collision-mediated signaling (Figure S2). Rule (iv) is not necessary for rippling, but it allows the model to reflect the cell reversal behavior exhibited at low densities when cell contacts are rare, and it does not significantly change the high-density motility patterns studies here. The mean value of the native reversal period is chosen to be about 8 min (Figure S3 A) to achieve agreement with experimental observations by us here and others [11].
Within the framework of the proposed model, each rule (i)–(iii) is necessary to generate rippling behavior. Specifically, rule (i) is necessary because eleminating intercellular reversal-generated signaling abolishes rippling motility (data not shown) and eliminating the assumptions that signaling occurs only between counter-moving agents has the same effect (Figure S4 A and B). It is noteworthy that rippling motility is robust to the minimal overlap between agents that is required for them to engage in side-to-side signaling (Figure S5). Hereafter, an arbitrary value of 50% as a minimal overlap threshold is assumed in all simulations. Moreover, Figure S4 C vs D show that rippling motility occurs regardless of whether each signaling event is bidirectional (when cell #1 signals to cell #2, cell #2 also signals to cell #1) or unidirectional (cell #1 signaling to cell #2 and cell #2 signaling to cell #1 are independent events). In our simulations we use unidirectional signaling assumptions for the reasons explained below. The refractory period (rule ii) is also required for ripples, as reducing it to a very short duration leads to the dissapearance of the waves (Figure S4 E and F). In our simulations, the refractory period is a stochastic quantity with a mean value of 2.7 min and standard deviation of 0.7 min (Figure S3 B). Side-to-side signaling and rippling motility can only occur in a locally aligned cell population, and thus, physical interaction aligning cells, rule (iii), is necesary to maintain the cells' long axes approximately parallel.
Since in our simulations the rules (i)–(iii) induce rippling motility, we addressed the question of which rule is modulated to ensure that rippling motility is observed only when prey cells or the macromolecules associated with their lysis are present. The initiation and maintance of ripples seems to depend on the probability of reversal-inducing signaling events (Figure S6), which must exceed a threshhold value of ∼5–10%. If the probability is below 5%, then the ripples will not form and the agents will remain uniformly distributed on the 2-D surface. When the signaling probability exceeds the threshold value, the uniform distribution becomes unstable and the agents self-organize into ripples. Therefore, we hypothesize that the presence of prey-associated macromolecules indirectly stimulates rippling by increasing the probability that side-to-side contact generates successive signaling events (reversals). Although the biochemical mechanism of this induction is unknown, various macromolecular substrates, such as peptidoglycan, bovine serum albumin, and salmon testes chromosomal DNA, have been shown to induce rippling motility [13], [15]. Thus, we predict that the presence of these substrates should increase the probability of reversal-inducing signaling. Although our experimental arrangement does not allow direct testing of this prediction, we can quantitatively compare the emergent properities of the rippling patterns in the model and in the experiments.
It should be noted that the experiments demonstrating side-to-side signaling were preformed in the absence of prey cells or prey-associated macromolecules [8]. However, the results reported by Mauriello et al. [8] are consistent with a low probability of side-to-side signaling and the assumption that signaling is unidirectional. This is because in their observations only one of the cells engaged in side-to-side contact signaling reverses its gliding direction [8] (see also Figure S7). If the probability of signaling is low, it is unlikely that two signaling events will occur simultaneously. Furthermore, once one of the cell reverses, both cells will then be moving in the same direction and as a result, they will no longer be capable of signaling one another.
To test the modeling predictions experimentally, we observed cell motility on a solid nutrient agar surface in the presence of prey cells. The ripples were observed with fluorescence and differential interference contrast (DIC) time-lapse microscopy, allowing us to track cell density changes and the motility of a small percentage (0.5%) of GFP expressing cells in a wild-type population (see Materials and Methods section and Video S2). These images allowed us to calculate the global properties of the ripples: wavelength (distance from one wave crest to the next) and wave-crest width, and at the same time to measure the behavioral properties of individual cells: coordinates, velocity, reversal period, and the time/position of cellular reversals.
These data provided crucial input into the model and allowed us to test our modeling predictions. It is clear that the experimental ripple patterns appear very similar to those produced in the simulation (Figure 1). To compare the timing of wave initiation between the mathematical model and the experimental results, the time point when M. xanthus cells fully cover the prey in the field of view was chosen as the starting time (0 hrs in Figure 1; Video S3). Snapshot images at 0, 1, 3 and 5 hrs were selected to show the process of ripple formation in both systems. The experimental process of wave initiation appears to follow the same dynamics as the simulations. Initially, the cells homogeneously cover the field of view and the cells align as they cover the prey. During the first 3 hrs the reversals of individual cells become synchronous and result in the formation of ripples. By 5 hrs the ripples are pronounced and are easily discernible.
These results indicate that the ABM is capable of qualitatively reproducing the dynamics of rippling motility observed under our experimental conditions. Interestingly, waves generated with the ABM appear somewhat more pronounced than experimentally observed ripples, which have a smaller cell density gradient from crest to trough. This observation suggests that not all the cells in the biofilm participate in rippling behavior.
To compare the rippling patterns produced by the ABM to those of the experiments, we quantitatively characterized the ripples and related their patterns to the behavior of individual cells. Previous models of rippling motility [20], [21] proposed a simple equation, which relates wavelength (λ), individual agent speed (v), and agent reversal period (τ):(1)This equation indicates that cells in two colliding crests (relative speed ) reverse their directions every time the crests are superimposed. This prediction was confirmed by both the ABM and experimental results of developing cells [19]. However, our analysis of the measurements by Berleman et al. [13], showed that wavelengths of their predatory ripples were ∼50% larger than those predicted by Eq. (1). Using their experimental values of v = 3 µm/min and τ = 8 minutes, the calculated λ should be 48 µm, however their observed λ was ∼70 µm.
To determine if the wavelength relationship, Eq. (1), works for our new ABM of rippling motility, two sets of simulations were conducted. First, the agent speed was fixed at 6 µm/min, while the spontaneous reversal period was varied between 5 min and 30 min (corresponding to the variation between 3 and 12 min of an actual average reversal period, which is smaller due to early reversals triggered by side-to-side contact signaling; Figure 2A, solid line). Second, the spontaneous reversal period was fixed at a value corresponding to an average reversal period of approximately 6.6 min and the cell speed was varied between 2 µm/min and 12 µm/min (Figure 2B, solid line). These fixed values correspond to the experimental cell motility parameters. As shown in Figure 2A and 2B, the wavelength (λ) scales linearly with agent speed (v) and average reversal period (τ). However, when no-intercept linear regression was used, regression coefficients of 15.2 µm/min for Figure 2A and 16.1 min for Figure 2B were obtained. Both values are slightly larger than the predicted coefficients of 2v (12 µm/min) and 2τ (13.2 min), respectively. When we tracked the reversal points of individual agents, we observed that the agent reversals were initiated as soon as the leading edge of each crest came into contact (Test S1; Figure S6). This indicates that as the agents at the front of each crest reverse, they signal to the other cells in their crests, leading to a “chain-reaction” of signaling and reversals. Given the wave crest width Δ, the cells in each crest only move an average distance of λ−2Δ before reversing again, which results in the average reversal period τ = (λ−2Δ)/2v. Thus, we modified our wavelength equation to be:(2)To test the modified expression in our simulations, we automatically computed the average wave-crest width Δ from the simulation results (see Text S1) and used it to compute the wavelength with Eq. (2). The results demonstrate good agreement between the simulated and predicted wavelengths (Figure 2 A and B, solid vs. dashed line).
To test the Eq. (2) prediction experimentally for predatory rippling motility, we tracked 37 GFP-labeled individual cells within ripples for about 2 hr (or until the cells left the field of view). Continuous 1-D wavelet transform of the microscopy images (see Text S1) was used to compute the wavelength and wave-crest width by fitting a Gaussian function to the wave crest calculations. The distributions of average speed and reversal period are shown in Figures 2 C and D; and the ABM-predicted wavelengths are in agreement with the experimentally observed wavelength (denoted by the stars in Figures 2A and B). The prediction of Eq. (2) is also in good agreement with the data from Berleman et al. [13]. Using their experimentally derived values of v = 3 µm/min, τ = 8 min, Δ∼10–15 µm, the wavelength, λ, is calculated at ∼70–80 µm, which matches their published values. Rippling motility simulated with these parameters is shown in Video S4. To further test modeling predictions, we attempted to alter rippling wavelengths with changes in agar density and initial prey-cell concentration. We have selected two plates displaying reduced wavelength for detailed analysis and cell tracking. The results show that predictions of Eq. (2) also hold for these data: a reduced wavelength resulted from a reduction in the cell speed in both movies (∼3 µm/min) and a reduction of the reversal frequency (∼4.5 min) in one of the movies. Table S2 summarizes our experimental tests of Eq. (2).
According to our ABM assumptions and predictions, most of the rippling cells should travel with the wave crest and reverse, essentially as a group, when the leading edges of the two opposing wave crests collide. To test this prediction, we observed reversals of individual cells in the context of wave-crest movement by plotting cell trajectories on the space-time florescence intensity of ripples (Figure 2E). The space-time image illustrates the timing and location of the wave crests (see the dark gray ridges in Figure 2E). By examining trajectories of GFP-labeled cells (colored lines), we observe that the tracked cells travel with the high-density crests and reverse when and where two crests collide. Statistical analysis of the position and timing of cell reversals (dots) show that 75.0% (±2.6%) of all tracked reversals occur during wave crests collisions, matching ABM prediction (Figure 2F). Interestingly, some cells move through a counter-propagating wave crest without reversing and subsequently reverse with the next crest. This “wave-hopping” pattern explains the small peak at ∼12 min (twice the average reversal time) in Figure 2D and the more pronounced second peak in the distribution of the average distance travelled per reversal (Figure S8 E).
Myxococcus xanthus cells self-organize into periodic bands of traveling waves, termed ripples, during multicellular fruiting body development and predation on other bacteria. Here we have used an approach that combines mathematical modeling with experimental observations to investigate the mechanistic basis of rippling behavior and its physiological role during predation. The resulting new mathematical model, which is more robust than previous models, is based on the recent observation of Mauriello et al. [8], that when counter-moving cells come into side-to-side contact, clusters of chemotaxis-like FrzCD receptors within the cells transiently align and thereafter one of the cells reverses. Our model shows that this side-to-side contact-mediated signaling is sufficient to induce rippling self-organization in a locally aligned cell population, assuming that there is a minimal refractory period during which the cells can not reverse again regardless of their signaling state. The existence of the refractory period has also been assumed in our previous model [19], [20] and this assumption is plausible as reversals are anticipated to require a significant reorganization of the cell-motility machinery [7], [34]. The existence of a refractory period also naturally follows from the dynamic properties of a negative-feedback oscillator (Frzilator), which was previously hypothesized to regulate cell reversals [35]. Altogether our modeling results suggest that the self-organization of cells into ripples during predation can be explained by the increased efficiency or higher probability of side-to-side signaling induced by the presence of prey macromolecules. This prediction is not tested directly in our experiments, but the emergent properties of simulated waves quantitatively match those in our predation experimental approach.
Our model builds on the detailed characterization of M. xanthus predatory rippling behavior by Berleman et al. [15], which showed that rippling motility occurs during predation on the variety of microorganisms and is induced by the presence of macromolecular substances. However, our model differs from the concept promoted by Berleman et al. [15] that ripples originate solely as an interaction of individual cells with macromolecules without any self-organizing intercellular interactions. In contrast, we propose that ripples result from the self-organization of cells into traveling wave patterns, which result from the intercellular signaling that is stimulated or facilitated by the presence of macromolecules. Indeed, in our experimental approach the macromolecules are likely to be distributed uniformly and their concentration is expected to vary very little during the typical wave period (∼10 min). Moreover, even if macromolecules induce the periodicity of M. xanthus cell motility as suggested by Berleman et al. [15], this would not be sufficient to induce ripples because their formation requires temporal and spatial synchronization of cellular behavior that is unattainable without cell-to-cell signaling.
Based on the previous modeling of M. xanthus developmental rippling behavior, one is prompted to ask: does the same mechanism control predatory and developmental rippling motility? Certainly this new model is similar to the previous mathematical models of developmental rippling, as they each consider that self-organization occurs when counter-moving cells interact to induce reversals [19], [20]. As expected, the new model is in good general agreement with the experimental patterns that were previously observed for developmental rippling motility [16], [18]. However, our tests reveal that this new side-to-side contact-mediated signaling model is much more robust, in that it can withstand realistic levels of variability in cell speed and reversal times (Figure S5 left panels). Specifically, when the level of randomness in cell motility consistent with the single-cell tracking experiments (fluctuations of velocity and reversal period over 30% of the mean value) is used in the pole-to-pole collision-mediated signaling model, the cells do not form ripples (Figure S5, bottom right panel). It is noteworthy that pole-to-pole signaling can result in more robust waves, if the cells are able to accumulate signals from multiple collisions and if signaling during the refractory period leads to a reduced reversal rate as the Frzilator model predicts [19]. However, for the new side-to-side contact-mediated signaling model, realistic rippling can be observed assuming only that single successful signaling events result in cellular reversals.
Furthermore, the experiments of Berleman et al. [13], [15] provided evidence indicating that developmental rippling occurs as a side effect of cell lysis during aggregation, which suggests that rippling motility is likely to be a response to the released macromolecules. Thus, we propose that our new side-to-side contact-mediated signaling model of rippling describes both predatory and developmental rippling. The new model therefore explains ripples without requiring the pole-to-pole exchange of the starvation-induced C-signal. This may be biologically justified for a number of reasons. First, to date no C-signaling receptor has been identified. Second, localization of CsgA to the cell poles has not been demonstrated directly. Third, the robustness of pole-to-pole signaling-mediated mechanism is questionable as the probability of this type of collision is low. However, as C-signaling mutants fail to display rippling motility [16], it would be interesting to investigate in future studies how C-signaling affects the FrzCD cluster alignment and whether C-signaling plays a role in predatory rippling.
The main hypothesis of this new computational model is that rippling behavior is initiated by side-to-side contact-mediated signaling in the presence of prey cells. This hypothesis cannot be directly tested at this time, since we do not have a complete understanding of the specific biochemical mechanisms involved. However, we can rigorously test the model by comparing the model predictions with experimental data collected by us and others.
An important prediction of the model is that M. xanthus cells will reverse more frequently when prey is present. This agrees with our experimental observations (Table S3) and that of Berleman et al. [13]. Moreover, the resulting self-organization of cells into ripples provides various ways to quantitatively and qualitatively compare in silico-generated rippling motility with experimental observations.
A second prediction is that if the presence of prey stimulates this side-to-side contact-mediated signaling, then the rippling would only be observed in the regions where signaling is sufficiently probable, i.e. only in the regions covering prey. This is in good agreement with our observations (Video S6) and those of Berleman et al. [13], [15]. Indeed, our simulations show that the signaling probability can serve as a bifurcation parameter that induces a transition between the homogeneous cell distribution and the formation of ripples (Figure S6).
A third prediction is based on the timescale of rippling self-organization, which can be defined as the time it takes to generate ripples that consist of well-focused wave patterns, starting from an initially homogeneous cell population. Our model predicts that time to be of the order of 3 hrs, which is remarkably consistent with our experimental observations (Figure 1). The qualitative comparison of the time-lapse dynamics (Videos S1 vs. S2, S3 and S6) is also in good agreement. Interestingly, the time-scale of rippling origination in the experiments of Berleman et al. [13] is significantly longer (∼12 hrs). Although it is hard to pinpoint the source of this discrepancy, our model indicates that the cell density and the amount of noise in cell orientation can significantly affect the wave synchronization time.
A fourth prediction of the model is based on measuring the rippling wavelengths and correlating them to the parameters of individual cell motility. Our new model predicts a slightly modified relationship (Eq. (2) between wavelength, wave-crest width, individual cell speed, and reversal time as compared to the previously established [19], [20]. This new relationship is confirmed by our simulations and is in excellent agreement with the experimental measurements of wavelength (Figure 2 A and B, Table S2). The wavelength prediction is also compatible with previously reported measurements [13] and with the observations of Sliusarenko et al. [15], which show that cells moving in opposite directions tend to inter-penetrate one cell length before a reversal is triggered. Figure S7 shows the sequence of events that occur during two-crest collisions. This cartoon model indicates that once the cells at the front of each crest reverse, they signal to the cells following them, which results in a chain-reaction of signaling and reversal events. This cartoon also illustrates the importance of the refractory period, because once the cells at the front of the crest reverse, it is essential for them to keep signaling to other cells to reverse without reversing themselves.
A fifth prediction of the new model is based on tracking the cell reversals and locations of wave-crest collisions in time and space. Just as the model predicts (Figure 2F), the experimental results (Figure 2E) indicate that most reversals occur when and where two wave crests collide.
Our previous model of developmental rippling motility suggested [28] that periodic travelling waves can ensure a more regular distribution of fruiting-body aggregates at the colony edge, as seen in the submerged culture system of Welch et al. [18]. However, the physiological implications of this observation are unclear as the developmental aggregate distribution can be well organized even without rippling [36]. Furthermore, if rippling motility is predominantly a response to predation, what is its role in these situations? Berleman et al. [37] proposed two hypotheses. The first, termed the “grinder model” speculates that the movement of the waves of M. xanthus cells during rippling motility causes a physical disruption of the prey colony. The second, termed the “population control model” suggests that waves maximize the prey-predator contact area and push excess predator cells to the edges of the rippling area. Neither of these hypotheses is likely to be correct, based on the biophysics of this environment in which the very-low Reynolds number hydrodynamics will not allow temporary periodic perturbations to affect mixing or transport [38]. Nevertheless, our mathematical model suggests several alternatives for physiological benefits of rippling to predatory cells. These predictions are consistent with the experimental observations reported here and previously.
First, the model is in agreement with the observations of Berleman et al. [13] that during the expansion over prey, the presence of side-to-side contact-mediated signaling significantly facilitates the rate of M. xanthus cell spreading (Figure 3). As a result these cells cover their prey faster. This has obvious physiological benefits in the competitive soil environment. The observation is also consistent with our own experiments. Notably, this result does not require ripples per se, but only reversal-inducing signaling. However, our model indicates that side-to-side contact-mediated signaling is key for rippling self-organization and the other model ingredients can easily be justified by what is known about the biophysics of M. xanthus motility [7]. Furthermore, the increase in spreading also takes advantage of the cell-density gradient of M. xanthus cells that is generated by spreading at the leading edge. It is important to note that the rippling behavior does not require a density gradient of prey cells, as the alternative chemotaxis-based explanation would predict.
Second, the model predicts that cells that ripple in the absence of a cell-density gradient (i.e. when they are behind the leading edge of the swarm or once the prey is fully covered), would engage in less noisy and more periodic motion and as a result will have less of a random drift (Figure 4A). This effect would help the predatory cells to remain in the prey area for a longer time and to reduce random movement away from the prey. This prediction was confirmed by the cell-tracking assays (Figure 4B). Notably, this effect requires ripple formation, as the collective interaction of cells in the ripples leads to their synchronization. This effect is analogous to the well-known mathematical phenomena in which a collection of coupled noisy oscillators is less noisy than each oscillator on its own [39].
Third, it is likely that the formation of the ripples increases the cell alignment due to an increase in steric interactions in the denser crests. This prediction agrees with our observations and those of Berleman et al. [13]. However, it is worth noting that the causal relationship between rippling and alignment is not obvious, as ripples also require cell alignment. Therefore, it is likely that there is a positive self-reinforcing feedback loop between the formation of ripples and cell alignment: as cells align, ripples become more pronounced and their crests become more dense leading to further cell alignment. Although the physiological benefit of better alignment is not obvious, it may further enhance the rate of spreading, which contributes to the effects discussed above.
Uncovering the mechanistic basis of spatial and temporal multicellular self-organization is a daunting task and a full understanding has not been achieved for even the best-studied model systems. Here, agent-based modeling, time-lapse fluorescence microscopy, and image quantification have been used synergistically to provide new insights into the mechanisms of M. xanthus self-organization into ripples. Our modeling demonstrates that a simple set of ingredients based on experimental observations is sufficient to produce rippling patterns. The subsequent experiments have tested a number of predictions based on the model and have allowed us to refine the model to achieve quantitative agreement with the experimental data. This type of combined approach is essential to further our understanding of self-organization in more complex systems such as development of multicellular organisms.
ABM are widely used to computationally simulate emerging patterns formed by multiple agents. The ABM of M. xanthus rippling presented here is kept simple yet sufficiently flexible to accurately describe the experimentally observed behaviors of M. xanthus cells. The model is an extension of the earlier ABM [19] of M. xanthus self-organization that now incorporates a side-to-side contact-mediated signaling mechanism.
In this ABM, each agent represents a cell – a self-propelled rod on a 2-D surface with length of L, width of w, with a center position of (x(t),y(t)), and orientation 0≤θ(t)≤2π. Specifically, the agent length and width are constant throughout all simulations, whereas the center position and direction of movement are changed at each time step as the cells move and align. For each simulation, the time is updated by constant increments δt. The simulations are conducted on a fixed 2-D area in which all simulated moving agents are bounded. For most simulations periodic boundary conditions are imposed.
|
10.1371/journal.pcbi.1003077 | Towards Systematic Discovery of Signaling Networks in Budding Yeast Filamentous Growth Stress Response Using Interventional Phosphorylation Data | Reversible phosphorylation is one of the major mechanisms of signal transduction, and signaling networks are critical regulators of cell growth and development. However, few of these networks have been delineated completely. Towards this end, quantitative phosphoproteomics is emerging as a useful tool enabling large-scale determination of relative phosphorylation levels. However, phosphoproteomics differs from classical proteomics by a more extensive sampling limitation due to the limited number of detectable sites per protein. Here, we propose a comprehensive quantitative analysis pipeline customized for phosphoproteome data from interventional experiments for identifying key proteins in specific pathways, discovering the protein-protein interactions and inferring the signaling network. We also made an effort to partially compensate for the missing value problem, a chronic issue for proteomics studies. The dataset used for this study was generated using SILAC (Stable Isotope Labeling with Amino acids in Cell culture) technique with interventional experiments (kinase-dead mutations). The major components of the pipeline include phosphopeptide meta-analysis, correlation network analysis and causal relationship discovery. We have successfully applied our pipeline to interventional experiments identifying phosphorylation events underlying the transition to a filamentous growth form in Saccharomyces cerevisiae. We identified 5 high-confidence proteins from meta-analysis, and 19 hub proteins from correlation analysis (Pbi2p and Hsp42p were identified by both analyses). All these proteins are involved in stress responses. Nine of them have direct or indirect evidence of involvement in filamentous growth. In addition, we tested four of our predicted proteins, Nth1p, Pbi2p, Pdr12p and Rcn2p, by interventional phenotypic experiments and all of them present differential invasive growth, providing prospective validation of our approach. This comprehensive pipeline presents a systematic way for discovering signaling networks using interventional phosphoproteome data and can suggest candidate proteins for further investigation. We anticipate the methodology to be applicable as well to other interventional studies via different experimental platforms.
| Signal transduction is a ubiquitous and essential mechanism regulating cellular functions, including responses to environmental stress. Dysfunction of signaling pathways results in a variety of diseases, including cancer, diabetes, and cardiovascular disease. Phosphorylation regulates the activity of signaling and target proteins at different cellular locations and controls activation and inactivation of signal pathways. Here, we provide an analysis of phosphoproteome datasets from yeast, utilizing kinase mutants versus wild type strains. In order to provide an objective approach to identify candidate proteins involved in the transition to a filamentous growth form, we proposed and applied a comprehensive pipeline incorporating statistical and mathematical methods to investigate the phosphoproteome data from multiple perspectives. This included phosphorylation variation in response to a single mutant, phosphorylation variation patterns over multiple mutants, and the relationships represented by these patterns. We make an effort to discover the components and targets of the signaling network, infer the network structure, and to find the relationships of changes of protein phosphorylation to cellular functions, specifically in response to stress in the context of filamentous growth.
| Cells exchange and receive information from the environment through signaling pathways, which are crucial for cells to maintain normal functions and properly respond to stress and stimuli. Dysregulation of these processes is a major factor in the emergence of many diseases, including cancer, diabetes, and cardiovascular disease. Reversible phosphorylation is one of the major forms of signal transduction and can affect protein function and gene expression [1]–[7]. Investigations into phosphorylation provide insight into signaling pathways by providing the target sites of phosphorylation and the quantitative changes in phosphorylation level in response to genetic or environmental perturbations. Effective, sensitive identification of candidate proteins for further studies remains a challenge in the face of experimental limitations of current technologies which have a high cost component, provide incomplete coverage of the phosphoproteome, and have sampling limitations which affect replicate runs.
Large-scale phosphoproteomics studies on a number of organisms have been carried out using mass spectrometry (MS)-based approaches (reviewed in [8]–[10]). These include two recent global phosphoproteomic studies of the budding yeast (Saccharomyces cerevisiae) [5], [6]. In the study carried out by Bodenmiller at al. [5], protein kinases and phosphatases were systematically perturbed through gene deletions. The system-wide responses to the perturbations were measured by label-free MS-based quantification, and the results evaluated to determine their contributions to understanding the relationships between these signal transduction proteins and cell pathways. Another global interaction study focused on kinase and phosphatase interactions [6] by capturing protein-protein interactions by affinity capture-immunoblot and identifying the isolated protein complexes by mass spectrometry. These two global studies both adopted label-free, cost-effective quantitative approaches. However, label-free methods typically increase variance relative to isotope enrichment methods [11]. For the purpose of this study, we have used isotope labeled SILAC (Stable Isotope Labeling with Amino acids in Cell culture) method [12], [13] to increase sensitivity to change.
The general scope of this manuscript encompasses a comprehensive pipeline, incorporating statistical and mathematical methods for investigating and evaluating quantitative phosphoproteomic data, the elucidation of candidate proteins, and the identification of processes to be pursued in subsequent molecular biology and genetic studies. The phosphoproteome data utilized in this analysis was obtained from interventional experiments of a subset of yeast kinases involved in filamentous growth. Filamentous growth is a developmental transition observed in S. cerevisiae where yeast cells form elongated and connected multicellular filaments; these filaments resemble hyphae but lack the parallel-sided walls and structure of true hyphal tubes. This pseudohyphal growth transition is induced in response to several cell stresses, including nitrogen stress, growth in the presence of short-chain alcohols, and glucose stress [14]–[17]. The filamentous growth form presumably represents a foraging mechanism enabling non-motile yeast to better survive cell stress [14]. During pseudohyphal growth, yeast cells elongate due to a delay in the G2/M transition, exhibit an altered budding pattern, and remain connected after cytokinesis [18], [19]. The resulting pseudohyphal filaments extend superficially from a colony over an agar substrate and invasively downward into the solid substrate below the colony. In liquid culture under inducing conditions, a filamentous strain of yeast exhibits elongated cells and multicellular filaments encompassing typically 3–4 cells. It is important to note that most laboratory strains of S. cerevisiae are non-filamentous and that studies of filamentous growth are typically performed in the ∑1278b strain, which we employ here.
The molecular basis of filamentous growth in S. cerevisiae is broad in scope. Classic studies have identified key kinase-based signaling networks that regulate the filamentous growth transition. In particular, yeast filamentous growth is regulated by mitogen-activated protein kinase (MAPK) and protein kinase A (PKA) pathways [15], [20], [21] as well as being impacted by other signaling pathways. MAPK pathways are evolutionarily conserved across phyla and consist of three-kinase cascades serving central roles in signal transduction in eukaryotic cells [22]; the yeast filamentous growth MAPK cascade terminates in the MAPK Kss1p. In S. cerevisiae, PKA consists of the regulatory subunit Bcy1p and one of three catalytic subunits Tpk1p, Tpk2p, or Tpk3p; Tpk2p is known to be required for filamentous growth [23]–[25]. It should be noted that the Kss1p MAPK pathway is required for pseudohyphal growth induced by both nitrogen stress and butanol, while the genes GPR1, MEP2, and GPA2, acting upstream of PKA, are not required for butanol-induced filamentous growth [16]. In our experiments, we treated cells with 1% (vol/vol) butanol to induce filamentous growth [26]. A graphical illustration of currently recognized budding yeast filamentous growth pathways, integrating information from authoritative pathway databases and reviews, is shown in Figure 1. While these core signaling units are well defined, the downstream scope of their signaling networks is unclear.
We have generated phosphoproteomic datasets indicating kinase-dependent phosphorylation events underlying the filamentous growth transition. Specifically, we generated kinase-dead mutations (also called kinase-inactivating mutations) for a set of eight kinases that we have identified as components of the yeast filamentous growth response: Ksp1p, Kss1p, Sks1p, Ste20p, Snf1p, Tpk2p, Elm1p and Fus3p [20], [26]–[28]. Each of these kinases exhibits a filamentous growth deletion phenotype, with the deletion of KSP1, KSS1, SKS1, STE20, SNF1, and TPK2 yielding a loss of filamentous growth and the deletion of ELM1 and FUS3 yielding enhanced filamentation. The kinase-dead alleles of these proteins were constructed by site-directed mutagenesis. The system-wide phosphorylation responses of the mutant strains were determined using SILAC approach, and we used the Mascot search engine [29] followed by MaxQuant software [30] to identify and quantify peptides and proteins. We obtained phosphorylation level changes from the MaxQuant analysis for mutants versus wild type control for the comprehensive quantitative analyses.
The broad focus of the filamentous growth kinase networks in particular has made it difficult to tease out important kinase targets (direct or indirect). Bioinformatics methods provide a promising avenue with which local kinase signaling relationships can be identified. While traditional cluster analyses associated with functional enrichment analysis are useful tools, their performance might be affected by the missing value issue. We need to deal with it in order to obtain reliable clusters and enriched functions. Furthermore, a more integrative and extensive analysis is necessary to find new components of the pathways, uncover relationships between the pathway components, and to elaborate the signaling network structure. Thus we propose this comprehensive quantitative analysis pipeline customized for SILAC data, and partially compensate the missing value issue. The major building blocks include phosphopeptide meta-analysis, correlation network analysis, causal relationship discovery, and validation by literature mining. We have successfully applied the pipeline to analyze our current yeast data. Candidate proteins predicted to contribute to the filamentous growth response were selected by phosphopeptide meta-analysis and correlation network analysis. Causal relationship discovery was performed on candidate proteins identified from our analysis and validated proteins from the literature. The inferred causal relationships, along with the interactions inferred from phosphorylation changes in response to individual mutants, have suggested potential proteins that can be further intervened and studied in the future.
An overview of the analytical workflow is shown in Figure 2. Following peptide identification and quantification, the comprehensive post-identification analyses performed consisted of phosphopeptide meta-analysis, correlation network analysis, and literature mining, followed by causal relationship discovery to infer signaling network characteristics. The inferred protein-protein relationships involving hub proteins were backed up by literature, and suggested potential proteins to be intervened in the future studies of yeast filamentous growth pathways. Details of the methodologies are described in Materials and Methods. Table 1 lists several important summary numbers of this dataset and subsequent analyses.
The relationships of the eight kinase mutants and their effects on global phosphorylation patterns were subjected to correlation analysis (see Overview of the influences inferred from kinase-dead mutations in Materials and Methods). The results were visualized in a correlation heatmap (Figure 3). The negative correlation between kinase mutants of SKS1 and ELM1 are apparent from Figure 3 as are the similarities between some of the mutants (e.g., SNF1 and TPK2). SKS1 mutants inhibit filamentous growth and ELM1 promotes it, while SNF1 and TPK2 have similar phenotypes. The general correlations between kinases are consistent with their filamentous growth phenotypes and reinforce the identification of core target proteins.
We need to be cautious when interpreting the correlations for partially multiplexed data, such as in triplex SILAC experiments. Because a peptide quantified for one sample is highly likely to be quantifiable for the other two samples in the same triplex, the identification and quantification of phosphopeptides in a triplex experiment tend to be linked. In other words, the overlap within a triplex run should be near 100% but the overlap between different triplex runs will be lower due to instrument sampling limitations. A high number of replicates may help minimize missing data, and compensate for the possible bias introduced by tied identification and quantification; but it is rarely performed due to the high cost of these analyses.
A total of 882 phosphopeptides representing 486 proteins were commonly identified in 4–8 kinase-dead (KD) mutants. After the missing values were imputed, the tight clustering method [31] was used to assign those phosphopeptides into groups, and identify the most informative, tight and stable clusters (see Clustering phosphopeptides in Materials and Methods). The results are illustrated in Figure S1 in Text S1. The assignment of proteins and peptides in the top 8 tight clusters is provided in Table 2 and Dataset S1. We also surveyed enriched functions in the tight clusters (Table 2), in terms of functional categories, Gene Ontology, pathways and proteins Domains [27], [32]–[34]. In summary, similar phosphorylation change patterns over multiple mutants (compared to wild type) tends to suggest involvement in similar biological functions. Enriched functional terms include nucleotide phosphate-binding domains, ribosome biogenesis, fructose and mannose metabolism, and glycolysis. Differential carbohydrate metabolism is consistent with the invasive nutrition forage observed under environmental stresses leading to filamentous growth.
We observed examples of multiple phosphorylation domains on the same protein that share similar phosphorylation change patterns and thus end up in the same cluster. For example, “_KGS(ph)FTTELSCR_” (position of the phosphorylated serine: 520) and “_RSS(ph)YISDTLINHQMPDAR_” (position of the phosphorylated serine: 238 or 239) on Psp1p in Cluster 3. It is possible that those phosphorylation sites are co-regulated by the same biological process. They might be closely located in protein tertiary structure or share sequence similarities that allow them to be phosphorylated by the same kinase. Another example where two phosphorylation sites are in the same domain and thus physically close in the protein sequence, “_DQDQSSPKVEVTS(ph)EDEK_” (position of the phosphorylated serine: 495) and “_VEVT(ph)SEDEKELESAAYDHAEPVQPEDAPQDIANDELK_” (position of the phosphorylated threonine: 494) on Leu1p in Cluster 4. Both of these phosphorylation sites were identified in a WT/SNF1/TPK2 experiment, where the serine (S) at position 495 in the former has phosphorylation probability 0.999 (reported by MaxQuant), while the threonine (T) at position 494 in the latter has phosphorylation probability 0.96. These two sites might be alternative phosphorylation sites having similar effects; or the dominancy of either site might be affected by protein cellular localization or kinase activity patterns.
On the other hand, we also found examples of the same protein (e.g., Spt6p) to be clustered in multiple functional groups. Those different sites do not necessarily change phosphorylation in a similar pattern, since they might have different functions and be regulated by different biological processes. All the above observations are worth further investigation.
A total of 863 unique phosphopeptides representing 452 proteins were identified to have significant phosphorylation changes in at least one kinase-dead mutant. We can then infer the downstream proteins regulated by the kinases and the inferred regulation might be direct or indirect. A total of 1,299 significant kinase-phosphopeptide regulation pairs were identified (Dataset S2). We incorporated the corresponding proteins and generated an extended pathway map (Figure 4) based on the known map (Figure 1).
A total of 28 phosphopeptides representing 26 proteins from the entire dataset were found to have globally significant phosphorylation changes (Dataset S3). These candidates were picked out without using prior knowledge. The Fisher's probability test [35] was extended to allow missing values (see Materials and Methods), and it was used for detecting global significance. Each selected phosphopeptide satisfies the following criteria: the combined p-value < 0.05, q-value < 0.05 for controlling false discovery rate (FDR) [36], and the significance B value < 0.05 in at least 4 out of 8 kinase-dead mutant (KD) versus wild type (WT) conditions. The combined p-value is a measure of global significance, while the significance B value [30] is a measure of significance in an individual experiment. Five of the globally significant phosphopeptides, Nth1p, Hsp42p, Pbi2p, Rcn2p and Pdr12p, were identified with complete measurements (Table 3). We consider them high-confidence candidates. Another adaptively weighted statistic [37] was applied to all complete measurements for validation. Adopting the same selection criterion as above, Nth1p, Pbi2p, Rcn2p and Pdr12 were again identified as globally significant. Both retrospective and prospective validation was performed on selected predictions.
Nth1p is a key enzyme in the trehalose pathway which plays a crucial role in glucose homeostasis and stress responses [38], [39] and is a substrate phosphorylated for both Tpk1p and Tpk2p [40]. The NTH1 gene also has been reported to have genetic interactions with the TPK1 and TPK2 genes [41]. It has been reported to physically interact with the kinase Sks1p [1] and with Bmh1p [42]. The above direct interactors of Nth1p, i.e., Tpk1p, Tpk2p, Sks1p and Bmh1p, are all known to play roles in filamentous growth [26], [43]–[47]. The Rcn2p protein was also reported to physically interact with Bmh1p [42], which associates with the Ste20p protein involved in filamentous growth [47], [48]. Bmh1p may also interact with Tpk1p [49]–[51]. Thus, Nth1p and Rcn2p have been closely associated with a number of proteins known to be involved in filamentous growth. Hsp42p has a physical association with Fus3p [42], and its expression is induced under starvation [52]. The remaining two proteins in Table 3 have not yet been closely linked to filamentous growth but play roles in other stress responses and represent new leads.
We also searched the STRING database [53], [54] to investigate the inner connections between the 26 globally significant proteins (shown in Figure S2 in Text S1). STRING assigns the confidence of protein-protein interactions integrating high-throughput experiments, genetic context, co-expression and other previous knowledge. In Figure S2 in Text S1, 17 proteins, including Nth1p, Hsp42p, Rcn2p, Pbi2p, Hsp26p, Bfr1p, YGR250C protein, Leu1p, Lys20p, Cdc19p, Fol2p, Pil1p, Abp1p, Cdc11p, Shs1p, YLR413W protein and Pxr1p, have direct or indirect connections. It presents a closely inter-connected sub-network embodying Nth1p, Pbi2p, Rcn2p, Hsp42, YGR250C protein and Hsp26p.
All possible pairs among the 73 common phosphopeptides with complete measurement were tested using the Pearson correlation. A total of 45 strongly correlated phosphopeptide pairs were identified, each satisfying the following criteria: the correlation test p-value < 0.05, and the stringent requirement of |Pearson correlation coefficient| ≥ 0.9. Detailed information on the 45 pairs of phosphopeptides is provided in Dataset S4. Twenty-seven of the pairs have positive correlations, while 18 pairs have negative correlations. A stringent protein correlation network containing 35 proteins (Figure 5) was generated by connecting the strongly correlated peptide pairs and then tracing the peptides back to their parent proteins.
In the protein correlation network, proteins with the highest degrees of connectivity are considered core components in the network. The 19 proteins having degrees greater than 1 (protein self-connection ignored) in the stringent protein correlation network were predicted to be core components of the network. Detailed descriptions and evidence of the proteins are summarized in Table S2 in Text S1. Kem1p, Spa2p and Spt6p have been reported to be directly involved in filamentous growth in previous literature. Six other proteins, Are2p, Dcp2p, Hsp42p, Ssd1p, Sum1p, and Ufd1p, have reported evidence in terms of genetic and/or physical interactions with known components of filamentous growth. The remaining proteins have been implicated in various stress responses, including the unfolded protein response (e.g., sensitivity to tunicamycin), osmotic shock, and thermal shock, but not previously linked to filamentous growth. Pbi2p has not been reported previously as being involved in filamentous growth; however, our experimental results indicate that a haploid strain of S. cerevisiae deleted for PBI2 exhibits decreased invasive growth relative to wild type (see Experimental validation in Results).
Gpd2p and Lys21p are two self-connected proteins. The self-connection was built up by two distinct phosphorylation sites on the protein. Gpd2p has not been related to filamentous growth in Saccharomyces cerevisiae. Its homolog Gpd2p in Candida albicans, is involved in core stress responses, and GPD2 is induced upon pseudohyphal growth in S. cerevisiae [42]–[49].
In addition to the candidate proteins predicted from our dataset, we retrieved from the literature and authoritative databases [20], [26]–[28], [55], [56] a list of proteins involved in filamentous growth. A total of 69 unique proteins, not all being phosphoproteins, were extracted (Table S1 in Text S1), and 20 of them have been detected in our phosphoproteome dataset. Among those, 15 proteins, including Bcy1p, Cdc28p, Cyr1p, Dig1p, Dig2p, Flo8p, Kem1p, Ras2p, Sfl1p, Snf1p, Spa2p, Ste20p, Ste50p, Tpk3 and Tpm1p, showed significant phosphorylation changes in at least one kinase-dead mutant, and are displayed in our extended pathway map (Figure 4).
The interactions retrieved from the differentially phosphorylated proteins in individual kinase-dead mutants (the dashed edges in Figure 4) did not make use of phosphorylation change pattern over different kinase-dead mutants, and the protein pairs must contain a mutated kinase. In contrast, the correlation network is a network of the common peptides, taking into account the protein responses in all the kinase-dead mutants, and the correlated protein pairs do not necessarily contain the mutated kinases. Note that this network is not directed and more information may be gleaned from a causal analysis. We implemented causal relationship discovery to detect the direction of influences between proteins with the understanding that the relationships may be direct or indirect. A total of 46 unique proteins, including the kinases we mutated, the predicted high-confidence globally significant proteins and hub proteins, and other literature reported proteins, were selected to construct the network. All are listed in Table 4.
Bayesian network modeling identified causal influences for 22 protein pairs (44 phosphopeptide pairs) (Table S3 in Text S1), satisfying the posterior probability of the relationship greater than 0.5. The network comprising all the causal relationships is presented in Figure 6. Among those, only 6 protein pairs have the posterior probability higher than 0.7. The other protein pairs do not have high probability since the samples available for training the model is limited due to the missing data issue caused by instrument limitation. The arrows in Figure 6 only indicate the existence of causal influence, but do not specify whether the influence is activation or inhibition. The causal relationship discovered might be between proteins that are not immediately adjacent in pathways so the relationship could be quite indirect. For example, the causal relationship between Rcn2p and Ste20p might be indirect: Rcn2p and Bmh1p have physical interaction captured by affinity capture-MS [42], while Bmh1p associates with Ste20p to influence filamentous growth [47].
Through another inspection of the phosphorylation change patterns of the peptide pairs detected with relatively strong causal influences (posterior probability higher than 0.7), we observed that: Ste20p has opposing phosphorylation changes compared to Are2p, Pdr12p and Sec21p; two phosphopeptides (the same amino acid sequence but different phosphorylation sites) on Hsp42p present opposing phosphorylation changes compared to Ste20p; and Pbp1p presents consistent phosphorylation change compared to Ste20p. With caution we predict that the opposing pattern implicates an inhibitive influence of Are2p, Pdr12p and Sec21p to Ste20p; and similarly, inhibition of Hsp42p to Ste20p; while Pbp1p shed activating influence to Ste20p. Again, we emphasize that the influence might be quite indirect and even be influenced by multiple pathways.
Our computational analyses highlight the proteins Nth1p, Pbi2p, Pdr12p, and Rcn2p as undergoing globally significant phosphorylation changes. To determine if these proteins do in fact impact filamentous growth in S. cerevisiae, we constructed haploid strains singly deleted for each gene and assayed for filamentation phenotypes. Precise gene deletions were carried out using a standard PCR-based strategy, and resulting haploid strains were assayed for invasive growth by standard plate-washing assays under normal growth conditions [17]. Invasive growth, characterized by filament penetration into the agar substrate, was decreased upon deletion of PBI2 relative to wild-type. In addition, the deletion of NTH1, PDR12, and RCN2 yielded exaggerated invasive growth relative to an otherwise isogenic wild-type strain. Results are shown in Figure 7. All four proteins have been previously implicated in various yeast stress responses, but not specifically with respect to filamentous growth [38], [52], [57]–[63]. Nth1p, i.e. neutral trehalase, is involved in the trehalose pathway, which is a glucose storage pathway [64]. Pbi2p is a cytosolic inhibitor of vacuolar proteinase B, and is involved in the regulation of proteolysis [65]–[69]. Rcn2p, a regulator of calcineurin [70], is induced in response to DNA-damaging agents [59]. Pdr12p is a multidrug transporter inducible by weak acid, and is required for weak organic acid resistance [71], [72]. These four proteins are not reported to be signaling molecules themselves, but we demonstrate that they appear to play roles in filamentous growth and are likely downstream targets of the filamentous growth pathways.
In summary, all four deletions result in differential invasive growth compared to the wild type control, providing prospective validation for our approach to identification of candidate proteins in this biological system from phosphoproteomics data alone.
In this study, we demonstrate that interventional phosphoproteome studies can provide new insight into signaling pathways involved in biological processes such as yeast filamentous growth. In order to increase sensitivity to smaller changes in phosphorylation relative to previous yeast global phosphoproteome studies [5], [6], we used SILAC, an isotope labeling approach. Isotope labeling approaches are generally more precise relative to label-free approaches [11], but require greater resources to implement, resulting in trade-offs between precision and missing data due to sampling limitations inherent to current instruments. We proposed and developed a comprehensive computational and statistical analysis pipeline for the post-identification studies of phosphoproteome data. The analyses are aimed at discovering candidate components of significant pathways involved in filamentous growth as well as the potential targets of the pathways, and to provide more information on the signaling network structure by monitoring changes in phosphorylation in response to mutational interventions. We applied the pipeline to analyze our interim high mass accuracy yeast phosphoproteome datasets and a total of 882 unique phosphopeptides representing 486 proteins were identified as significantly influenced by at least one out of 8 kinase-dead mutants. Twenty-eight unique phosphopeptides having globally significant phosphorylation were identified from the whole dataset among which 5 peptides representing 5 proteins, Nth1p, Pbi2p, Rcn2p, Pdr12p and Hsp42p, were identified as high-confidence candidates. Nineteen candidate proteins with relatively high degrees of connectivity were selected as hub proteins in the stringent correlation network (Pbi2p and Hsp42p were identified as hub proteins too). Among the high-confidence candidate proteins, 3 proteins have been previously reported to be directly involved in filamentous growth and another 6 proteins were also supported, in terms of genetic and physical interactions with known components involved in filamentous growth. The remaining proteins have been implicated in various other stress responses and may play roles in filamentous growth or may be secondary stress responders. In particular, we validated four candidate proteins, Nth1p, Pbi2p, Pdr12p and Rcn2p as impacting invasive growth. Causal relationship discovery was further performed on the candidates and validated proteins. The inferred causal relationships, along with the interactions inferred from phosphorylation changes in response to individual mutants, form phosphoprotein interaction networks, which suggested potential proteins to be intervened in future studies.
Each of the kinases mutated in this study had previously been implicated in filamentous growth. Many of these kinases are known to also affect pathways that are not involved directly in filamentous growth. However, the proteins which change phosphorylation level in response to multiple mutants are reasonable candidates involved in filamentous growth. The sensitivity of such detection is constrained by the degree of overlap between pathways, the coverage of pathways by the mutants, and the extent of missing data. Upstream components of isolated pathways may be missed, while downstream core components are more likely to be identified.
A remaining challenge for quantitative phosphoproteome analysis arises from the sampling limitations and resolution of current mass spectrometers [11]. This feature of tandem mass spectra of complex mixtures results in poor overlap of peptides identified across samples unless a relatively large number of replicate experiments are carried out (which is time consuming and often economically impractical for large-scale projects). For this reason, a significant number of missing values exist in these datasets which can obscure potential candidates for further validation studies. This is likely to be alleviated to some extent in the future as mass spectrometry technologies continue to improve, but we have developed methods to partially compensate for the missing data issue. In the phosphopeptide meta-analysis, an extension of Fisher's combined probability test was made to relax the restrictions of complete measurements. The causal network modeling component was also developed to allow missing values without excluding the incomplete measurements. We also performed cluster analysis of phosphopeptides. Instead of adopting traditional clustering methods, we directly identified the most stable clusters using missing value-imputed data. Our approach was able to pick out significantly enriched functions, and identify a number of reliable candidate proteins for further validation of which four were validated.
This analysis pipeline has been developed to study yeast filamentous growth pathways; however, the methodology is not limited to yeast or this biological process. It can be applied to other complex organisms to facilitate investigation into various biological processes. We anticipate the methodology to be applicable as well to other interventional studies via different experiment platforms.
Tandem mass spectrometry data were generated from a series of triplex SILAC [12], [73], [74] experiments of kinase-dead mutant (KD) strains versus the wild type (WT) haploid filamentous yeast ∑1278b strain. Eight yeast kinases, KSP1p, KSS1p, SKS1p, STE20p, SNF1p, TPK2p, ELM1p and FUS3p, all known to be involved in filamentous growth [20], [26], [27], were chosen to generate kinase-dead mutations (inactivated alleles) individually. We investigated the yeast phosphoproteome from the eight mutants vs. wild type. We have obtained 2–3 replicates for 7 (out of 8) kinase-dead mutants. The dataset constitution is listed in Table S4 in Text S1. Because mass spectrometry experiments are time-consuming and costly, most recent studies in proteome research perform two [75], [76] or three replicates [77], [78] which contributes to the missing data problem in proteomics.
All strains were auxotrophic for Lys and Arg, and were grown on defined medium supplemented with the appropriate isotopic forms of Lys and Arg. The cultures were grown to log phase, and treated with 1% (vol/vol) butanol to induce filamentous growth [26]. The treated samples were incubated for another 16 hours to obtain enough proteins for mass spectrometry analysis. The final O.D. at 600 nm reached a high value usually between 1.0 and 1.5. Different Lys and Arg isotope forms were used to label the three samples in a triplex SILAC experiment: light (Lys0/Arg0) for WT control sample, medium (Lys4/Arg6) and heavy (Lys8/Arg10) for two different mutant samples. Cells were harvested by centrifugation and lysed in the presence of protease and phosphatase inhibitors. In SILAC experiments, samples were pooled at the harvest stage before protein extraction. Samples pooled at this early stage can reduce both systematic and random errors that may occur in later sample preparation [79], [80], thus the results have smaller variance compared to unpooled samples. Small sample sizes (two or three replicates) is acceptable for the low-variance SILAC experimental design. We observed in our SILAC experiments that the majority of the ratio “variability” in the data was less than 20. (The “variability” is reported by MaxQuant and is defined as the standard deviation of all log ratios used for obtaining the reported ratio value multiplied by 100 [30], [81].)
Protein levels were determined by the Bradford protein assay and the proteins from the triplex labeling were then pooled, and were digested by trypsin. The digest was separated into fractions using strong cation-exchange (SCX) fractionation, followed by selective enrichment of phosphorylated peptides using titanium dioxide [82], [83] and then analyzed by LC-MS/MS using a Thermo Fisher Orbitrap XL mass spectrometer. Peptides were identified using MaxQuant software [30] following the Mascot search engine [29], and filtered requiring peptide identification FDR<1%. For Mascot searches, enzyme specificity was set to trypsin. Carbamidomethyl cysteine was set as a fixed modification. N-terminal carbamyl, oxidized methionine, as well as phosphorylation of serine, threonine, and tyrosine were set as variable modifications. Some missed cleavage was observed and could potentially contribute to the variance. The method for calculating peptide identification FDR based on concatenated databases was described by Cox J and Mann M [30]. A total of 3,312 phosphopeptides representing 1,063 proteins were identified. Among those, 73 unique phosphopeptides representing 66 common proteins were commonly identified in all the 8 kinase-dead mutants; while, 882 phosphopeptides representing 486 proteins were common to at least half of the kinase-dead mutants.
|
10.1371/journal.pntd.0005813 | A dynamic model for estimating adult female mortality from ovarian dissection data for the tsetse fly Glossina pallidipes Austen sampled in Zimbabwe | Human and animal trypanosomiasis, spread by tsetse flies (Glossina spp), is a major public health concern in much of sub-Saharan Africa. The basic reproduction number of vector-borne diseases, such as trypanosomiasis, is a function of vector mortality rate. Robust methods for estimating tsetse mortality are thus of interest for understanding population and disease dynamics and for optimal control. Existing methods for estimating mortality in adult tsetse, from ovarian dissection data, often use invalid assumptions of the existence of a stable age distribution, and age-invariant mortality and capture probability. We develop a dynamic model to estimate tsetse mortality from ovarian dissection data in populations where the age distribution is not necessarily stable. The models correspond to several hypotheses about how temperature affects mortality: no temperature dependence (model 1), identical temperature dependence for mature adults and immature stages, i.e., pupae and newly emerged adults (model 2), and differential temperature dependence for mature adults and immature stages (model 3). We fit our models to ovarian dissection data for G. pallidipes collected at Rekomitjie Research Station in the Zambezi Valley in Zimbabwe. We compare model fits to determine the most probable model, given the data, by calculating the Akaike Information Criterion (AIC) for each model. The model that allows for a differential dependence of temperature on mortality for immature stages and mature adults (model 3) performs significantly better than models 1 and 2. All models produce mortality estimates, for mature adults, of approximately 3% per day for mean daily temperatures below 25°C, consistent with those of mark-recapture studies performed in other settings. For temperatures greater than 25°C, mortality among immature classes of tsetse increases substantially, whereas mortality remains roughly constant for mature adults. As a sensitivity analysis, model 3 was simultaneously fit to both the ovarian dissection and trap data; while this fit also produces comparable mortality at temperatures below 25°C, it is not possible to obtain good fits to both data sources simultaneously, highlighting the uncertain correspondence between trap catches and population levels and/or the need for further improvements to our model. The modelling approach employed here could be applied to any substantial time series of age distribution data.
| Trypanosomiasis, spread by tsetse flies (Glossina spp.), is a disease that is fatal for both humans and livestock if left untreated, and is a serious threat to public health in many regions of sub-Saharan Africa. In order to understand the dynamics of the disease it is important also to understand tsetse population dynamics. Tsetse fly mortality estimates are central to this understanding, but are difficult to acquire from wild populations. Previous methods for estimating mortality from age-distribution data assume a stable age structure and age-invariant mortality and capture probability. Based on prior fieldwork, none of these assumptions appears justified. Building on previous mortality estimation techniques, and incorporating what is known about tsetse population dynamics, we develop simulation techniques to estimate mortality for tsetse populations where the age distribution is not necessarily stable. We fit our models to age-distribution data produced in 1991 and 1992 at Rekomitjie Research Station in the Zambezi Valley in Zimbabwe. Our final model produces mortality estimates consistent with those of mark-recapture studies performed in other settings. We find that mortality increases with temperature, a result consistent with field and laboratory findings, and that the temperature effects are much more severe for pupae and newly emerged adults than for mature adults. Our dynamical modelling approach could be used for mortality estimation for any population where substantial age distribution data are available: specifically, it could be used to answer substantive questions about tsetse flies in other settings.
| Human and animal trypanosomiasis, which is spread by tsetse flies (Glossina spp), is a major health concern in much of sub-Saharan Africa [1–3]. Research on trypanosomiasis and tsetse has been carried out for nearly 60 years at Rekomitjie Research Station in the Zambezi Valley (16° 18' S, 29° 23' E; altitude 500 m) [4]. Studies at the station have provided improved understanding of vector and disease dynamics, with the aim of improving disease control [5–7]. It has been shown that the basic reproduction number of vector borne diseases such as trypanosomiasis is strongly dependent on vector mortality rates [8,9]. Accurate estimates of adult tsetse mortality constitute an essential element of that understanding [10].
Laboratory animals and wild populations can differ greatly in their life expectancies [11]. To understand population dynamics in the wild, it is thus essential to obtain data from free-ranging field populations rather than laboratory animals [11]. Mark-recapture can provide good estimates of tsetse population parameters in closed situations, particularly where it is feasible to recapture the same flies many times during their lifetimes [12,13]. Mark-recapture studies are, however, logistically demanding, costly, and time consuming [10]. Moreover, in open populations subject to in- and out- migration, the results are often difficult to interpret, and researchers have accordingly developed alternate methods for estimating mortality [10,14].
As described more extensively elsewhere [15,16], ovarian dissection of tsetse can show the number of times a fly has ovulated. Since tsetse ovulate at approximately regular intervals, these data can be used to estimate the age distribution of female tsetse populations. It has been argued that it should then be possible, in principle, to determine how female mortality rates change with season, and over time, by analysing changes in age-distributions of female tsetse [17–20]. Current techniques for estimating female mortality from ovarian dissection data rely on three important assumptions [17]: first, that sampling probability is not dependent on the age of the fly; second, that mortality rates are independent of the age of the fly; third, and crucially, that the population under study has a stable age distribution.
A recent study shows, however, that these assumptions are often violated, leading to unrealistic mortality estimates. For example, standard techniques predict that mortality decreases with increasing temperature, which contradicts data from mark-recapture studies and is unlikely on biological grounds [10]. We therefore need new methods, for estimating female mortality from ovarian dissection data, which allow for unstable age distributions and for age-related changes in mortality and capture probability.
We develop dynamic models that simulate female tsetse populations, and the associated changes in their age distribution. At Rekomitjie, instability of the age distribution appears to result from large seasonal variation in temperature. Elsewhere, such instability could, however, result from other factors, such as seasonal changes in host density or rainfall. Our alternative approach to mortality estimation will be appropriate however the instabilities arise. Our models estimate or incorporate published estimates of temperature-dependent mortality in adults and pupae, temperature-dependent development rates in pupae, and density-dependent mortality in pupae [16,21,22].
The present paper is concerned with adult female G. pallidipes captured between 1 July 1991 and 30 June 1992 at Rekomitjie using stationary mechanical traps [23], baited with artificial host odour, consisting of acetone (dispensed at 500 mg/h), 1-octen-3-ol (0.4 mg/h), 4-methylphenol (0.8 mg/h) and 3-n-propylphenol (0.1 mg/h) [24]. The capture and processing methods have been described in detail elsewhere [5]. While fly populations and capture probabilities vary seasonally, trap catches were approximately the same at the start and end of the study period.
Female tsetse flies were subjected to ovarian dissection, and assigned to an ovarian category, depending on the relative sizes of the oocytes in the left and right ovaries [15,16]. For flies that have ovulated fewer than four times the ovarian category is equal to the number of times that the fly has ovulated. For flies that have ovulated more than three times, the ovarian category provides only the number of ovulations, or ovarian age, modulo four: that is, a fly in ovarian category 4 may have ovulated 4, 8, 12, 16 … etc. times, and analogous statements apply to flies in ovarian categories 5, 6 and 7. Flies too damaged to assign an ovarian category were excluded from the current analysis. During the study, 19,323 female G. pallidipes were dissected and assigned an ovarian category. The data were aggregated into monthly counts of flies in each category, as detailed in the Supporting Information.
Daily maximum and minimum temperatures are routinely measured using mercury thermometers housed in a Stevenson screen at the station. We incorporated smoothed monthly mean temperatures (estimated as the monthly average of the maximum and minimum readings for each day of the month) in our model of female tsetse populations. Daily mean temperatures are given in the Supporting Information.
We develop a deterministic compartmental model of female tsetse populations based on an understanding of tsetse biology acquired from field and laboratory data. We model only the female population since this is the productive part of the population, for which we have age distribution data. Male and female G. pallidipes at Rekomitjie accrue mouthpart and mid-gut trypanosome infections at similar rates but, since females also live significantly longer than males, they are more likely to transmit trypanosomes (3). There are always sufficient G. pallidipes males at Rekomitjie such that about 98% of females are inseminated by the age of 8 days [16]. We do not, therefore, need to model the male population in order to model the dynamics of the female population. Our model assumes that all females are inseminated and ovulate at 6 days in age. The schematic of the model is shown in Fig 1 and parameters are described in Table 1. Tsetse could be in one of nine groups: the pupal stage and adult ovarian ages 0 to 7.
The assumed pattern of mortality among adult female tsetse is based on field experiments from Zimbabwe [17,18,21]. The experiments showed that adult female G. m. morsitans suffered mortality in excess of 10% per day immediately after emergence: mortality then declined rapidly to about 3% by age 8 days, and then to about 1% per day at age 20 days. Thereafter, mortality increased steadily but slowly with age, such that the mortality was still less than 2.5% per day at age 100 days. Female adult mortality is thus clearly a function of age, but the major changes are restricted to the immature stages, while the fly’s thoracic musculature is still developing, and before it has ovulated for the first time. Notice that, since we are fitting our models only to ovarian age data from adult females, and we have no data on pupal numbers or losses, we cannot separate between deaths that occur in the pupal stage proper and those that occur among flies that have just emerged from the pupa and are not yet available to traps. Thus, while for convenience, we formally apply temperature and density-dependent mortality to pupae in the modelling, pupal mortality needs, strictly, to be interpreted as mortality in all stages prior to the mature adult stage. In subsequent text and tables, we refer to this mortality as immature mortality. We assumed, as a first approximation, that mortality was constant for all mature adult females. We also include, however, a model in which mature adult female mortality was allowed to vary with age in the Supporting Information.
Adults in a given age category are assumed to progress to the next age category at a constant rate, with flies of ovarian age 0 progressing to ovarian age 1 at a rate of (6 days)-1 and flies of all other ages progressing to the subsequent ovarian ages at a rate of (9 days)-1[6,7,16,25]. These rates are all temperature dependent but, at the field temperatures observed at Rekomitjie, the times between ovulations are unlikely to vary by more than 2 days from the assumed mean value and should not thus be a source of major error in the mortality estimates.
Pupae are deposited at a rate equal to the number of flies in ovarian categories 1 to 7 divided by the expected amount of time (9 days) which flies remain at these ovarian ages. Since only half of these pupae are female, this quantity is then divided by two to obtain the number of female pupae deposited. In other settings a pupal mortality of around 1% per day [26] has been assumed; in order to encompass this estimate in our models, we set the minimum pupal mortality rate below this at 0.001 per day. There is evidence for density-dependent mortality in pupae in the wild [26]: accordingly, we model a density-dependent pupal mortality rate proportional to the number of female pupae. The pupal mortality rate at temperatures under 25°C is given by:
μp,T≤25=0.001+dP,
(1)
Where d parameterizes the density dependence and P is the pupal population in a given, but undefined, area.
Pupal duration is temperature-dependent in tsetse and female pupa emerge to become adults in age category zero. Laboratory studies show that pupal emergence rates and temperature are related by [27]:
h(T)=k/(1+exp(a+bT))
(2)
where T is the mean daily temperature and k, a, and b are constrained to be the values given in Table 1. We assume that Eq (2) also provides acceptable estimates of pupal duration for female G. pallidipes at Rekomitjie.
Mark-recapture studies suggest that mortality in adult female G. pallidipes increases exponentially with temperature (T) for T > 25°C [28]. Accordingly, we model adult fly mortality using:
μa(T,T>25°C)=μa,T≤25expβ(T−25)
(3)
Where μa,T≤25 is the mature adult mortality for temperatures at or below 25°C, and β parameterizes the increase at higher temperatures.
Pupal mortality is also thought to be temperature dependent and we also allow this quantity to increase exponentially with temperatures above 25°C:
μp(T,T>25°C)=μp,T≤25expα(T−25),
(4)
Where μp,T≤25 is the adult mortality rate for temperatures at or below 25°C given in Eq 1, and α parameterizes the increase in the mortality rate at higher temperatures.
Field evidence from Zimbabwe indicates the probability with which a female tsetse is captured in a trap increases with her age, with the greatest bias against flies in ovarian categories 0 and 1 [5,29]. We model this by allowing a relative risk of capture less than 1 for these young flies.
All simulation and analysis was performed using R 3.3.0. Differential equations specifying the three models, based on the compartmental model diagram in Fig 1, are given in the Supporting Information. Differential equations were solved numerically using the Livermore solver in the deSolve package. Since starting conditions were unknown, models were run for three years, which allows enough time for a stable pattern in population levels to be achieved. For the last year of the simulation, the average ovarian age distribution was calculated for each month of the study and compared with the observed ovarian age distribution for that month. For the main analyses, the log-likelihood for a given month was calculated assuming multinomial sampling of the available population. For the fits to both ovarian dissection and population data, the log-likelihood of the population data given the model was calculated using binomial sampling probabilities and added to the log-likelihood of the ovarian dissection data. To do this, catches per trap were scaled such that equal weight was given to both datasets, the true population was scaled such that it was large compared to the catch population (10,000 times on average), and the sampling probability was assigned to be the mean ratio of the scaled catch population to the true population. The overall log-likelihood was calculated as the sum of the log-likelihoods for each month. Since the population was roughly the same at the beginning and the end of the study period, a simulation was given a likelihood of zero if the simulation failed to produce a pupal population on July 1, 1991 within 3% of the pupal population on June 30, 1992.
Rates of pupal development, of transition between ovarian categories, and of minimum pupal mortality were fixed at the values suggested in the literature: all other parameters were determined using model fits. For each model, the maximum log-likelihood of the model was determined using the Nelder-Mead downhill-simplex algorithm, as implemented using the optim function in R. The optimization was performed to determine the optimum parameter set for that model, using a logit scale for some parameters (S0, S1, μa) and a log scale for the remainder of the unknown parameters. The Hessian was calculated and then inverted to give the Fisher information matrix, which was used to obtain confidence limits for the variables. The confidence limits were then detransformed to obtain the confidence intervals for each parameter. Nested models that have an AIC at least 6 units larger than the best performing model were considered to confer a significantly worse fit than the best performing model (less than 0.05 times as probable as the best performing model) [30].
Standard techniques for estimating female tsetse mortality from ovarian dissection data are biased, especially during the hot-dry season (November to March) [10]. The bias associated with standard techniques varies with temperature variation: where, as in Zimbabwe, this variation is large, the bias will be most severe. Tsetse near the equator, where temperature variations are small, may have age distributions that are approximately stable—although it is possible other environmental factors may lead to nonstable age distributions in these settings [10,25]. To quantify this bias, using simulated data from our model together with our the maximum likelihood parameters, we estimated daily mortalities for the study period using standard techniques, as described elsewhere [15,17,26,29] and reviewed by Hargrove and Ackley [10].
Using maximum likelihood parameter estimates, we generated simulated female tsetse populations with varying magnitudes of temperature variation, but the same mean temperature. For a parameter f, varying between 0 and 1, we created counterfactual daily temperature Tnew(f) for each day based on the true temperature in Zimbabwe, T, as follows:
Tnew(f)=f(T−T¯)+T¯,
(5)
Where T¯ is the mean annual temperature. We then determined the mean bias in mortality estimation using standard techniques as a function of f.
For the main analysis, we fit models 1–3 to the age distribution data, only constraining the population to be approximately the same at the beginning and end of the experimental period, but not taking any account of differences between predicted population levels and observed trap catches in the period between the endpoints. Maximum likelihood fits for model 3 provide a good fit to the age distribution data, but show large differences between predicted trends in simulated population and trap catches (Fig 2). Maximum likelihood fits are shown for models 1 and 2 in the Supporting Information. Table 2 summarizes the estimates of maximum likelihood parameters and their confidence intervals. Model 3, which allows for differential mortality between mature adults and immature tsetse, confers a significantly better fit than models 1 and 2. For all three models, the estimated mortality rate at temperatures less than 25°C is consistently 0.03 per day; 3% attrition per day is considered plausible for a non-decreasing tsetse population [12,31], and the maximum adult mortality before the population starts to decline is estimated to be approximately 4% per day [32]. For model 3, we estimate a mortality rate of 0.028 per day at temperatures less than 25°C, with mortality increasing for temperatures above 25°C. Figs 3 and 4 illustrate that mortality among mature adults hardly changes with time of year and temperature, whereas there is significantly more variation in the mortality of immature tsetse, which exhibit peaks in about November and March. Pupal mortality exceeds mature adult mortality except during the dry, generally cooler, months of May to September.
When model 3 was fit to both the ovarian dissection data and the catch data simultaneously there was a poorer fit to the ovarian dissection data, as apparent in the considerably lower likelihood and from examination of the fits (Table 3, Fig 5). Using this model, it is not possible to simultaneously obtain good fits to both the age distribution and the trap catch data. Nonetheless, fitting to both data sources produces a mortality rate estimate for mature adults of 0.027 per day at temperatures under 25°C, comparable to estimates from the models fit only to the ovarian dissection data (Table 2). Given the challenges in fitting both data sources and the uncertainty of the correspondence between trap catches and population levels, we used model fits to only the age distribution data for further investigations.
We used model 3, the minimum AIC model in our main analysis, and corresponding parameter estimates (Table 2) to quantify the possible biases inherent in standard mortality estimation techniques. We generated daily ovarian category distributions using observed smoothed monthly mean temperatures at Rekomitjie. With these distributions, we estimate the mortality among mature adult tsetse for that day. Fig 6 shows the model-generated mortality (solid line) and the mortality estimated using standard techniques (dashed line) as a function of time of year at Rekomitjie, assuming the maximum likelihood parameter estimates for model 3 (Table 2). Standard techniques are biased at all times of year, but the bias is largest during the hot-dry season. In this season, standard techniques give the lowest mortality estimates when mortality is in fact highest. Our modelling suggests, however, that this increased mortality occurs almost exclusively among recently emerged, immature, adults. Since data are typically aggregated by month to estimate a monthly mortality, Fig 6 also shows the monthly mean of mortalities generated using our model (closed dot) and estimated using standard techniques (open dot). Aggregating the data by month does not significantly alter the magnitude of the bias.
Fig 7 shows the mean bias of standard mortality estimation techniques as a function of temperature variation. This bias is plotted for various mortalities (text annotations), using model 3 and the maximum likelihood parameter estimates for the remaining parameters. As shown in Eq 5, f = 0 corresponds to no temperature variation, and f = 1 corresponds to the temperature variation at Rekomitjie. The general trend is that bias in standard techniques is greater when there is greater temperature variation. However, the magnitude of the bias and the extent to which it depends on temperature variation depends on the mortality rate.
A major finding of this study is that mortality in immature tsetse is higher, and increases much more rapidly with increasing temperature, than in mature adults. These results are consistent with results from a mark-recapture study showing that mortality among recently emerged female tsetse is markedly higher than for all older flies [16,21]. It may be objected, however, that the high mortality estimated in the mark recapture experiment, in newly emerged—and newly marked and released—flies might merely reflect stress due to their handling and marking. This possibility was acknowledged in the original analysis [33], but it was argued that the continual decrease in the loss rate over the first 18 days of life was consistent with high (natural) losses in young flies. This conclusion is also consistent with published evidence that a large percentage of newly emerged tsetse can die before they become available for capture in the field and that this percentage can increase dramatically at high temperatures [31,33–35].
The results are, admittedly, at variance with published data on other insects where there have been no reports of mortality being higher in recently emerging insects than in older adults. In part, however, this may reflect the difficulties attendant on estimating insect mortality in the field. The case of age-dependent mortality in mosquitoes provides a good example of the problems involved. The following analysis of published methods for estimating age-specific mortality in mosquitoes suggests that they would not allow the detection of increased mortality in newly emerged field mosquitoes even if it existed.
To our knowledge, nobody has yet carried out on mosquitoes, or indeed on any other insect, the equivalent of the experiment on tsetse where insects were marked uniquely, released in field at birth, and where their recapture history was recorded for the rest of their lives [16,21]. Early analysis of the age structure of mosquitoes captured in the field showed that mortality rates increased with age [36]. There was no evidence for increased mortality in the youngest mosquitoes, but this method obviously cannot estimate the number of mosquitoes that die before they become available for sampling and thus cannot provide any estimate of mortality among the youngest mosquitoes.
This objection does not apply to a study where the survival of large samples (>10,000) of mosquitoes was followed from birth [37]. For both sexes, mortality was low at young ages (< 10 days old), steadily increased among middle-aged mosquitoes, and decelerated at older ages. Again, therefore, this experiment produced no evidence for increased mortality among young mosquitoes. However, this was a study of laboratory-bred and raised mosquitoes and, again, says nothing about the rate at which newly emerged mosquitoes might die in the field. As observed above and elsewhere, mortality among young tsetse in the laboratory is much lower than estimated for field flies [21]: the same may well be true for mosquitoes.
The “captive cohort method” for estimating population age structure in the wild can be used to estimate age-specific mortality rates. Since, however, the method involves following the survival of samples captured in the field [38–40], the problem referred to above arises: the method cannot provide mortality estimates for insects that die before they can be sampled. Moreover, while the method is valid for stationary populations (stable age distribution and zero growth rate), violations of this assumption require more complex modelling approaches, with quantification of population birth rates or immigration/emigration rates [39].
Thus, while the studies reviewed here produced no evidence for increased mortality in very young mosquitoes, the methods used would not anyway be able to detect increased mortality among newly emerged mosquitoes in the field. Nonetheless, it is not unreasonable to expect that the risks faced by young tsetse, relative to mature adults, are much greater than those faced by their mosquito counterparts. Teneral tsetse have low levels of fat, poorly developed flight musculature, relatively weak flight capability and, being obligate blood feeders, need to locate a vertebrate host and feed off it safely before they starve [41–43]. The problems for newly emerged mosquitoes are less severe: the flight performance of Aedes aegypti, for example, is highest during the first 14 days of life [44]: since, also, mosquitoes can feed off nectar and plant juices, their feeding risks should be much lower than for tsetse.
Since we were only fitting our models to age distributions of adult females, we had no way of separating death rates in immature tsetse between deaths that occurred in the pupal stage and those occurring in very young adults, before they became available for trap capture. Regardless of how these deaths among immature classes are counted, however, if they occur in large numbers they will contribute to destabilisation of the population age structure. This effect has bedevilled past efforts to estimate female tsetse mortality from ovarian dissection data, which assumed that the population age distribution was stable [10]. We developed dynamic models that allow for age distributions that are unstable, in our case due to fluctuations in temperature, and also allow for age-related changes in mortality and capture probability.
For each of the three models, mortality estimates are around 0.03 per day for temperatures under 25°C, consistent with estimates from other areas. Mark-recapture studies at the nearby Antelope Island gave an adult female mortality rate of 0.023 at lower temperatures, but with a larger increase in mortality with increasing temperature than predicted by model 3 [33]. Previous mark-recapture work has shown that the increase in mortality for adult G. pallidipes females at higher temperatures is exponential and characterized by a coefficient of 0.106 [33], which is much larger than the value of the analogous parameter (β) estimated for model 3. The disagreement could be due to the incorporation in the mark-recapture estimates of some young flies that still have higher natural mortality rates than fully mature flies. Mortality is highest among flies that have just emerged, but only declines to mature levels over the first 10 days of adult life [33].
Our modelling produces mortality estimates that appear more reliable than those from classical analyses in that mortality is predicted to increase with temperature, as expected. Our technique also offers insight into the factors driving the dramatic changes in age distribution observed during the hot-dry season. Model 2, where mortality among mature adults and all immature tsetse have the same dependence on temperature, does not perform significantly better than model 1, where mortality is independent of temperature. Temperature-dependent mortality for immature stages and mature adults cannot thus explain the observed changes in age distribution if it is applied evenly across all stages. Model 3, which allows for differential dependence of mortality on temperature for mature adults and immature stages, performs significantly better than models 1 and 2. This differential dependence of mortality on temperature causes the observed ratio of young flies to older flies to decrease at high temperatures, which would explain why a greater fraction of flies sampled during the hot-dry season are older than at other times of year.
Our results suggest that increases in temperature affect the mortality of recently emerged adult G. pallidipes more than all older flies. Age-dependent mortality has been documented in the field for G. m. morsitans [21] and in the laboratory for various tsetse species [22,45,46], and we suggest that these differences may not remain constant with changes in temperature. Using standard mortality estimation techniques to estimate fly mortality can demonstrate the bias inherent in these techniques and show how this bias varies seasonally for locations near and far from the equator. We find that during the hot-dry season, the mortality estimates using standard techniques are most biased, whereas during the rest of the year, the bias is smaller, though still significant (Fig 5). We also find that the bias in standard techniques will tend to decrease with decreasing annual variation in temperature. Standard techniques, which assume a population that is declining or growing exponentially [17,29], may nonetheless give biased mortality estimates, and the magnitude of this bias is not easily predicable since it may depend on parameters such as the mature adult mortality. An unpredictable bias can make it impossible to compare mortalities at different times or in different regions.
Elsewhere in Africa, particularly at sites close to the Equator, it may be true that temperature variations are not sufficient to produce age structure instability. However, the possibility cannot be excluded that age structure instability could result elsewhere from other climatological effects: any such effects will cause errors in mortality estimates where these are derived using the classical approach. It is thus incumbent on investigators to convince themselves whether or not their study population does indeed exhibit a stable age structure.
Our work has several strengths: (i) Extension of traditional mortality estimation techniques from ovarian dissection data. Our model is an extension of traditional mortality estimation techniques from ovarian dissection data that allows for non-stable age distributions and explicitly constrains population growth. This allows us to make mortality estimates that are directly comparable to those from traditional techniques, while avoiding incorrect assumptions. (ii) Dynamical modelling. Dynamical modelling techniques, such as the compartmental modelling we employ, have been used to estimate key demographic and disease transmission parameters in many settings, including for vector populations [47] and tsetse populations [12]. (iii) Model based on an understanding of tsetse biology acquired from field and laboratory data. Our models include temperature-dependent pupal emergence, pupal (and immature) mortality, and mature adult mortality, as well as density-dependent pupal mortality.
Our work also has several limitations. (i) Unknown populations of adults and pupae over the course of the year. The most serious challenge we have in our modelling arises from our ignorance regarding the way in which the true population numbers of adult and puparial tsetse vary with time and season. Differences in tsetse trap catches from month to month are undoubtedly related to population changes; however, they also plausibly reflect changes in fly behaviour, and thus capture probability, with changing temperature, and potentially reflect changes in age structure and seasonal in- and out-migration [48]. For our main analysis, we did not constrain our models to fit population numbers, whether for adult or immature stages. The different models presented in Tables 2 and 3 give very different estimates for how the pupal and adult fly populations change over the course of the year. Nonetheless, the mortality estimates for temperatures under 25°C did not vary significantly between the three models, and, for model 3, with or without fitting to trap data. The mortality estimates for temperatures less than 25°C do not, therefore, appear to be particularly sensitive to the way in which the total population is predicted to change. However, more accurate estimates of population changes over the course of the year would be required to determine how high temperatures affect mortality.
(ii) Classification of teneral deaths. As detailed in the Methods section, we could not separate pupal deaths from those occurring in newly emerged adults before they could be trapped. Indeed, it is not always clear how to categorise deaths among immature classes. Pupae that are predated or parasitized are clearly true pupal deaths. A major loss of immature flies is, however, due to the excessive use of fat at extreme temperatures during the pupal phase [34]. If so much fat is used during the pupal phase that the emerging fly has insufficient energy to find its first meal—or even fails to emerge—it is unclear whether this should be characterised as a pupal or a teneral adult death. For convenience, we have modelled temperature-related deaths in immature stages as occurring with equal probability across the pupal stage but, in reality, most of these deaths will actually only occur at, or very shortly after, emergence. In addition, at high temperatures, the abortion rate increases to about 2% and the high pupal mortalities produced by our model fit may be accounting for this as well [49].
(iii) Assumptions about the pupal mortality rates. We assumed a density dependence of a specific form with a minimum pupal mortality of 0.001. Since the maximum likelihood estimate of the pupal population was always large enough such that the pupal mortality was much greater than this minimum, our model was not sensitive to this assumption. Previous research has indicated that pupal mortality increases sharply for temperatures < 18°C [50]. During the period of our study, however, the 20-day running mean temperature at Rekomitjie never dropped below 20°C. Accordingly, for simplicity, we only model increases in pupal mortality with increasing temperature (see the Supporting Information).
(iv) Age independent mortality for mature adult females. We do not include age-dependent mortality for adults in models 1, 2, or 3. While mortality does increase with age among mature adult female tsetse, our dynamic models capture the spirit of the field-estimated changes in mortality with age among female tsetse in having: (i) very high mortality in immature adults (inasmuch as a high pupal mortality approximates this); (ii) low mortality in mature adults; (iii) a rate of increase in mortality with age that is very small—and in fact taken as zero in our model. As an additional sensitivity analysis, we have included in the Supporting Information a model where mortality among mature adults is allowed to vary with age. This model produces similar mortality estimates to those produced by model 3, and also predicts a very low rate of increase in mortality with the age of mature adult females.
(v) Other factors omitted from the model. Tsetse biology is complicated and our models did not include every possible factor that could affect mortality estimation. For example, we assumed that the time taken for a female to produce her first larva, and the subsequent interlarval periods, were independent of temperature. This is known not be the case at Rekomitjie, but the variation amounts only to a few days over the whole temperature range and errors arising from this approximation may be expected to be small relative to other errors. In addition, meteorological factors other than temperature may affect tsetse mortality and these factors were not included in the model. The inclusion of additional factors could help to address the difficulty in obtaining good fits to catch and ovarian data simultaneously.
In conclusion, we developed a dynamical model that produces adult tsetse mortality estimates and temperature-dependent increases in mortality consistent with mark-recapture studies. Mortality estimates at temperatures under 25°C are consistent across a number of sensitivity analyses. This model provides important insights into the changing ovarian age-distribution over the course of the year: differential increases in mortality for pupae and young adults versus older adults lead to higher mean ovarian ages at the hottest times of the year. In addition, standard techniques for mortality estimation may be highly problematic in areas with large variation in temperature, such as Rekomitjie. In areas near the equator with less marked variations in temperature, standard techniques may still be significantly biased. Lastly, with longitudinal ovarian dissection data, or indeed any age distribution data, the dynamic modelling approach we employ could be used to estimate mortalities in other settings, and could be modified to answer other research questions, such as the effects seasonal changes in host density.
|
10.1371/journal.pgen.1002810 | Gene Conversion Occurs within the Mating-Type Locus of Cryptococcus neoformans during Sexual Reproduction | Meiotic recombination of sex chromosomes is thought to be repressed in organisms with heterogametic sex determination (e.g. mammalian X/Y chromosomes), due to extensive divergence and chromosomal rearrangements between the two chromosomes. However, proper segregation of sex chromosomes during meiosis requires crossing-over occurring within the pseudoautosomal regions (PAR). Recent studies reveal that recombination, in the form of gene conversion, is widely distributed within and may have played important roles in the evolution of some chromosomal regions within which recombination was thought to be repressed, such as the centromere cores of maize. Cryptococcus neoformans, a major human pathogenic fungus, has an unusually large mating-type locus (MAT, >100 kb), and the MAT alleles from the two opposite mating-types show extensive nucleotide sequence divergence and chromosomal rearrangements, mirroring characteristics of sex chromosomes. Meiotic recombination was assumed to be repressed within the C. neoformans MAT locus. A previous study identified recombination hot spots flanking the C. neoformans MAT, and these hot spots are associated with high GC content. Here, we investigated a GC-rich intergenic region located within the MAT locus of C. neoformans to establish if this region also exhibits unique recombination behavior during meiosis. Population genetics analysis of natural C. neoformans isolates revealed signals of homogenization spanning this GC-rich intergenic region within different C. neoformans lineages, consistent with a model in which gene conversion of this region during meiosis prevents it from diversifying within each lineage. By analyzing meiotic progeny from laboratory crosses, we found that meiotic recombination (gene conversion) occurs around the GC-rich intergenic region at a frequency equal to or greater than the meiotic recombination frequency observed in other genomic regions. We discuss the implications of these findings with regards to the possible functional and evolutionary importance of gene conversion within the C. neoformans MAT locus and, more generally, in fungi.
| Recombination has been thought to be repressed within sex chromosomes, as well as within the mating-type (MAT) loci in many fungi, due to the highly diverged and rearranged nature between alleles defining opposite sexes or mating-types. However, it has long been appreciated that recombination can occur within these presumptive recombinational “cold spots,” and recent studies reveal that recombination, including gene conversion, can occur at a frequency higher than previously appreciated and could play important roles in shaping evolution of these chromosomal regions. Here, we provide evidence that, during sexual reproduction of the human pathogenic fungus Cryptococcus neoformans, recombination (gene conversion) occurs across a GC-rich intergenic region within the MAT locus. The frequency of this gene conversion is comparable to those of typical meiotic recombination events observed in other chromosomal regions. This is in accord with population genetics analyses, which indicate homogenization between alleles of opposite mating-types within the intergenic region. Gene conversion within these highly rearranged chromosomal regions may serve to ensure proper meiosis and/or rejuvenate genes/chromosomal regions within MAT that are otherwise facing irreversible evolutionary decay. In conclusion, our study provides further experimental evidence that at least some recombinational “cold spots” are not that cold, after all.
| During meiosis, recombination occurs to promote genetic exchange and ensure the proper segregation of homologous chromosomes. This process is initiated by the introduction of genome wide DNA double strand breaks (DSBs), followed by strand invasion and elongation. If the second end of the DSB is captured, a double Holliday Junction (double-HJ) will form. Resolving of the resulting double-HJ usually results in exchange of genetic information between the two homologous chromosomes. Depending on the way that the double-HJ is resolved (i.e. resolution or dissolution), this exchange can be either reciprocal (i.e. crossing-over), or unidirectional (i.e. gene conversion). It should be noted that although all recombination events are accompanied by gene conversion, only a fraction of recombination events actually result in crossing-over [1], [2], [3],[4],[5]. Alternatively, after the initial strand invasion and elongation, DSBs can also be repaired through a synthesis-dependent strand-annealing (SDSA) pathway, which only produces gene conversion [6], [7].
Studies have shown there are homeostatic controls acting on crossover formation during meiosis in both yeast and mouse [8], [9]. Additionally, recombination frequency is not evenly distributed across genomes, with certain regions being “hot spots”, where recombination occurs at higher frequencies, and other regions being “cold spots” that experience recombination at comparatively lower frequencies [10]. Some examples of recombination “hot spots” include the major histocompatibility complex (MHC) in humans [11], [12], as well as the regions flanking the MAT locus in the human pathogenic fungus Cryptococcus neoformans [13]. On the other hand, it has been shown that in some areas of the genome, such as centromeric regions and sex chromosomes, recombination is repressed during meiosis, largely due to the existence of repetitive sequences and extensive chromosomal rearrangements within these regions that prevent the proper pairing between the homologous chromosomes during meiosis. However, a recent study by Shi et al. reported that recombination, in the form of gene conversion, is widespread within the centromeric regions of maize [14], and studies have shown that meiotic recombination can occur within the less complex yeast centromere [15]. Additionally, gene conversion has also been reported to occur within human and ape Y chromosomes, as well as between X-Y homologues located within the nonrecombining region of the Y chromosome in Felidae [16], [17], [18]. Nevertheless, recombination frequencies within these presumed “cold spots” are still considered to be much lower than those within typical chromosomal regions.
Recombination frequencies have been shown to be positively correlated with the local GC content in a variety of species, including humans [19], [20], the yeast Saccharomyces cerevisiae [21], and the human pathogenic fungus C. neoformans [13]. Several studies support the view that recombinational activity is driving the local GC composition [22], [23], [24], [25], [26], possibly through a process in which recombination-associated gene conversion introduces a bias in favor of GC, also called GC-biased gene conversion (gBGC) [27], thus increasing the local GC content around the recombination hot spots. However, a recent study of the yeast S. cerevisiae suggests that local GC content is not driven by recombination [21]. Additionally, the local GC content may also influence the susceptibility of a given chromosomal region to mutagenic factors, such as UV radiation [28], and thus indirectly affect the local frequency of recombination required for DNA damage repair processes.
Sex chromosomes, such as those in mammals and birds, and larger mating type (MAT) loci of several fungi (e.g. the MAT locus of C. neoformans), usually have extensive sequence divergence and chromosome rearrangements between opposite alleles, and are thus thought to be suppressed for recombination during meiosis. This repression of recombination protects the integrity of the linked sex determining genes, and prevents generation of deleterious abnormal chromosomes resulting from recombination within rearranged chromosomal regions. For organisms with heterogametic sex, the proper segregation of the sex chromosomes is usually ensured by recombination occurring within a defined region within the sex chromosomes, such as the pseudo-autosomal region (PAR) of the mammalian sex chromosomes. Studies have shown that crossing-over occurring within these regions is crucial for the proper segregation of the two sex chromosomes during meiosis [29]. Similarly, recombinational hot spots have also been identified flanking the non-recombining MAT locus in C. neoformans [13]. Despite the facts that recombination within the PAR maintains homology between the opposite sex chromosomes [30], as well as that certain selection processes, such as purifying selection, operate to ensure the proper function of these regions, the non-recombining nature of the majority of the sex chromosomes, as well as the fungal MAT loci, threatens these regions with gradual deterioration, as suggested for the mammalian Y chromosomes [31].
A previous study by Hsueh et al. [13] showed that recombination hot spots exist in the regions flanking the MAT locus of the human pathogenic fungus C. neoformans, and these regions have an unusually high GC content. In the same study, a minor GC peak was identified that is located within the intergenic region of the RPO41 and BSP2 genes in the MAT locus. However, whether or not this minor GC peak is also associated with a higher recombination frequency was not analyzed due to the lack of polymorphic markers within this region. The objective of this study was to investigate whether this minor GC rich region also shows unique features in recombination frequency during meiosis. To address recombination in this GC rich region, we employed both population genetics analyses of natural isolates, as well as analysis of meiotic progeny generated from laboratory crosses. We found that recombination, in the form of gene conversion, occurs at a frequency that is at least comparable to the meiotic recombination frequencies in other chromosomal regions, and this observation is also supported by results from population genetic analyses showing homogenization spanning this intergenic GC rich region in natural strains. These observations have implications for the evolution of mating type loci and sex chromosomes, and the nature and formation of recombination in other complex genomic regions, such as centromeres.
The GC rich intergenic region between the RPO41 and BSP2 genes in the MAT locus of C. neoformans var. neoformans was first identified by Hsueh et al [13]. Taking advantage of the existing genomic sequences of species of the pathogenic Cryptococcus species complex [32], [33], we further investigated whether this GC rich region is present in other lineages that are closely related to C. neoformans. GC plots of the homologous regions from strains 125.91 (C. neoformans var. grubii, MATa), H99 (C. neoformans var. grubii, MATα), E566 (Cryptococcus gattii, VGI, MATa), WM276 (C. gattii, VGI, MATα), R265 (C. gattii, VGII, MATα), NIH312 (C. gattii, VGIII, MATa), and B4546 (C. gattii, VGIII, MATα) all showed GC peaks at similar positions (Figure 1), reflecting the common origin of this region in these lineages. It also suggests this GC rich intergenic region may be functionally important, such that the high GC content is being maintained through either natural selection (e.g. purifying selection) or other molecular mechanisms (e.g. gene conversion).
We sequenced a region that encompasses the 5′ ends of the RPO41 and BSP2 genes, and the GC rich intergenic region between the RPO41 and BSP2 genes from a group of natural C. neoformans strains, including both var. grubii and var. neoformans (Table 1 and Figure 2). We also PCR amplified and sequenced serotype A and serotype D specific alleles from a group of AD hybrid isolates (Table 1 and Table S1).
Overall, this region showed a serotype specific phylogeny when all of the alleles are compared together, as all of the serotype A alleles (from both haploid var. grubii isolates and serotype AD hybrids) grouped within a well supported cluster that showed considerable divergence from the cluster that included all of the serotype D alleles (Figure 3). This pattern still held when the regions belonging to the three sections (the two genes, RPO41 and BSP2, and the intergenic regions) were analyzed separately (Figure 3), consistent with the view that serotypes A and D are well separated lineages that diverged from each other long ago. Among the three sections, the intergenic region showed the highest level of divergence between the two clusters of serotype A and serotype D alleles (Figure 3). Within each cluster, the level of polymorphism within the intergenic region is comparable to the other two gene coding regions among serotype D alleles, and is higher than the other two sections among serotype A alleles (Table 2). No signal of positive selection (dN/dS>1) was detected in either RPO41 or BSP2 gene coding region for either serotype A or serotype D alleles.
When only serotype A alleles were considered, well supported clusters containing all and only the MATa specific alleles were observed when all of the sites were included in the analyses (Figure 4A), as well as when the regions corresponding to the two genic regions were analyzed (Figure 4B and 4D). However, when only the intergenic region was analyzed, this mating type specific topology no longer held. Instead, MATα alleles from six haploid strains were placed within a well supported cluster that otherwise contained only MATa specific alleles (Figure 4C). The phylogeny of the intergenic region was shown to be statistically different (p<0.01) from those of the two flanking genes by the Shimodaira-Hasegawa test [34].
When only serotype D alleles were considered, again a well supported cluster containing all and only the MATa specific alleles was observed when all of the sites were analyzed (Figure 5A), as well as when the region corresponding to the gene RPO41 was analyzed separately (Figure 5B). When the regions corresponding to the intergenic region and the BSP2 gene were analyzed separately, this mating type specific phylogeny was no longer well supported. Specifically, when only the region within the BSP2 gene was analyzed, four MATa specific alleles formed a well supported cluster, while the other five MATa specific alleles were grouped together with most of the MATα specific alleles within a well supported cluster, reflecting a sharing of polymorphisms between alleles from the two mating types in this region. For the intergenic region, other than the two well supported clusters, one containing two MATα haploid alleles and the other containing two haploid MATα alleles and nine MATα alleles from hybrid AD isolates, all of the other MATa and MATα alleles grouped together, indicating a lack of divergence between MATa and MATα alleles at the GC rich intergenic region (Figure 6). Similar to serotype A alleles, the topology of the intergenic region was shown to be statistically different (p<0.01) from those of the RPO41 and BSP2 genes by the Shimodaira-Hasegawa test [34].
We further looked for possible gene conversion tracts in this region in the natural population of C. neoformans. For serotype D alleles, we did not detect any statistically significant gene conversion tracts, possibly due to the low level of polymorphism present around the GC rich intergenic region. However, when serotype A alleles were analyzed, we indeed found several statistically significant gene conversion tracts encompassing the GC rich intergenic region (Figure 6; p<0.05 based on 10000 permutations). The lengths of the gene conversion tracts ranged from 560 to 2100 bp, consistent with previous studies in yeast showing most gene conversion tracts were between 1800 to 2000 bp [35]. These gene conversion tracts collectively spanned a 2.5 kb region encompassing the GC rich intergenic region (Figure 6).
Taken together, our population genetic analyses revealed that: 1) compared to the two genic regions, there was higher divergence between serotypes A and D alleles in the intergenic regions; 2) within serotype D alleles, the intergenic region showed a low level of polymorphisms compared to the flanking genic regions, as well as other intergenic regions within the MAT locus, even though no evidence of purifying selection was detected in the flanking genic regions of the RPO41 and BSP2 genes; 3) within serotype A and serotype D alleles, respectively, a mating type specific topology was observed for regions corresponding to the RPO41 and BSP2 genes, albeit the pattern was less well resolved for the serotype D BSP2 region; 4) there was no mating type specific topology at the intergenic regions for both serotypes A and D alleles, and the phylogeny of this GC rich intergenic region is statistically different from those of the flanking genic regions of RPO41 and BSP2, suggesting sequence exchange and homogenization of this GC rich intergenic region in both lineages; and 5) statistically significant gene conversion tracts were detected in serotype A alleles across the GC rich intergenic region. These observed patterns could be explained by a process of ongoing gene conversion at the GC rich intergenic region within each lineage to prevent divergence between the two mating types, while allowing polymorphisms to accumulate independently between the two serotypes.
We hypothesized that if gene conversion at the GC rich intergenic region occurs at a frequency high enough to produce the population structure observed among the natural isolates, we should be able to detect it in the meiotic progeny generated from laboratory crosses. To test this hypothesis, we crossed two fertile serotype D strains of opposite mating type, JEC169 (MATa) and S13 (MATα) and collected 260 recombinant progeny (i.e. the progeny whose phenotype with respect to auxotrophic mutations differed with both parental strains) by screening progeny for auxotrophic mutations present in the two parental strains (see Materials and Methods). We then PCR amplified and sequenced from each of these meiotic progeny the same region analyzed for the natural isolates and looked for gene conversion events that might have occurred around the intergenic region during meiosis.
Among these 260 recombinant progeny recovered, 259 of them had the recombinant genotype ADE2 ura5, while the other one was ade2 URA5. Additionally five of the recombinant progeny were filamentous when grown on YPD solid medium. Further analysis by FACS showed that these five progeny were diploid (data not shown). Thus, these five progeny were likely diploid fusion products of the two parental strains that underwent loss of heterozygosity at the auxotrophic markers (rather than haploid meiotic progeny), and consequently they were excluded from the following analyses.
The remaining 255 recombinant progeny grew as yeast (i.e. no filamentation) on YPD solid medium and our analyses indicated that they were haploid. For chromosome 4, where MAT is located, only the a or α allele was present, but not both. Mating assays by backcrossing each of these 255 progeny to the two parental strains confirmed that the majority are fertile. Specifically, 92 (36.1%) progeny typed as MATa (i.e. successful mating with S13 but not JEC169); 139 (54.5%) progeny typed as MATα (i.e. successful mating with JEC169 but not S13); and 24 (9.4%) progeny were sterile (i.e. no mating was observed with either parental strain).
The RPO41-BSP2 intergenic region of these 255 F1 progeny was then PCR amplified and sequenced. Based on the polymorphic sites within this region between the two parental strains (Figure 2), 109 progeny inherited the alleles from the MATa parent (strain JEC169), and 141 progeny inherited the alleles from the MATα parent (strain S13). These genotypes are in accord with the mating phenotypes of the progeny determined by mating assays (see above). However, two progeny, F1N44 and F1N251, showed evidence of gene conversion at the RPO41-BSP2 intergenic region (Figure 2). Specifically, these two isolates inherited alleles from JEC169 (MATa) at the two markers located within the intergenic region (sites 4023 and 4413, Figure 2), and both inherited alleles from S13 (MATα) at the loci located within the RPO41 and BSP2 genes. The most parsimonious explanation is that gene conversion events occurred across the GC rich intergenic region (Figure 2). In both cases, the two polymorphic sites within the intergenic regions were converted from A to G, consistent with the biased gene conversion in favor of G/C that has been reported previously [27]. Our analyses using PCR-RFLP markers located on other regions of chromosome 4 suggests these two isolates arose from independent gene conversion events, as they inherited different alleles at other markers (Table 3 and Table S2; also see below). Additionally, these two progeny still mate as MATα, suggesting the conversion of the intergenic region from MATα to MATa does not have direct effects on the mating phenotype.
Assuming DSBs occur only within the intergenic region, the maximum frequency of this gene conversion around the GC rich intergenic region could be estimated to be:In which 0.614 kb is the size of the intergenic region. Alternatively, using the distance between the two polymorphic sites flanking the intergenic region, the minimum gene conversion frequency could be estimated as:in which 2.134 kb is the distance between sites 2966 and 5100 in Figure 2.
To put the observed frequency of gene conversion around the GC rich intergenic region into perspective, we calculated meiotic recombination frequencies at other chromosomal regions by constructing a genetic linkage map using the same meiotic progeny population and markers located on the same chromosome, chromosome 4, as the MAT locus. We then compared these meiotic recombination frequencies directly with the gene conversion frequency observed involving the GC rich intergenic region.
Specifically, we identified eight PCR/PCR-RFLP markers between the two parental strains (Table 3 and Table S2) that were located between ∼100 kb and ∼540 kb away from the SXI1α gene located at one end of the MAT locus. We then used these markers to screen all of the recombinant progeny. Five of the eight markers were co-dominant PCR-RFLP markers (Figure 7A), and we did not observe heterozygosity in any of the 255 recombinant progeny at any of these five markers, further corroborating that these isolates are haploid for chromosome 4.
The majority of the F1 progeny (>90%) had 0 to 2 crossing-overs within this region (Table 3, Table 4, and Table S2), and there is extensive genetic diversity among the F1 progeny (Figure 8). The order of the markers in the linkage map is consistent with that in the physical map (Figure 7B). Our results showed that the recombination frequencies (calculated as “kb/(recombination event/100 progeny)”) between adjacent marker pairs ranged between 2.07 and 6.44, and the average recombination frequency within the chromosomal region covered by these markers (calculated as [Total Physical Distance]/[Total Genetic Distance]) was 3.11. Of the seven chromosomal regions, six of them had a recombination frequency that was considerably lower than the frequency of the gene conversion at the GC peak region (estimated to be between 0.78 and 2.72) (Figure 7). Not surprisingly, the only region that showed possibly higher recombination frequency than the gene conversion frequency was between the CND05310 and SXI1/SXI2 genes that encompass the region flanking the MAT locus, which was previously shown to exhibit an elevated recombination frequency during meiosis [13]. Although it is not possible to perform statistical analyses given the small number of gene conversion events observed, these results provide evidence that the frequency of the gene conversion at the GC peak region was at least comparable to, and likely greater than the average recombination frequency in other regions on the same chromosome during meiosis.
Meiotic recombination is thought to be repressed over the majority of the heterogametic sex chromosomes (e.g. mammalian X and Y chromosomes), as well as the MAT loci in some fungi (e.g. the MAT locus in C. neoformans). Although this ensures the integrity of these highly rearranged chromosomal regions and maintains the divergence between alleles from opposite sexes (mating types), the lack of meiotic recombination also means these regions are effectively asexual, and thus under the threat of gradual deterioration due to 1) the accumulation of deleterious mutations that can no longer be eliminated efficiently through recombination, and 2) the irreversible manner in which the deleterious mutations accumulate within these regions (i.e. Muller's ratchet effect) [36], [37].
In the current study, we showed that within the MAT locus there is a GC rich intergenic region between the RPO41 and BSP2 genes that exists in all of the lineages of C. neoformans and C. gattii (Figure 1). The nucleotide composition of this GC rich region is diverged among different lineages while relatively conserved within each lineage. These observed population genetics patterns reflect the descent from a common ancestor of this GC rich intergenic region in all of the C. neoformans and C. gattii lineages, and the observed divergence among these lineages is the result of independent accumulation of mutations within this intergenic region in different lineages after their split from a common ancestor.
Recombination is thought to be repressed within the MAT locus of C. neoformans, due to sequence divergence and chromosomal rearrangements existing within the MAT locus between the two opposite mating types. If this is the case, the phylogenies of the three sections (the two flanking genes and the intergenic region) should have consistent topologies. In addition, the alleles from opposite mating types should exhibit independent accumulation of mutations, and thus their phylogeny should have a mating type specific topology; that is, alleles cluster together based on the mating types of the strains in which they reside. Furthermore, because no signal of positive selection was detected in the two flanking genic regions, the intergenic region should have accumulated more polymorphisms than regions corresponding to the two flanking genes, in which possible mechanisms such as purifying selection could slow down the accumulation of mutations and maintain sequence identity. However, this is in contrast to the results from our analyses. Specifically, the topologies of the two flanking genes were statistically different from those of the intergenic region when serotype A or serotype D alleles were analyzed. Additionally, among the serotype D alleles, the intergenic GC rich region showed a comparable level of polymorphism when compared to the two flanking genic regions (RPO41 and BSP2), and a lower level of polymorphism when compared to other genes and intergenic regions that are also located within the MAT locus (Table 2). Furthermore, we found evidence of allele sharing between strains of opposite mating types in both the serotype A and D populations (Figure 4, Figure 5), as well as gene conversion tracts around the GC rich intergenic region among serotype A strains (Figure 6). Thus, our results instead support a model in which among serotype A and D C. neoformans strains, respectively, there is still ongoing gene flow between the two mating types at the GC rich intergenic region within the MAT locus.
This hypothesis is supported by observations from the laboratory sexual cross. By collecting and analyzing a large number of meiotic progeny from a laboratory cross, we found that recombination, in the form of gene conversion, occurred around the GC rich intergenic region during sexual reproduction in serotype D C. neoformans. Interestingly, in both of the gene conversion events identified among meiotic progeny, the direction of the gene conversion was A→G, consistent with previous studies indicating that gene conversion at recombination hot spots shows a bias favoring G/C over A/T [25], [27], [38]. However, it could also be that this uni-directional gene conversion is actually favoring MATa alleles over MATα alleles. This could be investigated by analyzing meiotic progeny of a laboratory cross, in which the MATa parent has A or T, while the MATα parent has G or C alleles at the polymorphic sites within the intergenic GC rich region.
The frequency of this gene conversion event was estimated to be between 0.78 and 2.72 kb/(events/100 progeny). This is at least comparable to the meiotic recombination frequencies observed in other regions of chromosome 4 (Figure 7). The genome wide meiotic recombination of serotype D C. neoformans has been previously estimated to be ∼13.2 kb/cM [39], which is considerably lower than the meiotic recombination frequency that we observed at the other regions of chromosome 4 (ranged between 2.07 and 6.44 kb/cM). However, in the study by Marra et al. [39], several of the chromosomes were each composed of multiple linkage groups, suggesting an underestimation of the genetic distances across the genome (those recombination events between the linkage groups), and thus an overestimation of the genomic average recombination frequency. Unfortunately, chromosome 4 was one of the chromosomes that were composed of two separate linkage groups, preventing us from comparing directly the results from the two studies. Another possibility is that chromosome 4 could experience a higher than average recombination frequency during meiosis. It has been shown that in C. neoformans during serotype A and D hybridization, although most of the chromosomes experience significantly reduced recombination frequency, the regions on chromosome 4 still showed levels of recombination that are comparable to those observed in intra-variety mating [40]. It is possible that there is an intrinsic mechanism that promotes recombination along chromosome 4 during sexual reproduction, which could result from the MAT locus residing on chromosome 4. This could be investigated by genome wide analysis of the meiotic recombination frequencies at markers from different chromosomes and with different GC content profiles. Taken together, our analyses support the conclusion that gene conversion is occurring around the GC rich intergenic region at a frequency that is at least comparable to, and likely higher than the typical meiotic recombination frequencies in other genomic regions during sexual reproduction of C. neoformans.
It is not yet clear how gene conversion occurs at this intergenic GC rich region. The two flanking genes, RPO41 and BSP2, are among a group of genes that constitutes a syntenic cluster with the same orientation between MATa and MATα alleles in serotype A, whereas this gene pair is oppositely oriented between MATa and MATα alleles in serotype D (Figure 9). A previous study has hypothesized that this gene cluster represents a strata that was recruited into the MAT locus most recently during evolution [41]. Thus, it could be that the repression of recombination is less severe, and recombination can still be initiated within this region. It has been shown that recombination hot spots flank the MAT locus of C. neoformans, and these hot spots are associated with chromosomal regions with a high GC content [13]. Thus, it is possible that the factors responsible for initiating a high frequency recombination at the flanking regions of MAT locus could also recognize the GC rich intergenic region between the RPO41 and BSP2 genes within the MAT locus, and induce lesions such as double-strand breaks (DSBs) that promote recombination within this region. Additionally, DSBs could also be induced in this region because of the high susceptibility of this region to mutagenic factors due to their high GC content [28]. The DSB could then be repaired through either crossing-over or gene conversion. However, due to the rearranged chromosomal locations within the MAT loci, as well as the opposite orientations (in serotype D) of the MATa and MATα alleles, typical crossing-over within the GC rich intergenic region would result in abnormal chromosomes, such as dicentric or acentric chromosomes, and/or chromosomes with duplications and deletions of a variety of essential genes or genes involved in mating and meiosis (Figure 9). As a consequence, the progeny inheriting these abnormal chromosomes would likely be inviable. On the other hand, DSBs repaired through gene conversion do not produce abnormal chromosomes, thus resulting in the bias that only gene conversion events can be recovered from the meiotic recombination events that had actually occurred at this GC rich intergenic region during sexual reproduction. If this is the case, our estimation of the recombination frequency at this GC rich intergenic region would be an underestimation of the actual recombination frequency (including both nonviable cross-overs and viable gene conversions) within this region.
It is also possible that recombination, in the form of gene conversion, could actually be initiated and carried out actively during meiosis within the GC rich intergenic region, and maybe even in other regions of the MAT locus of C. neoformans. One of the best studied examples of active initiation of gene conversion during mitosis is mating type switching in the yeast S. cerevisiae through gene conversion initiated by the HO endonuclease, which enables haploid Saccharomyces cells to have the potential to change mating type as often as every mitotic generation [42], [43]. Active initiation of gene conversion within the C. neoformans MAT locus could have two consequences. First, it might help align the opposite mating type alleles during meiosis. The MAT locus of C. neoformans is unusually large (>100 kb), and is highly rearranged between alleles of opposite mating type. Although it has been widely accepted that meiosis in almost all species requires at least one crossover event per chromosome, the highly diverged and rearranged nature between alleles of opposite mating types could still pose a problem for the proper alignment between opposite alleles during meiosis. Thus, a recombination event within the MAT locus could help in generating the tension required between homologous chromosomes for their proper segregation during meiosis I. Second, full repression of recombination within the MAT locus of C. neoformans would render this region asexual, and thus subject the MAT locus to gradual deterioration due to the irreversible accumulation of mutations within the MAT (i.e. Muller's ratchet [36]). It has been shown that five of the 20 genes within the MAT locus are essential (including the RPO41 gene) [41], and although mechanisms such as purifying selection could act to prevent mutations from accumulating within these genes, the maintenance of sequence identity of these essential genes could nonetheless be facilitated by gene conversion events occurring in these regions. This is similar to situations where gene conversion that is biased against new mutations has been proposed to slow the observed mutation rate in plants and bacteria [44], [45]. This is also consistent with recent simulation-based studies that suggest that even low levels of gene conversion are sufficient to maintain the sequence integrity of the genes located on the human Y chromosome [46], [47]. Alternatively, this GC rich intergenic region may represent an ancient recombination hot spot that was adjacent to the essential gene, RPO41, before it was recruited into the MAT locus, as demonstrated by a study showing that purging of deleterious mutations in essential genes may be an important factor driving meiotic crossover [48].
Recombination, including gene conversion, has been reported to occur at chromosomal regions within which recombination has been thought to be repressed, such as centromeric regions [15]. Recent studies provide further empirical evidence that recombination is more universal, and is occurring at considerably higher frequencies within these “non-recombining” regions. It has been shown that in mouse crossing-over within the PAR region located on the otherwise non-recombining sex chromosomes is critical for male meiosis [29], [30]. Additionally, gene conversion has been discovered to be widespread within the centromeric regions of maize [14]. Furthermore, occasional X-Y recombination, as well as gene conversion, has been found to be the most plausible explanation for the observed sex-chromosome homomorphy in European tree frogs [49]. These recombination events play important roles in shaping the evolutionary trajectories of the chromosomal regions that are involved. A recent study by Sloan et al proposed that changes in recombination frequency (including gene conversion) are a central force driving the evolution of mitochondrial genome structure [50]. It is possible that many recombination events occurring in these supposed “non-recombining” regions might have gone unnoticed, partly due to the fact that in many cases, recombination might have taken the form of gene conversion, which can be detected only when sequences involved are non-identical, and thus is less likely to be recorded than crossing-over. Our study provides further evidence that, indeed, some presumed recombination “cold spots” are actually not that cold, after all.
The strains used in this study are listed in Table 1, and their VN types have been determined in previous studies [51], [52], [53] Strain JEC169 is an ade2 ura5 auxotrophic isolate derived from JEC20 [54]. The ADE2 and URA5 genes are located on chromosomes 5 and 7, respectively. All of the other strains are clinical or environmental isolates. All of the strains were grown and maintained on YPD agar plates unless mentioned otherwise.
A cross between strains JEC169 (MATa ade2 ura5) and S13 (MATα ADE2 URA5) was conducted on V8 agar medium plates as previously described [55]. After 2 weeks of incubation at room temperature in the dark, the plate was inspected by light microscopy. The area outside the edge of the mating mixture containing abundant basidiospores was excised, and suspended in water. The suspension was then diluted and spread onto YPD agar plates. After two days of incubation at 30°C, the colonies that appeared on the YPD plates were transferred onto testing plates, SD-ade and SD-ura [56], to confirm their genotypes at the auxotrophic markers. The colonies that only grew on one of the two testing plates were considered to be recombinant, and were then stored in 35% glycerol at −80°C for the following studies.
The mating type of each recombinant progeny was determined by backcrossing the progeny with the two parental strains. The mating assay was performed in the same way as the laboratory cross described above. Mating was determined to be successful if hyphae, basidia, and basidiospore chains were observed after 2 weeks of incubation.
DNA isolation and PCR amplification were carried out as previously described [57]. Primers used in this study are listed in Table S1. For serotype AD strains, different primer pairs were used to amplify the serotype A and serotype D specific alleles of the RPO41-BSP2 region, respectively (Table S1). All of the PCR reactions were carried out using an annealing temperature of 60°C. Restriction enzyme digestions were performed according to the manufacturer's instructions (New England Laboratory Casework Co., Inc.). Sequencing reactions were conducted at the Genome Sequencing & Analysis Core Facility at the Duke Institute for Genome Sciences & Policy.
Sequence alignments were carried out using ClustalX 2.1 [58]. The aligned sequences were then curated manually using the software MacClade 4.06 [59]. MEGA 5 [60] was used for the phylogeny constructions and population genetics analyses. Bootstrap support for the phylogenetic trees was determined by performing 1000 bootstraps. Kishino-Hasegawa test of topologies was carried using the SHTest implemented in the software PAUP 4.0 [61]. Gene conversion events were detected using program GENECONV v.1.81 following the software's instruction [62]. Significant gene conversion tracts were identified if the Global-P value was smaller than 0.05 based on permutation.
A linkage map was constructed using the software program MapMaker [63]. The chromosomal location of each marker was estimated as the position of the midpoint of the PCR product of that marker in the sequence of strain JEC21 chromosome 4 (GenBank: AE017344).
|
10.1371/journal.pntd.0002629 | Mass Vaccination with a New, Less Expensive Oral Cholera Vaccine Using Public Health Infrastructure in India: The Odisha Model | The substantial morbidity and mortality associated with recent cholera outbreaks in Haiti and Zimbabwe, as well as with cholera endemicity in countries throughout Asia and Africa, make a compelling case for supplementary cholera control measures in addition to existing interventions. Clinical trials conducted in Kolkata, India, have led to World Health Organization (WHO)-prequalification of Shanchol, an oral cholera vaccine (OCV) with a demonstrated 65% efficacy at 5 years post-vaccination. However, before this vaccine is widely used in endemic areas or in areas at risk of outbreaks, as recommended by the WHO, policymakers will require empirical evidence on its implementation and delivery costs in public health programs. The objective of the present report is to describe the organization, vaccine coverage, and delivery costs of mass vaccination with a new, less expensive OCV (Shanchol) using existing public health infrastructure in Odisha, India, as a model.
All healthy, non-pregnant residents aged 1 year and above residing in selected villages of the Satyabadi block (Puri district, Odisha, India) were invited to participate in a mass vaccination campaign using two doses of OCV. Prior to the campaign, a de jure census, micro-planning for vaccination and social mobilization activities were implemented. Vaccine coverage for each dose was ascertained as a percentage of the censused population. The direct vaccine delivery costs were estimated by reviewing project expenditure records and by interviewing key personnel.
The mass vaccination was conducted during May and June, 2011, in two phases. In each phase, two vaccine doses were given 14 days apart. Sixty-two vaccination booths, staffed by 395 health workers/volunteers, were established in the community. For the censused population, 31,552 persons (61% of the target population) received the first dose and 23,751 (46%) of these completed their second dose, with a drop-out rate of 25% between the two doses. Higher coverage was observed among females and among 6–17 year-olds. Vaccine cost at market price (about US$1.85/dose) was the costliest item. The vaccine delivery cost was $0.49 per dose or $1.13 per fully vaccinated person.
This is the first undertaken project to collect empirical evidence on the use of Shanchol within a mass vaccination campaign using existing public health program resources. Our findings suggest that mass vaccination is feasible but requires detailed micro-planning. The vaccine and delivery cost is affordable for resource poor countries. Given that the vaccine is now WHO pre-qualified, evidence from this study should encourage oral cholera vaccine use in countries where cholera remains a public health problem.
| Cholera – an acute life-threatening diarrheal illness – continues to disrupt public health in resource poor countries. The devastating outbreaks in Haiti and Zimbabwe – to name just two of many occurrences – calls for the use of available oral cholera vaccines as an additional tool in the arsenal of cholera control measures. An oral cholera vaccine (Shanchol) has been licensed in India since 2009; however, there has only been limited use of this vaccine in government public health programs. A vaccination campaign using 2 doses of Shanchol was conducted in Odisha, India, during May and June, 2011, where 31,552 persons (61% of the target population) received the first dose and 23,751 of them completed their second dose. The vaccine delivery cost was $0.49 per dose. Through our findings and experience, we discuss the organization of the cholera vaccination campaign in Odisha, the challenges met for conducting the campaign and the strategies designed to overcome those challenges, and the delivery costs incurred in the use of this vaccine, the first of its kind, in a public health setting. We believe that evidence from this study is of significant interest and use to policymakers from countries where cholera remains a public health problem.
| Cholera continues to pose a public health threat in resource- poor countries. Estimates suggest that 1.4 billion people are at risk for cholera, with 2.8 million cases and 91,000 deaths occurring annually in cholera-endemic countries worldwide [1]. The devastating and prolonged outbreaks of cholera in Haiti (with 682,475 cases and 8,328 deaths as of October 9, 2013), and in Zimbabwe (with >98,000 cases and 4,000 deaths as of July 2009) [2], [3] demand the use of cholera vaccine as an additional tool in the arsenal of cholera control measures. Given the potential of cholera outbreaks to disrupt health systems, the World Health Organization (WHO) recommends that available oral cholera vaccines (OCVs) be used in conjunction with other preventive and control strategies in areas where the disease is endemic and in areas at risk for outbreaks [3].
Until recently, Dukoral - a monovalent, whole cell killed vaccine with recombinant B sub unit cholera toxin (WC/rBS) had been the only WHO-prequalified OCV available for use. However, due to its relatively high cost (about US$ 5.3/dose for public sector), the use of Dukoral has been primarily limited to travelers from developed and higher income countries. A bivalent, killed, whole-cell OCV that was reformulated by the International Vaccine Institute (IVI) was licensed in India in 2009 based on results from a phase III trial in Kolkata, India. This vaccine, called Shanchol, is safe and confers 65% protective efficacy after 5 years post-vaccination, as measured by the reduction in the number of culture-confirmed cholera cases [4]. Shortly after its licensure, recommendations were made at a national-level policymakers' meeting in Delhi, India [5] to conduct pilot vaccine introductions in endemic areas such as one in Orissa (now Odisha) in India, distributing the two-dose OCV by utilizing the existing public health infrastructure. It is worth to mention that about 514 million people are at risk for cholera, with 834,000 cases and 25,000 deaths occurring annually in India [1].
Odisha, which is adjacent to the Bay of Bengal, is one of the most natural disaster-prone states in India [6]. It is severely affected by seasonal floods and droughts, creating conditions that facilitate the spread of cholera. Almost every year, from May to November, coastal areas in Odisha experience cyclones and floods. During this time, outbreaks of diarrheal illness often due to cholera occur [7]–[10]. In a ten year review of reported and published cholera cases in India, Odisha had the highest number of affected individuals in cholera outbreaks and had reported cholera in seven out of ten years [11]. A three-year diarrheal disease surveillance study (2004–2006) conducted by the Regional Medical Research Center (RMRC), in Bhubaneswar, Odisha, which involved taking stool specimens or rectal swabs from admitted cases on a weekly basis at three hospitals, found that, among a total of 1,551 stool and rectal collected swab samples, up to 17.3% tested positive for cholera [12]. A large outbreak in tribal districts (Koraput, Kalahandi, and Rayagada) of Odisha between August and September, 2007, was caused by a new hybrid strain which is believed to cause more severe disease [8].
While Shanchol is licensed in India, there has only been limited use of the vaccine and no documentation on how the vaccine would be deployed using government public health resources. The primary objective of this study was to evaluate the feasibility, acceptability and costs of using a less expensive oral cholera vaccine delivered through the government's public health infrastructure. We describe the organization of the vaccination campaign in Odisha, the challenges met for conducting the campaign, and the strategies designed to overcome those challenges. We also present vaccine coverage by age groups and sex, and the delivery costs incurred in the use of this vaccine, the first of its kind, in a public health setting.
The study protocol was approved by all of the following: the Health and Family Welfare Department, Government of Odisha; the Human Ethical Committee of the Regional Medical Research Center (RMRC) in Bhubaneswar, Odisha; the Health Ministry Screening Committee, Government of India; and the Institutional Review Board of the International Vaccine Institute in Seoul, Korea. This study was registered as number NCT01365442 with clinicaltrials.gov. Informed consent was obtained both at the community level through meetings with community leaders, and at the individual level through verbal informed consent just before vaccination.
The state of Odisha in India, with a population of about 37 million, is composed of 30 districts where each district is divided into 3–26 blocks [13]. In the public health system, blocks are further sub-divided into sub-centers (the lowest public health unit) of various sizes. Each sub-center is supported by a midwife nurse. In the villages within each sub-center, public health activities are also supported by volunteers called ASHA (Accredited Social Health Activist) and AWW (Anganwadi workers).
The Directorate of Health Services (DHS), in consultation with RMRC, suggested conducting a mass vaccination in Satyabadi block of Puri District because DHS data from 2005 to 2007 suggested that this block had the highest number of severe diarrhea cases, presumably due to cholera. Diarrhea cases increase during the monsoon season (usually July to September) every year in the study area. We therefore decided to complete the vaccination before the start of rainy season. Of the 19 sub-centers in Satyabadi block, 10 sub-centers with 145 villages and hamlets encompassing approximately 50,000 people were targeted for vaccination (Figure 1). Four supervisors, thirteen midwives, forty-nine ASHA and sixty-seven AWW (133 health providers in total) implement immunization activities – both regular vaccination and campaigns -in the catchment area of the selected ten sub-centers.
A baseline census to determine the target population was carried out from February 9 to April 2, 2011. Trained project staff made house-to-house visits to collect demographic (e.g., age, sex) and social (e.g., marital status, educational level) information of all members in each household. In addition, data on primary occupation, access to water, sanitation, and hygiene practices for each household was also collected. A unique number was assigned to each household and all of its members. From the census database, a household identification (ID) card was generated containing information on the total number of household members, name, age, sex and marital status of each member in that particular household. Laminated ID cards were then distributed to each household by the community health volunteers (ASHA and AWW). The household members were requested to bring the ID cards at the time of vaccination.
Shanchol is a modified bivalent killed whole cell-based oral cholera vaccine given in two doses at least 14 days apart [4]. The antigens are provided in a 1.5 ml liquid formulation in a 2.5 ml glass vial. The vaccine was presented in single-dose vials contained in small cardboard boxes. During vaccination, a cardboard box was opened and the removed vial was shaken before its liquid contents were directly poured into the vaccinee's mouth. This was at times followed by a drink of clean water; no buffer was required.
A detailed micro-plan was developed in consultation with health volunteer and community leaders to identify the location of each vaccination booth. Each booth was selected after ensuring that no villager traveled more than 10 to 15 minute by foot to reach a booth. The number of vaccination days, staffing of immunization booths (their working hours and supervision structure), and transport of vaccines and ice-packs from the central storage facility to each booth, were assessed during a series of meetings with public health officials at the state, district, block, and sub-center levels.
We assessed the vaccine storage capacity and ice-pack production/storage facilities at the state, district, and block levels to identify any potential cold-chain gaps during a vaccination campaign. This assessment was done by meeting with public health officials and by making site visits before the campaign. Based on the number of booths and cold chain boxes/vaccine carriers required at each booth, we calculated the projected number of ice-pack required for each day during each phase of the campaign.
We also conducted various community mobilization activities to raise awareness about the importance of the campaign and to encourage participation. Before the campaign, meetings were organized to inform community leaders and health care providers at the state, district and block levels about cholera, the oral cholera vaccine profile and the upcoming mass vaccination activities in the area. Prior to and throughout the campaign, information was disseminated within the study area using local newspapers, posters, leaflets, banners and mobile announcements (‘miking’). In addition, a door-to-door outreach campaign was also carried out by the local health volunteers.
All healthy, non-pregnant (as ascertained by verbal screening) residents from the study area aged 1 year and above were invited to participate in the mass vaccination. A vaccination registry (vaccination record book) with pre-printed information for each participant from the baseline census database was used to record dosing status. Considering the public health implications, individuals who wished to receive the vaccine but lived outside the study area were also given an opportunity to participate in the vaccination campaign. A separate vaccination registry was maintained to record vaccination data for persons who could not be found in the vaccination record book, either because they were from outside the study area, or because they had not been accounted for during the baseline census survey. A vaccination card (different from the household ID card described earlier) was issued to each participant, whether they were from the study area or not, at the time of administration of the first dose. Each participant was requested to bring his/her vaccination card at the time of second dose administration. Similarly, persons who took a first dose in the second round, were asked to present to any of the two public health facilities in the area after 14 days to receive a second dose on two fixed dates.
The information from the vaccination record book was doubly entered into a password-protected computerized database developed using Microsoft Visual FoxPro 7.0. Vaccine coverage for each dose was ascertained as a percentage of the eligible censused population (one year and older). Drop-out rates in the second dose were calculated based on 1st dose of vaccination. Vaccine coverage was also stratified by age groups (children: 1–5, older children: 6–17, adults: 18–60, and older adults: 61+ years) and sex (male, female). Vaccine wastage rate was estimated by comparing the delivered number of doses with the vaccine coverage for the censused population.
The input cost items required for the vaccination campaign, along with respective quantities, were identified and listed onsite by a health economist; cost items related to research were excluded. Subsequently, financial receipts and records maintained at the field office were matched against the listed input items to estimate unit costs. In the case of items for which expenditure invoices were not available, the costs were collected by interviewing management and finance staff involved in the mass vaccination campaign. At the end, to confirm that all the expenses were included, financial costs collected at the field were cross-verified with the itemized expenditure reports submitted by RMRC to IVI.
The primary cost items included special activities conducted for the mass vaccination such as: vaccine price, freight and shipment, storage and transport, cold-chain maintenance and logistic support, sensitization meetings and various social mobilization activities, training of staff, incentives and travel support for vaccinators, supervisors and cold chain handlers, surveillance activities for the management of adverse events following immunization and vaccine procurement. For the cost estimation, although the vaccine was obtained at a subsidized price of US $1 per dose for this study, we have used US $1.85 per dose, which is the current market price of Shanchol for public health programs in less-developed countries. Only costs of vaccine delivered to target and non-target population and wasted vials were taken into account. Excluded cost items were: staff time spent on program planning, costs of vaccine storage equipment and utilities, and costs of unused vaccines. Similarly, rental costs for training rooms and vaccination booths were excluded because the campaign employed the existing government infrastructure. Cost of waste management was excluded as it was absorbed within existing government waste management system. Costs were presented based on the mean exchange rate between US dollars and Indian National Rupees (1 USD = 46.7 INR) and on data from the International Monetary Fund [14] as of 2011.
A total of 51,865 persons residing in 9,166 households in the study area were enumerated during the baseline census survey (Table 1). After excluding children below one year of age, 51,488 persons were defined as the targeted population for the mass vaccination. The population was predominantly Hindus (98%) with a density of 508 inhabitants per square kilometer (km2) and dispersed on small plots of land, approximately 30 km away from the sea (Indian Ocean). The majority of adults (80%) and nearly half of all children (46%) used open field defecation. Two-thirds of the populations were dependent on community tap/hand pump for drinking water. It was observed that bathing and washing clothes/utensils usually took place in ponds distributed around the community (data not collected).
A total of 77,000 doses of the vaccine (assuming 80% coverage with first dose, 15% drop out and 5% wastage) were transported from Hyderabad, the capital city of the Indian state of Andhra Pradesh, to Bhubaneswar, the capital city of the Indian state of Odisha. Transport required cold boxes by special delivery van to maintain the cold-chain. Ten single-dose glass vials of Shanchol, each contained within a small cardboard box, were packaged in an outer carton; 54 of these cartons were placed in a thermochol box (dimensions: 0.46 m, 0.38 m, 0.29 m). A seven-cubic meter space was required to store 77,000 vials. Since there were only 4 small refrigerators available (each with a volume of 0.09 cubic meter) at the Primary Healthcare Center (PHC) serving the catchment population, vaccines were stored in a ‘walk-in cooler’ at the State Drug Management Unit at Bhubaneswar. The walk-in cooler temperature was monitored and maintained between +2° to +8°C. The routine Expanded Program on Immunization (EPI) cold boxes were used for transportation of vaccines to the study area on a daily basis during the campaign. Similarly, since ice-pack production and storage facilities were very limited at the PHC level (48 ice-packs per day), we used the ‘walk-in freezer’ facility at the state level to meet the ice-packs requirement for the campaign. At the time of delivery, about 50 vials, each with a small outer cardboard box, could be placed in one routine EPI vaccine carrier. To accommodate more vials in the vaccine carrier, we removed the outer cardboard box while still in the cold room so that about 90 vials could be kept in 1 vaccine carrier. Approximately 450 and 700 ice-packs were required daily for the first and second phases, respectively.
Two phases of the vaccination campaign, each with two dosing rounds (3 vaccination days in each round), were conducted (Table 2) from May 5 to June 4, 2011. Of the 62 vaccination booths that were established in the community, 59 were located in schools and 3 were established in local clubs. We conducted the campaign in 2 phases to overcome the deficit in number of staff, cold boxes and ice-packs that needed to be at each of the vaccination booths. For example, for all 62 booths to operate in a single phase, with at least 5 workers at each booth, 310 workers would have been needed. Each booth was led by a midwife and supported by 5–6 community health workers/volunteers. A total of 260 health workers (midwives and volunteers) were provided with a one-day training session on vaccination. Training was held five times between April 29 and May 3, 2011.
Each team performed the following activities on vaccination days: screening for eligibility, obtaining verbal consent from each participant, administering vaccine, filling tally sheets and vaccination registration books, monitoring for immediate adverse events for up to 30 minutes, issuing vaccination cards, collecting remaining vaccine vials and wastes (aluminum and rubber lids, used vaccine vials) at the end of each session, and bringing waste back to the designated health facility. Used vaccine vials were destroyed by incineration while other wastes were buried at the PHC; unused vials were donated to the DHS.
Each booth was open daily from 7.00 am to 5.00 pm for three consecutive days in each round. Eight vehicles in the first phase and twelve vehicles in the second phase were used to transport staff, cold boxes with vaccines and ice-packs, and other supplies. Each vehicle was manned by one supervisor. These mobile vans (mini-centers) were also used to replenish vaccines and ice-packs during the campaign.
A total of 31,552 eligible censused persons (61% of the target population) received the first dose of vaccine and 23,751 (46%) of these completed their second dose, accounting for a 25% drop-out between the two doses. In addition, 4,446 persons who were either not captured during census or were from outside the study area, received the first dose and 2,170 of these completed the second dose. Thus, 55,303 doses of vaccine were delivered to the eligible censused population and 12% of vaccine (6,616 doses) was given to people outside the censused population. An additional 6% of vaccine (3,312 doses) was wasted. The main reasons for wastage were: broken vials, empty vials, spillage, or persons failing to swallow.
Vaccine coverage, stratified by age groups and sex, is shown in Table 3. The highest coverage rate was achieved for 6- to 17-year-olds, while adults below 60 years of age had the lowest coverage. Males had lower coverage for both first and second dose. The lowest drop-out rates were observed among females (23%) and among the 6- to 17- year-olds (21%).
The total cost of the vaccination program was US$ 149,574 or US$ 2.7 per dose delivered to the target population (Table 4). Vaccine cost at market price (US$1.85) was the largest cost item. Omitting vaccine shipment (from Hyderabad to Bhubaneswar) cost (US$0.04, not shown), the vaccine delivery cost was US $0.49 per dose, or US $1.13 per fully vaccinated person. Vaccines provided to persons outside of the census population were considered a public health good, and the delivery cost per dose was reduced to US $0.44, or US $1.04 per fully vaccinated person while accounting for them.
This was the first project to be undertaken to collect empirical evidence for mass vaccination campaign using the new, less expensive oral cholera vaccine (Shanchol) in a government-run, public health program. This was also the first opportunity to conduct a mass vaccination campaign using Odisha's public health system, which is already capable of supporting mass vaccination campaigns against other diseases, like polio. However, unlike the usual polio campaigns, in which the target population is roughly 13% (0- to 5-year-olds), the catchment population for this cholera campaign was almost the entire community. In addition, in terms of outreach comparisons, community residents were well aware of the polio vaccine and their health providers/volunteers were generally trained on its delivery, as polio campaigns are conducted regularly. In contrast, our cholera campaign had to raise awareness of a new vaccine. Our campaign also differed from polio campaigns in terms of the number of booths: whereas only 25 vaccination booths are generally established in our catchment area for polio campaigns, we established 62 booths to cover a wider population for the cholera vaccine. This had implications for additional requirements of human resources and of vaccine carriers/ice-packs.
In terms of cold-chain infrastructure comparisons, the polio vaccine volume is small (2 drops or 0.1 mL) in multi-dose vials compared to the 1.5 mL single-dose vials used for Shanchol; therefore, the space required for cholera vaccine storage and transportation is greater. Further, conducting a mass vaccination campaign for an entire community poses considerable challenges for the public health infrastructure, particularly human resources and cold chain capacity at the peripheral health facilities. For example, there were only 129 midwives and volunteers in the catchment area, while we needed a total of 161 booth members for the first phase and 234 for the second phase of the campaign. Fortunately, since we conducted the mass vaccination in a catchment area of half of a primary health care (PHC) facility, additional human resources could be mobilized from the other half of the same PHC and from nearby PHCs. Typically, for other mass vaccination campaigns, like polio campaigns, two to three persons at each vaccination booth would accomplish the entire set of activities in Odisha, India, and in most of other developing country settings. However, this project required additional manpower to obtain verbal informed consent, perform registration, and issue vaccination cards, all activities which are not typically done during vaccination campaigns conducted in a public health setting. This requirement for additional manpower also increased the delivery costs, a phenomenon which was observed in an earlier study [15].
The limited capacity to produce and store ice-packs at the peripheral health care facilities was overcome by the availability of cold chain facilities located at the central level, which were within a two hours' drive from the study area. Further, receiving the vaccines in two lots also made it easier to store them. During administration, one has to remove the aluminum lid first, shake the vial well, and then remove the rubber lid before pouring the vaccine into the participant's mouth. Simpler packaging of the vial, in plastic tubes, for example, and production of multi-dose vials without outer cardboard box would greatly facilitate its storage and delivery and should be considered for future production and use of this vaccine.
According to the vaccine package insert, the temperature must be maintained between +2° to +8°C; thus, storage of vaccines and their distribution with cold chain maintenance posed substantial challenges under hot and humid weather conditions, with temperatures reaching up to 42°C on vaccination days. Since this is an inactivated vaccine, recommendations from the manufacturer or from the regulatory authority on waiver of strict cold chain requirements (e.g., during vaccine delivery at a minimum) would greatly facilitate its use in future campaigns. A similar recommendation for the use of killed whole cell/recombinant B-subunit (WC/rBS) cholera vaccine, Dukoral, was made based on vaccine delivery experiences in Indonesia in 2005 [16]. The recent approval of a meningitis vaccine (MenAfrivac) to be stored and transported without ice-packs/refrigeration for four days could also facilitate similar recommendation for Shanchol in the future [17].
Additional vaccine carriers and larger-sized cold boxes (10 to 20 liters in capacity) were mobilized from nearby PHCs and central level institutions. The use of a phased approach during the implementation of the campaign greatly helped to overcome the challenges due to human resources and cold chain infrastructure and should be considered in future large-scale vaccination campaigns. The absence of buffer preparation and co-administration with this new vaccine also minimized associated logistic challenges that were observed in previous studies with WC/rBS [15], [16], [18].
Our findings suggest that mass vaccination using two doses of OCV, where almost the entire community is targeted, is doable but requires detailed micro-planning, additional human resources, modifications in cold-chain capacity and modifications in the number and the location of vaccination booths. In our case, we showed that it was feasible to install vaccine booths within an average distance of 267–283 meters from the households. According to our booth location plan, 1,087 meters was the maximum possible distance a person had to walk to reach a booth. This micro planning was made in an effort to maximize coverage.
Nonetheless, the level of vaccine coverage achieved during this campaign (46–61%) was lower compared to the coverage observed (59–83%) during previous studies with other OCVs [16], [18], [19]. The lower vaccine coverage reported here may be due to several reasons. Due to ongoing routine weekly public health activities (e.g., immunization days, nutrition days etc.) conducted by the PHC, we could not implement the campaign beyond three days for each round. In comparison, vaccination campaigns with other OCVs were usually held for about10 days [16], [18], [19]. In addition, we speculate that a low turn-out in the hot and humid weather (with temperatures reaching up to 42°C) was a factor associated with the relatively lower coverage in this study. Further, since adults (18–60 years) were observed to have the least coverage, innovative strategies (e.g., operating vaccination booths until late in the evening or early in the morning) to catch this group should be considered for future campaigns. This was noted as well for the previous studies concerning the use of WC/rBS [15]. At the end of each vaccination day, unused vials with cold chain and wastages from vaccination needed to brought back to the PHC; therefore, we had to stop vaccination by 5 pm to allow for unused vial collection and transportation. Invariably, there were people still queuing for vaccination at the end of the day, but booths needed to be closed. They were requested to visit again the following day.
We also observed that some participants did not like the ‘taste and smell’ of this vaccine, which they described as ‘fishy’ or ‘rotten egg’ in nature. Vaccination days were from Thursdays to Saturdays in each round of each phase, and since Thursdays are ‘complete vegetarian days’ for Hindus in the study area, the vaccine's taste and smell could potentially influence lower coverage in this particular community. However, further studies are needed to substantiate or refute this taste-influence hypothesis in this community. In addition, in spite of raising cholera vaccine awareness during community outreach programs, the perception that “vaccines are only for children” remained prevalent in the communities (data not collected); this could explain, at least in part, the lower coverage observed among adults compared to children. Since the oral cholera vaccine was being introduced for the first time, it was regarded as a ‘new’ vaccine to both the providers and the community residents and this perception could have also influenced the lower coverage. The relatively high drop-out rates could have been due to adverse weather conditions and unpleasant taste/smell of vaccine, which merit further investigation. We also observed that people who did not take any dose of vaccine tended to be older, were male and belonged to high socio-economic status of the community (data not shown).
The public sector vaccine delivery cost of $0.49 per dose excluding vaccine freight and shipment was similar to the vaccine delivery cost estimate for a campaign-based delivery from the “WHO comprehensive Multi-Year Plans Guidelines for EPI Vaccines in 2006” [19] and is within the range of estimates from previous studies using other OCV [15], [18], [20]. The delivery cost was higher than the $0.09 and $0.23 per dose estimates reported from campaigns in Vietnam in 1998 [20] and in a refugee camp in Uganda in 1997 [18]. When the delivery costs are readjusted to 2011 price based on country inflation consumer prices [21], the delivery cost in Odisha is still higher than that in Vietnam ($0.23) but lower than that in Uganda ($0.56). Further, the delivery cost estimate was lower than the $0.94 estimated in 2003/2004 ($1.97 in 2011 prices) per dose in Beira, Mozambique, where the very high vaccine transportation cost was a factor [15].
A cholera vaccination economic model using incidence estimates for high-risk populations in India from a recent global burden study [1] estimated a cost effectiveness ratio of $785 per DALY averted for programs targeted to ages 1 year and above in South East Asia Region, where cholera vaccine coverage is assumed at 80% and 50% of measles vaccine coverage for populations 1–14 and 15+ years, respectively [22]. Applying these estimates, cholera vaccination would be considered “very cost effective” based on WHO criteria [23]. Without cholera incidence data, it is not possible to estimate the Odisha-specific cost-effectiveness of vaccination. However, we observed that the vaccine coverage for two-dose recipients was 62% in 1–14 years and 41% in 15+ years which are, respectively, more than 80% and 50% of measles vaccine coverage (76%) in Odisha in 2010–2011 [24].
Since we conducted the vaccination campaign in a cholera-endemic setting, we believe that our methods and our findings provide a model that may be extrapolated to other endemic settings in India – a country which accounts for an estimated 30% of global cholera burden - and beyond. Given that the vaccine is now WHO pre-qualified, evidence from this study is of significant interest and use to policymakers from countries where cholera remains a major public health problem. The vaccine could be a viable, affordable, and effective tool in public health programs to control cholera in these countries.
|
10.1371/journal.ppat.1005191 | Myeloid Cell Arg1 Inhibits Control of Arthritogenic Alphavirus Infection by Suppressing Antiviral T Cells | Arthritogenic alphaviruses, including Ross River virus (RRV) and chikungunya virus (CHIKV), are responsible for explosive epidemics involving millions of cases. These mosquito-transmitted viruses cause inflammation and injury in skeletal muscle and joint tissues that results in debilitating pain. We previously showed that arginase 1 (Arg1) was highly expressed in myeloid cells in the infected and inflamed musculoskeletal tissues of RRV- and CHIKV-infected mice, and specific deletion of Arg1 from myeloid cells resulted in enhanced viral control. Here, we show that Arg1, along with other genes associated with suppressive myeloid cells, is induced in PBMCs isolated from CHIKV-infected patients during the acute phase as well as the chronic phase, and that high Arg1 expression levels were associated with high viral loads and disease severity. Depletion of both CD4 and CD8 T cells from RRV-infected Arg1-deficient mice restored viral loads to levels detected in T cell-depleted wild-type mice. Moreover, Arg1-expressing myeloid cells inhibited virus-specific T cells in the inflamed and infected musculoskeletal tissues, but not lymphoid tissues, following RRV infection in mice, including suppression of interferon-γ and CD69 expression. Collectively, these data enhance our understanding of the immune response following arthritogenic alphavirus infection and suggest that immunosuppressive myeloid cells may contribute to the duration or severity of these debilitating infections.
| Mosquito-transmitted chikungunya virus (CHIKV), Ross River virus (RRV), and related alphaviruses cause epidemics involving millions of persons, such as on-going CHIKV outbreaks in the Caribbean and Central and South America. Infection with these viruses results in severe pain due to inflammation of musculoskeletal tissues that can persist for months and even years. There are no specific therapeutics or licensed vaccines for these viruses. Suppressive myeloid cells have been shown to inhibit anti-pathogen immune responses, including T cell responses, which can promote chronic disease. We showed previously that a gene associated with suppressive myeloid cells, arginase 1 (Arg1), was induced in musculoskeletal tissues and macrophages of mice infected with RRV or CHIKV, and mice that lacked Arg1 expression in myeloid cells had reduced viral loads at late times post-infection. Here, we demonstrate that Arg1 is induced in PBMCs isolated from CHIKV-infected patients, and Arg1 expression is associated with viral loads. Moreover, we found that Arg1-expressing myeloid cells inhibit the activation and function of antiviral T cells in RRV-infected mice. These studies underscore the role of suppressive myeloid cells in modulating the T cell response to arthritogenic alphaviruses and provide a therapeutic target to enhance viral clearance and potentially limit chronic disease.
| Arthritogenic alphaviruses, including chikungunya virus (CHIKV) and Ross River virus (RRV), are re-emerging, mosquito-transmitted alphaviruses that cause both endemic and explosive epidemics of debilitating musculoskeletal inflammatory disease [1]. CHIKV has caused outbreaks of unprecedented scale involving millions of persons in the Indian Ocean Islands [2], India [3], Southeast Asia [4,5,6] and Europe [7]. Most recently, CHIKV has emerged in the Western Hemisphere where ongoing epidemics on multiple islands in the Caribbean as well as in Central and South America have resulted thus far in more than one million suspected cases [8,9,10]. RRV, which causes ~4,000–7,000 cases in Australia and Papua New Guinea annually, has similarly caused explosive outbreaks [11]. For example, an RRV epidemic occurred in 1979–1980 with > 60,000 cases, where RRV spread from Australia to multiple islands in the Pacific Region including Fiji, the Cook Islands, and America Samoa [12,13,14]. Currently there are no specific therapies for the treatment of alphavirus-induced rheumatological disease and no licensed vaccines.
CHIKV/RRV-induced disease is characterized by fever, intense pain and inflammation in joints, tendons, and muscles, and an impaired ability to ambulate [11]. This acute stage lasts for 1 to 2 weeks and is typically followed by convalescence. However, some disease signs and symptoms—such as joint swelling, joint stiffness, arthralgia, and tendonitis/tenosynovitis—can last for months to years, with up to 60% of patients reporting persistent rheumatological symptoms three years after initial diagnosis [15,16,17,18,19,20,21]. This chronic phase of the disease has been linked in both humans and animal models to persistent CHIKV/RRV infection in the affected musculoskeletal tissues [22,23,24,25].
Monocytes and macrophages can be activated by a variety of stimuli, resulting in a spectrum of activation phenotypes [26]. Macrophages that promote tissue repair/remodeling during wound healing and have immunoregulatory functions express arginase 1 (Arg1), an enzyme that hydrolyzes L-arginine [27]. High Arg1 expression has been associated with a variety of diseases such as chronic inflammation [28], asthma [29], and infectious diseases [30,31,32,33,34]. The expression of Arg1 by human and murine monocytes/macrophages, neutrophils, and myeloid-derived suppressor cells (MDSCs) has emerged as a major regulator of immune responses [35,36,37]. Indeed, Arg1 activity in myeloid cells impairs effective immunity against intracellular pathogens such as Mycobacterium tuberculosis and Toxoplasma gondii [38], exacerbates tumor growth by suppressing T cell function [39,40], and limits T cell-driven inflammatory tissue damage [41]. Arg1 activity has also been associated with higher viral loads and lower CD4+ T cell counts from HIV-seropositive patients [33,42] and with inhibition of CD8+ T cell responses in hepatitis B virus (HBV) and hepatitis C virus (HCV)-infected patients [34,43].
We previously showed that Arg1 was induced in the musculoskeletal inflammatory lesions and tissue-infiltrating macrophages of RRV- and CHIKV-infected mice [44]. We further showed that mice specifically deleted for Arg1 in myeloid cells had reduced viral loads, as well as improved tissue pathology, at late, but not early, times post-RRV infection, indicating that Arg1+ macrophages prevent efficient host control of RRV infection in musculoskeletal tissues [44]. We sought to expand our previous findings in the mouse model to human CHIKV infections. Here, we show that Arg1 is induced in peripheral blood mononuclear cells (PBMCs) from CHIKV-infected patients and that higher Arg1 expression levels were associated with higher viral loads and more severe disease. Using the RRV and CHIKV mouse models, we further investigated the mechanism(s) by which Arg1 regulates clearance of arthritogenic alphavirus infection. We found that Arg1-expressing myeloid cells inhibited the antiviral T cell response in RRV-infected mice, resulting in reduced expression of the antiviral cytokine interferon (IFN)-γ and modulation of T cell activation markers.
We have previously shown that Arg1 is highly induced in inflamed and infected musculoskeletal tissues and infiltrating macrophages of mice infected with RRV or CHIKV [44]. Moreover, mice specifically deleted for Arg1 in myeloid cells had reduced viral loads at late, but not early, times post-RRV infection, suggesting that Arg1 activity in macrophages inhibits efficient host control of RRV infection in musculoskeletal tissues [44]. To evaluate the role of Arg1 during CHIKV infection in patients, we analyzed the gene expression profile of PBMCs collected at four times post-illness onset (PIO) in a cohort of CHIKV-infected patients in Singapore: (1) acute phase (median, 4 days PIO); (2) early convalescent phase (median, 10 days PIO); (3) late convalescent phase (4–6 weeks PIO); and (4) chronic phase (2–3 months PIO) [45]. Compared to healthy controls, we found that Arg1 expression levels were significantly increased in circulating PBMCs during the acute phase and remained elevated into the chronic phase of the disease (Fig 1A). By segregating the cohort into high and low viral load groups (HVL and LVL, respectively), as previously defined [45], higher Arg1 expression was observed to associate with higher viral loads in these patients (Fig 1B). High viremia was also previously shown to correlate with increased disease severity during the acute phase of infection (ref. [45] and S1A Fig). By segregating on disease severity, we were able to analyze Arg1 gene expression in PBMCs isolated from patients with mild or severe clinical disease throughout the disease time course. We found that although Arg1 expression levels were equally elevated in PBMCs isolated from patients with mild or severe disease during the acute, early convalescent, and chronic phases of disease (S1B, S1C and S1E Fig), PBMCs isolated from patients with severe disease during the late convalescent phase (4–6 weeks PIO) expressed Arg1 at significantly higher levels than PBMCs isolated from patients with mild disease (P = 0.002) (S1D Fig). Due to the low number of patients that complained of persistent arthralgia 2–3 months PIO in this study (6 of 23 patients), there was no association between Arg1 expression levels in PBMCs isolated from patients with persistent join pain during the chronic phase of disease (2–3 months PIO) compared to those who had fully recovered.
In addition to comparing Arg1 expression levels to viral loads and clinical disease, we compared Arg1 gene expression in PBMCs to plasma concentrations of C-reactive protein (CRP) and to the expression of other genes in PBMCs collected during the acute phase (median, 4 days PIO) or early convalescent phase (median, 10 days PIO). Positive correlations were observed between Arg1 gene expression and the plasma concentration of CRP, IP-10, and IL-4 (Fig 1C) as well as the gene expression of type I interferons (IFNs), IL-6, IP-10, IL-10, and NADPH oxidase 1 (Nox1) in PBMCs isolated during the acute phase (Fig 1D). This suggests that these factors are induced as part of the early response to CHIKV infection. Overall, these data show that, consistent with our studies in mice, Arg1 expression is induced in PBMCs isolated from humans infected with an arthritogenic alphavirus, and Arg1 expression levels are associated with viral load and disease severity. Additionally, other genes that have been shown to have suppressive functions, such as Nox1, or to be associated with polarizing suppressive myeloid cells, such as IL-6, are also elevated in PBMCs from CHIKV-infected patients. In sum, these data suggest that suppressive myeloid cells may be induced in human CHIKV infection.
Since Arg1 is inducible in myeloid cells, we sought to test if exposure to CHIKV would induce Arg1 or other genes with immunoregulatory functions in human monocytes ex vivo. To do this we utilized the human fibrosarcoma cell line HS 633T, which has been shown to be susceptible to CHIKV infection [46]. To recapitulate the in vivo setting whereby monocytes infiltrate infected joint and muscle tissue, human monocytes isolated from PBMCs via negative selection were co-cultured with mock or CHIKV-infected HS 633T cells for 24 hpi, at which time all of the cells were harvested for gene expression analysis. The transcriptional profile of these groups was compared to mock or CHIKV-infected fibroblasts cultured in the absence of monocytes. Expression of the Arg1, IL-6, and Nox1 genes was significantly upregulated in the CHIKV-infected co-cultures (Fig 2A). These data suggest that CHIKV-infection of human fibroblasts in vitro induces a transcriptional profile in co-cultured human monocytes that has features similar to myeloid suppressor cells.
Next, we sought to determine if direct contact was required to induce Arg1 expression in monocytes. Monocytes were isolated from PBMCs and stimulated for 24 hours with supernatant from mock- or CHIKV-inoculated HS 633T cells collected at 24 hpi. Virus present in the cell culture supernatant was not inactivated or removed prior to incubation with monocytes. Arg1, TGF-β, and Nox1 transcripts were induced in the monocytes cultured in the CHIKV-infected cell supernatant compared to those cultured in culture supernatants collected from the mock-inoculated cells (Fig 2B). Following 24 h culture with fibroblast supernatant harvested at 48 hpi, IL-6 expression was significantly upregulated in the monocytes (Fig 2B). These data indicate that direct contact with infected fibroblasts is not required for Arg1 induction in monocytes; instead, soluble factor(s), which could include virus and/or other soluble factors such as cytokines, in the supernatant mediate activation of this transcriptional profile.
To investigate if the presence of live virus in cell culture supernatants from alphavirus-infected cells is required for the induction of Arg1 in myeloid cells, we performed experiments utilizing J774 murine macrophages and C2C12 myoblasts that had been differentiated into myofibers to mimic muscle cells, which are a target cell of RRV and CHIKV infection in vivo (S2 Fig). In these studies, supernatant transfer experiments were performed to determine whether RRV-infected muscle cells produce factors that induce the expression of Arg1 in macrophages. Stimulation of J774 macrophages with IL-4 (a positive control) or culture supernatants from RRV-infected differentiated C2C12 muscle cells induced Arg1 expression (representative blot in S2A Fig, quantified in S2B Fig). In contrast, stimulation of macrophages with cell culture supernatants from mock-infected differentiated C2C12 muscle cells did not induce Arg1 expression above levels detected in control cells. To determine if the virus present in these cell culture supernatants was required for Arg1 induction in macrophages, we stimulated J774 macrophages with cell supernatant that was left untreated, treated with ultraviolet light (UV) or heat to inactivate the virus, or ultracentrifuged to eliminate virus. Additionally, macrophages were treated with virus that was purified via density gradient centrifugation. The ultra-centrifugation, UV, and heat inactivation treatments of the C2C12 cell culture supernatants effectively reduced the amount of live virus (S2D Fig). Regardless of the presence of live or inactivated virus, the infected cell supernatant induced Arg1 expression in the macrophages, whereas purified virus did not (representative blot in S2C Fig, quantified in S2E Fig). These data suggest that a factor(s) present in culture supernatants other than the virus itself mediates induction of Arg1 expression.
Previously, we showed that Arg1 is expressed in infiltrating macrophages present in musculoskeletal tissues of CHIKV- or RRV-infected mice [44]. Since exposure of primary human monocytes to supernatants collected from CHIKV-infected cells induced Arg1 expression (Fig 2), we sought to further define the tissue compartments in which Arg1 was induced in CHIKV- and RRV-infected mice. The gastrocnemius muscle, circulating blood leukocytes, bone marrow, spleen, and draining popliteal lymph node were harvested from mock-, RRV-, or CHIKV-inoculated mice at 7 dpi for RT-qPCR analysis of Arg1. As demonstrated previously [44], Arg1 transcript was highly induced in inflamed muscle tissue of RRV-infected mice (245-fold increase, P = 0.002) as well as ankle joint tissue (8.2-fold increase, P = 0.06) (Fig 3A). Additionally, Arg1 was induced in circulating blood leukocytes (8.6-fold increase, P = 0.02). In contrast, Arg1 expression was not induced in lymphoid tissues such as bone marrow cells, cells from the draining popliteal LN, or spleen cells (Fig 3A). Similar results were observed in tissues collected from CHIKV-infected mice, although the highest level of Arg1 induction was observed in ankle-joint associated tissue (54-fold increase, P = 0.002) as opposed to the gastrocnemius muscle in RRV-infected mice (Fig 3B). These data indicate that, similar to our analysis of Arg1 in PBMCs from CHIKV-infected humans, Arg1 expression was elevated in circulating cells from CHIKV- and RRV-infected mice, albeit to a lesser extent.
Arg1-expressing myeloid cells have immune-suppressive activity [41,47,48]. Additionally, we have previously shown that Arg1-expressing CD11b+F4/80+ macrophages present in the inflamed musculoskeletal tissues of RRV-infected mice suppressed T cell proliferation ex vivo in a dose-dependent manner by a mechanism that was partially Arg1-dependent [44]. Furthermore, mice specifically deleted for Arg1 in myeloid cells (LysMcre;Arg1F/F mice) have significantly lower viral loads in musculoskeletal tissues, suggesting that Arg1 activity in macrophages prevents efficient host control of RRV infection in musculoskeletal tissues [44]. Based on these data, we hypothesized that tissue-infiltrating myeloid cells inhibited RRV clearance by suppression of the antiviral T cell response. To investigate this hypothesis, WT and LysMcre;Arg1F/F mice were treated with anti-CD4 and anti-CD8α antibodies, or an isotype control antibody, on days 7 and 12 pi to deplete CD4+ and CD8+ T cells. Experiments were terminated at 14 dpi, spleens were analyzed for efficiency of T cell depletion, and muscle tissue was harvested for absolute quantification of RRV genome levels. Administration of anti-CD4 and anti-CD8α antibodies depleted ≥ 98% of both CD4+ and CD8+ T cell subsets as demonstrated by staining spleens for CD3, CD4, and CD8β (Fig 4A and 4B). As with untreated mice [44], control antibody-treated LysMcre;Arg1F/F mice had significantly lower viral RNA levels compared to control antibody-treated WT mice (3.2-fold decrease, P < 0.05) (Fig 4C). The difference between control antibody-treated LysMcre;Arg1F/F mice and WT mice is not as substantial as we previously published in untreated LysMcre;Arg1F/F mice and WT mice [44]; this may be due to unanticipated effects of the isotype control antibody treatments. Interestingly, following depletion of both CD4+ and CD8+ T cells, LysMcre;Arg1F/F mice had similar viral loads to T cell-depleted WT mice (P > 0.05) (Fig 4C). Taken together, these data suggest that Arg1-expressing myeloid cells have suppressive activity and inhibit antiviral T cells.
One mechanism by which myeloid cell Arg1 has been shown to suppress T cells is by inhibiting T cell proliferation [47,49]. To determine if myeloid cell Arg1 inhibits T cell stimulation and proliferation in lymphoid tissues, T cell populations in the spleen and draining popliteal lymph node (LN) were analyzed by flow cytometry at 7, 10, and 14 days post-RRV inoculation of WT and LysMcre;Arg1F/F mice. After gating on lymphocytes, CD8+ T cells were identified by staining positive for CD3 and CD8; CD4+ T cells were identified by gating on CD3+CD8- cells followed by gating on CD3+CD4+ cells (S3A Fig). We found that WT and LysMcre;Arg1F/F mice had similar frequencies and total numbers of CD4+ and CD8+ T cells in both the spleen and draining popliteal LN at all of the time points analyzed (S3B–S3E Fig).
Since Arg1 is highly induced in inflamed musculoskeletal tissues but not lymphoid tissues of RRV-infected mice (ref. [44] and Fig 3), we hypothesized that the inhibitory effects of Arg1-expressing myeloid cells could be localized to sites of pathology in the musculoskeletal tissues. Thus, we next assessed the presence of CD4+ and CD8+ T cells in inflamed muscle tissue of WT versus LysMcre;Arg1F/F mice at 7, 10, and 14 dpi. CD4+ and CD8+ T cells were identified as described above (Fig 5A). We again found similar frequency and number of total CD4+ and CD8+ T cells in the quadriceps muscle of both WT and LysMcre;Arg1F/F mice at all time points analyzed (Fig 5B and 5C). In total, these data suggest that although myeloid cell Arg1 inhibits the antiviral T cell response, this is not through the suppression of total T cell numbers at immune inductive sites or the sites of infection and inflammation.
Arg1-mediated suppression has been shown to inhibit antigen-specific T cell responses [50]. For example, in Leishmania-infected mice, parasite-specific T cell proliferation was suppressed at the site of pathology, where high arginase activity was detected, but not in the draining LN, where arginase activity was low or undetectable [48]. Thus, we next quantified virus-specific T cell responses in both lymphoid tissues and inflamed muscle tissue of RRV-infected mice. For these experiments we utilized a recombinant Ross River virus (“RRV-LCMV”) that was recently developed in our laboratory [51] and encodes a CD8 (gp33) TCR immunodominant epitope of LCMV, enabling the identification and quantification of virus-specific CD8+ T cells in tissues via tetramer staining (see Materials and Methods for description of virus generation). WT and LysMcre;Arg1F/F mice were inoculated with RRV-LCMV and gp33+CD8+ T cells were analyzed by flow cytometry at 10 dpi, a time point associated with extensive skeletal muscle inflammation and damage as well as high Arg1 expression [44]. To confirm the specificity of the gp33 tetramer staining, leukocytes from the spleens of mock- and WT RRV-infected mice were also stained for gp33-specific T cells. Similar to the analysis of bulk CD8+ T cells in WT RRV-infected mice (S3 Fig), we found that the total number of CD8+ T cells in the spleen of RRV-LCMV-infected LysMcre;Arg1F/F mice was mildly elevated in comparison to WT mice (Fig 6C). However, the total number of virus antigen-specific gp33+CD8+ T cells in the spleen of RRV-LCMV-infected LysMcre;Arg1F/F mice was similar to WT mice (Fig 6E). Since Arg1 is highly induced in the inflamed muscle tissue of RRV-infected mice (Fig 3 and ref. [44]), we next analyzed the bulk and virus-specific CD8 T cell response in muscle tissue of WT and LysMcre;Arg1F/F mice. The gp33+ gate was set such that ≤ 0.1% of CD8+ T cells from the muscle tissue of a WT RRV-inoculated mouse stained positive with the gp33 tetramer (Fig 6F). Interestingly, RRV-LCMV-infected LysMcre;Arg1F/F mice had significantly increased frequency and number (2.5-fold) of CD8+ T cells in the muscle tissue compared to RRV-LCMV-infected WT mice (Fig 6G and 6H). This suggests that loss of myeloid cell Arg1 may have a more profound effect on CD8 T cells in the presence of a RRV expressing the CD8 T cell immunodominant epitope from LCMV, gp33. Indeed, there was a slight but non-significant difference in the frequency of gp33-specific CD8+ T cells (Fig 6I); moreover, there was a significantly greater number (3.2-fold) of gp33+CD8+ T cells detected in the inflamed quadriceps muscle tissue of RRV-LCMV-infected LysMcre;Arg1F/F mice than WT mice (Fig 6J). These data are consistent with the data presented in Fig 3, which shows elevated expression of Arg1 in skeletal muscle tissue but not lymphoid tissue. In addition, these data suggest that loss of myeloid cell Arg1 increased virus-specific T cell trafficking, proliferation, and/or avoidance of deletion at the site of inflammation.
In addition to inhibiting T cell proliferation and trafficking, Arg1-expressing myeloid cells can also inhibit T cell activation [36,37]. We next sought to investigate the activation of virus-specific CD8+ T cell responses in WT and LysMcre;Arg1F/F mice. The three activation markers that we evaluated are known to be upregulated on antigen-experienced T cells: CD44, CD11a, and CD69 [52]. Of the CD44+gp33+CD8+ T cells in WT and LysMcre;Arg1F/F mice, essentially all were also CD11a+ (representative histogram shown in Fig 7A; quantified in Fig 7B). However, significantly more CD44+gp33+CD8+ T cells in LysMcre;Arg1F/F mice stained for CD69 than those cells in WT mice (representative histogram shown in Fig 7A; quantified in Fig 7B), suggesting specific modulation of this activation marker on virus-specific T cells in the absence of arginase activity. These data suggest that the inflamed musculoskeletal tissue environment alters the T cell activation phenotype in an Arg1-dependent manner.
To further explore the effects of Arg1 on T cell function, we investigated the cytokine expression levels in T cells sorted from muscle tissue of RRV-infected WT and LysMcre;Arg1F/F mice. At 10 dpi, quadriceps muscle tissue was dissected and enzymatically digested, and infiltrating leukocytes were stained for CD3, CD19, CD4, and CD8. CD4+ and CD8+ T cells were FACS-sorted by gating on CD3+CD19- cells and then gating individually on CD4+ T cells and CD8+ T cells (Fig 8A). Gene expression analysis of the sorted T cell subsets was compared to respective T cells sorted from the spleen of a mock-infected mouse. Both CD4+ and CD8+ T cell subsets isolated from LysMcre;Arg1F/F mice expressed increased levels of IFN-γ compared to T cells sorted from WT mice (Fig 8B). T cells sorted from the spleens of RRV-infected WT or LysMcre;Arg1F/F mice showed no difference in IFN-γ expression (S4 Fig). CD4+ T cells sorted from LysMcre;Arg1F/F mice also expressed elevated levels of TNF-α (Fig 8C) and IL-10 (Fig 8D) transcripts. In contrast, CD8+ T cells sorted from WT or LysMcre;Arg1F/F mice expressed similar levels of TNF-α, and IL-10 expression was not detected in this T cell subset. Additionally, IL-2 expression was not detected in either T cell subset (Fig 8E). To further confirm cytokine expression, T cells from RRV-infected WT mice isolated at 10 dpi were restimulated ex vivo with anti-CD3 and anti-CD28 antibodies followed by intracellular cytokine staining. A subset of CD4+ T cells produced both IFN-γ and IL-10, whereas CD8+ T cells produced IFN-γ but very little IL-10 (S5 Fig). These data suggest that Arg1 activity in macrophages inhibits cytokine expression, including IFN-γ, by T cells in musculoskeletal tissues, which may be one mechanism by which Arg1 influences viral loads.
Suppressive myeloid cells have been shown to mediate T cell suppression through the induction or expansion of regulatory T (Treg) cells [53], which can indirectly inhibit CD8+ T cells. Interestingly, we found that at 10 days post-RRV infection a subset of muscle-infiltrating CD4+ T cells produce IL-10 (Fig 8D and S5 Fig), a cytokine that has anti-inflammatory functions and has been shown to regulate Arg1 expression levels in myeloid cells [53]. To determine if CD4+ T cells were contributing to Arg1 induction in musculoskeletal tissues following RRV infection, we treated mice with an anti-CD4 Ab or a control Ab on day 4 pi and harvested muscle tissues on day 7 pi for analysis of CD4 T cell numbers, macrophage numbers, RRV loads, and Arg1 expression. Administration of anti-CD4 Ab depleted ≥ 95% of CD4+ T cells in the spleen and muscle tissue as demonstrated by staining for CD3 and CD4 (S6A–S6E Fig). Importantly, the depletion of CD4+ T cells did not reduce the total number of macrophages present in skeletal muscle tissue (S6A, S6F and S6G Fig). Mice that were depleted of CD4+ T cells had reduced Arg1 expression in muscle tissue at 7 dpi compared to control Ab-treated mice (146-fold increase versus 271-fold increase), however these differences were not statistically significant (S6H Fig). Consistent with viral loads in LysMcre;Arg1F/F mice at 7 dpi, mice depleted of CD4+ T cells had similar viral loads as control Ab-treated mice (S6I Fig). These data suggest that Arg1 expression levels in tissues are predominantly regulated by a CD4+ T cell-independent mechanism(s).
Other groups have shown that IFN-γ production from T cells was critical for clearance of SINV RNA from neurons [54], demonstrating a role for IFN-γ-mediated clearance of alphavirus RNA. To further investigate the role for T cell-derived IFN-γ in control of RRV infection, we adoptively transferred T cells from WT and Ifng-/- mice into Rag1-/- mice, which lack B and T cells, one day prior to RRV inoculation and analyzed RRV RNA levels at 14 dpi in muscle tissues compared to Rag1-/- mice that received media alone. T cell reconstitution was confirmed by flow cytometric analysis of spleen and muscle tissue at 14 dpi for the presence of cells staining positively for CD3 and CD4 or CD3 and CD8 and negatively for B220 and GR-1 (Fig 9A). Although T cell engraftment varied between individual mice, the frequencies (Fig 9B) and total numbers (Fig 9C) of CD4 and CD8 T cells in the spleen and muscle tissue of mice receiving WT or Ifng-/- T cells were comparable. RRV RNA levels were lower in muscle tissue of mice that received WT T cells compared to mice that received media alone (9-fold decrease, P < 0.001) (Fig 9D). Moreover, RRV RNA levels in Rag1-/- mice that received Ifng-/- T cells were similar to mice that received no T cells, which was significantly greater than RRV levels in Rag1-/- mice that received WT T cells (7.1-fold increase, P < 0.001) (Fig 9D). These data indicate that IFN-γ production is critical for the antiviral effects of T cells following RRV infection. In sum, our data suggest that Arg1-mediated inhibition of T cell activation and IFN-γ production results in enhanced viral loads, perhaps contributing to viral persistence and chronic disease in humans.
We previously showed that RRV and CHIKV infection in mice resulted in elevated expression of Arg1 in inflamed musculoskeletal tissues and tissue-infiltrating macrophages [44]. Here, we demonstrate that Arg1 is also highly induced in PBMCs isolated from CHIKV-infected patients during the acute phase and remained elevated in PBMCs isolated from patients 2–3 months post-illness onset. These data suggest that CHIKV infection may result in the expansion of immunoregulatory myeloid cells that express high levels of Arg1. Moreover, higher Arg1 transcript levels were associated with higher viral loads and with more severe disease, suggesting a relationship between viral levels, disease severity, and cells that express Arg1. Immune-suppressive myeloid cells are heterogeneous populations of myeloid cells that are functionally defined by their potent ability to suppress T cell functions via a variety of mechanisms, including Arg1 activity [53]. We previously demonstrated that specific ablation of Arg1 in myeloid cells enhanced the clearance of RRV from musculoskeletal tissues and diminished muscle tissue pathology at late times post-RRV infection [44]. Here, we show that the enhanced control of RRV infection in LysMcre;Arg1F/F mice is likely due to more effective antiviral T cell responses. Several lines of evidence support this conclusion. First, in our previous study we found that CD11b+F4/80+ macrophages sorted from muscle tissue of RRV-infected mice suppressed T cell proliferation ex vivo in a mixed leukocyte reaction via a mechanism that was partially Arg1-dependent [44]. Here, depletion of CD4+ and CD8+ T cells from WT and LysMcre;Arg1F/F mice increased viral loads compared to the respective control antibody-treated mice at 14 dpi in inflamed muscle tissue, consistent with our previous studies that demonstrated a role for T cells in controlling RRV infection in muscle tissue [51]. However, viral loads in the muscle tissue of T cell-depleted WT and LysMcre;Arg1F/F mice at 14 dpi were not significantly different, suggesting that Arg1 activity in macrophages inhibited the antiviral activity of T cells at the sites of infection.
Arginase-expressing myeloid cells can inhibit T cell functions via a variety of mechanisms, including reducing the bioavailablity of L-arginine, increasing the production of reactive nitrogen species such as peroxynitrite, and expanding regulatory T cell populations [53]. For instance, depletion of extracellular L-arginine levels from the local microenvironment via the activity of myeloid cell Arg1 has been shown to suppress T cells, including inhibiting T cell proliferation and suppressing other T cell functions such as IFN-γ production [47,49]. We found that WT and LysMcre;Arg1F/F mice had similar frequencies and numbers of CD4+ and CD8+ T cells in the spleen and draining popliteal LN at 7, 10, and 14 dpi, suggesting that myeloid cell Arg1 did not affect T cell numbers in lymphoid tissues. These data are consistent with our findings showing that Arg1 expression is highly induced in inflamed musculoskeletal tissues but not lymphoid tissues of RRV- and CHIKV-infected mice. This expression pattern is similar to mice infected with the Leishmania major parasite where high arginase activity was detected at the site of pathology but not in the draining lymph node [48]. Moreover, Arg1-expressing myeloid cell-mediated suppression has been shown to inhibit antigen-specific T cell responses [50], as was shown for L. major-specific T cells at the site of pathology [48]. Thus, we hypothesized that virus-specific T cells, rather than bulk T cells, would be inhibited specifically at the site of pathology in musculoskeletal tissue. To this end we found that CD4+ and CD8+ T cell frequencies and numbers were similar in muscle tissue of WT and LysMcre;Arg1F/F mice at 7, 10, and 14 dpi. However, utilizing a recombinant RRV encoding the CD8 immunodominant epitope of LCMV, we found that LysMcre;Arg1F/F mice, as compared to WT mice, had significantly more virus-specific CD8+ T cells in the muscle tissue at 10 dpi, a time point preceding the significant difference in viral loads detected at 14 dpi. Further studies will be required to determine if this is a result of deletion of virus-specific T cells due to depletion of amino acids, the inability to detect gp33-specific T cells at the sites of pathology in WT mice, and/or a result of enhanced virus-specific T cell proliferation in muscle tissue of LysMcre;Arg1F/F mice. Preliminary studies show that a similar proportion of bulk and gp33-specific CD8+ T cells in spleen and muscle tissue stain for the proliferation marker Ki67 at 7 dpi (S7 Fig), suggesting that arginase activity may not inhibit T cell proliferation in RRV-infected WT mice.
Alternatively or additionally, an effect of myeloid cell Arg1 on CD4+ T cells, such as through the induction or expansion of Treg cells, another mechanism by which MDSCs mediate T cell suppression [53], could indirectly inhibit virus-specific CD8+ T cells. If this mechanism of suppression was occurring in the context of RRV or CHIKV infection, loss of CD4+ T cells would result in less severe acute disease. Consistent with that hypothesis, Cd4-/- mice are protected from CHIKV-induced arthritis/swelling [55], suggesting that CD4+ T cells may have a pathogenic role in acute CHIKV disease. In this study, we found that during RRV infection a subset of muscle-infiltrating CD4+ T cells produce IL-10, a cytokine that has anti-inflammatory functions and has been shown to induce Arg1 expression in myeloid cells via signaling through STAT3 [53]. However, depletion of CD4+ T cells resulted in minimal effects on Arg1 expression levels in musculoskeletal tissues at 7 days post-RRV infection, suggesting that Arg1 expression levels are primarily regulated by CD4+ T cell-independent mechanisms, and studies are ongoing in our laboratory to define the role of specific cytokines, such as IL-10 and IL-6, and other factors in the regulation of Arg1 expression in musculoskeletal tissues of RRV- and CHIKV-infected mice. In addition, further studies are required to determine if other cell(s) besides T cells are also inhibited by Arg1-expressing myeloid cells in the context of arthritogenic alphavirus infection.
In addition to the increased number of gp33-specific CD8+ T cells in muscle tissue of RRV-LCMV-infected LysMcre;Arg1F/F mice, we also found that a greater frequency of these muscle-infiltrating virus-specific T cells stained for the activation marker CD69 compared to virus-specific T cells from WT mice. Importantly, no difference in the expression of another activation marker—CD11a—was seen on virus-specific T cells WT and LysMcre;Arg1F/F mice. These data suggest a modulation of specific activation markers on virus-specific CD8+ T cells in the presence or absence of arginase activity. CD69 is an early activation marker found on T cells that recently received stimulation through the TCR and has been shown to be persistently expressed at inflammatory foci [56]. We found that a larger proportion of virus-specific CD8+ T cells in muscle tissue of LysMcre;Arg1F/F mice were CD69+ at 10 dpi, suggesting that a greater number of T cells are restimulated in the muscle tissue of Arg1-deficient but not Arg1-sufficient mice, augmenting their activation and antiviral functions (e.g., IFN-γ production). Additionally, Cd69−/− T cells are not efficiently retained in lymphoid tissues and also fail to establish or sustain tissue residency [57,58]. Thus, arginase activity may inhibit virus-specific T cell retention in musculoskeletal tissues, resulting in reduced viral control. Additional studies are required to delineate the cause(s) of this differential T cell activation in WT versus LysMcre;Arg1F/F mice.
Studies with Sindbis virus (SINV), an alphavirus that causes encephalomyelitis in mice, have shown that mice unable to make antibodies can clear infectious virus from the brain stem and spinal cord but not the brain [54]. This was shown to be due at least in part through the action of IFN-γ produced by both CD4+ and CD8+ T cells, resulting in site-specific non-cytolytic clearance of virus from the CNS [54]. Consistent with a role for IFN-γ in controlling alphavirus infection, CHIKV-infected IFN-γ-/- mice had increased serum viral RNA levels compared to WT mice [55]. Here, we demonstrated that muscle-infiltrating CD4+ and CD8+ T cells express IFN-γ transcripts, and IFN-γ expression is higher in T cells isolated from muscle tissue of RRV-infected LysMcre;Arg1F/F mice compared to T cells isolated from muscle tissue of WT mice, suggesting that myeloid cell Arg1 activity inhibits cytokine expression by T cells in inflamed musculoskeletal tissues. Increased IFN-γ mRNA expression combined with an increased frequency of CD69+ T cells in musculoskeletal tissues of Arg1-deficient mice suggests that a lack of arginase activity may lead to more efficient T cell restimulation within the inflamed and infected musculoskeletal tissues. This results in more effective antiviral T cells and thus better viral clearance. The mechanism by which IFN-γ acts may include direct antiviral effects as well as regulatory functions important for other immune effector mechanisms, such as increasing expression of MHC class I and class II. Further studies demonstrated that adoptive transfer of naïve WT T cells but not T cells lacking IFN-γ could control RRV infection in muscle tissue of infected Rag1-/- mice, supporting a direct role for IFN-γ. Since uncontrolled cytokine production can be highly toxic, IFN-γ expression by T cells is tightly regulated at the transcriptional level [59]. Indeed, studies with effector CD8+ T cells during virus infection have shown that cytokine production terminates immediately following loss of antigen contact but is quickly initiated again after antigen contact is restored [59]. The increased IFN-γ expression by muscle-infiltrating T cells from RRV-infected LysMcre;Arg1F/F mice is another indication that the Arg1-driven immunosuppressive environment inhibits T cell responses.
These studies provide important evidence for the role of Arg1-expressing myeloid cells in the control of arthritogenic alphavirus infection in humans and mice. Sustained expression of Arg1 throughout the course of disease suggests that activation of immunosuppressive myeloid cells may contribute to the duration of disease and/or the development of chronic disease. Thus, therapeutics that target the induction or activity of Arg1 could limit the severity or duration of these debilitating virus-induced diseases.
All mouse studies were performed in strict accordance with the recommendations in the Guide for the Care and Use of Laboratory Animals of the National Institutes of Health. All mouse studies were performed at the University of Colorado Anschutz Medical Campus (Animal Welfare Assurance #A 3269–01) using protocols approved by the University of Colorado Institutional Animal Care and Use Committee. All studies were performed in a manner designed to minimize pain and suffering in infected animals. CHIKV human PBMC samples were collected from 23 patients that were admitted to the Communicable Disease Centre at Tan Tock Seng Hospital during the 2008 Singapore CHIKF outbreak. All patients were diagnosed with CHIKF and blood was collected with written informed consent obtained from all participants. The study was approved by the National Healthcare Group’s domain-specific ethics review board (DSRB Reference No. B/08/026).
A total of 23 PCR-confirmed CHIKV-positive individuals from the 2008 Singapore CHIKV outbreak were included in this study [45]. Peripheral blood mononuclear cell (PBMC) samples were isolated from patients admitted with acute CHIKV disease to the Communicable Disease Centre at Tan Tock Send Hospital during the outbreak from August 1 to September 23, 2008 in Singapore. For a control group, PBMC samples were isolated during the same time period from 8 healthy volunteers (controls) residing in Singapore [60]. PBMCs from patients and controls were isolated using standard Ficoll-Paque density gradient centrifugation method and stored in -80°C until use. Samples were taken at 4 different time points: (a) acute phase (median 4 days post illness onset), (b) early convalescent phase (median 10 days post illness onset), (c) late convalescent phase (4–6 weeks post illness onset), and (d) chronic phase (2–3 months post illness onset). On the basis of their viral loads, quantified upon admission to hospital, the patients were classified into either high viral load (HVL) group (n = 11) or low viral load (LVL) group (n = 12) [45,60]. Based on the clinical parameters defined in earlier studies [45,61], illness was defined as “severe” if a patient had either a maximum temperature greater than 38.5°C, or a maximum pulse rate greater than 100 beats/min, or a nadir platelet count less than 100 x 109/liter. Patients who did not fulfill these criteria were classified as “mild” [45,61]. Ten (91%) of 11 patients with HVL (median viral load, 9.97 x 105 pfu/mL; range, 1.42 x 105–5.62 x 108 pfu/mL) presented with “severe” clinical illness, compared with 1 (19%) of 12 patients with LVL (median viral load, 2.02 x 104 pfu/mL; range, 1 x 102–5.36 x 104 pfu/mL) (see S1A Fig).
The T48 stain of RRV was isolated from Aedes vigilax mosquitoes in Queensland, Australia [62]. Prior to cDNA cloning, the virus was passaged 10 times in suckling mice, followed by two passages on Vero cells [63,64]. The SL15649 strain of CHIKV was isolated from a serum sample collected from a febrile patient in Sri Lanka in 2006. This virus was passaged two times in Vero cells prior to cDNA cloning [23].
Gradient-purified RRV was generated as previously described [65]. Briefly, virus particles were banded on a 60% to 20% discontinuous sucrose gradient by centrifugation at 24,000 rpm for 2.5 h at 4°C in a Beckman SW-28 rotor. Banded virus was collected and centrifuged through 20% sucrose at 24,000 rpm for 6 h 4°C in a Beckman SW-28 rotor. Virus pellets were then resuspended, aliquoted, and stored at −80°C.
The recombinant “RRV-LCMV” was generated by inserting a tandem sequence, similar in design to a sequence inserted in the influenza virus genome that encodes the LCMV CD8 T cell receptor epitope gp33–41 (KAVYNFATC) and CD4 T cell receptor epitope gp61–80 (GLKGPDIYKGVYQFKSVEFD) [66] in-frame with the RRV structural polyprotein as previously described [51]. Stocks of infectious RRV, RRV-LCMV, or CHIKV (SL15649) were generated from cDNAs and titered by direct plaque assay on BHK-21 cells as previously described [44].
The CHIKV isolate used in the experiments involving HS 633T cells and monocytes was originally isolated from a French patient returning from Reunion Island during the 2006 CHIKV outbreak (IMT strain) [67]. Virus stocks were prepared via numerous passages in Vero-E6 cultures, titered, washed, and precleared by centrifugation before storing at –80°C. These virus stocks were titered by plaque assay on Vero-E6 cells.
The HS 633T human fibrosarcoma cell line is a kind gift of Philippe Gasque and his team at the University of La Réunion. HS 633T cells were grown in DMEM supplemented with 10% FBS. HS 633T cells were inoculated with CHIKV at a MOI of 1 in serum-free medium for 1.5 h. Inoculum was removed and fresh DMEM containing serum was added. Supernatant was harvested 24 h later. Fresh human PBMCs were isolated from whole blood by gradient centrifugation using Ficoll-Paque. Untouched monocytes were isolated using an indirect magnetic labeling system (Monocyte Isolation Kit II, Miltenyi Biotec). Following selection, monocytes were either plated in serum-free IMDM for 1 h to adhere prior to stimulation with HS 633T cell supernatant for 24 hours post-inoculation (hpi) or co-cultured with mock or CHIKV-inoculated HS 633T cells for 24 hpi. Following the 24 h incubation, supernatant was removed and all of the cells were resuspended in TRIzol (Life Technologies) and stored at –80°C prior to RNA isolation.
C2C12 cells (ATCC CRL-1772) were grown in DMEM (Sigma) containing 10% FBS. To differentiate into myotubes, C2C12 myoblasts were plated in 12 well plates in DMEM containing 2% horse serum which was replaced every-other day. Myotubes were inoculated 6 days after plating and cultured for 48 hours in DMEM containing 2% FBS. C2C12 supernatant was collected into 1.5 mL Eppendorf tubes and centrifiuged at 13,000 rpm for 10 min at 4°C to remove any cellular debris. J774A.1 murine macrophages (ATCC TIB-67) were grown in DMEM containing 10% FBS. For co-culture and supernatant transfer experiments, J774A.1 cells were plated in 48 well plates and cultured in DMEM containing 2% FBS. J774A.1 cells treated with recombinant mouse IL-4 (5 ng/ml; R&D Systems) were used as a positive control for Arg1 induction. J774A.1 cells were inoculated with gradient-purified RRV at a multiplicity of infection (MOI) of 10, which is approximately how much virus was present in the RRV-infected C2C12 supernatant. Prior to addition to J774A.1 cells, identical samples of cell supernatant or purified virus were UV-treated for 15 min (short wave) or heat-treated at 56°C for 1 h to inactivate virus, or were centrifuged at 30,000 rpm in a swinging bucket (SW50.1 Beckman rotor) for 4 h at 4°C to remove live virus. J774A.1 cells were cultured in these conditions for 24 hours. The cells were then washed with PBS and resuspended in RIPA buffer [H2O containing 50 mM Tris (pH 8.0), 150 mM NaCl, 1% NP-40, 0.5% sodium deoxycholate, 0.1% sodium dodecyl sulfate, and 1x protease inhibitor complex (Sigma)] for protein analysis.
Protein lysates were separated by Tris-HCl–buffered 10% SDS-PAGE, followed by transfer to polyvinylidene difluoride membranes. Membranes were blocked in 5% milk in PBS containing 0.1% Tween and incubated in the appropriate Abs against the indicated proteins. GAPDH expression was used as a loading control. Anti-mouse Arg1 Ab (V-20) was obtained from Santa Cruz; anti-mouse GAPDH Ab (clone 71.1) was obtained from Sigma-Aldrich. Membranes were imaged on a ChemiDoc XRS Plus imager (Bio-Rad), and bands for Arg1 and GAPDH were quantified using Bio-Rad Image Lab software.
C57BL/6 wild-type (stock # 000664), Rag1-/- (stock # 002216), LysMcre (stock # 004781), Arg1F/F (stock # 008879), and Ifng-/- (stock # 002287) mice were obtained from The Jackson Laboratory and bred in house. LysMcre;Arg1F/F mice were generated as previously described [44]. Animal husbandry and experiments were performed in accordance with all University of Colorado School of Medicine Institutional Animal Care and Use Committee guidelines. All mouse studies were performed in an animal biosafety level 3 laboratory. Three-to-four week old mice were used. Mice were inoculated in the left rear footpad with 103 PFU of RRV in diluent (PBS/1% bovine calf serum) in a 10-μl volume. Mock-infected animals received diluent alone. Mice were monitored for disease signs and weighed at 24-h intervals. Disease scores were determined by assessing grip strength, hind limb weakness, and altered gait, as previously described [68].
On the termination day of each experiment, mice were sacrificed by exsanguination, blood was collected, and mice were perfused by intracardial injection of 1x PBS. PBS-perfused tissues were removed by dissection and homogenized in TRIzol Reagent (Life Technologies) with a MagNA Lyser (Roche). Alternatively, quadriceps muscles were dissected, minced, and incubated for 1.5 h with vigorous shaking at 37°C in digestion buffer (RPMI 1640, 10% FBS, 15 mM HEPES, 2.5 mg/ml Collagenase Type 1 [Worthington Biochemical], 1.7 mg/ml DNase I [Roche], 1x gentamicin [Life Technologies], 1% penicillin/streptomycin). Following digestion, cells were passed through a 100-μm cell strainer (BD Falcon) and banded on Lympholyte-M (Cedarlane Laboratories) to isolate infiltrating leukocytes. Additionally, spleens and draining popliteal lymph nodes were dissected from mice and passed through a 100-μm cell strainer. Following red blood cell lysis (spleens only), cells were washed in wash buffer (1x PBS, 15 mM HEPES, 1x gentamicin, 1% penicillin/streptomycin), and total viable cells were determined by trypan blue exclusion. Sera samples were titered by direct plaque assay on BHK-21 cells.
Leukocytes isolated from enzymatically digested tissues were incubated with anti-mouse FcγRII/III (2.4G2; BD Pharmingen) for 20 min on ice to block nonspecific Ab binding and then stained in FACS staining buffer (1x PBS, 2% FBS) with the following Abs: anti-CD3-fluorescein isothiocyanate (FITC, clone 145-2C11), anti-CD3-allophycocyanin (APC, 145-2C11), anti-CD4-PerCP-Cy5.5 or Pacific Blue (RM4-5), anti-CD8α-phycoerythrin (PE, 53–6.7), anti-CD44-Pacific Blue (IM7), anti-CD69-FITC (H1.2F3), anti-B220-PE-Cy7 (RA3-6B2), and anti-CD19-FITC (6D5) (all from BioLegend); anti-CD8β-PE (H35-17.2), and anti-Gr-1-PE-Cy7 (RB6-8C5) (all from eBioscience). gp33-APC H-2Db KAVYNFATM tetramer was kindly provided by the National Institutes of Health Tetramer Core Facility. Cells were fixed overnight in 1% paraformaldehyde and analyzed on an LSR II using FACSDiva software (Becton Dickinson). Further analysis was done using FlowJo Software (Tree Star). Doublets were excluded using side- and forward-scatter height and width parameters.
For intracellular cytokine analysis, wells in a 96-well plate was coated overnight with anti-CD3 Ab (10 μg/mL). Muscle-infiltrating leukocytes from RRV-infected mice were isolated at 10 dpi. 100 μl RPMI containing 7% FBS, Brefeldin A, and anti-CD28 Ab (2 μg/mL) was added to the 96-well plate followed by addition of muscle-infiltrating leukocytes in 100 μl RPMI containing 7% FBS. Cells incubated in the absence of anti-CD3 and anti-CD28 Abs were used as a control. After 5 h incubation, cells were harvested, and incubated with anti-mouse FcγRII/III (2.4G2) for 20 min on ice followed by surface marker staining in FACS buffer for an additional 20 min on ice. After one wash step, cells were fixed and permeabilized in a 1% paraformaldehyde and saponin solution for 15 min at room temperature. Cells were washed with PBS containing saponin and then stained for intracellular cytokines for 45 min on ice in PBS with saponin. Finally, cells were washed 1X with PBS containing saponin, 1X with FACS buffer, and then stored overnight in 1% paraformaldehyde.
The following antibodies from Bio X Cell were used for depletion studies: rat IgG2b control antibody (anti-KLH, clone LTF-2), CD4-depleting antibody (rat IgG2b; clone YTS 191), and CD8-depleting antibody (rat IgG2b; clone YTS 169.4). Mice were treated on days 7 and 12 post-inoculation via the intraperitoneal route (i.p.) with 200 μg of each CD4- and CD8-depleting antibody, or 400 μg of the control antibody, diluted in PBS to a final volume of 150 μl. Depletion efficiency was determined by flow cytometric analysis of spleen tissue at 14 dpi as described in the text for the presence of cells staining positively for CD3 and CD4 or CD8β. In separate experiments, mice were treated i.p. with 200 μg of CD4-depleting antibody or a control antibody diluted in PBS to a final volume of 150 μl on day 4 pi and harvested on day 7 pi. Depletion efficiency was determined by flow cytometric analysis of spleen and quadriceps muscle tissue as described in the text.
For cell sorting, mice were sacrificed at 10 dpi and the quadriceps muscles were processed as described above. Cells were stained in FACS staining buffer with anti-CD19-FITC, anti-CD3-APC, anti-CD4-Pacific Blue, and anti-CD8α-PE antibodies (BioLegend). Cells were sorted under BSL2 conditions on a FACSAria cytometer using FACSDiva software (Becton Dickinson). CD19-CD3+ cells were gated on first, then CD4+CD8- or CD4-CD8+ cells were sorted separately. Cells were resuspended in TRIzol (Life Technologies) and stored at –80°C prior to RNA isolation.
For T cell adoptive transfer experiments, T cells were isolated from the spleens of naïve wild-type C57BL/6 mice or Ifng-/- mice via negative selection using a pan-T cell isolation kit (Miltenyi Biotec). Following isolation, cells were counted and 2.5 x 106 T cells were resuspended in RPMI containing 2% FBS in a total volume of 200 μl for i.p. injection into Rag1-/- mice one day prior to RRV infection. Control Rag1-/- mice received 200 μl of media alone. Of the transferred T cells, ≥ 95% of the cells were CD3+ and ~60% of the CD3+ T cells were CD4+ and ~35% were CD8+. Additionally, ≤ 1% of CD19+ B cells remained in the injected cell preparations. T cell transfer was confirmed by flow cytometric analysis of spleen and muscle tissue at 14 dpi as described above for the presence of cells staining negatively for B220 and Gr-1 and positively for CD3 and CD4 or CD8α.
For analysis of gene expression in mouse tissue samples or cells, RNA was isolated using a PureLink RNA Mini Kit (Life Technologies), and 1 μg of total RNA was reverse-transcribed using Superscript III (Life Technologies), random oligo(dT) primers, and RNaseOUT. Real-time qPCR experiments were performed using Taqman gene expression assays and a LightCycler 480 (Roche). 18S rRNA was used as an endogenous control to normalize for input amounts of cDNA. The relative fold induction of amplified mRNA were determined by using the Ct method [69].
For analysis of gene expression in human cells, total RNA was extracted using RNeasy Mini Kit (QIAGEN) according to the manufacturer’s instructions. Quantification of total RNA was performed using a NanoDrop 1000 Spectrophotometer (Thermo Scientific); following quantification, RNA samples were diluted to 10 ng/μl. qRT-PCR was performed using QuantiFast SYBR Green RT-PCR Kit (QIAGEN) according to the manufacturer’s recommendations in a 12.5 μl reaction volume. All reactions were performed using 7900HT Fast Real-Time PCR System machine (Applied Biosciences) with thermal cycling conditions as described [70]. As above, the relative fold change for each gene between CHIKV-infected and mock-infected was calculated using the Ct method after normalization to GAPDH [70]. See S1 Table for forward and reverse primers used.
RNA was isolated using a PureLink RNA Mini Kit (Life Technologies) as described above. Absolute quantification of RRV RNA was performed as previously described [44]. Briefly, a sequence-tagged (small caps) RRV-specific RT primer (4415 5’-ggcagtatcgtgaattcgatgcAACACTCCCGTCGACAACAGA-3’) was used for reverse transcription. A tag sequence-specific reverse primer (5’-GGCAGTATCGTGAATTCGATGC-3’) was used with a RRV sequence-specific forward primer (4346 5’-CCGTGGCGGGTATTATCAAT-3’) and an internal TaqMan probe (4375 5’-ATTAAGAGTGTAGCCATCC-3’) during qPCR to enhance specificity. To create standard curves, 10-fold dilutions, from 108 to 100 copies of RRV genomic RNAs, synthesized in vitro, were spiked into RNA from BHK-21 cells, and reverse transcription and qPCR were performed in an identical manner. The limit of detection was 100 genome copies.
All data were analyzed using GraphPad Prism 5 software. Data were evaluated for statistically significant differences using a two-tailed, unpaired t test with or without Welch’s correction, a one-way analysis of variance (ANOVA) test followed by Tukey’s multiple comparison test, or a two-way ANOVA followed by a Bonferroni multiple comparison test. Comparison between the high viral load and low viral load group in the patient cohort was performed by two-tailed Mann Whitney U test. Similarly, for the in vitro infection studies, pair-wise comparison was performed using a two-tailed Mann Whitney U test. A P-value < 0.05 was considered statistically significant. All differences not specifically indicated to be significant were not significant (P > 0.05).
|
10.1371/journal.pcbi.1003264 | Virus Encoded MHC-Like Decoys Diversify the Inhibitory KIR Repertoire | Natural killer (NK) cells are circulating lymphocytes that play an important role in the control of viral infections and tumors. Their functions are regulated by several activating and inhibitory receptors. A subset of these receptors in human NK cells are the killer immunoglobulin-like receptors (KIRs), which interact with the highly polymorphic MHC class I molecules. One important function of NK cells is to detect cells that have down-regulated MHC expression (missing-self). Because MHC molecules have non polymorphic regions, their expression could have been monitored with a limited set of monomorphic receptors. Surprisingly, the KIR family has a remarkable genetic diversity, the function of which remains poorly understood. The mouse cytomegalovirus (MCMV) is able to evade NK cell responses by coding “decoy” molecules that mimic MHC class I. This interaction was suggested to have driven the evolution of novel NK cell receptors. Inspired by the MCMV system, we develop an agent-based model of a host population infected with viruses that are able to evolve MHC down-regulation and decoy molecules. Our simulations show that specific recognition of MHC class I molecules by inhibitory KIRs provides excellent protection against viruses evolving decoys, and that the diversity of inhibitory KIRs will subsequently evolve as a result of the required discrimination between host MHC molecules and decoy molecules.
| Human natural killer (NK) cells patrol peripheral tissue, monitoring changes on the surface of body cells. They express a network of activating and inhibitory receptors called the killer immunoglobulin-like receptors (KIRs). The main ligands of inhibitory KIRs are MHC class I molecules, which present viral peptides to other immune cells. Several herpes viruses interfere with MHC expression, and when a virus down-regulates MHC class I, NK cells loose an inhibitory signal, become activated and kill the infected cell. The KIR family has a large genetic diversity. However, for the recognition of “missing” MHC molecules this diversity seems redundant as one set of receptors should be sufficient. To study why the KIR system has evolved such a high complexity, we developed an in-silico model, simulating the evolution of populations infected with a herpes-like virus. Next to down regulating MHC-I molecules, these viruses are able to escape the NK cell response by expressing MHC-decoys engaging the inhibitory KIRs. We show that specific KIR-MHC interactions protect against viruses expressing decoys. Because of the provided protection, specific inhibitory KIRs have an evolutionary advantage, giving rise to a high level of diversity. We propose that herpes-like viruses evolving decoys affect in the evolution of KIRs.
| Natural killer (NK) cells constitute 5–25% of the lymphocytes circulating in human peripheral blood [1]. Being part of the innate immune response, they play an important role in the defense against viral infections and in tumor surveillance [2]. In contrast to T and B cells, NK cells do not use somatic gene rearrangements to generate a diverse repertoire of cells expressing unique receptors. Instead, they sample a subset of receptors from a repertoire of activating and inhibitory receptors encoded by the germline.
Individual NK cells express several inhibitory and activating receptors that recognize, among others, major histocompatibility complex (MHC) class I and MHC class I related molecules as their ligands [3]. The interaction between these receptors and ligands generates signals that either allow the NK cell to attack target cells or prevent it from harming healthy tissue. Several viruses down-regulate the expression of host MHC class I molecules, and since these molecules are often inhibitory ligands of NK cell receptors, loss of their expression on the infected cell induces NK cell activation. This mechanism by which NK cells attack MHC-class I deficient cells was coined by Kärre et. al [4] as “missing-self” detection.
In humans there are two main receptor families contributing to missing-self detection. The inhibitory receptor CD94/NKG2A binds to complexes of the human leukocyte antigen (HLA)-E, presenting peptides derived from the leader sequences of HLA-A, -B, and -C [5], [6]. In this inhibitory interaction both receptor and ligand are highly conserved, and the down-stream effects are remarkably similar in different individuals [7]. In contrast, killer immunoglobulin-like receptors (KIR), recognizing the highly polymorphic HLA-A, -B, and -C molecules, can be both inhibiting and activating, are very diverse, and rapidly evolving [8]. Engagement of either inhibitory KIR or NKG2A inhibits the activity of an NK cell, preventing target cell lysis. Phylogenetic studies have shown that the CD94/NKG2 system is relatively old, and that the KIR genes have evolved more recently [9]. Thus, there are two NK cell receptor systems, one conserved and one highly diverse, detecting abnormalities in MHC expression on cell surfaces.
KIRs are encoded by a large family of genes exhibiting a remarkable variability in gene content and allelic polymorphism. The KIR complex in humans contains up to 14 KIR genes and pseudogenes [10] that are arranged into two main groups of haplotypes, A and B, differing in size, gene content, function, and disease association [11]. Since the MHC and KIR loci are on different chromosomes (in humans, on chromosome 6 and 19, respectively), a tremendous number of possible receptor-ligand combinations exists on the population level. Moreover, KIR-HLA interactions are rather specific, with four mutually exclusive epitopes on HLA molecules (A3/11, Bw4, C1 and C2) so far identified as inhibitory KIR ligands [12]. KIR interactions with HLA-C are sensitive to polymorphisms at distal positions [13] and to bound peptides [14], affecting KIR binding, and with that the functionality of NK cells. It is widely accepted that the fine specificity and vast diversity of B and T cell receptors per individual render each host the capacity to recognize many different pathogens, and to distinguish them from healthy tissue. But how does the specificity and much smaller diversity of NK cell receptors per individual contribute to the host's survival? If missing-self detection were the main function of inhibitory KIRs, and since this can also be achieved by the conserved receptor NKG2A, why have these more recent NK cell receptors evolved to become specific, polymorphic, and diverse?
Specific KIR alleles have been associated with particular infections such as HIV, HCV, cerebral malaria, and with several pregnancy disorders [15]–[21]. Indeed, population genetic studies have suggested that a high degree of KIR diversity is necessary for surviving epidemic infections and population bottlenecks [22], but no explicit evolutionary mechanism selecting for novel KIR alleles has been proposed so far. Why polymorphic KIRs would be required to just detect MHC down-regulation remains puzzling.
Cytomegaloviruses (CMV) and other viruses from the herpes family have large genomes that encode for a series of immuno-evasive mechanisms, targetting key molecular steps necessary for a successful immune response [23]–[25]. Particularly important for the evasion of NK cell surveillance are MHC-I like molecules that can engage inhibitory NK cell receptors, like the mouse CMV (MCMV) encoded glycoprotein m157 binding to Ly49 receptors [26], [27], and the human CMV (HCMV) UL18 engaging the inhibitory leukocyte immunoglobulin-like receptor LIR–1 [28]. Not all of these evasion strategies have been elucidated yet, and it remains unclear whether m157 and UL18 are the only decoy molecules evolved by herpes viruses. Recent studies have revealed a strong imprint in the KIR repertoire of CMV seropositive individuals [29], [30], suggesting that additional CMV evasion mechanisms interacting directly with KIRs (e.g. novel decoy molecules yet to be identified) exert a strong selection pressure.
We investigated whether the presence of viral decoys like MCMV m157 or HCMV UL18 can drive the expansion of specific, inhibitory NK cell receptors, such as KIRs. We performed our study with an agent-based computer model of co-evolving hosts and viruses. Our results show that specific MHC recognition by inhibitory KIRs provides excellent protection against viruses evolving decoy molecules, and that diversity in the receptor system can be a consequence of this specific interaction between MHC and KIR molecules.
To investigate the evolution of the KIR genes we developed an agent-based model consisting of a host population infected with a non-lethal herpes-like virus causing chronic infections. In this model, individuals were randomly selected during every time step to be confronted with one of the randomly chosen events: birth, viral infection, and death. We modeled a host population of simplified humans carrying one MHC locus and one KIR haplotype composed of five genes. Here, we only modeled inhibitory KIRs and work on activating receptors is in progress. All hosts were initialized with the same randomly generated KIR haplotype, but with different MHC genes. While we allowed for mutation of novel KIR genes during the birth event, mutation of MHC molecules was not considered, and the initial MHC polymorphism (14 alleles, mimicking the number of common HLA–C molecules) remained constant throughout the simulation. Although the host population considered here is inspired on humans, the model remains simple to allow it to be general and also apply to the evolution of KIR in ‘higher’ primates in which the KIR genes also expanded and diversified [31].
KIR, MHC, and viral MHC decoys were modeled with bit strings (i.e., randomly generated sequences of zero and ones), as a simplified representation of amino acid sequences (see Material and Methods). Receptor-ligand binding depended on the longest adjacent complementary match between their bit strings (Fig. 1 A). If the length of the longest match reached a threshold , the molecules could interact. Table 1 depicts the relation between the specificity, i.e., the likelihood of such an interaction, and .
In every host, those KIRs that failed to recognize any of the two MHC molecules present in the individual were deleted from the host's repertoire, leaving each host with only a “licensed” KIR repertoire. Only those KIR molecules that were licensed participated in the immune response, mimicking the MHC-dependent education process during NK cell development [32]. The expected number of licensed KIRs per host, and consequently the probability of the host being protected, depends on the specificity of the KIR molecules, and can be calculated as described in Material and Methods (see Table 1).
Infection of a host started with a short acute phase, after which the individual could either recover, or become chronically infected. We considered 16 different viruses: one “wild type” virus, one “MHC down-regulating” virus, and 14 “MHC decoy” viruses (i.e., one for each of the 14 MHC molecules in our model). We did not allow for superinfection and hosts could be infected with one of the 16 viruses only. The evolution of decoy proteins was modeled by allowing the virus to adopt a randomly selected MHC molecule from its host. Therefore, each decoy protein was actually an MHC molecule. The host populations were first inoculated with the “wild-type” virus, which was typically cleared after the acute phase because of the assumed immune response of both, cytotoxic T and NK cells (Fig. 1 B). The effect of the immune response was modeled by a parameter describing the probability of clearing the infection. For the wild-type virus, this parameter was set relatively high, i.e. 85% of the wild type infections were cleared, resulting in approximately 30% of the individuals becoming chronically infected (Fig. 1 B, and Fig. 2 blue line). The clearance probability was lower for the down-regulating and decoy viruses, resembling their immune escape. Viruses establishing chronic infections spread over much longer periods of time than those that do not; and we therefore expect viruses capable of MHC down-regulation and carrying decoy molecules to outcompete wild-type viruses. We used this model to investigate the evolution of the KIR system, and compared the selection pressures imposed by different viral variants.
If detection of MHC class I is the main function of inhibitory KIRs, we expect that KIRs do not need to be specific nor diverse because missing self detection can be achieved by a limited set of monomorphic receptors. To address this, we analyzed the effects of KIR specificity on populations infected with a virus that is capable of MHC down-regulation to escape T cell response. The immune escape of this mutant was modeled by decreasing the probability of clearing the infection (from 85% to 70%), resulting in a better spread of the virus and a larger fraction of individuals becoming chronically infected (Fig. 1 B, and Fig. 2 A red line).
We screened the average population size as an indicator of the hosts' protection after an infection. By comparing simulations with degenerate KIRs with those with specific KIRs, we observed significant differences in population sizes (from 4300 in to 4100 in , , Mann-Whitney U test). Although the effect of KIR specificity on protection against an MHC down-regulating virus was small, it clearly indicated that hosts with highly specific KIR–MHC interactions were more vulnerable than those having degenerate KIRs (Fig. 2 A,B,E). Why is a high KIR specificity disadvantageous during an infection with an MHC down-regulating virus? Since degenerate KIRs (i.e. ) are likely to recognize any MHC in the population, these receptors are perfectly capable of detecting the presence (and hence absence) of MHC molecules within one individual. But if the KIR-MHC interaction is specific enough (i.e. ), the chance of a KIR to recognize any MHC within the same individual is small, impeding the host to detect MHC down-regulation, i.e. missing-self. Thus, the potential to recognize the absence of MHC molecules, and with it to clear the infection, decreases with a higher specificity of KIR-MHC interactions. Note that the inability of a specific inhibitory KIR to recognize missing-self is independent of the education process we implemented in the model. These results were consistent in all simulations we ran for each specificity setting (, Fig. 2 E red line), confirming our reasoning that for missing-self detection, inhibitory NK cell receptors do not need to be specific.
To avoid elimination by the host immune response, viruses like CMV code decoy MHC molecules that can engage inhibitory NK cell receptors [26], [27]. As KIR specificity did not have a large effect on missing-self detection, we wondered whether high KIR specificity can be an adaptation to a CMV like virus. In our model, a virus down-regulating the MHC expression in one individual, can randomly select one of the MHC molecules of its host, incorporate it in its “genome”, and express it as a decoy protein in the current and subsequent hosts. While in the current host this decoy is always successful, in the subsequent hosts its success will depend on the specificity of the KIRs. Viruses carrying successful MHC decoys can escape the immune response of both T and NK cells. The fitness cost of a host infected with one of these successful viruses was modeled by decreasing the probability of clearing the infection to zero (Fig. 1 B). Thus, each individual with KIRs recognizing a foreign viral decoy like self MHC, became chronically infected in the model.
The better adaptation of a decoy virus compared to the MHC down-regulating virus was reflected in a higher fraction of chronically infected individuals and in a lower population size (Fig. 2 C). But, opposite to what we observed with the virus down-regulating MHC, the effect of KIR specificity was drastic. The average population size increased from 2500 individuals in a degenerate system to 4100 in a very specific system (, Mann-Whitney U test). Populations having specific KIR-MHC interactions were thus much better protected than those with degenerate or cross-reactive KIRs (Fig. 2 C, D, E).
Why is a highly specific KIR-MHC interaction advantageous in this CMV like infection? To protect the host, KIRs face the challenge to detect MHC down-regulation but not recognize the viral decoy masking MHC down-regulation. As seen in the previous section, a host with degenerate KIRs always has a large repertoire of licensed KIRs, and therefore always succeeds in detecting missing-self. But because of the same low specificity, the KIRs within that individual are expected to also recognize foreign decoy molecules as self MHC. On the other hand, a specific KIR system results in a smaller repertoire of licensed KIRs per individual, impeding the host's ability to detect missing-self (see Table 1 and previous section). However, because of their high specificity, it is also unlikely for any licensed KIRs to recognize foreign decoys. Therefore, a decoy virus typically fails to escape NK immune responses, allowing the infection to be cleared. Again, these results were consistent in all simulations we ran for each specificity setting (, Fig. 2 E black line), showing that KIR specificity helps protecting individuals against viruses evolving MHC-like molecules.
We next studied the effect of specificity on the evolution of KIRs. To estimate the diversity of KIR molecules in the population, we calculated the Simpon's Reciprocal Index (SRI) [33]. The SRI is a diversity measure that is equal to the total number of KIR alleles if they are all equally distributed in the population, whereas the SRI is lower than that in a population where some alleles dominate (described in Material and Methods). This measurement of diversity has the advantage that it is not sensitive to fluctuations in the frequencies of rare KIR alleles in the population.
KIR polymorphism remained low in populations having degenerate and cross-reactive KIRs (i.e. ), whereas it increased significantly in populations with specific KIR-MHC interactions (Fig. 3 A–B). Why is there only selection for diversity in those populations having specific KIRs? Since every host needs to recognize at least one of its MHC-molecules to have a licensed KIR repertoire, and this is guaranteed with degenerate KIRs, there is hardly any selection pressure in these populations to evolve novel KIR molecules. But with specific KIRs, individuals do not always recognize their own MHC and hence they are more vulnerable during the infection with an MHC down-regulating virus. The chance of recognizing self MHC is higher in individuals carrying two different haplotypes of inhibitory KIRs. Therefore, heterozygous hosts have an advantage over homozygous hosts, an effect that becomes larger with increasing KIR specificity. We conclude that this “heterozygous advantage” is the main selection pressure driving the evolution of novel KIR haplotypes, and that the selection pressure is largest in populations with specific KIR-MHC interactions.
We argued that heterozygous advantage drives the selection of novel KIR molecules in populations with specific KIR-MHC interactions (i.e. ). Yet, the KIR diversity differed significantly between simulations with MHC down-regulating and decoy viruses for exactly the same specificity (, and , Mann-Whitney U test, Fig. 3 C). This result was surprising, because the heterozygous advantage in the likelihood of recognizing self MHC should be equally strong in both types of infection. To address the possible mechanisms underlying this result, we studied the KIR molecules that were being selected after an infection with the CMV-like virus.
First, we analyzed the specificity of the KIR molecules. Although the specificity threshold was fixed, some KIRs happened to recognize more MHC molecules than others. In populations with highly specific KIR-MHC interactions (e.g. ), the initial haplotype was composed of 5 KIRs, each of them recognizing a different number of MHC in the population (Fig. 4 A,B). This distribution remained constant until the mutant viruses emerged. When the CMV-like virus was introduced, there was a clear selection for those KIR molecules that were most specific, i.e. KIRs recognizing only one MHC molecule in the population (Fig. 4 B). Similarly, populations infected with an MHC down-regulating virus evolved more cross-reactive KIR molecules (Fig. 4 A). Thus, although the specificity threshold was fixed, the system exploited the stochastic variation in cross-reactivity among KIR molecules, evolving towards the specificity that rendered most protection (Fig. 4 C,D). During an infection with decoy viruses, this selection started already in populations having intermediate specific KIRs (i.e. ). Surprisingly, a higher specificity was achieved by haplotypes implementing duplicate KIR genes, which effectively decreased the number of loci (Fig. S2). The fact that the evolution of an even higher specificity affects the heterozygote advantage (Fig. S1) explains the variation in KIR diversity between the infections with MHC down-regulating and decoy viruses (Text S-S).
If evolution selects for the most specific KIRs to protect against viruses evolving decoys, the challenge of recognizing self MHC is even larger. Is there any mechanism that allows for a higher chance of detecting self despite a high KIR specificity? To address this question, we studied the KIR haplotypes before and after the infection with a decoy virus, and again measured the number of recognized MHC per haplotype. In populations having specific KIRs (e.g. ), a randomly generated initial haplotype recognized an average of 8 MHC molecules, reflecting the expected cross-reactivity of its KIRs. Upon infection with decoy viruses, more specific haplotypes evolved due to the evolution of more specific KIR molecules. At the end of the simulation, approximately 80% of the haplotypes recognized only five different MHC molecules in the population (Fig. S3); a surprising result because this property is only expected for 45% of the randomly created KIR haplotypes with . Hence, there was a clear selection for haplotypes that overlapped as little as possible in their MHC recognition, while keeping the highest specificity per KIR molecule. By evolving such “orthogonal” haplotypes, the paradox of recognizing as many MHC in the population as possible without detecting foreign decoy molecules was solved. Together, our analysis suggests that, if the specificity of KIR increases, it becomes beneficial to have more loci to be able to detect missing-self, which provides an explanation for the observed polygeny in the KIR complex.
The exact evolutionary advantage of the highly diverse KIRs has remained intriguing, especially because MHC class I detection, i.e. “missing-self” detection, can also be achieved by a limited set of monomorphic receptors. Our results show that for simple detection of MHC down-regulation, degenerate KIR molecules are advantageous, while a specific KIR-MHC interaction protects hosts against viruses evolving decoy molecules. In the presence of viruses expressing decoy molecules, the KIR became very specific, while at the same time the number of recognized MHC molecules per haplotype was maximized. The more specific the system becomes, the stronger the selection pressure on hosts to carry two different KIR haplotypes. As a result of this heterozygote advantage, KIR haplotypes evolve a high degree of diversity.
The results with viruses evolving decoy molecules depend strongly on the implemented MHC dependent NK education process, as we allow only for the “licensed” KIRs to participate in the immune response. Inhibitory receptors for MHC class I are very important for the education, repertoire development, and response of NK cells, and there is indeed good evidence that the failure to engage inhibitory receptors during development results in peripheral NK cells that are hyporesponsive [32], [34]–[37]. Yet, recent studies [38]–[40] showed that NK cell populations that cannot ligate their inhibitory receptors–either because they are unlicensed cells, or because they have been transferred into a different MHC class I environment–respond in a normal inflammatory manner after viral infection. This response of unlicensed NK cells appeared to be even more robust and protective than that of licensed NK cells. In all these studies, the NK cells were stimulated via their activating receptors. However, we only considered inhibitory receptors, modeling “functionality” as the capacity of the mature NK cells to respond to cells in which the expression of self MHC class I is decreased. MHC independent mechanisms for NK cell activation, such as activation via cytokines, is implicit in the model, and is taken into account in the overall probability of clearing the infection. Also note that we do not model the KIR molecules on individual NK cells, but define which KIRs are licensed in a host's whole repertoire.
The KIR system has evolved unusually rapidly, resulting in different levels of specificities across species. While KIRs in rhesus macaque have a broad specificity, orangutans, chimpanzees, and humans have evolved more specific KIR systems [41], [42]. KIR recognition in humans is restricted to at least four epitopes (HLA-A11,-Bw4,-C1, and -C2), where HLA-C1 and -C2 have the highest avidity. The evolution of these particular MHC epitopes have left an imprint on the evolution of the KIR system. This is clearly shown in the differences of the KIR haplotypes starting from old world monkeys to humans [43]. Humans, chimpanzees, bonobos, gorillas, orangutans, and rhesus macaques share four lineages of KIR genes, which expanded approximately 35–40 million years ago [31]. Within these lineages, each species has independently evolved different numbers of KIRs with either an inhibiting or activating function, and their emergence is related to the evolution of their ligands. The expansion of lineage III KIRs in orangutans, chimpanzees, and humans is associated with the emergence of the C1-, and later with the C2 epitopes. In contrast, rhesus macaques have expanded lineage II KIR genes, corresponding to their complex MHC system, which is composed of several subsets of differentially expressed MHC-A, and -B genes in the absence of an MHC-C locus. Here, we do not model the four KIR epitopes present on several MHC alleles, but have randomly made MHC molecules. The composition of KIR haplotypes, as well as the evolution of novel MHC molecules, is not the focus of this manuscript. Our main question is why KIRs have evolved a degree of specificity, and our approach clearly reveals that specificity has a selective advantage because of its protective effect against CMV-like viruses evolving MHC decoys. Some of the high specificity values used in our model might seem contradictory to the small number of MHC epitopes identified as main KIR ligands. Yet, all results presented here are already obtained at specificity values , which corresponds to a recognition of 20% of MHC molecules in the population (see Table 1), and is in agreement with the four MHC epitopes that have been identified so far in human KIRs.
Hosts are exposed to multiple challenges during their life span, and the immune system has evolved to respond to all of them rather than adapt to only one particular virus. For simplicity, we here consider only one type of infection at a time. Nevertheless, CMV seems to have an important role in NK cell mediated immunity. Recent studies revealed that there is a strong imprint in the NK cell repertoire of CMV seropositive individuals because a particular subset of NK cells with “self-specific” inhibitory KIRs is expanded [29], [30]. Furthermore, it has been shown that CMV plays an important role in viral driven evolution of NK cell receptors in mice [44], [45]. Mice possess the Ly49 receptor system, which is functionally similar to KIR but evolutionary and structurally different. The Ly49 receptors exhibit also high genetic diversity and also have mouse MHC-class I molecules as ligands. Mouse strains that are resistant to MCMV carry an activating receptor, Ly49H, binding to the “MHC-class I decoy” m157 with high affinity. Mice susceptible to MCMV lack the Ly49H gene but possess the inhibiting receptor Ly49I also binding strongly to the m157 glycoprotein. Because Ly49H evolved from its inhibitory homologue, Ly49I [46], it seems that the m157 induced immune pressure led to the evolution of a new activating NK cell receptor, conferring resistance to the virus. Our results agree with this data, showing that a CMV encoded MHC-like decoy imposes a selection pressure to drive the evolution of novel NK cell receptors.
Fighting pathogens and successful reproduction are two crucial functions for survival. By their contribution to immune defense and reproduction, KIRs reveal various selection pressures imposed on NK cells, emphasizing the importance of diversity for surviving population bottlenecks and infections. For these reasons, it may seem intuitive that receptor diversity is beneficial for viral control. But we have seen that the mere detection of missing self is achieved best with degenerate KIRs.
Our agent-based model provides a solid explanation for one selection pressure driving the evolution of specific KIRs, namely viruses expressing MHC decoys. This does not need to be the only explanation, and our findings call for further studies into other possible mechanisms. The evolution of the specificity and number of loci per haplotype, as well as the evolution of activating receptors or other viral strategies, should now be integrated in our model to address additional questions.
We developed an agent-based model consisting of two types of actors (hosts and pathogens) and three types of events (birth, death, and infection). The basic time step of the model is one week, during which we run through all hosts in a random order and confront them to one of the randomly chosen events. Hosts age over time, and after each time step, their age, infection state, and infection type is updated. The cycle is repeated for many hosts generations to model the long-term evolution. All model parameters are fully described in Table 2. The following is a detailed description of the actors and the events:
The model was initialized with a host population of 4500 hosts, with a random age between 1 and 70 years. Gene pools for MHC and KIR alleles were created at the start of each simulation. The pool of MHC consisted of 14 alleles according to the most frequent HLA-C alleles in the European population (dbMHC Anthropology [49]). For each MHC allele, ten different KIR were randomly generated, which could bind to the MHC with a specificity of at least , resulting in a KIR pool of “functional” 140 alleles. To create the initial genome of each host, MHC and KIR genes were randomly drawn from the pools. The individuals were initialized with the same KIR haplotype, but with different MHC genes.
The Simpson's Index is a measurement of diversity that can be interpreted as the probability that two randomly chosen molecules from two random hosts in the population are identical. The lower the Simpson's Index, the higher is the diversity of molecules in the population, and the reciprocal of the Simpson's Index [33] defines a “weigthed” diversity. This diversity measure has the advantage over the total number of unique KIR molecules because it is less sensitive to fluctuations in molecule numbers caused by random neutral drift. For instance, if all molecules are equally frequent in a population, the SRI score is equal to the number of alleles in the population. A population dominated by a single molecule will have an SRI score close to 1. The SRI was calculated as follows: , where is the fraction of the molecule over all KIR molecules in the population, and is the total number of unique KIR molecules.
The probability that a host having a heterozygous diploid genome recognizes its own MHC molecules is defined by , where is the probability that a KIR recognizes a random MHC molecule in the population (which depends on , see Table 1), and is the number of KIR loci. The expected number of licensed KIR is determined by . In our model each host has a genome consisting of one MHC locus and five KIR loci, i.e. , hence that individual will recognize its own MHC molecules with a chance , and the expected number of licensed KIR for the same individual will be . The expected protection against a decoy virus, i.e. the probability of not recognizing the viral protein as self MHC molecule, depends on the size of the licensed KIR repertoire, , and is described by .
Heterozygous advantage is defined as: , where and represent the probability of recognizing self MHC molecules for a heterozygote and a homozygote individual, respectively. We obtained by measuring the fraction of MHC molecules detected by a single KIR haplotype. To obtain we measured the fraction of recognized MHC by all pairwise combinations of KIR-haplotypes. The population has heterozygote advantage if . Values of HA for different , , and are given in Table 1.
The model was implemented in the C++ programming language. The value of was varied in a range from one to ten. For each value , ten simulations were performed for 2000 centuries.
|
10.1371/journal.ppat.1000864 | The Type I NADH Dehydrogenase of Mycobacterium tuberculosis Counters Phagosomal NOX2 Activity to Inhibit TNF-α-Mediated Host Cell Apoptosis | The capacity of infected cells to undergo apoptosis upon insult with a pathogen is an ancient innate immune defense mechanism. Consequently, the ability of persisting, intracellular pathogens such as the human pathogen Mycobacterium tuberculosis (Mtb) to inhibit infection-induced apoptosis of macrophages is important for virulence. The nuoG gene of Mtb, which encodes the NuoG subunit of the type I NADH dehydrogenase, NDH-1, is important in Mtb-mediated inhibition of host macrophage apoptosis, but the molecular mechanism of this host pathogen interaction remains elusive. Here we show that the apoptogenic phenotype of MtbΔnuoG was significantly reduced in human macrophages treated with caspase-3 and -8 inhibitors, TNF-α-neutralizing antibodies, and also after infection of murine TNF−/− macrophages. Interestingly, incubation of macrophages with inhibitors of reactive oxygen species (ROS) reduced not only the apoptosis induced by the nuoG mutant, but also its capacity to increase macrophage TNF-α secretion. The MtbΔnuoG phagosomes showed increased ROS levels compared to Mtb phagosomes in primary murine and human alveolar macrophages. The increase in MtbΔnuoG induced ROS and apoptosis was abolished in NOX-2 deficient (gp91−/−) macrophages. These results suggest that Mtb, via a NuoG-dependent mechanism, can neutralize NOX2-derived ROS in order to inhibit TNF-α-mediated host cell apoptosis. Consistently, an Mtb mutant deficient in secreted catalase induced increases in phagosomal ROS and host cell apoptosis, both of which were dependent upon macrophage NOX-2 activity. In conclusion, these results serendipitously reveal a novel connection between NOX2 activity, phagosomal ROS, and TNF-α signaling during infection-induced apoptosis in macrophages. Furthermore, our study reveals a novel function of NOX2 activity in innate immunity beyond the initial respiratory burst, which is the sensing of persistent intracellular pathogens and subsequent induction of host cell apoptosis as a second line of defense.
| Mycobacterium tuberculosis, the causative agent of tuberculosis, is highly adapted to survive in macrophages of its human host. Host cell suicide is an ancient host cell defense mechanism employed by organisms to wall off invading pathogens. M. tuberculosis manipulates infected cells to inhibit host cell death but the molecular mechanism of this interaction has not been elucidated. Here we describe that M. tuberculosis uses an enzyme complex (NDH-1) usually needed for energy generation in order to neutralize the NOX-2 enzyme-mediated production of toxic oxygen radicals (ROS) by the host cell. We demonstrate that an M. tuberculosis mutant deficient in NDH-1 accumulates ROS inside the macrophage which induces the secretion of an inflammatory cytokine (TNF-α) and subsequent host cell death. The increase of ROS is dependent upon functional NOX-2, since host cells missing a NOX-2 component do not undergo cell death upon infection with the mutant. We propose that a novel function of the host cell NOX-2 complex is to allow sensing of intracellular pathogens by the host cell in order to commit suicide and thus limit bacterial survival.
| The phagocytic NADPH-oxidase (NOX2-complex or phox) resides on phagosomes and has been shown to be involved in microcidal activity in phagocytes. NOX2 is the original member of the NOX family of reactive oxygen species (ROS)-generating NADPH oxidases, which now includes NOX1-NOX5, DUOX1 and DUOX2 [1], [2]. The multicomponent NOX2 complex consists of two transmembrane proteins, gp91phox and gp22 phox, and three cytosolic components, p40 phox, p47 phox and p67 phox [1], [2]. Additionally, the cytosolic GTPase Rac has to be recruited in order to form a fully active NOX2 complex [1]. The gp91phox subunit, which is constitutively associated with gp22 phox, is a transmembrane redox chain that generates phagosomal superoxide by transferring electrons from cytosolic NADPH to phagosomal oxygen [1]. NOX2-generated superoxide can then be converted into a multitude of microcidal oxidants, including hydrogen peroxide and hypochlorous acid, which are important components of the bactericidal activity of the macrophage phagosome [3]. However, NOX2 activity seems to serve a different function in phagosomes of dendritic cells, where it is important for efficient crosspresentation of antigens [4]. The significance of the NOX2-complex for innate immune response is illustrated by the development of chronic granulomatous disease (CGD) in human subjects that have genetic defects in components of the complex. CGD is characterized by greatly increased susceptibility to fungal and bacterial infections [5]. Correspondingly, mice deficient in the NOX2 subunits are much more susceptible to infections with bacterial pathogens such as Salmonella typhimurium for example [3], [5]. Not surprisingly, some pathogens have developed strategies to counter the NOX2 response by either inhibiting NOX2 assembly on the phagosome, as is the case for S. typhimurium [3] and Helicobacter pylori [6], or reducing steady-state levels of NOX2 components as illustrated by Anaplasma phagocytophila [7] or Ehrlichia chaffeensis [8] (for review [9]).
Programmed cell death (PCD), or apoptosis, plays an important role in the innate immune response (IR) against pathogens, a defense strategy that is evolutionarily conserved and extends even into the plant world[10]. Inhibition of host cell apoptosis has been extensively studied and there are numerous examples of viral proteins directly interfering with host cell apoptosis signaling[11]. Furthermore, an increasing number of protozoal pathogens have been shown to manipulate PCD signaling of infected host cells[12]. Finally, prokaryotic pathogens such as Chlamydia, Legionella, Rickettsia, and Mycobacterium among others have the capacity to inhibit host cell apoptosis signaling [13], [14].
Mycobacterium tuberculosis (Mtb) is an extremely successful human pathogen that manipulates host cells via multiple pathways in order to achieve survival[15], [16], [17]. The inhibition of host cell apoptosis by Mtb has been implicated as a potential virulence mechanism[18]. Indeed, an inverse correlation between the virulence of a mycobacterial species and their capacity to induce apoptosis of primary human alveolar macrophages was demonstrated[19]. Cells infected with virulent Mtb have also been shown to be more resistant to various apoptosis stimuli when compared to uninfected or avirulent strains of Mtb[18]. For example, Mtb-infected macrophages secrete soluble TNF-α-receptor in order to inhibit TNF-α-mediated host cell apoptosis induction [20]. Mtb-infection reduces the cell surface expression of Fas receptors, resulting in the resistance of the host cells to Fas-ligand induce cell death[21]. Infection with Mtb also induces the upregulation of the anti-apoptosis gene mcl-1, which confers resistance of cells to apoptosis induction via the host cell mitochondria[22]. Finally, it has recently been shown that Mtb can manipulate the surface of infected macrophages in order to favor a necrotic, rather than apoptotic, cell death outcome[23]. In macrophages infected with virulent MtbH37Rv, but not avirulent MtbH37Ra, the amino-terminal domain of annexin-1 is removed by proteolysis, preventing completion of the apoptotic envelop [24]. Similarly, cells infected with Mtb are less likely to induce host cell membrane repair, which is important for the induction of apoptosis and supports the induction of necrotic cells death and the subsequent dissemination of bacteria[24], [25].
While there is substantial evidence supporting the ability of Mtb to inhibit host cell apoptosis, a causal link between apoptosis inhibition and virulence of Mtb had not been established due to the lack of defined pro-apoptosis mutants. We have recently performed a “gain-of-function” genetic screen and identified three independent regions in the genome of Mtb that contain anti-apoptosis genes[26]. The deletion of one of the identified genes, the nuoG gene of Mtb, which encodes one of the 14 subunits of the type I NADH dehydrogenase (NDH-1), increased infection-induced apoptosis of macrophages and significantly reduced bacterial virulence in mice. These findings support a direct causal relationship between virulence of pathogenic mycobacteria and their ability to inhibit macrophage apoptosis[26]. Our findings are consistent with the identification of another anti-apoptosis gene (superoxide dismutase A) that plays an important role in the virulence of Mtb[27]. Finally, a third gene (Protein kinase E) with anti-apoptosis capacity has recently been described, but the impact of the deletion of this gene on bacterial virulence has not been established[28]. Altogether, the identification of multiple anti-apoptosis genes suggests that Mtb utilizes several strategies to inhibit the apoptotic response of the host cell; however the molecular mechanisms of these interactions have not been investigated.
The present study describes the investigation of the molecular mechanisms by which NuoG of Mtb inhibits host cell apoptosis. The use of TNF-α-neutralizing antibodies and specific caspase inhibitors on human macrophage cell lines, as well as the infection of bone-marrow derived macrophages (BMDM) of wild-type and TNF−/− mice demonstrated that NuoG is involved in inhibiting an extrinsic, TNF-α-dependent, apoptosis pathway. Furthermore, the pro-apoptotic phenotype of the nuoG mutant was abolished in the presence of both ROS scavengers and in the absence of a functional NOX2 system as demonstrated in BMDM and primary human alveolar macrophages. Altogether, our results reveal a novel function of the NOX2 system in helping the host macrophage in sensing persistent intracellular mycobacteria via increased phagosomal ROS levels and the subsequent induction of host cell apoptosis. This may constitute a second line of defense of the macrophage and it is intriguing to speculate that this NOX2 mediated apoptosis induction is equally important in the defense against other intracellular pathogens.
We previously demonstrated that a ΔnuoG mutant of Mtb induced more apoptosis in host cells than wild type bacteria [26]. In order to analyze the mechanism of the NuoG/NDH-1 mediated host cell apoptosis inhibition, we first determined the involvement of caspases in the pro-apoptotic phenotype of MtbΔnuoG using specific caspase inhibitors. PMA-differentiated THP-1 cells were pre-treated with Caspase-3 inhibitor (C3I) or a chemical analog with no inhibitor activity (C3I-A) at 20 µM for 3 h before infection, during infection, and after infection. Cells were either left uninfected, or were infected with Mtb or MtbΔnuoG. After five days cells were harvested and stained for genomic DNA degradation using the fluorescent Terminal deoxynucleotidyl transferase dUTP Nick End Labeling (TUNEL) assay. The percentage of TUNEL positive cells was determined by flow cytometry analysis. This analysis revealed that the uninfected population contained low percentage of apoptotic cells (2.3±0.3%), Mtb infection slightly increased this amount to 11±0.6%. As expected from previously published results [26], cells infected with the nuoG mutant showed a very significant increase in apoptosis (67.5±7.0%). Interestingly, treatment of THP-1 cells with the C3I reduced the percentage of ΔnuoG induced apoptosis to 8.3±1.2%, whereas the C3I-A had no significant effect on apoptosis induction (65.3±9.7%)(Figure 1A). The C3I did not have an effect on the low level of Mtb-induced apoptosis, as 10.0±1.0% of C3I-treated Mtb-infected cells were TUNEL positive, suggesting that Mtb may be inducing low levels of apoptosis via a caspase independent mechanism. In order to determine if the nuoG mutant induces apoptosis via the extrinsic (i.e., death receptor mediated), or the intrinsic (i.e., mitochondrial) pathway [29], cells were treated with Caspase-8 and Caspase-9 inhibitors, respectively. The experimental conditions and analysis were identical to the previous experiment with the exception that cells were harvested 3 days post infection. Analysis of TUNEL staining after this shorter time period resulted in similar rates of apoptosis for Mtb infected and uninfected cells (2.1±0.1% and 1.1±0.1%, respectively). Treatment of these populations with either C8I or C9I had no effect. The nuoG mutant induced apoptosis in about 34.7±0.7% of cells, which was not significantly affected by the addition of C9I (33.2±0.6), but was significantly reduced by the addition of C8I to levels similar to uninfected cells (1.2±0.2%)(Figure 1B). These results indicated that the nuoG mutant induced host macrophage apoptosis via an extrinsic, caspase-8 dependent signaling pathway.
TNF-α is of major importance for a successful host defense against mycobacterial infections, and has also been implicated in the apoptosis response to mycobacterial infection by the macrophage [30], [31]. Since TNF-α receptor signaling can result in cellular apoptosis, we tested whether autocrine TNF-α production and signaling were involved in apoptosis of MtbΔnuoG infected THP-1 cells. We first determined if infection with MtbΔnuoG resulted in an increase of secreted TNF-α. Supernatants from infected THP-1 (Figure 2A) and BMDM cells (Figure 2B) were collected 3 days post infection and levels of TNF-α were measured by ELISA. In both systems, MtbΔnuoG infected cells secreted significantly more TNF-α than those infected with wild type (30 pg/ml to 2.1 ng/ml for Mtb and MtbΔnuoG in THP-1 cells; 0.2 ng/ml to 1 ng/ml for Mtb and MtbΔnuoG in murine cells). Having established the presence of higher levels of TNF-α, we next evaluated the effect of TNF-α signaling on the pro-apoptotic phenotype of MtbΔnuoG. This was first addressed by addition of human TNF-α-specific, neutralizing antibodies (5 µg/ml) to the culture media of THP-1 cells during and after infection. The anti-TNF-α-antibody significantly inhibited macrophage apoptosis induced by MtbΔnuoG infection, as the percentage of apoptotic cells was reduced from 62.3±9.6% to 7.2±1.96% after addition of antibody (Figure 2C). PCD in uninfected or Mtb infected cells was not significantly affected by the addition of antibodies (Figure 2C). The involvement of TNF-α in the pro-apoptotic phenotype of the nuoG mutant was further analyzed by utilizing BMDM from TNF-α−/− mice. The nuoG mutant induced apoptosis in 28±2.3% of wild type C57B/6 (B6) cells, as compared to 5.8±1.8% of Mtb infected cells (Figure 2D). In contrast, the pro-apoptotic phenotype of the nuoG mutant was partially complemented in TNF−/− BMDM, resulting in levels of apoptosis of 13.3±1.9%, where as Mtb infected cells were not significantly different at 4.3±1.6%. Overall, these experiments confirmed the involvement of TNF-α in the pro-apoptosis phenotype of the nuoG mutant.
Reactive oxygen species (ROS) are involved in shifting the balance of TNF-α-R1 mediated signaling from anti-apoptotic to pro-apoptotic [32], [33]. We investigated the role of ROS in MtbΔnuoG induced apoptosis by utilizing a general ROS scavenger (the antioxidant glutathione) and an oxidase inhibitor (diphenylene iodonium or DPI) during infections of THP-1 cells [1]. THP-1 cells were incubated with 15 mM glutathione or 10 µM DPI 3 hours prior to and throughout infection with Mtb and MtbΔnuoG. Untreated cells infected with the nuoG mutant induced apoptosis in about 40.95±3.8% in the population, as compared to 1.3±0.4% in uninfected, and 3.1±0.2% in Mtb infected cells (Figure 3A). The presence of DPI and glutathione reduced apoptosis induced by the mutant to 6.6±0.4% and 3.3±0.2% of cells, respectively (Figure 3A). Thus, both of these agents greatly suppressed apoptosis induced by MtbΔnuoG in THP-1 cells, a finding consistent with a strong dependence of the apoptotic death response on ROS accumulation (Figure 3A) [33]. These inhibitors can also potentially affect cellular NO levels but we determined, using the Griess assay, that THP-1 cells produce no significant increase in NO after infection with the bacteria (Figure S2). Increased ROS levels in the cytosol can also lead to increased gene transcription of an array of genes involved in oxidative stress and immunity, including TNF-α [32]. For that reason, the TNF-α levels in the supernatant of infected THP-1 cells were analyzed after 3 and 5 days by ELISA. Insignificant amounts of TNF-α were detected in supernatants of uninfected cells, and only low concentrations of TNF-α (below 50 pg/ml) were detected in supernatants of cells infected with Mtb or the complemented nuoG mutants strains (Figure 3B). In contrast, the nuoG mutant increased secretion of TNF-α by a factor of 10 to 0.5–0.6 ng/ml. This increase was partially reduced to about 0.1–0.2 ng/ml by treatment of the cells with DPI, and almost completely reduced by the addition of glutathione (0.02–0.03 pg/ml)(Figure 3B). Thus, the increase of intracellular ROS induced by infection of cells with the nuoG mutant is required for the increase in TNF-α secretion by infected cells.
Next, we addressed the question of the subcellular origin of ROS during MtbΔnuoG infection. The mitochondrial respiratory chain complex I is an important generator of cellular ROS that is shared by all cells types and might be at the origin of mitochondrial-induced host cell apoptosis. However, the NADPH oxidases are also potent inducers of cellular and extracellular ROS. In macrophages, the phagocyte NADPH oxidase, NOX2, is recruited to phagosomes and generates the production of superoxide in the lumen of the phagosome. These superoxides and their derivates are thought to be important for the killing of ingested bacteria, although their role in pathogenesis is not completely understood. In order to address the importance of NOX2 in the pro-apoptotic phenotype of the nuoG mutant, we utilized BMDM derived from mice deficient in NOX2 activity due to the deletion of the major transmembrane subunit of the NOX2 complex, gp91phox (gp91−/−). The nuoG mutant induced significantly more apoptosis than Mtb in macrophages of wild type C57Bl/6 mice, 28.6±3.4% versus 8.8±1.6%, respectively (Figure 3C). Importantly, this increase was abolished when gp91−/− BMDM were used as host cells, since only 5.7±1.1% of MtbΔnuoG infected cells were apoptotic compared to 4.1±1.1% of Mtb-infected cells. Therefore, the presence of functional NOX2 is required for the pro-apoptotic phenotype of the nuoG mutant of Mtb. Interestingly, the absence of NOX2 in infected primary macrophages did not result in the reduction of TNF-α secretion as nuoG infected cells secreted more TNF-α (day 3: 1.7±0.3 ng/ml; day 5: 1.1±0.2 ng/ml) than those infected with Mtb (day 3: 0.5±0.2 ng/ml; day 5: 0.2±0.1 ng/ml).
If the ROS responsible for the pro-apoptotic phenotype of the nuoG mutant originate from NOX2, then macrophages infected with MtbΔnuoG should have higher intracellular levels of ROS than those infected with Mtb. In order to address this hypothesis, two dyes for detection of ROS were used: DCFDA, which is more sensitive to H2O2, and DHE, which is more sensitive to O2-. Macrophages were infected and after 24 h the amount of ROS was detected in uninfected, Mtb and MtbΔnuoG infected cells using flow cytometry analysis. Mtb infection induced only slightly elevated levels ROS as detected either by DCFDA or DHE since the histogram overlays closely with that of uninfected cells (Figure 4A–4C). Conversely, both dyes detected a significant increase in ROS levels after infection of wild type cells with the nuoG mutant as depicted by the positive shift in fluorescence (Figure 4A–4C). The pro-apoptotic phenotype of the nuoG mutant was also observed under these conditions (Figure S1A). Importantly, this increase in ROS staining was abolished in gp91−/− BMDM, thus clearly indicating that ROS are being generated by the NOX2 complex (Figure 4A–4C). In order to directly observe ROS localization on a subcellular level, macrophages were infected with DiI-labeled mycobacteria (Figure 4D), stained with DCFDA, fixed, and analyzed by fluorescence microscopy. Only in the MtbΔnuoG infected macrophages were phagosomes stained with the ROS sensor DCFDA, whereas phagosomes of Mtb-infected macrophages remained DCFDA negative (Figure 4E). This data also revealed that the DiI fluorescence is quenched in the presence of ROS and thus the bacterial staining is lost during infection with MtbΔnuoG, but not during infection with Mtb (Figure 4E). Other dyes were used for external labeling of bacteria with similar results (Data not shown). These results not only confirmed the flow cytometry analysis in which an increase of ROS signal was detected only after infection of macrophages with the MtbΔnuoG mutant (Figure 4A), but furthermore localized this increase of ROS to the host cell phagosome (Figure 4E). Nitric oxide (NO) can also oxidize DCFDA to induce fluorescence[34]. However, BMDMS infected with ΔnuoG did not produce significantly more NO (0.97±0.3 µM) than those infected with wild type Mtb (0.82±0.09 µM) and both values were only very marginally elevated compared to uninfected cells (0.45±0.04 µM). In contrast, IFNγ-activated macrophages infected with non-pathogenic Mycobacterium smegmatis induced very significant increases in NO levels (6.77±0.41 µM at MOI 3 and 13.25±0.30 µM at MOI 10)(Figure 4F). The overall NO production in the human THP-1 cells was low, even after IFNγ activation (Figure S2). Thus, the visualized increase of DCFDA fluorescence is likely due to oxidation by ROS.
In order to analyze if the ROS-dependent mechanism of apoptosis induction upon infection with the ΔnuoG is conserved in human cells, primary alveolar macrophages were used as host cells. Due to the source of the cells, only a limited number of cells were available, and therefore the apoptosis assay was adapted to be performed on slides which were analyzed by fluorescence microscopy. For each donor triplicate wells were infected with Mtb, MtbΔnuoG, or were left uninfected. Cells were stained with TUNEL assay 3 days post infection (Figure 5A). Approximately 500 cells were counted on each slide in blinded fashion and the number of TUNEL positive cells was recorded (Figure 5B). Approximately 7.9±2.2% of uninfected macrophages were apoptotic, a percentage which was not significantly different from that of Mtb infected cells (8.5±1.7%). In contrast, there was roughly a 3fold increase in the percentage of apoptotic macrophages infected with MtbΔnuoG (26.9±3.3%). These results were pooled from five different donors, indicating that the phenotype of NuoG-mediated apoptosis inhibition is consistently conserved among different human subjects. The dependence of this pro-apoptotic phenotype on the generation of intracellular ROS was analyzed in two different donors using the inhibitor DPI as described above. Approximately 5 times as many human cells infected with the nuoG mutant underwent apoptosis as compared to those infected with Mtb (21.9±2.4% and 4.5±0.8% respectively). However, this difference between the two strains was abolished by the treatment of cells with the inhibitor DPI, as about 8.7±2.3% of Mtb and 8.1±0.3% of ΔnuoG infected cells were apoptotic under these conditions (Figure 5C). These data strongly suggests that in primary human alveolar macrophages, as in murine BMDM, the NOX2 complex is critical for the pro-apoptotic phenotype of the nuoG mutant. Lastly, the intracellular ROS levels in Mtb or MtbΔnuoG infected cells were analyzed using DCFDA staining. The percentage of infected cells containing one or more ROS-positive phagosomes was quantified from two donors. The amount of cells containing ROS-positive Mtb phagosomes was similar from both donors (19.1±2.9% and 21.8±1.1%). However, these percentages were increased at least 3 fold in MtbΔnuoG infected cells to be 69.3±1.9% and 69.5±7.4 for the two donors (Figure 6A). Also of note, cells infected with MtbΔnuoG contained many more ROS-positive phagosomes than those infected with Mtb (Figure 6B and data not shown).
Since the pro-apoptotic phenotype of MtbΔnuoG is dependent upon the accumulation of ROS in the phagosome, we hypothesized that neutralization or countering of phagosomal ROS may be a general mechanism of inhibition of apoptosis. If this hypothesis was true, other known ROS neutralizing proteins could potentially play a role in inhibition of PCD in host cells. M. tuberculosis contains several enzymes involved in the neutralization of ROS including a secreted Mg, Fe superoxide dismutase (SodA), an outer membrane bound Cu, Zn superoxide dismutase (SodC), and a secreted catalase (KatG). Interestingly, a previous report established the involvement of SodA in the inhibition of apoptosis [27]. To determine if SodC or KatG could likewise affect cell death pathways, THP-1 cells were infected with sodC and katG deletion mutants and stained with TUNEL after 3 days. MtbΔsodC did not induce more apoptosis than the wild type Mtb (strain Erdman) (Figure S3), possibly due to the redundant presence of secreted SodA. However, MtbΔkatG induced significantly more apoptosis than Mtb, both at day 3 (63±5.1% and 23±3.2%, respectively)(Figure 7A) and at day 1 (Figure S1B) post infection. Similar to cells infected with ΔnuoG bacteria, MtbΔkatG infected cells secreted 50-fold more TNF-α (0.5 ng/ml) than those infected wild type bacteria (16 pg/ml)(Figure 7B). Infection of murine macrophages with the katG knockout also resulted in the increase of NOX2-dependent phagosomal ROS (Figure 7C–7E). These results are consistent with the data obtained from the MtbΔnuoG analysis and reinforce the hypothesis that the NOX2-mediated accumulation of phagosomal ROS can lead to induction of host cell apoptosis.
The search for anti-apoptosis genes in the genome of M. tuberculosis led to the identification of nuoG as being important in host cell apoptosis inhibition and bacterial virulence [26]. Here we describe that primary human alveolar macrophages and murine BMDMs infected with the nuoG mutant responded with a NOX2-mediated increase in phagosomal ROS, which was essential to its pro-apoptotic phenotype when compared to wild-type Mtb-infected cells. The presence of TNF-α was necessary but not sufficient for the nuoG mediated apoptosis induction. Furthermore, the infection with the nuoG mutant led to an increase in TNF-α secretion in human and murine macrophages. It is to our knowledge the first time that a direct connection between phagocytosis of a pathogen, NOX2-generated phagosomal ROS levels, and TNF-α-mediated apoptosis signaling has been demonstrated in infected macrophages.
TNF-α receptor 1 (TNF-R1) mediated signaling has either pro-survival or pro-apoptotic consequences [32]. The ligation of TNFR-1 results in either activation of NF-κB, leading to survival of the cell, or activation of the c-Jun N-terminal kinase (JNK), which entails an apoptotic response [32], [35]. A major determinant in the outcome of TNF-α-mediated cell signaling is the concentration of cytosolic ROS [36]. High ROS levels lead to oxidation and inactivation of the MAP Kinase Phosphatases (MKPs), which in their active form inhibit JNK activity. Without active MKP, TNF-α signaling leads to prolonged activation of JNK and subsequent cell death [33]. We have clearly demonstrated that intracellular ROS levels are important for the apoptosis phenotypes of the nuoG and katG deletion mutants of Mtb (Figure 3+7). It will be of interest to determine if the increased phagosomal ROS during mutant Mtb-infection leads to the oxidation of MKPs, or if other components are involved to modify TNF-α signaling outcome. How the increase in phagosomal ROS actually affects cytosolic host cell signaling components is not a trivial question. NOX2-complex generated superoxide is impermeable to lipid bilayer of the phagosome[37]. In contrast, hydrogen peroxide is highly permeable and might thus quickly diffuse into the host cell cytosol to oxidize susceptible cysteines in signaling proteins. Indeed, the JNK phosphatases MKP-1, MKP-3, MKP-5 and MKP-7 all share a phosphatase domain that contains a cysteine which is oxidized upon increase of H2O2 to inactivated the phosphatase and thus lead to increased JNK activity[33]. Nevertheless, H2O2 diffuses rapidly and so it would be surprising that we could detect such a strong accumulation of ROS in the phagosome of infected cells (Figures 4D, 6B and 7E). An alternative hypothesis is that the increase in phagosomal ROS leads to a change in the signaling of receptors in the phagosomal membrane. Cell surface receptors such as TLRs, MARCO and TNF-R1 are phagocytosed together with bacteria in macrophages and the content of the phagosome influences outcome of receptor signaling [38], [39], [40], [41]. The highly oxidative environment of the phagosome containing mutant Mtb compared to wild-type Mtb, may lead to the modification of the ligand/receptor-interactions which could affect the outcome of the signals transmitted by the receptors. The signaling difference observed between non-oxidized and oxidized LDL may serve as an example of how the oxidative modification of a ligand affects receptor signaling [42].
Infection of macrophages with either the katG mutant or the nuoG deletion mutant of Mtb increased the amount of secreted TNF-α (Figures 2A, 2B and 7B). Infection of macrophages with wild-type Mtb induces a low basal level of TNF-α secretion which is induced by transcriptional upregulation of TNF-α mRNA expression[43]. The Mtb mutant mediated increase in TNF-α secretion in human THP-1 cells was inhibited by glutathione and DPI but was not affected in murine BMDMs derived from NOX2-deficient mice (Figures 3B and 3D). This would suggest that there are functional differences between the human macrophage-like cell line THP-1 and murine BMDMs. Alternatively, the TNF- α induction might be mediated via ROS generated in mitochondria which would be inhibited by glutathione and DPI but not by deletion of NOX2. How the nuoG mutant infection leads to an increase in TNF-α secretion by macrophages needs to be investigated in more detail but activation of transcription factors such as ATF-2, Elk-1 and c-Jun upon JNK activation have been reported and would lead to an increase in TNF-α gene transcription[35].
The respiratory burst associated with Mtb infection has been shown to rapidly induce a MAP kinase cascade and NF-κB activation in a NOX2-dependent manner during very early time points (<1 hr post infection)[44]. However, our data suggests that ROS signaling may also play a role at later stages of infection as NOX2-derived ROS are necessary for induction of apoptosis several days post infection (Figure 3). Comparing the effects of Mtb and the nuoG mutant should prove to be a useful model for elucidating the interactions of NOX2-generated phagosomal ROS levels on the host cell apoptosis signaling cascade after prolonged infection.
The specific mechanism by which NuoG inhibits ROS accumulation in the phagosome remains to be determined. However, one potential mechanism could be via the direct neutralization of NOX2 generated superoxides, since they are able to oxidize iron-sulphur ([Fe-S]) clusters with extremely high efficiency[37]. The Mtb NuoG protein contains four [Fe-S] clusters which could directly compete for NOX2 generated superoxides. Nevertheless, this model would require NuoG to enter the lumen of the phagosome, and to date there is no evidence that NuoG is being secreted by Mtb. NuoG does not have a signal peptide and structural analyses of other bacterial NDH-1 systems predict NuoG to be in the cytosol of the bacteria [45]. Furthermore, we have previously failed to detect secretion of a NuoG-phoA fusion protein[26] and in the current manuscript we also did not observe secretion of a myc-tagged NuoG protein into culture filtrate (Figure S4). These results are significant as they suggest that it is not NuoG by itself that is important for phagosomal superoxide neutralization, but that it is potentially the enzymatic activity of the NDH-1 complex that it is required. In order to address this question experimentally, deletion mutants of the NuoL and NuoM subunits of NDH-1 will be created. In homology with other prokaryotic NDH-1 complexes the L and M subunits are proposed to be important in translocation of protons across the membrane during the dehydrogenase activity of NDH-1 and thus their deletion should abolish the enzymatic activity of the NDH-1 complex[45]. If the hypothesis that the enzymatic activity of the NDH-1 complex is important for NOX2 neutralization is valid, then these deletion mutants should have a similar phenotype to the nuoG mutant in regard to ROS and apoptosis increases in host macrophages.
In the light of our results it is tempting to hypothesize that the NDH-1-encoding nuo-operon in M. tuberculosis might have acquired a different function when compared to other prokaryotes. Accordingly, regulation of the Mtb nuo-operon is opposite to that in E. coli. In Mtb, gene expression of the nuo-operon is down-regulated under hypoxic conditions in vitro and at late stage infections in the lungs of mice, whereas it is upregulated under these conditions in E.coli [46]. Interestingly, it is the type II dehydrogenase complex, NDH-2 (ndh, ndhA), of Mtb that is upregulated under hypoxic, nonreplicating conditions[46]. Under these conditions NDH-2 is crucial for maintaining a minimal PMF which is essential for survival[47], suggesting a possible alternative role for the Mtb NDH-1 system. The nuo-operon is under positive control by the two-component system PhoPR [48], which is important for virulence of Mtb and is one of the targets for attenuating mutations in Mtb H37Ra[49], [50]. The phoP mutant fails to replicate in macrophages and infected mouse organs; however bacteria are able to survive in a state of nonreplicating persistence, suggesting that the dormancy regulon is not affected by the phoP mutation and that the PhoPR system is important for early steps of Mtb infection[51]. This is consistent with a role of the NDH-1 complex during the replicative phase of Mtb infections when the host cell NOX2 system is the most active.
The NOX2 complex has been investigated and demonstrated to be of great importance for innate immune defense against a variety of pathogens[5]. In order for bacterial or protozoal pathogens to survive inside the macrophage they must have developed strategies to overcome NOX2 activity. One approach is to directly inhibit NOX2 activity by either perturbing the recruitment of the subunits to the phagosome[3], [52] or by decreasing the steady state levels of NOX2 complex subunits[7], [8]. A novel mechanism employed by Helicobacter pylori is to misdirect the assembly of functional NOX2 complex away from the membrane of phagosome to the plasma membrane, so that superoxides are being released into the extracellular space instead of the phagosomal lumen[6]. Furthermore, a common strategy used by several pathogenic bacteria, including M. tuberculosis, is the enzymatic detoxification of NOX2 generated superoxides via the secretion of enzymes such as superoxide dismutases and catalases[53]. In the case of Mtb, the secretion of large amounts of SodA and KatG may account for the relative insensitivity of the bacteria to bactericidal effects of NOX2 produced superoxides[54]. If our discovery that the NuoG-mediated neutralization of NOX2 activity is important for inhibition of host cell apoptosis is of general importance, one would predict that any mutant deficient in inhibition NOX2 activity should have a pro-apoptotic phenotype. There are few defined mutants for any pathogen described that are deficient in neutralizing NOX2 activity, and could thus be used to confirm or disprove this hypothesis. In the present study, we interrogated a Mtb deletion mutant of the only catalase in the Mtb genome (katG) and demonstrated that it had a similar phenotype to the nuoG mutant of Mtb in regard to an increase in phagosomal ROS and host cell apoptosis induction, both of which were dependent upon functional NOX2 (Figure 7). Interestingly, the katG mutant has been described as being attenuated and the attenuation was dependent on the presence of functional NOX2 complex in the host[55]. Furthermore, inhibition of SodA secretion by Mtb achieved either via deletion of secA2 or via inhibiting sodA transcription also leads to a pro-apoptotic phenotype of the bacteria[27]. This increase in apoptosis is likely to be due to increases in phagosomal ROS levels and dependent upon host cell NOX2 activity, although that has not been investigated to date.
Mycobacterium tuberculosis also contains the membrane bound superoxide dismutase SodC. We have found that the deletion of sodC does not result in a pro-apoptotic phenotype, likely due to the presence of secreted SodA (Figure S3). However, this deletion does render the mutant more susceptible to the bacteriacidal effects of ROS [56]. It is possible that the deletion of nuoG from Mtb may also create a bacterial strain that is more vulnerable to ROS mediated killing, in which case the pro-apoptotic phenotype of ΔnuoG may be due to decreased fitness of the mutant. However, the nuoG deletion mutant was not more susceptible to superoxides being added directly to bacteria using the hypoxanthine/xanthine oxidase system (Figure S5). Therefore, the increase in phagosomal ROS may be affecting apoptosis signaling rather than direct bacterial killing.
The identification of both SodA and KatG as anti-apoptotic proteins indicate that for Mtb, mutants deficient in countering host cell NOX2 activity are generally pro-apoptotic. It will be interesting to know if this mechanism can be extended to other pathogens such as Leishmania donovani, which is able to inhibit host cell NOX2 recruitment to the phagosome. This hypothesis is testable as a mutant deficient in producing the surface glycolipid lipophosphoglycan has lost the capacity to inhibit NOX2 recruitment [52].
Other pathogens, such as Listeria monocytogenes, may evade NOX2 activity by escaping from the phagosome into the cytosol. This is clearly a successful approach to evading the detrimental effects of increased proteolytic activity associated with phagosome maturation. Nevertheless, in the light of our results it is tempting to speculate that this strategy also helps to evade the NOX2-mediated apoptosis induction. It will interesting to test this hypothesis using bacterial mutants that fail to escape the phagosome such as the Listeriolysin O mutant of Listeria monocytogenes.
In conclusion, the investigation of the pro-apoptotic phenotype of a mutant of Mtb deficient in functional NDH-1 complex serendipitously revealed a novel important function of host cell NOX2 complex in macrophages. Our results demonstrate that continuous NOX2 activity will ultimately lead to host macrophage apoptosis induction. The classical respiratory burst is transient, since this generates sufficient amounts ROS to kill susceptible bacteria and thus reduce NOX2 activity. However, infection of macrophages with persistent pathogens, who have adapted to the macrophage as a survival niche and are able to survive this initial ROS burst, would thus potentially lead to continuous NOX2 activity. The results presented in the current manuscript enable us to formulate the following hypothesis: successful intracellular pathogens need strategies to inhibit prolonged activation of NOX2 and/or neutralize the generated superoxides since this will otherwise be sensed by the host cell and will lead to host cell apoptosis. This hypothesis expands the function of NOX2 from the previously described ROS generation for bactericidal activity, to postulate that the host cell macrophages use the NOX2 complex as a mechanism to detect persisting intracellular pathogens.
This study was conducted according to the principles expressed in the Declaration of Helsinki. The study was approved by the Institutional Review Board of the University of Maryland. All patients provided written informed consent for the collection of samples and subsequent analysis.
All animals were handled in strict accordance with protocols approved by the Institutional Animal Care & Use Committee of the University of Maryland (protocol #R-09-35).
C57/B6 and GP91 knockout mice were obtained from Jackson laboratories (www.jaxmice.jax.org). Caspase specific inhibitors and analogs were purchased from Calbiochem (www.emdbiosciences.com). Neutralizing anti human-TNF antibody (#500-M26), the biotinylated detection antibody (500-P31Abt) were purchased from Peprotech Inc (www.peprotech.com). Recombinant human and murine TNF-α, and anti murine-TNF-α antibodies were purchased from BD Pharmingen (www.bdbiosciences.com). CM-DCFDA, DHE, and Vybrant® DiI cell-labeling solution were purchased from Invitrogen (www.invitrogen.com). All other reagents unless otherwise noted were purchased from Sigma (www.sigma.com).
M. tuberculosis H37RV (ATCC 25618) was obtained from the American Type Culture Collection (www.atcc.org), MtbΔkatG was obtained from TARGET (http://webhost.nts.jhu.edu/target/Default.aspx), and MtbΔnuoG has been previously described [26]. GFP expressing Mtb and ΔnuoG were created by transfecting the GFP-pmV261 plasmid into competent cells by electroporation as previously described [26]. All mycobacteria, excluding ΔkatG, were grown in 7H9 media supplemented with 0.5%glycerol, 0.5% Tween-80, and 10% ADS. ΔKatG was grown in the same media supplemented with ADC in place of ADS. For selective media, 50 µg/ml Hygromycin or 25 µg/ml Kanamycin were added.
Human myelomonocytic cell line THP1 (ATCC TIB-202) was cultured in RPMI (ATCC) supplemented with 10% heat inactivated FCS (Hyclone) and differentiated using 20 ng/ml phorbol myristate acetate (PMA)(Sigma) as described [26]. Bacteria were grown to an OD600 ranging from 0.5 to 0.8 and the culture was allowed to settle for 10 minutes. Infections were carried out at a multiplicity of infection (MOI) of 5∶1 (5 bacilli to 1 cell) for 4 hours in infection media containing 10% human serum (Sigma) and 10% non heat inactivated FCS. After 4 hours, extracellular bacteria were removed by 2 washes with phosphate buffered saline (PBS) and the cells were incubated in chase media containing 100 µg/ml of gentamicin (Invitrogen). Cells were assayed for apoptosis by TUNEL staining 3 or 5 days post infection as detailed in the figure legends. The protocol to obtain normal human bronchoalveolar lavage fluid (BALF) was pre-approved by the IRB of the University of Maryland-Baltimore (H-23204). Normal, asymptomatic, non-smoking volunteers between the ages of 18 and 50 were anesthetized with topical and endobronchial lidocaine, and clinically standard fiberoptic bronchoscopy (FOB) was performed in the endoscopy suite at the University of Maryland Hospital. BALF was obtained using 200 mL of sterile normal saline infused in an identical manner into the right middle lobe of each subject, yielding 75–125 mL of BALF. 10–15 mL of BALF was filtered through sterile gauze to remove mucous, and the alveolar macrophages were washed 3 times with PBS before being used in experiments. Cells were resuspended in warm RPMI with 10% heat inactivated FCS, seeded on 8 well slides, and allowed to rest for 1–3 days. Infection was carried out as described above. Bone marrow macrophages were derived from the femur and tibia of C57B/6 and knockout mice and differentiated in DMEM media containing 20% L-929 supernatant. Murine cells were infected as described above using 10% FCS and 5–10% L929 supernatant in the infection and chase media. L929 supernatant was included in order to protect against cytokine withdrawal induced apoptosis. For experiments using caspase inhibitors or analog (20 µM), antioxidants (15 mM glutathione), and oxidase inhibitor (10 µM diphenylene iodonium, DPI), the cells were incubated with the reagents during infection and chase period. In experiments using TNF-α neutralizing antibody (#500-M26, Peprotech) the antibody was included only in the chase medium at a concentration of 5 µg/ml.
The TUNEL assay was preformed to reveal apoptosis-induced DNA fragmentation in tissue culture, primary human, or murine cells using the “In Situ Cell Death Detection Kit-Fluorescein or –TMR Red” (Roche Applied Sciences at roche.com). The assay was carried out as described by the manufacturer and the percentage of stained cells was analyzed using flow cytometry or quantification via fluorescent microscopy.
Reactive oxygen species in primary murine bone marrow cells and alveolar macrophages were detected at 24 hrs or 3 days post infection respectively using the ROS sensitive dyes 5-(and-6)-chloromethyl-2′,7′-dichlorodihydrofluorescein diacetate, acetyl ester (CM-DCFDA) and dihydroethidium (DHE) (Invitrogen). Bone marrow cells were deprived of L929 supernatant 16 hrs prior to infection and maintained in L929 free media without phenol red for the length of the experiment. Human alveolar macrophages were maintained in normal growth, infection, and chase media. In some cases bacteria were labeled with lipophilic red dye Vybrant-DiI (invitrogen). Bacteria were incubated in 7H9 media containing 5 µl/ml of DiI for 30 minutes, washed twice with PBS with 0.05%tween, and then used for infection as normal. Post infection, murine or alveolar macrophages were washed once in HBSS and then incubated in 10 µM DCFDA for 30 minutes or 2 µM DHE for 15 minutes. Cells were washed 3 times with HBSS, fixed with 4% paraformaldehyde, and analyzed by either flow cytometry or fluorescence microscopy.
Nitrite (NO) concentrations in supernatants from C57/B6 BMDMs were quantified via the Griess assay according to the manufacturer's protocol. In brief, supernatants were collected from macrophagess 3 days post infection with Mtb or ΔnuoG. Supernatants from macrophages primed for 16 hrs with IFNγ (100 units/ml), and infected with heat-killed M. smegmatis (MOI of 5 and 20), were used as positive controls. 150 µl of sample was mixed with 20 µl of Greiss reagent and 130 µl water, and the mixture was incubated at 30°C for 30 minutes before measuring absorbance at 548 nm. Nitrite concentrations were calculated from a standard curve with sodium nitrite as the reference.
ELISA was performed with the supernatants of bone marrow derived macrophages or THP-1 cells infected for 3 or 5 days and treated with or without glutathione or DPI as described above. For detection of human TNF-α, the ELISA-plates were coated with 2 µg/ml capture antibody (500-M26, Peprotech) for 2 hours at 37°C. 100 µl of cell supernatant was used for the reaction and recombinant human TNF-α (554618, BD Pharmingen) diluted in infection medium was used as a standard. TNF-α was detected using the secondary biotinylated anti human-TNF-α detection antibody at 200 ng/ml (500-P31Abt, Peprotech Inc), Streptavidin-alkaline phosphatase at 1 µg/ml (Zymed), and phosphatase substrate at 1 mg/ml (Sigma). The plate was read at an absorbance of 405 nm. Murine TNF-α ELISAs were preformed as above using recombinant mouse TNF-α standard, the capture antibody rat anti murine-TNF-α at 8 µg/ml, and the biotinylated detection antibody rat anti mouse-TNF-α antibody at 1 µg/ml (Catalog numbers 554589, 551225, 554415 respectively, BD Pharmingen).
Statistical analyses were performed on three independent experiments (ANOVA with Tukey post-test) unless otherwise noted in the figure legends. Significance indications are as follows: *, 0.01<p<0.05; **, 0.001<p<0.01; ***, p<0.001. Graphs and in –text citations are a representation of the population mean and standard error of mean. Percentages of DCFDA or DHE positive cells found in the sample and not the control (Figure 4 and Figure 7) were calculated by subtracting the histogram of uninfected cells from experimental histograms using Overton cumulative histogram subtraction (FlowJo version 8.8.6 DMV). Differences were compared via ANOVA.
|
10.1371/journal.pbio.1000302 | Regulatory T Cells and Human Myeloid Dendritic Cells Promote Tolerance via Programmed Death Ligand-1 | Immunotherapy using regulatory T cells (Treg) has been proposed, yet cellular and molecular mechanisms of human Tregs remain incompletely characterized. Here, we demonstrate that human Tregs promote the generation of myeloid dendritic cells (DC) with reduced capacity to stimulate effector T cell responses. In a model of xenogeneic graft-versus-host disease (GVHD), allogeneic human DC conditioned with Tregs suppressed human T cell activation and completely abrogated posttransplant lethality. Tregs induced programmed death ligand-1 (PD-L1) expression on Treg-conditioned DC; subsequently, Treg-conditioned DC induced PD-L1 expression in vivo on effector T cells. PD-L1 blockade reversed Treg-conditioned DC function in vitro and in vivo, thereby demonstrating that human Tregs can promote immune suppression via DC modulation through PD-L1 up-regulation. This identification of a human Treg downstream cellular effector (DC) and molecular mechanism (PD-L1) will facilitate the rational design of clinical trials to modulate alloreactivity.
| Graft-versus-host disease (GVHD) is the most serious complication of bone marrow transplants between individuals (so-called allogenic transplants). The class of suppressor immune cells called regulatory T cells (Tregs) inhibit GVHD by dampening the effects of donor immune cells in the grafted tissue. The cellular and molecular mechanisms involved in this process have not been fully characterized, particularly for human cells. In this study, we report that human Tregs, which we generated from precursor cells ex vivo, express high levels of a cell surface protein called PD-L1 (programmed death ligand-1) that is known to mediate immune suppression. Coculture of these Tregs with allogeneic antigen-presenting cells (APCs), which are known to initiate GVHD, increased, in turn, the amount of PD-L1 on the APCs. The Treg-conditioned APCs were then less able than unconditioned APCs to provoke GVHD in a mouse model of the condition, preventing the death of the animals after transplantation. We found that an antibody against PD-L1 blocked the immunosuppressive effects of Tregs or Treg-conditioned APCs, indicating that this protein is an important part of the molecular mechanism. These findings are potentially important for attempts to modulate immune responses in disease by transplanting T cells into patients.
| Regulatory T cells (Tregs) promote immune tolerance to self-antigens and alloantigens (reviewed in [1]). Genetic deficiency of Tregs mediated by lack of Foxp3 transcription factor yields autoimmunity in mice [2] and humans [3]. Numerical or functional deficiency of Tregs in murine models exacerbates autoimmune disease [4],[5], predisposes to solid organ and hematopoietic stem cell graft rejection [6],[7], and associates with acute and chronic graft-versus-host disease (GVHD) [8]–[10]. Importantly, clinical studies have demonstrated Treg defects in humans with autoimmune disease [11],[12] and GVHD [13]–[15]. Given this background, a rationale has been outlined to evaluate adoptive cell therapy using ex vivo–expanded Tregs as an approach to treat autoimmune [16] or alloimmune [17] conditions. Negative selection against the IL-7 receptor alpha chain (CD127) enriches for human Tregs [18] and thereby may represent a useful tool for such cell therapy efforts; however, there are currently no reports pertaining to the regulatory function of cells expanded from CD127-depleted human T cells. Given this information, our experiments focused on human Tregs generated ex vivo by enrichment for CD127-depleted CD4+ T cells and by culture in conditions demonstrated to promote Treg expansion, including CD28 costimulation IL-2, TGF-β [19], and rapamycin [20].
A more comprehensive understanding of cellular and molecular mechanisms of adoptively transferred Treg products would facilitate the rational design of clinical trials evaluating Tregs. Such an understanding may be difficult to ascertain given the varieties of Tregs [21] and numerous molecular mechanisms operational in murine Treg cells, including: CTLA-4 [22], TGF-β [23], PD-L1 [24], GITR [25], or IL-10 [9]. The cellular mechanism of Tregs also is complex and varied depending on the particular experimental model; importantly, recent evidence indicates that murine Tregs inhibit responder T cells indirectly via modulation of dendritic cells (DC) [26],[27].
Identification of cellular and molecular mechanisms of human Tregs, in particular ex vivo–generated Tregs, has been relatively elusive. For example, ex vivo–generated human Tregs suppressed an allogeneic mixed lymphocyte reaction (allo-MLR) by an undefined mechanism that operated independent of IL-10 or TGF-β [28]. Indeed, the role of antigen-presenting-cell (APC) modulation as a human Treg mechanism has been somewhat neglected in part because published studies have typically utilized APC-free suppressor assays. Nonetheless, one recent study determined that freshly isolated Tregs inhibited myeloid DC inflammatory cytokine secretion and costimulatory molecule expression; such Treg-conditioned DC had reduced capacity to stimulate alloreactivity in vitro [29]. In light of this relative paucity of information relating to the mechanism of ex vivo–generated human Tregs, our primary objective was to elucidate the cellular and molecular pathways associated with human Treg cell suppressor function. Because of our focus on allogeneic hematopoietic stem cell transplantation (HSCT), the role of Tregs in GVHD protection, and the role of host APC for GVHD induction [30], we elected to study human Tregs in vitro using an allo-MLR driven by a defined population of myeloid DC and in vivo using a xenogeneic GVHD (x-GVHD) model similar to that previously utilized to study human Tregs [31].
Total CD4+ and CD4+CD127− T cells were costimulated and expanded in medium containing IL-2, TGF-β1, and rapamycin to generate control bulk “CD4” and “Treg” populations that were directly compared in each experiment. Expanded T cells maintained their CD127− status, were comparable in terms of expansion (Figure S1A), coexpression of CD62L with CCR7 (Figure S1B (i)) and Foxp3 expression (Figure S1B (ii)). Because Foxp3 is expressed in human Tregs and transiently expressed in human effector T cells [32], we reasoned that bulk CD4 cell Foxp3 content may represent a marker of effector differentiation. To address this, we compared ex vivo–expanded T cells for simultaneous expression of Foxp3 and effector cytokines, including IL-2 (Foxp3+IL-2+ events) and IFN-γ (Foxp3+IFN-γ+ events). Indeed, relative to Tregs, control CD4 cells had increased coexpression of Foxp3 with IL-2 (Figure S1C (i)) and Foxp3 with IFN-γ (Figure S1C (ii)). Furthermore, relative to control CD4 cells, expanded Tregs mediated increased suppression of CD4+ and CD8+ T cell alloreactivity (Figure S1D (i) and S1D (ii)); suppression was observed at a Treg cell to responder T cell ratio of 1∶20 that approximates the physiologic ratio (see dose-response curve, Figure S1E).
Further experiments were performed to characterize the mechanism of immune modulation mediated by expanded Tregs generated from CD4+CD127− cells. Blockade of TGF-β, IL-10, IDO, CTLA4, or LAP did not abrogate Treg suppression in the allo-MLR (unpublished data). However, experiments utilizing transwell plates indicated that Treg suppression in the allo-MLR was contact dependent (unpublished data). Programmed death (PD) ligand 1 (PD-L1, or B7-H1) is expressed on DC [33], human tumor cells [34], and normal human tissue [35] and interacts with PD receptors on T cells to modulate the balance of tolerance and immunity (reviewed in [36]). In murine systems, Treg cell expression of PD-L1 associates with suppressor function [24]; in addition, endothelial cell [37] or CD8α+ DC [38] expression of PD-L1 promotes murine Treg generation. In humans, intratumor Tregs directly inhibited responder T cell proliferation through PD-L1 [39]. Because Tregs in our experiments expressed increased PD-L1 (Figure 1A; representative flow plot (i) and (ii); summary (iii)), we reasoned that Tregs might modulate DC via the PD-1 pathway. Indeed, allogeneic DC isolated from the Treg-containing MLR expressed increased PD-L1 relative to DC isolated from the standard MLR (Figure 1B; representative plot (i) and (ii); summary (iii)); remarkably, DC harvested from control CD4-containing MLR failed to up-regulate PD-L1. Of note, Treg-conditioned DC did not have increased expression of PD-1 (CD11c+PD-1+ cells, <1%).
PD-L1 inhibits T cell function via the PD-1 receptor and B7-1 (CD80) [40]. To determine PD-L1 binding pathways in our system, we first measured effector T cell expression of PD-1 and CD80 after incubation with three types of allogeneic myeloid DC (control, Treg conditioned, or control CD4 conditioned). Effector CD4+ T cells (Figure 1C; representative flow plot (i)and (ii); summary (iii)) and CD8+ T cells (Figure 1C representative flow plot (iv) and (v); summary (vi)) up-regulated PD-1 expression, but not CD80 expression, upon exposure to Treg-conditioned DC, but not CD4-conditioned DC. We next utilized a PD-L1 fusion protein to characterize binding pathways. Using laser scanning cytometry (LSC), we found that effector T cells up-regulated total PD-L1 binding partners in the presence of Treg-conditioned DC, but not CD4-conditioned DC (Figure 1D, left panel); importantly, effector T cell PD-L1 binding was abrogated by T cell preincubation with anti-PD1, but not anti-CD80 (Figure 1C, right panel). And finally, effector T cell PD-L1 binding was quantified by flow cytometry (Figure 1E (i)–(iii)). Remarkably, PD-L1 binding was greatly increased on effector T cells exposed to Treg-conditioned DC (% effector T cell PD-L1 binding increased from 7.3±0.4 to 92.6±2.8, p = 0.001); similar to results using LSC, effector T cell PD-L1 binding was abrogated by T cell preincubation with anti-PD1, but not anti-CD80 (Figure 1E (iv)).
Secondary transfer experiments were performed to evaluate whether Tregs mediated suppression in part through DC modulation (experimental scheme, Figure 2A). Indeed, allogeneic DC conditioned with Tregs yielded reduced levels of CD4+ and CD8+ responder T cell proliferation relative to CD4-conditioned allogeneic DC (representative results, Figure 2B; pooled results, Figure 2C). Importantly, blockade of DC expression of PD-L1 partially corrected the observed stimulatory deficit of Treg-conditioned DC on CD4+ and CD8+ T cell proliferation (representative results, Figure 2B; pooled results, Figure 2C).
Next, we utilized an in vivo xenogeneic transplantation model to further characterize the ability of Tregs or Treg-conditioned DC to modulate the PD1 pathway. As expected, recipients of Treg-conditioned DC, which expressed increased PD-L1 in vitro prior to adoptive transfer, had an increased in vivo number of dendritic cells in the spleen that expressed PD-L1 (Figure 3A; representative flow plots (i), (ii), and (iii); summary data, 3b (i)); relative to recipients of control DC, recipients of Treg-conditioned DC also had an increase in PD-L1–expressing DC in the bone marrow (p = 0.006). Remarkably, recipients of Treg-conditioned DC also had increased numbers of effector CD4+ and CD8+ T cells in the spleen that expressed PD-L1 in vivo (Figures 3B (ii) and (iii), respectively); such recipients also had increased numbers of T cells that expressed PD-L1 in the bone marrow (p = 0.003). In marked contrast, recipients of control CD4-conditioned DC did not have increased responder T cell PD-L1 expression. Interestingly, recipients of Treg-conditioned DC also had increased numbers of effector CD8+ and CD4+ cells in the spleen that expressed PD-1 in vivo (Figures 3B (iv) and (v), respectively); in the bone marrow, such recipients also had increased numbers of CD8+PD-1+ cells (p = 0.02) and CD4+PD-1+ cells (p = 0.009).
Further experiments were performed to assess the functional significance of this sequential increase in PD-L1 expression from Treg cell, to conditioned DC, and then to responder T cells in vivo. Recipients of Treg-conditioned DC that were incubated with anti–PD-L1 prior to adoptive transfer had lower numbers of PD-L1–expressing DC in vivo, although cohort comparisons did not reach statistical significance (Figure 3C (i)); a repeat experiment yielded similar findings (unpublished data). Blockade of PD-L1 on Treg-conditioned DC yielded a reduction in the in vivo number of effector CD8 cells expressing PD-L1 (Figure 3C (ii)). Blockade of PD-L1 on Treg-conditioned DC also reduced the number of PD-L1–expressing responder effector CD4+ cells in the spleen (Figure 3C (iii)). Finally, PD-L1 blockade of Treg-conditioned DC reduced the in vivo number of effector CD8+ cells in the spleen that expressed PD-1 (Figure 3C (iv)); the number of CD4+PD1+ T cells in the spleen was not significantly altered by PD-L1 blockade (Figure 3C (v)). In sum, these data indicate that PD-L1 expression on Treg-conditioned DC was functionally significant in vivo, particularly with respect to up-regulating downstream expression of PD1 and PD-L1 on effector CD4+ and CD8+ T cells.
Next, we evaluated whether human Treg and Treg-conditioned DC might modulate xenogeneic GVHD in a PD-L1–dependent manner. Previous xenogeneic GVHD models have utilized human peripheral blood mononuclear cells (PBMCs) that contain unmanipulated human T cells [41],[42], or more recently, ex vivo costimulated T cells [43]. Our initial xenogeneic GVHD experiments utilized PBMC or purified lymphocytes as the human T cell inocula. However, despite following the protocol utilized by previous publications, we found an unacceptably low rate of lethal GVHD (<10% lethality by day 45 postinfusion); an inability to consistently generate lethality was associated with a low level of human T cell engraftment (see Figure S2A). Subsequent experiments were designed to identify human inocula that yielded enhanced human T cell engraftment and a resultant increase in lethal xenogeneic GVHD incidence. An initial experiment found that engraftment of purified human T cells was enhanced by coinfusion of a human, but not murine, source of APC (unpublished data). Based on these data, in a subsequent experiment, immune-deficient murine hosts received one of five distinct human T cell–containing inocula: (1) PBMC; (2) lymphocytes plus monocytes; (3) lymphocytes plus DC; (4) ex vivo–activated effector T cells plus monocytes; and (5) ex vivo–activated T cells plus DC. At day 30 postinfusion, recipients of the ex vivo–activated T cells plus DC had the highest levels of human CD4+ and CD8+ T cell engraftment (Figure S2A (i) and (ii), respectively); furthermore, recipients of ex vivo–activated T cells plus DC had the highest capacity for secretion of human IFN-γ at day 30 postinfusion (Figure S2B). Therefore, in order to evaluate the effects of Tregs, Treg-conditioned DC, and the PD-1 pathway in a more stringent model of xenogeneic GVHD, subsequent experiments utilized human inocula that contained ex vivo–activated T cells and DC.
Further in vivo experiments were performed to evaluate the effect of Tregs and Treg-conditioned DC on human T cell engraftment, cytokine activation, and induction of lethal xenogeneic GVHD. Recipients of human inocula that contained either Tregs or Treg-conditioned DC had reduced absolute numbers of human T cells as measured in the spleen at day 45 posttransplant (Figure 4A); the absolute number of human T cells present in vivo was also reduced when the evaluation was performed in the bone marrow for recipients of both Tregs (p = 0.03) and Treg-conditioned DC (p = 0.01). Tregs and Treg-conditioned DC transfer resulted in reduced absolute numbers of both human effector CD8+ and CD4+ cells (Figure 4B; representative data (i); summation of data in (ii) and (iii), respectively). Human CD4+ T cell numbers in the bone marrow was also reduced for recipients of Tregs (p = 0.01) but not significantly reduced in recipients of Treg-conditioned DC (p = 0.08); human CD8+ T cell numbers in the bone marrow were also reduced for recipients of Tregs (p = 0.02) but not significantly reduced in recipients of Treg-conditioned DC (p = 0.09). Of note, recipients of Tregs, but not recipients of Treg-conditioned DC, had a statistically significant reduction in the absolute number of posttransplant CD8+Tc1 and CD4+Th1 cells capable of IFN-γ secretion (Figure 4C (i) and (ii), representative flow plots; 4C (iii) and (iv), summation of data). In sum, these data indicated that both Treg cells and Treg-conditioned DC were capable of inhibiting human T cells in vivo, with Treg therapy manifesting more potent regulation both in terms of limiting T cell numbers and T cell effector function.
Xenogeneic GVHD was evaluated by weight loss measurement, survival analysis, and histology evaluation of GVHD target tissues. Recipients of Tregs or Treg-conditioned DC were uniformly protected against lethal xenogeneic GVHD (Figure 5A (i)); importantly, recipients of control CD4-conditioned DC uniformly died of xenogeneic GVHD. Posttransplant weight loss, which is a more sensitive clinical parameter of xenogeneic GVHD, was moderated by Treg-conditioned DC therapy and virtually eliminated by Treg therapy (Figure 5A (ii)). In a second experiment, we confirmed the ability of Treg-conditioned DC to completely abrogate the generation of lethal xenogeneic GVHD; importantly, protection against lethal xenogeneic GVHD conferred by the Treg-conditioned DC was completely abrogated by anti–PD-L1, but not by isotype control antibody (Figure 5B). Of note, both control DC and Treg-conditioned DC engrafted and persisted in vivo; importantly, such numbers were not substantially influenced by Treg therapy or anti-PDL1 antibody. That is, at day 25 posttransplant, the absolute numbers of CD11c+ DC per spleen (each value, ×103; n = 5 per cohort) in transplant recipients that received effector human T cells in combination with the indicated specific type of human DC were 136±11 (control DC), 107±6 (control DC and Treg therapy), 418±98 (Treg-conditioned DC), 163±63 (Treg-conditioned DC, anti–PDL1-treated), and 279±77 (Treg-conditioned DC, isotype antibody treated) (each comparison, p = NS by ANOVA test). GVHD control mice uniformly developed a diffuse skin rash and hair loss; skin histology analysis documented cutaneous acanthosis and hyperkeratosis in GVHD controls, but not in Treg recipients (representative histology; Figure 5C (iii) and (iv), respectively). Furthermore, GVHD controls, but not Treg recipients, developed diffuse lymphocytic infiltration of the liver (representative histology; Figure 5C (i) and (ii), respectively).
The rational design of adoptive cell therapy protocols using ex vivo–expanded Tregs would be facilitated by an improved understanding of their cellular and molecular mechanism of action, which has been difficult to ascertain, particularly with respect to human Tregs. In this report, utilizing a novel method of generating human Tregs based on CD127 negative selection, we have elucidated a unique Treg mechanism of immune suppression analogous to previously described models of infectious tolerance [44] that is mediated at least in part by modulation of allogeneic dendritic cells through the PD-L1 pathway. This mechanistic understanding is particularly pertinent to efforts that will utilize Tregs for the prevention or treatment of GVHD, which is driven by allogeneic DC [30] and is amenable to suppression through PD-1 [45].
Our results are the first, to our knowledge, to describe a mechanism of human Treg action that involves potent in vitro and in vivo suppression of effector T cells through a secondary cellular messenger, myeloid dendritic cells. A similar biology has been described in murine models, whereby Tregs create a weak stimulator DC through induction of immunosuppressive indoleamine 2,3-dioxygenase (IDO) via a CTLA-4– or IFN-γ–dependent pathway [26]. Interestingly, the reverse biology has also been described in murine models, whereby murine plasmacytoid DC that produce IDO promote the generation of immunosuppressive Tregs that express PD-L1 [24]. Of note, in our experiments, inhibition of IDO by 1-MT treatment did not abrogate suppression mediated by Treg-conditioned myeloid DC (unpublished data). Similar to a previous study using freshly isolated human Tregs [29], we found that ex vivo–generated human Tregs inhibited myeloid DC secretion of the proinflammatory cytokines IL-6 and TNF-α (unpublished data) and induced a DC phenotype with greatly reduced capacity to induce responder T cell proliferation in vitro. Most importantly, we have significantly extended this prior work through our discovery that myeloid DC conditioned by Tregs were effective in vivo for the complete elimination of posttransplant lethal xenogeneic GVHD induced by effector T cells. In addition to this apparent DC-mediated mechanism of GVHD protection, other non-APC mechanisms are likely operative for the Tregs that we studied, because transplant cohorts that received Tregs had more robust protection against xenogeneic GVHD than recipients of Treg-conditioned DC (lowest CD4+ and CD8+ T cell engraftment, lowest posttransplant IFN-γ secretion, and lowest degree of weight loss posttransplant).
Furthermore, this is the first demonstration that human Tregs mediate immune suppression in vivo through modulation of the PD-1 pathway. First, we observed that ex vivo–expanded human Tregs expressed increased PD-L1 relative to control expanded CD4+ T cells. Second, Treg-conditioned DC expressed greatly increased PD-L1 relative to DC conditioned with control CD4 cells. As such, Treg PD-L1 appeared to directly induce DC PD-L1 expression; the potential existence of such a PD-L1 “positive feedback loop” adds to the known complexity of PD-1 pathway regulation [36] and to our knowledge has not been previously described for murine or human Tregs. Third, this feedback appeared to extend to the distal stage of effector T cell regulation because effector CD4+ and CD8+ cells under the influence of Treg-conditioned DC, but not control CD4-conditioned DC, had nearly universal expression of PD-L1 binding partners. Finally, we determined that such effector T cell binding to PD-L1 was preferentially mediated through PD-1 rather than the other receptor associated with this pathway, CD80. It is interesting to note that a recent study found that PD-1 expression on Treg cells in patients with viral hepatitis played a negative regulatory role for Treg cell function via limitation of STAT-5 phosphorylation [46]. In our experiments, the ex vivo–activated Treg cells expressed a high level of PD-1, yet were able to mediate potent suppression of effector T cells in vivo at relatively dilute Treg to effector T cell ratio; as such, it does not appear that the PD-1 pathway exerted a functionally significant down-regulatory effect on the Tregs utilized in our model. It is interesting to note that the Treg-conditioned DC did not express significant PD-1; it is thus possible that the capacity of this cell population to effectively prevent xenogeneic GVHD may reside in part on a limited susceptibility to PD-1–mediated suppression.
Importantly, this biology was functional in vivo because: (1) Treg-conditioned DC maintained expression of PD-L1 after adoptive transfer; (2) effector (Teff) cells up-regulated both PD-L1 and PD1 in vivo in the presence of Treg-conditioned DC; and (3) a significant proportion of this immune modulation was abrogated if Treg-conditioned DC were blocked with anti–PD-L1. Remarkably, the survival advantage conferred by Treg-conditioned DC was fully abrogated by anti–PD-L1. In sum, these data demonstrate that modulation of the PD-1 pathway represents a significant mechanism of action of ex vivo–expanded Tregs that involves an interaction between Tregs, DC, and effector T cells in an apparent positive feedback loop. Further experiments will be required to better understand this process of intercellular PD-1 pathway modulation. Potentially, the PD-L1 suppressor phenotype might be transferred from Tregs to DC and then to effector T cells by a process of trogocytosis [47], which results in the generalized transfer of cell membrane proteins, including costimulatory molecules [48]. However, because we observed that Tregs up-regulated DC expression of PD-L1, but not other cell surface molecules such as PD-1, we speculate that alternative mechanisms of intercellular regulation may be operational.
These findings have several implications for ongoing efforts to utilize ex vivo–generated Tregs for adoptive cell therapy. First, we have found that CD127− selection represents a suitable alternative to CD25+ selection for attempts to enrich for Tregs prior to ex vivo expansion; further experiments will be required to directly compare these two methodologies to determine whether such methods result in differential modulation of APC function through the PD-1 pathway. It is perhaps important to emphasize that the regulatory T cell or Treg-conditioned DC modulation of xenogeneic GVHD was robust because it occurred at the relatively low regulatory cell to effector cell ratio of 1∶20, which is considered to represent a physiologic ratio. Second, our demonstration that ex vivo–generated Tregs operate to a significant degree indirectly through allogeneic myeloid DC may help guide protocol design, particularly in the setting of allogeneic HSCT. One theoretical limitation to Treg cell therapy is the transfer of “contaminating” effector T cells or the conversion of Tregs to proinflammatory Th17 cells [49] that are known to induce GVHD [10]. The APC mechanism we have identified offers a solution to this potential limitation: that is, one could harvest host-type monocytes pretransplant and generate myeloid DC in a manner similar to the methods that we utilized, condition such DC with ex vivo–generated Tregs, and then transfer only the conditioned host DC prior to allogeneic HSCT. Such an approach would be analogous to that proposed for type II DC (DC2 cells) that promote Th2 cytokines and prevent murine GVHD [50]. Our results also indicate that the capacity of adoptively transferred Tregs to modulate GVHD may relate in part to the bioavailability of host-type myeloid DC. This consideration may have relevance to the choice of host conditioning for Treg protocols: predictably, non-myeloablative regimens may be favorable in this regard because such regimens would preserve host myeloid DC as a key secondary cellular mediator of the Treg therapy.
It should be stated that xenogeneic models of GVHD likely do not fully reflect the biology of clinical GVHD, and as such, the potential clinical implications of the findings in our model must be interpreted with caution. Specifically, the xenogeneic transplantation model that we utilized did not incorporate a human hematopoietic stem cell component, and as such, the potential effect of the regulatory T cells or Treg-conditioned DC that we evaluated on stem cell engraftment was not assessed. However, we found that human dendritic cell and effector T cell engraftment was persistent at the relatively late time points of day 25 and day 45 posttransplant, respectively; thus, it is possible that human hematopoietic progenitor cell engraftment would also be durable under the conditions that we evaluated. Such a possibility would be consistent with data emanating from murine models of MHC-disparate transplantation, which have found Treg adoptive transfer to augment allogeneic hematopoietic stem cell engraftment [7],[51]. Our findings relating to PD-1 pathway modulation may also hold clinical implications. Adoptive cell therapy using Tregs or Treg-conditioned DC may be conceptualized as a vehicle for PD-L1 delivery. Such a cell therapy approach may have immediate practical benefit for the treatment of the myriad of diseases that may benefit from PD-1 agonism [52]. That is, although an antibody-based method of PD-1 antagonism has already been investigated in phase I clinical trials [53], it is unclear whether agonistic PD-1 antibodies will be available or safely administered. Finally, the mechanisms we have identified will provide a rationale for monitoring PD-1 and PD-L1 expression on posttransplant T cells and DC as a biological marker for in vivo activity of the administered Tregs or Treg-conditioned DC; in addition, surface PD-L1 expression may be utilized as a marker to facilitate a functionally defined release criteria for the experimental cell therapy products.
In conclusion, ex vivo expansion of CD127-negatively selected CD4+ T cells yielded a human Treg product that inhibited alloreactivity in vitro and in vivo, in large part due to modulation of myeloid DC and a multifaceted promotion of the PD-1 pathway in Tregs, DC, and effector T cells. As such, we have identified two distinct cell therapy vehicles, Tregs and Treg-conditioned myeloid DC, each of which show promises as a novel approach to modulate human effector T cells through the PD-1 pathway.
Female RAG2−/−γc−/− mice were obtained from Taconic and utilized at 8–12 wk of age. Experiments were performed according to a protocol approved by the National Cancer Institute Animal Care and Use Committee. Mice were housed in a sterile facility and received sterile water and pellets. As in previously reported methods [31],[41], mice were injected with 0.1 ml of chlodronate-containing liposomes (Encapsula Nanoscience) for macrophage depletion and given low-dose irradiation (350 cGy).
X-VIVO 20 media was obtained from BioWhitaker and AB serum was from Gem Cell. CD4 microbeads were from Miltenyi Biotec. Sheep anti-mouse (SAM) IgG dynabeads were from Dynal. Anti-CD3, anti-CD28 coated tosyl-activated magnetic beads were manufactured as previously described [54]. Rapamycin was from Wyeth (Rapamune). Recombinant human (rh) IL-2 and rhIL-12 were from PeproTech, and rhTGF-β1, αTGF-β1, -β2, -β3, and purified αPD-L1 were from R&D Systems. All other antibodies (unless otherwise stated) were provided by BD Biosciences; anti-human Foxp3 APC was from eBioscience. Luminex kits for detection of IFN-γ and TNF-α were from Bio-Rad. 5-(and-6)-Carboxyfluorescein diacetate, succinimidyl ester [5(6)-CFDA, SE; CFSE] was from Invitrogen.
Normal donor peripheral blood cells were collected by apheresis on an IRB-approved protocol. Total lymphocytes were isolated by elutriation [55]. Total CD4+ T cells were then enriched by CD4 microbeads according to manufacturer instructions. To isolate CD127-depleted CD4+ T cells: (1) elutriated lymphocytes were adjusted to 100×106 cells/ml and incubated with anti-CD127 (10 µg/ml, 30 min, 4°C); (2) cells were washed, mixed with SAM dynabeads (bead∶cell ratio, 4∶1), incubated (30 min, 4°C), separated (hand-held magnet, Dynal); and (3) CD127-depleted cells were subjected to CD4 cell isolation by microbeads.
Total CD4+ and CD4+CD127− T cells were cultured in polystyrene tissue culture flasks (Corning). Cells were activated by anti-CD3, anti-CD28 costimulation (bead∶cell ratio, 3∶1), and cultured in X-VIVO 20 with 5% heat-inactivated (HI) AB serum containing rapamycin (1 µM), TGF-β1 (20 ng/ml), rhIL-2 (100 IU/ml). rhIL-2 alone was added at days 2, 4, and 6. Cultures were started at 1.5×106 cells/ml, maintained at 1×106 cells/ml through day 7, and then split daily to 0.5×106/ml by addition of IL-2 and rapamycin-replete medium through day 12.
T cells were washed with PBS supplemented with 0.1% BSA and 0.01% azide, and stained using anti-: CD4 PE-cy7 (clone S3.5; Caltag), Foxp3 APC (clone 249D; eBioscience), CCR7 PE (clone 150503; R&D), CTLA-4 Biotin (clone BN13), CD27 FITC (clone M-T271), and CD62L APC-cy7 (clone DREG-56; Biolegend). For intracellular (IC) flow cytometry, fixation and permeabilization buffer was utilized (eBioscience); four-color IC flow cytometry was performed with combinations of anti-: IL-2 biotin (clone B33-2), IFN-γ APC (clone B27), CD4 Pe-Cy5 (clone RPA-T4), and Foxp3 PE (clone PCH101; eBioscience). DC were evaluated using anti-: CD80 Bio (clone L307.4), CD86 APC (clone 2331), CD14 PE (clone M5E2), CD83 FITC (clone HB15e), CD40 APC (clone 5C3), and PDL1 PE-cy7 (clone MIH1).
Monocytes from four healthy, randomly selected donors were obtained by apheresis and elutriation; HLA typing confirmed that the donors did not share major haplotypes. Each monocyte population was cultured in X-VIVO 20 medium with 5% HI-AB serum, rhGM-CSF (50 ng/ml), and rhIL-4 (20 ng/ml). On day 5, each DC culture was enumerated and subjected to flow cytometry to document a DC phenotype (CD14−, CD11c+, CD83+, CD80+, CD86+; unpublished data). The four separate DC populations were pooled in equal proportions, and aliquots of the final product were cryopreserved and utilized for each experiment.
Normal donor lymphocytes (“responder T cells”; 2×105 cells) were cocultured with allogeneic DC (5×104 cells) in 96-well round-bottom plates (T cell to DC ratio, 20∶1). To detect proliferation, responder T cells were CFSE-labeled before coculture. From the same normal donors, Tregs were generated from CD4+CD127− cells, or as a control, from total CD4+ cells. Initial experiments determined that a Treg to responder T cell ratio of 1∶20 consistently yielded suppression of proliferation. Proliferation of CD4+ and CD8+ responder T cells was evaluated by CFSE dye dilution; percent suppression of CD4 and CD8 responder T cell suppression was calculated, with values representing the ratio of total divided peaks to both divided and nondivided peaks, normalized to the sham-treated experimental group.
During the MLR, neutralizing antibodies were added, including anti-: CTLA-4, IL-10, TGF-β1, TGF-β2, TGF-β3, LAP, and their respective isotope controls. All antibodies were used at 20 µg/ml. A combination of anti-CTLA-4, anti-TGF-β, and anti-LAP was also tested. 1-methyl-d-tryptophan (1-MT, 1 mM; Sigma) was utilized to inhibit indoleamine 2,3-dioxygenase (IDO). Transwell plates with a 4-mm membrane (Corning LifeSciences) were utilized to assess Treg contact dependency.
For the secondary transfer experiments, DC were incubated with Tregs for 48 h (Treg to DC ratio, 1∶1). Tregs were then removed using T cell–positive selection (anti-CD3 microbeads and subsequent magnetic column separation; Miltenyi); the resultant population was >99% pure for DC content, as determined by flow cytometry using CD11c in combination with CD80, CD86, and CD40. Such Treg-conditioned DC were then used as stimulator cells, with degree of proliferation determined relative to DC conditioned with control T cells (cells generated ex vivo from total CD4 cells) or sham-treated DC. MLR assays using preconditioned DC were also performed with anti–PD-L1 (20 µg/ml) or isotype control antibody.
On day 5, responder T cells were evaluated for expression of PD-L1 binding partners PD1 and CD80. The responder T cells were blocked with a specific αPD1 (1 µg/1×106 cells) and αCD80 (1 µg/1×106 cells) antibody and then PD-L1 binding was studied by incubation with recombinant PD-L1-Fc fusion molecule (R&D); secondary incubation was performed with FITC-labeled rabbit anti-human IgG, Fc-fragment antibody (Jackson Laboratory). Stained T cells were delivered to 96-well plates with a plastic #1 cover slip bottom (1×105 cells in 200 µl) and analyzed (iCys Laser Scanning Cytometer; Compucyte Corporation). Cells were scanned (488-nm laser) and fluorescence was detected (530/30-nm band-pass filter). Scan images and fluorescence data were generated (iGeneration and innovator software; Compucyte). Images were collected at 0.5-µm scan resolution.
Human effector CD4+Th1/CD8+Tc1 (Teff) cells were generated by T cell culture for 6 d by costimulation and expansion of T cells in rhIL2 (20 IU), αIL-4 (100 ng/ml), rhIL-12 (20 ng/ml), and rapamycin (1 µM). On day 6 of culture, Teff cells were harvested and injected (i.v. by retro-orbital method, as previously described [43]) into Rag2−/−γc−/− mice conditioned with chlodronate and radiation; Teff cell dose was either 1 or 3×107 cells/recipient (higher dose used for evaluation of posttransplant lethality). Specific cohorts additionally received ex vivo–generated Tregs (generated from CD4+CD127− cells) or control T cells (generated from total CD4 cells) at a dose of 0.5 or 1.5×106 cells/recipient such that the in vivo ratio of effector Teff cells to Tregs always matched that utilized in the allogeneic MLR (20∶1). As indicated, cohorts additionally received pooled allogeneic DC (complete mismatch as compared to Teff and Tregs) utilized in the MLR (DC dose, 0.5 or 1.5×106 cells/recipient) to maintain constant ratios; allogeneic DC were either not conditioned or conditioned with ex vivo–generated Tregs or control T cells. For blocking experiments, conditioned DC were incubated with anti-PD-L1 (20 µg/ml) or isotype control antibody prior to adoptive transfer. In some cases, anti-PD-L1 was injected following cell transfer (i.p.; 100 µg/recipient). After adoptive transfer, human engraftment was calculated using flow cytometry data from splenic single-cell suspensions (% huCD45+ = [huCD45+ (huCD45++ mCD45+)]×100%). Surface or intracellular flow cytometry was performed at indicated days after adoptive transfer to assess in vivo modulation of responder human CD4 and CD8 T cells and human DC.
Flow cytometry and cytokine data were analyzed using Student 2-tailed t-tests. Comparison values of p<0.05 were considered statistically significant. Survival was determined using Kaplan-Meier test. For three pairwise cohort comparisons, statistical analyses was performed using the Holm method [56].
|
10.1371/journal.pntd.0006077 | Low cost, low tech SNP genotyping tools for resource-limited areas: Plague in Madagascar as a model | Genetic analysis of pathogenic organisms is a useful tool for linking human cases together and/or to potential environmental sources. The resulting data can also provide information on evolutionary patterns within a targeted species and phenotypic traits. However, the instruments often used to generate genotyping data, such as single nucleotide polymorphisms (SNPs), can be expensive and sometimes require advanced technologies to implement. This places many genotyping tools out of reach for laboratories that do not specialize in genetic studies and/or lack the requisite financial and technological resources. To address this issue, we developed a low cost and low tech genotyping system, termed agarose-MAMA, which combines traditional PCR and agarose gel electrophoresis to target phylogenetically informative SNPs.
To demonstrate the utility of this approach for generating genotype data in a resource-constrained area (Madagascar), we designed an agarose-MAMA system targeting previously characterized SNPs within Yersinia pestis, the causative agent of plague. We then used this system to genetically type pathogenic strains of Y. pestis in a Malagasy laboratory not specialized in genetic studies, the Institut Pasteur de Madagascar (IPM). We conducted rigorous assay performance validations to assess potential variation introduced by differing research facilities, reagents, and personnel and found no difference in SNP genotyping results. These agarose-MAMA PCR assays are currently employed as an investigative tool at IPM, providing Malagasy researchers a means to improve the value of their plague epidemiological investigations by linking outbreaks to potential sources through genetic characterization of isolates and to improve understanding of disease ecology that may contribute to a long-term control effort.
The success of our study demonstrates that the SNP-based genotyping capacity of laboratories in developing countries can be expanded with manageable financial cost for resource constraint laboratories. This is a practical formula that reduces resource-driven limitations to genetic research and promises to advance global collective knowledge of infectious diseases emanating from resource limited regions of the world.
| Although genetic characterization of pathogenic organisms is a powerful tool for investigating outbreak origins and transmission, associated high upfront costs and demanding technological maintenance exclude this tool for many under-resourced laboratories. Paradoxically, resource constrained regions commonly suffer from high rates of infectious diseases and could benefit most from genetic tracking tools. One such country is Madagascar, which lacks resources to acquire high tech genetic typing equipment, yet suffers from seasonal human plague outbreaks. A serious disease, plague is caused by the clonal bacterium, Yersinia pestis, and is capable of causing human outbreaks. Using plague as a model organism, we developed a genetic typing method that requires only basic, widely used molecular machinery. Our tools target unique single mutations in the Y. pestis genome to assign isolates to distinct phylogenetic groups with known geographical distributions. Transfer of this technology to Madagascar permits genetic characterization of strains from current outbreaks. This eliminates the need for external genetic analysis and expands the research capacity of this resource-constrained laboratory by allowing rapid, in-house strain typing. Ultimately, our goal is to help improve the ability of local institutes to genetically characterize circulating strains, link outbreaks to originating sources, and improve our understanding of the ecology of tropical diseases in resource-limited regions of the world.
| Single nucleotide polymorphisms (SNPs) are highly valuable genetic markers that have advanced our knowledge of diverse biological fields such as human health [1,2], infectious disease epidemiology [3–5], agriculture [6], and ecology [7], among others. In the study of infectious diseases SNPs can be informative of bacterial phenotype, such as antibiotic susceptibility [8,9], and also can be used to classify unknown strains. For non-recombining bacterial pathogens, most of their SNPs become fixed in the genome and are faithfully replicated throughout future generations [3,10]. These stable signatures can be used to classify unknown strains into known phylogenetic groups according to SNP profiles [3,11,12]. Within the context of epidemiological investigations, these SNP profiles can link isolates from active outbreak sites to possible sources and help track disease transmission patterns [5,13,14].
Genotyping assays that use real-time PCR to identify single SNPs remain in demand despite the wide-scale availability of whole genome sequence (WGS) data and continued reductions in WGS costs. For many research facilities that are interested in small-scale studies or face resource limitations, a WGS-based approach to SNP typing is not a feasible nor a desirable option. A variety of other technological platforms have been employed for SNP typing and have been extensively described in several publications [15–18]. Popular platforms for SNP typing use real-time PCR instruments that employ Dual Probe TaqMan assays or melt-MAMA SNP assays [19–21]. But real-time platforms are not commonly available in resource constrained laboratories, due to their high upfront costs and the need for ongoing highly technical instrument maintenance. However, a more simplified method for SNP genotyping that employs conventional PCR coupled with standard agarose gel electrophoresis (agarose-MAMA) is a viable alternative in these settings. The advantage of this alternative method is that it utilizes relatively inexpensive instruments that are almost universally available even in developing nations where it is used for a variety of molecular applications. Much of this is due to the simplicity of the agarose gel electrophoresis platform, in contrast to the complex instrumentation of the real-time platform [20,22].
To illustrate the effectiveness of agarose-MAMA as a SNP genotyping tool in resource constrained laboratories, we developed Y. pestis assays for use at the Institut Pasteur de Madagascar (IPM). Y. pestis is the bacterium infamously known as the causal agent of the disease plague. Y. pestis is ecologically established on nearly every inhabited continent [12,23] and remains a particularly significant threat to human health in developing nations in Africa and especially the island country of Madagascar [24,25]. Primarily a zoonotic agent, Y. pestis has a complex ecological cycle involving rodent-host populations and flea vectors. In unfortunate circumstances, humans are incidental hosts [26]. Without prompt antibiotic treatment, human death rates vary from 30–60% to 100% depending on the route of exposure, with pneumonic plague being the most deadly [24,25]. Within the last two decades, Madagascar has reported some of the highest incidences of human plague infections throughout the world [24]. Given that Y. pestis is a re-emerging public health threat in Madagascar [27–29] and other developing African nations, performing SNP typing studies through the use of agarose-MAMA tools could lead to greatly improved epidemiological investigations and to more effective disease management.
Here, we describe how we resolved technological limitations that prevented Y. pestis SNP genotyping studies at IPM by: 1) re-designing a SNP-based Y. pestis genotyping system (from melt-MAMA real-time platform to agarose-MAMA) to be compatible with existing resources in Madagascar, and 2) validating these genetic tools at IPM for adoption. Collectively, these achievements have removed a research barrier at IPM once imposed by technological disparities and serve as a promising model for building similar research capacities in other developing nations affected by plague and/or other pathogens common to low-resource settings. In addition, these tools generate data that can be compared to other sequence-based methods and can feed into existing global databases. This ability can strengthen scientific exchange between affected countries and the international scientific community, thereby advancing the global collective knowledge of dangerous infectious diseases.
We present a SNP genotyping tool that uses standard genetic equipment and reagents that are accessible to nearly any laboratory. The successful development of agarose-MAMA tools removes the dependence on real-time instruments for conducting SNP-based epidemiological studies. Due to emerging infectious diseases being most globally prevalent in resource challenged countries [30,31], building research capacities in these regions has been a goal for the World Health Organization and other agencies charged with global biosafety, response, and biosecurity efforts [32,33]. It is now possible for researchers in resource constrained locations to conduct SNP studies to obtain bacterial phenotype information such as antibiotic resistance or phylogenetic information to more robustly understand the dynamics of disease transmission and potentially identify sources of human infections.
The phylogeny of Y. pestis in Madagascar has recently expanded [34] to include more phylogenetic groups (Fig 1) and the 18 assays developed for this study define a subset of these phylogenetic groups or lineages as illustrated on a simplified phylogeny (Fig 1). The genotyping results generated from each agarose-MAMA was identical to independent SNP genotyping technologies [34] when tested across the same diverse panel of 16 Y. pestis DNA strains. Assay specificity remained intact when these agarose-MAMAs were tested on diverse types of negative controls (high levels of human, Leptospira spp., and Bacillus anthracis DNA, or a no template water control). When the assays were employed in a stepwise, hierarchical order (Fig 2), each isolate could be assigned to a single lineage or one subgroup as depicted in the simplified Malagasy Y. pestis phylogeny (Fig 1). The equivalent performance of our agarose-MAMA tools to melt-MAMAs [11] and other independent SNP technologies [34] demonstrate that MAMA tools preserve their genotyping accuracy independent of the technology platform used.
For the 18 selected SNPs, we were able to successfully design agarose-MAMAs to achieve 100% genotyping accuracy through a few key optimization steps. Only one of our assays failed initial optimization efforts; however, when we redesigned the assay to target the reverse complement of the reference template DNA (strain CO92), the assay was rescued to full functionality. Essential validation steps included the identification of the optimal ratios for the concentration of the two forward primers (ancestral:derived) as described [20], ideal annealing temperature for each assay, appropriate number of PCR cycles, and an occasional need for MgCl2 concentration alteration in the reaction mix. These optimization strategies worked well on 8 other bacterial pathogens [20] and should be applicable to many other pathogenic organisms as well.
The GC-clamp (21–28 oligo base), added to derived MAMA forward primers but excluded from ancestral MAMA forward primers, was sufficient to provide visible size differences between the two allelic states when visualized on an agarose gel (Fig 3). A repeated pattern of 5’-cgggttcgggttcgggttcgggttcggg-3’did not appear to induce non-specific binding nor interact with template DNA in an inhibitory manner. However, we observed that the GC-clamp on the derived MAMA primer did frequently confer a competitive advantage over the ancestral MAMA, as previously described for MAMA tools [20]. This competitive advantage resulted in cross-hybridization of primers for a subset of assays. This phenomenon may be based on the ability for GC rich DNA regions to anneal at lower temperatures compared to GC poor regions. The GC-clamp of the derived MAMA primer likely anneals to its target amplicon at lower temperatures compared to the no-clamp ancestral primer. As a consequence, the GC-clamp primer anneals to the template at an earlier time point per PCR cycle than the no-clamp primer, resulting in a competitive kinetic advantage. To correct for this competitive advantage, we followed published guidelines [20] by increasing the ancestral primer concentration relative to the derived primer concentration of the affected assays. In these skewed reaction mixes, the ancestral primer had >1x-4x concentration compared to its paired derived primer. We have detected best results with 4:1 and 2:1 (ancestral:derived) primer ratios, depending on the assay.
We obtained maximal assay specificity by a combination of customizing the annealing temperature per assay and/or adjusting the MgCl2 concentrations. Most of our assays accurately genotyped SNP alleles at annealing temperatures around 60°C; however, some assays had non-specific product at this condition. For these assays, non-specific PCR products were present in our negative controls and/or in our sample PCR product, visualized as extra banding on a gel. A reduction of MgCl2 concentration by 0.5 mM was sufficient to greatly increase the specificity of most assays, as evidenced by the elimination of non-specific amplification. But this reduction in MgCl2 decreased the PCR robustness in a subset of assays. To address this loss, the annealing temperatures were reduced in affected assays while still preserving SNP specificity. For the few assays that did not respond to lowered MgCl2 strategy, the raising of annealing temperatures while still maintaining the normal 2.0 mM concentration of MgCl2 yielded improvements in reducing non-specific products.
The assays described here are capable of genotyping directly from complex clinical samples if pathogen DNA levels are sufficient (Fig 4). This capability is highly important as nearly 43.32% of 775 F1 RDT [36] positive human plague cases between 2011–2015 in Madagascar and many environmental samples (rodents and fleas) do not yield live isolate culture (IPM records). Using two agarose MAMA tools (Mad-05 and Mad-43) to demonstrate proof of principal, we were able to determine that the three complex clinical samples belonged to group I lineage and not group II. This genetic assignment for these samples matched the results of a recent publication [34]. These samples were not further tested on additional agarose-MAMAs. Out of five complex clinical samples positive for the high copy plasmid pla gene [37] only four were positive for Y. pestis chromosomal DNA, assessed by a TaqMan assay targeting the 3a gene (Fig 4 and S3 Appendix). Of the four 3a-positive samples, one isolate (Yp3182) gave a late amplification with the 3a assay resulting in a Ct value of 36 using real-time PCR. This indicates a very low concentration amount for this clinical extract, near a single copy of pathogen DNA [20,21]. The other three samples (Yp2483, Yp2486, Yp2485) (Fig 4) generated Ct values ranging from 24–27, which indicates higher concentrations of Y. pestis chromosomal DNA in these clinical samples. These more concentrated Y. pestis clinical samples were successfully genotyped by agarose-MAMA tools following an increase in the number of PCR cycles. The low-level sample (Yp3182) failed to amplify PCR product using agarose-MAMA tools. Together, these results indicate that agarose-MAMAs can genotype directly from complex clinical samples if the pathogen target (Y. pestis DNA) is of sufficient concentration. Published studies show that TaqMan assays are highly sensitive [20,21] and can readily detect minute amount of template not detectable using melt-MAMAs [20]. Our results suggest that the same is true for agarose-MAMAs (Fig 4).
Following the transfer of our agarose-MAMA tools to the IPM facility, a subset of our agarose-MAMAs was validated for genotyping accuracy. The differing instruments and reagents used at IPM introduced very little variability on assay performance when tested on the same DNA panel previously used at NAU (Fig 5). The optimized PCR conditions identified at NAU for the assays were suitable for most of the assays when conducted at IPM.
The most apparent difference in assay performance between the two laboratories was an increase in non-specific amplification at high fragment size (Fig 5) and, for some assays, also in our negative controls at IPM, which was not observed at NAU. These non-specific fragments did not affect the MAMA’s capability to accurately genotype isolates. Band profiles were not distinctly in line with either the ancestral or derived product but rather appeared either as a smear across both profiles or as a much fainter band falling between the ancestral and derived fragment sizes. We suspect that the basis of this performance difference is due to different Taq polymerases used between the two institutions. NAU employed an antibody-immobilized Taq polymerase (Invitrogen, Carlsbad, CA), which has been characterized as having no polymerase activity prior to a hot-start step in the thermal cycle protocol [38]. However, IPM utilizes regular Taq polymerase, which may begin product synthesis with primers and template in the master mix prior to PCR thermal cycling [38]. Although master mix preparation was done on ice to suppress premature Taq polymerase activity, additional bands above the target PCR product and faint bands in the NTC amplification suggest that pre-PCR Taq activity was not completely suppressed (Fig 5). We therefore reduced the concentration of MgCl2 in five of our assays (Mad-43, Mad-46, Mad-12, Mad-36 and Mad-58) to 1.5mM and observed the elimination of amplification in the negative controls in most of the affected assays. Reduction of MgCl2 also imparted a general improvement of assay specificity for positive control and test isolates. Surprisingly, the reduction of MgCl2 did not necessitate a corresponding increase in the number of PCR cycles for most assays, contrary to the results observed in our laboratory at NAU. Once again, this may be a result of differences in Taq polymerase activity between the two institutions although we have not found any published evidence of this occurring elsewhere. As a way to rule out possible contamination as a source of non-specific banding, all surfaces used for the preparation of PCR reaction mixtures were sterilized with UV light for 15 minutes and decontaminated with 70% ethanol. Following slight modifications, we found that results of the majority of our assays at IPM aligned very well with results produced at NAU (Fig 5 and S5 Appendix).
Our success in developing agarose-MAMA tools and transferring them to IPM facility in Madagascar demonstrates that this SNP genotyping strategy can be achieved with existing technologies routinely used in developing nations (Fig 5). The transferred agarose-MAMA technology is now in regular use at IPM and was recently used to infer the source of a 2015 pneumonic outbreak [39]. There is ample evidence that this same assay design strategy would transfer well to many other pathogenic organisms [20]. This would allow institutions in developing countries to perform molecular studies in-house and on local infectious organisms. The current methods largely used for short-term control efforts are presence/absence assays (PCR based [40] or protein based [36]) that diagnose the causative agents of disease outbreaks but provide no genetic resolution. Having in-house capabilities to genetically discriminate goes beyond what presence/absence assays can provide, therefore, resource-constrained laboratories will be able to advance their epidemiological capabilities over the status quo. The SNP data they generate locally can be shared among research institutes and compared to existing global databases. These advanced capabilities would accelerate the understanding of plague ecology, persistence, and evolution; which in turn could beneficially inform strategies for disease control.
Building research capacity in low-resource settings is an important endeavor for global biosafety, response, and biosecurity preparedness [32,33]. Our study is a successful model of achieving this goal pragmatically. We successfully developed and transferred genetic tools to a developing nation. The design principles for this technology, which we detail above, can be applied to diverse pathogen species [20]. The use of MAMA technology will give the scientific community the means to gain insight into the genetic patterns and population structure of many neglected diseases. This is a practical formula that will advance global collective knowledge of infectious diseases emanating from more impoverished regions of the world.
The 16 archival strains and five clinical samples were not subject to IRB regulations because they did not meet the federal definition of human subjects research according to 45 CFR 46.102 (f). All samples underwent de-identification of patient information prior to Northern Arizona University (NAU) transfer. They were collected as part of the medical workup mandated by the Ministry of Health in Madagascar and not for the purpose of this study. For this reason the strains used in this study does not meet the federal definition of human subjects research according to 45 CFR 46.102 (f) and therefore are not subject to review from NAU Institutional Review Board.
DNA samples utilized in this study were obtained from 16 Y. pestis isolates and 5 clinical samples (bubo aspirates and sputum) collected from suspected human plague cases (Table 1). The 16 archival strains and 5 clinical samples (Table 1) originated from diverse geographic locations in Madagascar and were collected as described in a recently published study [34]. Our assays worked well on DNA concentrations that ranged from 1 ng to100 pg. Molecular confirmation of Y. pestis in five clinical samples was based on PCR detection of the pla gene located in a high copy number plasmid PCP1 in addition to positive F1 RDT [28,36,37,41,42].
Eighteen previously published SNPs [11,34,35] specific, or canonical [3], for a subset of distinct phylogenetic groups within the Y. pestis Malagasy phylogeny (Fig 1) were selected as the targets of agarose MAMA genotyping assays following published guidelines [20]. These selected SNPs can be used in a hierarchical way (Fig 2; Table 2) to assign an unknown strain to one of the most common lineages or phylogenetic subgroups in Madagascar [11,34,35,43].
The common reverse primer and two forward allele-specific MAMA primers for each assay were designed using NetPrimer analysis software (Premier Biosoft, Palo Alto, CA). One forward primer represents the original SNP allele, referred to as “ancestral” and the other represents the mutated SNP allele, referred to as “derived”. The MAMA primers for each assay were designed to compete for the same SNP locus on the template and the resulting amplicon product is generated by the allele-specific MAMA primer that most closely matches the template.
To differentiate between the amplicon products of the derived and ancestral genotypes, additional length was added on the derived amplicon product but not the ancestral product. This was achieved by adding 21–28 oligonucleotides rich in GC content at the 5’end of the derived MAMA forward primer (GC-clamp) (Fig 3). Since the primers are incorporated in the final amplicon product, the addition of the GC-clamp on the derived MAMA forward primer resulted in derived amplicon products that were 21–28 bp longer than their ancestral amplicon counterparts.
To maximize the visible size differences between the two allelic-specific amplicons when viewed on an agarose gel, the size of the PCR amplicon was restricted to ≤ 80 bases total length. Amplicons within ~80 bases show the greatest migration difference on a gel when small size differences exist. This is the case with our derived and ancestral allele-specific PCR products, which differ between 21–28 bases (Fig 3, Table 3, S2 Appendix). Additionally, through the use of the GC-clamp, our assays retained the capability of SNP genotype discrimination on a real-time PCR platform, which is based on differential melt-curve properties of each SNP-specific PCR product [20].
Initial PCR conditions were identified at NAU. PCR conditions for different assays varied and are described in Table 3. PCR amplification per assay was carried out in 20 μL volume with the following reagents (see S1 Appendix for volume of each reagent): for one reaction, 1x PCR buffer without MgCl2, MgCl2 range of 1.5–2.5 mM, 0.30 mM deoxynucleoside triphosphate, 1.6 units of platinum Taq DNA polymerase (Invitrogen, Carlsbad, CA), both sets of forward MAMA primers (derived and ancestral allele-specific) with one common reverse primer at 0.40 μL each (for a 1:1 ratio), molecular grade water to achieve 18 μL total volume and 2.0 μL of diluted DNA template at ~1ng/μL per reaction. We did not directly test our assays on genomic DNA concentrations below 100 pg but previously published work suggests that the MAMA approach is sensitive to DNA amounts below 100 pg [20]. For each set of reactions, at least one of each ancestral and derived allele templates were used as positive controls as well as at least two no-template controls (NTC). Thermal cycling parameters for the eighteen assays are as follows: initial denaturation at 94°C for 5 min followed by 30–40 cycles of 94°C for 30 s, 51°C -67.3°C (Table 3) for 30 s, and 72°C for 30 s, with a final extension at 72°C for 5 min. All PCR amplifications were performed with a MJ Research PTC 200 thermal cycler (BioRad, Hercules, CA).
Conditions of agarose gel electrophoresis for PCR amplicons included adding 4 μL of 6x loading dye 0.25 w/v xylene cyanol FF and 30% v/v glycerol, water (Thermo Fisher scientific, Waltham, MA) to individual PCR products to achieve a 1x final dye concentration. Individual reactions (20 μL) mixed with loading dye were loaded onto a 2% agarose gel matrix; 100 bp DNA ladder (Invitrogen, Carlsbad, CA) was used for size referencing. Gels were prepared in 1x lithium borate buffer (S2 Appendix) and stained with SybrSafe dye (Life Technologies, Carlsbad, CA). Electrophoresis was conducted at 300V for 25–30 minutes and viewed under UV transillumination.
The genotyping accuracy of our SNP assays was validated using positive DNA controls that represented known ancestral and derived allele states for each SNP target. Assay accuracy was further assessed by testing them on 16 genomic DNA extracts of Malagasy Y. pestis strains belonging to known diverse phylogenetic groups (Table 1) based on amplicon sequencing [34]. To assess assay capability to genotype Y. pestis directly from complex DNA samples (containing high levels of host DNA), we tested the performance of each assay on a positive control sample comprised of high concentrations of human DNA with our positive control Y. pestis DNA extract (strain A1122). To further assess this capability, we tested two assays (Mad-05 and Mad-43) on five human complex samples confirmed to be plague positive. This confirmation was based on Y. pestis-specific TaqMan assay targeting a high copy plasmid pla gene [37]. We assessed the limit of detection of our MAMA tools on these five human clinical samples by testing them on a new TaqMan assay designed for the chromosomal 3a gene (Fig 4 and S3 Appendix).
To confirm the specificity of the assays to the Y. pestis genome we tested assay performance on DNA extracts of Bacillus anthracis, human DNA background, and no template water control. To compare assay performance at different research facilities, NAU and IPM jointly conducted a second validation study on the IPM laboratory premises using IPM PCR reagents and instrument (Fig 5). Thirteen of the eighteen assays were selected for validation (Table 3). At IPM, specificity was confirmed by testing on Leptospira interrogans serovar Canicola and no template controls. PCR was prepared for each assay using the Taq Core Kits 10, Cat# EPTQK300 PCR reagents (MP Biomedicals, Santa Ana, CA) and amplification was conducted using AB Applied Biosystems Veriti 96 Well Thermal Cycler thermal cycle (ThermoFisher Scientific, Waltham, MA). Electrophoresis was performed using 2% agarose gels visualized on a Gelscan (Bio-Rad, Hercules, CA).
|
10.1371/journal.pcbi.1004765 | A Multi-Method Approach for Proteomic Network Inference in 11 Human Cancers | Protein expression and post-translational modification levels are tightly regulated in neoplastic cells to maintain cellular processes known as ‘cancer hallmarks’. The first Pan-Cancer initiative of The Cancer Genome Atlas (TCGA) Research Network has aggregated protein expression profiles for 3,467 patient samples from 11 tumor types using the antibody based reverse phase protein array (RPPA) technology. The resultant proteomic data can be utilized to computationally infer protein-protein interaction (PPI) networks and to study the commonalities and differences across tumor types. In this study, we compare the performance of 13 established network inference methods in their capacity to retrieve the curated Pathway Commons interactions from RPPA data. We observe that no single method has the best performance in all tumor types, but a group of six methods, including diverse techniques such as correlation, mutual information, and regression, consistently rank highly among the tested methods. We utilize the high performing methods to obtain a consensus network; and identify four robust and densely connected modules that reveal biological processes as well as suggest antibody–related technical biases. Mapping the consensus network interactions to Reactome gene lists confirms the pan-cancer importance of signal transduction pathways, innate and adaptive immune signaling, cell cycle, metabolism, and DNA repair; and also suggests several biological processes that may be specific to a subset of tumor types. Our results illustrate the utility of the RPPA platform as a tool to study proteomic networks in cancer.
| Pan-cancer proteomic datasets from The Cancer Genome Atlas provide a unique opportunity to study the functions of proteins in human cancers. Such datasets, where proteins are measured in different conditions and where correlations are informative, can enable the discovery of potentially causal protein-protein interactions, which may in turn shed light on the function of proteins. However, it has been shown that the dominant correlations in a system can be the result of parallel transitive (i.e. indirect) interactions. A wide suite of computational methods has been proposed in the literature for the discrimination between direct and transitive interactions. These methods have been extensively tested for their performance in gene regulatory network inference due to the prevalence of mRNA data. However, the understanding of the performance and limitations of these methods in retrieving curated pathway interactions is lacking. Here, we utilize a high-throughput proteomic dataset from The Cancer Genome Atlas to systematically test different families of network inference methods. We observe that most methods are able to achieve a similar level of performance provided their parameter space is sufficiently explored; but a group of six methods consistently rank highly among the tested methods. The protein-protein interactions inferred by the high-performing methods reveal the pathways that are shared by or specific to different cancer types.
| The Cancer Genome Atlas (TCGA) Research Network has recently profiled and analyzed large numbers of human tumors both within and across tumor lineages to elucidate the landscape of cancer associated alterations at the DNA, RNA, protein, and epigenetic levels [1]. Integrated analyses of the resulting rich genetic and epigenetic data types have already started to shed light on commonalities, differences and emergent themes across tumor lineages [2,3]. Analysis of TCGA samples from 11 tumor types indicated that whole protein and phosphoprotein levels in these tumors, as measured by antibodies on reverse phase protein arrays (RPPA), capture information not available through analysis of DNA and RNA [4]. The RPPA platform used in the Akbani et al. analysis was subsequently expanded to include 187 high-quality antibodies, 51 of which are phosphospecific and 136 of which are non-phosphospecific. These antibodies were selected with a focus on cancer-related pathway and signaling events and analyzed with the intent to discover new therapeutic opportunities. This dataset is available for download from The Cancer Proteome Atlas [5], and referred to as PANCAN11 from here on.
The availability of proteomic datasets such as PANCAN11 where protein levels are measured across different conditions provides a unique opportunity to study the functions of proteins. However, the analysis of function requires knowledge of interactions. For instance, in the protein-folding domain, the function of a single residue during folding can be determined only by having knowledge about the residues it is interacting with. Similarly, the function of a protein in the cell can only be understood by determining its interaction partners. Therefore, the units of analysis are not the individual protein expression levels, but the interactions of proteins with other cellular entities.
Statistical techniques such as correlation can be used to study the interactions of proteins. However, correlation between two proteins does not imply that they directly interact, because correlation may also be induced by chaining of correlation between a set of intervening, directly interacting proteins. Such indirect correlations are called transitive interactions. It was previously shown that the dominant correlations in a system can be the result of parallel transitive interactions [6].
There are three main network motifs that lead to transitive interactions: fan-in, fan-out and cascade. A fan-in is a case where there are direct interactions from proteins A and B to a third protein C but there is no interaction between A and B. A fan-out is the situation where there is a direct interaction from protein C to both A and B but there is no interaction between A and B. A cascade, on the other hand, is a chain event where there are direct interactions from A to B, and from B to C, but not from A to C. In all these three cases, if the two direct interactions are in the same direction (both increasing or both decreasing), there is a transitive influence observed between the proteins that do not have a direct interaction. Since biological pathways and signaling events contain many fan-in, fan-out and cascade network motifs, transitive effects occur widely across the network and have previously been shown to be a systematic source of false positive errors for many computational network inference methods [7]. Thus, it is crucial to minimize transitive interactions when building network models from high-throughput datasets.
A wide suite of computational methods has been proposed in the literature for the identification of direct interactions in networks. The common objective of many of these methods is to call a direct interaction between two entities if they are ‘not conditionally independent’ of each other given a set of other entities. One simple example is the regression-based partial correlation approach. Consider a three-variable system consisting of A, B, and C. When testing the existence of a direct interaction between A and B in this approach, measurements on A and B would first separately be regressed on the measurements on C, the residual vectors would be computed, and then the correlation between the residual vectors would be found. If this ‘partial’ correlation is significantly different from zero, a direct interaction is called between A and B.
Despite the similarity in the objective, these methods employ diverse inference procedures such as mutual information [8–11], regression [12–14], Gaussian graphical models [15,16], and entropy maximization [17,18]. The diversity of algorithms for inferring direct interactions, coupled with the absence of a robust off-the-shelf method, creates challenges for investigators that aim to generate hypotheses and eventually discover novel functional interactions among proteins. We address this challenge by testing different families of network inference methods towards the goal of deriving guidance for the better-performing methods.
The RPPA platform, first introduced in Paweletz et al. [19] stands a good chance of becoming a widely used proteomics platform as greater numbers of reliable antibodies are being developed. Here, we present a rigorous comparison of the performance of 13 commonly used network inference algorithms based on PANCAN11, a pan-cancer RPPA dataset which contains levels of many proteins in a large number of samples, such that reasonably meaningful protein-protein correlations can be computed. The goal of this comparison is two-fold: To investigate 1) if the signal-to-noise ratio of the RPPA technology allows the discovery of known and novel protein-protein interactions (PPIs), and 2) to what extent algorithms that were originally developed for gene regulatory network inference accomplish the inference of PPIs.
Performance evaluation of PPI network inference for different cancers requires a ‘gold standard’ for each cancer type. However, a true gold standard for human PPIs does not exist, let alone a separate one for each tumor type. Most protein interactions in in vivo systems remain unknown or unproven and/or depend on physiological context. Yet public knowledgebases that store collections of curated pathway and/or interaction data contain useful information. For instance, Pathway Commons is a collection of publicly available and curated physical interactions and pathway data including biochemical reactions, complex assembly, transport and catalysis events [20], aggregated from primary sources such as Reactome, KEGG and HPRD and conveniently represented in the BioPAX pathway knowledge representation framework [21–24].
In this study, we adopted Pathway Commons as a benchmark, and evaluated the performance of 13 network inference methods (Table 1) in their capacity to retrieve ‘true’ PPIs from RPPA datasets of 11 cancer types. We then used a group of high-performing methods to investigate the similarities and differences among the 11 cancer types in our dataset. The workflow of this study (Fig 1), involves the parallel generation of two PPI network models, one from computational inference and one from the pathway knowledgebase. On the inference side, multiple antibodies are assayed on an RPPA platform (Step 1) and the resulting dataset is normalized to generate a proteomic profile of the cohort such as PANCAN11. Computational network inference methods are then employed to create a network model with the inferred PPIs (Step 2). On the knowledgebase side, various wet-lab experiments are performed to generate data, and the resulting information is stored in the scientific literature (Step 3). Curators sift through the literature to distill multiple-layered information on PPIs (Step 4), and then this information is catalogued in knowledgebases such as Pathway Commons (Step 5). A comparison of the PPI network models from the two sides reveals the level of fidelity at which the ‘true’ network is constructed by the computational methods (Step 6).
The mutual information-based methods listed in Table 1 infer unsigned undirected edges, whereas the edges inferred by the ‘correlation’ and ‘partial correlation’ family of methods are undirected but signed (positive or negative). An undirected positive edge between A and B means that the direction of influence is not known, but A and B generally exist at both high or both low levels among all tested experimental conditions. A negative sign, thus, means that one is generally high when the other is low (tending towards mutual exclusivity). The use of positive/negative edges in this study refers to positive/negative signs of the weight. Also, we use the words ‘edge’ and ‘interaction’ interchangeably throughout the manuscript.
The workflow of performance evaluation as described above involves certain caveats. These are discussed in detail in the Discussion section and in S1A Text. Here we discuss one of the caveats, the ascertainment bias in pathway knowledgebases (Step 5 in Fig 1). Wet-lab experiments for PPI plausibly have over-representation of certain proteins due to the perceived interest in the field and ease of study. In a recent paper, a Pearson correlation of 0.77 was reported for the correlation between the number of publications in which a protein was mentioned and the number of interactions reported for that protein in literature-curated data [25]. This implies the potential existence of an ascertainment bias in pathway knowledgebases. More documented interactions of a certain protein will exist if that protein is studied more intensively by the community. The ascertainment bias in Pathway Commons precludes our benchmark network from being a perfect gold standard. This and other caveats challenge the comparability of pathway models from a knowledgebase and network models from a computational algorithm. Thus, it is necessary to be mindful of these caveats when interpreting the performance evaluation results in this study.
In this study, we evaluated the performance of 13 different network inference methods on the PANCAN11 RPPA dataset by using Pathway Commons as a benchmark. The PANCAN11 dataset is comprised of 3,467 samples and 187 antibodies. The total number of possible non-self interactions with 187 antibodies is 17,391. However, the number of interactions in Pathway Commons (version 2) involving any two antibodies from this set of 187 is 1,212 as determined by PERA[26] (Methods). This Pathway Commons benchmark subnetwork of 1,212 interactions forms the gold standard for this study, and has only 162 of the 187 antibodies as interaction partners (meaning interactions between the remaining 25 antibodies and any one of the full set of 187 antibodies are not included in Pathway Commons). As 162 antibodies can form a total of 13,041 non-self interactions, the gold standard network has a density of 9.29% (1212/13041).
We obtained network predictions for 11 tumor types listed in Table 2 by using the 13 network inference methods listed in Table 1. We employed precision–recall curves to first find the optimal parameter values for each method, and then to compare the performance of methods using their optimal values. The precision-recall (PR) curves were constructed by first ranking an edge list based on significance, and then plotting precision and recall on the y and x axis respectively for cumulatively increasing numbers of the top (the most significant) edges from the list. The trade-off between precision and recall at different cutoffs gives a reliable idea about the performance of a method, and this performance can be quantified with the area under the precision-recall curve (AUPR).
The performance comparison for 13 methods was done separately for each tumor type. For a given tumor type, our procedure involved two steps. In the first step, we aimed to put all methods on an equal footing by finding each method’s optimal parameter values. This was achieved by running each method multiple times with different parameter values obtained from a one- or two-dimensional grid, computing the AUPRs for the resulting gene lists, and then finding the parameter or parameter combination with the highest AUPR. The parameters of each method and the design of the grid search are listed in Table A in S1E Text. In the second step, the highest AUPR values from all methods were compared to determine the method with the best performance. This procedure was repeated for each one of the 11 tumor types. Therefore the best-performing method may be different for each one of the tumor types.
There is, however, a caveat concerning the computation of AUPRs from the entire span of the PR curves. We observe in PR curves that (1) there is no significant difference among methods beyond a 10% recall level, and (2) the precision level of network predictions is very low when recall is 10% or higher, suggesting that network predictions are more likely to be affected by noise. The PR curves for BRCA and GBM are shown in Fig 2A as representative examples of these two phenomena. Therefore, we chose to use AUPR only from the 0–10% recall range (i.e. limited-recall), and not from the entire recall range (i.e. full-recall) for the comparison of parameter configurations or the comparison of methods. As the parameter configuration that optimizes AUPR in the limited-recall span can be different from that in the full-recall span, some methods were observed to have different PR curves for the limited-recall case (Fig 2B). The subsequent analysis is carried out with network predictions from the limited-recall case. The optimal parameter values and the number of edges in the limited-recall case for each method and tumor type are shown in S5 and S6 Figs respectively.
After identifying the PR curves to compare the methods, we asked whether any particular method is a clear winner by being the best in all of the 11 tumor types. The AUPR values in Fig 3A indicate that there is no single method that performs the best for all investigated tumor types. The tumor types in this figure are ordered from left to right according to increasing coefficient of variation. The differences in the tumor-wise AUPR means and variances indicate that the 11 tumor types are not equally amenable to network inference with RPPA data. These differences could partially be explained by the different statistics of inferred networks such as average-node-degree and network density, which we found to be negatively correlated with AUPR (Spearman r = –0.626 and –0.453 respectively) (S1B Text, S1 Fig).
Given the absence of a clear winner among the methods, we next asked what the overall best-performing methods were. To achieve an overall comparison of the methods, we ranked them across all tumor types based on (1) overall AUPR and (2) overall AUPR rank. For these two criteria, we computed respectively the sum of a method’s AUPR values in the investigated tumor types (Fig 3B, left panel), and the sum of its AUPR ranks in the same tumor types (Fig 3B, right panel). The different-colored segments in horizontal bars correspond to tumor types as shown in the legend. The numbers next to the horizontal bars indicate the rank of the method for the relevant criterion. Higher AUPR values but lower AUPR ranks indicate better performance. Therefore, the best rank of 1 is given to the highest overall AUPR and the lowest overall AUPR rank.
We observe in Fig 3B that the overall AUPR values (left) did not show as wide a variability across methods as the overall AUPR ranks (right). This might be due to the overfitting of the methods to the benchmark network, as each method was run with parameters that optimize performance (AUPR) against the same benchmark. The small differences in overall AUPR values suggest that these methods may have a general capacity to achieve similar performance in other contexts as long as their respective parameter space is sufficiently explored. However, such similarity in performance does not preclude the possibility that some methods consistently outperform others even if by small margins. To investigate this possibility, we ordered the methods from top to bottom according to increasing overall AUPR rank. This choice in the ordering shows that RIDGENET is the best-performing method overall. Broken down by tumor type, RIDGENET is the best for BRCA, OV, UCEC, BLCA and KIRC; but is not as good as ARACNE variants for HNSC, LUSC, LUAD, GBM, COAD, and READ. On the poor performance side, SIMPLEPARCOR has the worst rank according to both the overall AUPR and the overall AUPR rank (Fig 3B).
We next investigated the level of similarity among the network predictions of all 13 methods. One question here is whether the network predictions, as given by the inferred edge weights, would cluster the methods according to shared properties, such as the regularization technique, or the algorithm family listed in Table 1. To this end, we created one vector for each method by stacking the relevant edge weights from all 11 tumor types. We then computed the Spearman correlation between each pair of methods, and also performed dimensionality reduction on the same vectors using principal component analysis (PCA). Unsupervised clustering on the Spearman correlation matrix (hierarchical clustering with complete linkage and Euclidean distance) and PCA on the edge weight matrix reveal concordant results in terms of the grouping of the methods (Fig 3C and 3D). We observe three major groups of methods in Fig 3C: (1) Mutual information-based methods ARACNE (variants), CLR, MRNET, (2) correlation-based methods SPEARMANCOR and PEARSONCOR, and (3) partial correlation-based methods. SIMPLEPARCOR from the third group can be considered an outlier compared with the other partial correlation methods. Therefore, if we remove it as a separate group, the remaining partial correlation methods RIDGENET, LASSONET, ELASTICNET, PLSNET, GLASSO, GENENET can also be categorized as ‘regularized methods’.
In the PCA plots, the 1st principal component (PC) primarily separates the correlation-based methods SPEARMANCOR and PEARSONCOR from the others, accounting for 53% of the variance (Fig 3D). Correlation methods are fundamentally different from other investigated methods because they do not attempt to eliminate transitive edges in any way. This defect could predict poor performance for both SPEARMANCOR and PEARSONCOR. However, the superior overall performance of the rank-based SPEARMANCOR compared with the value-based PEARSONCOR and several regularized methods (Fig 3B) could be due to the ability of SPEARMANCOR to capture nonlinear relationships and/or its robustness against outliers.
The 2nd PC (23.4% variance) separates SIMPLEPARCOR, a method that is based on Gaussian graphical models and that employs the sub-optimal pseudo-inverse technique when the covariance matrix is singular. Even when the covariance matrix is non-singular, the inversion of the covariance matrix without any regularization is known to introduce defects into the inference procedure unless the number of samples is at least twice the number of features [16]. As the cohort sizes in this study are less than twice the number of antibodies (2*187 = 374) for 7 of the 11 tumor types (Table 2), it is not surprising that SIMPLEPARCOR has poor performance in these tumor types, hence the poorest overall performance by a margin (Fig 3B). Indeed, we can observe that the tumor types where SIMPLEPARCOR achieves relatively better ranks are BRCA, OVCA, KIRC, and UCEC, the four tumor types that have cohort size greater than 374 (Fig 3A and 3B and Table 2).
The 3rd PC (8.1% variance) achieves the separation of mutual information methods from regularized methods. Mutual information-based methods have the capability to model nonlinear relationships, but are not able to infer the direction of the relationship. These two fundamental differences may account for the clear separation of these methods from the others. Principal components can achieve a separation of regularization-based methods only at the 5th and 6th PC, which account for as little as 4% and 1.4% of the variance respectively (Fig 3D).
The modest differences between overall AUPR values in the left panel of Fig 3B, and also the lack of a consistently best-performing method in all tumor types are reasons to refrain from recommending one method as the best off-the-shelf method for PPI inference. Therefore, we propose a set of high performers by taking into consideration both the overall AUPR and the overall AUPR rank criteria. The methods that rank in the top six according to both of these criteria are the same six methods: RIDGENET, ARACNE-M, ARACNE-A, LASSONET, CLR, and SPEARMANCOR (Fig 3B). This set of high performers, referred to as TOP6 from here on, includes representative methods from all algorithm families in Table 1 except for inverse covariance-based partial correlation methods. This may be indicative of inverse covariance being a poor framework to model PPIs in cancer especially if the cohort size is not several times as large as the number of proteins. In contrast, linear measures such as correlation and (ℓ1- or ℓ2- regularized) partial correlation, and also nonlinear measures such as mutual information are all represented in the set of high performers. Although ARACNE-M and ARACNE-A differ only in the form of the threshold (i.e. multiplicative or additive) used to remove the weakest edge in a triplet, the networks inferred by these methods are a function of the user-specified threshold values (S1F Text), and thus are not necessarily similar.
We next asked how the network predictions from the TOP6 methods cluster the 11 tumor types. However, similar to the reduction from 13 methods to the TOP6 methods, it was necessary to apply a significance threshold for edges before performing the clustering. P-values were not a viable option as significance scores because several methods did not return p-values. Even if p-values were obtained from all methods, it would not be possible to combine the p-values in this study in a statistically sound way because all methods used the same data, hence violating the independence requirement. Therefore, we resorted to an alternative method and used edge ranks as a nonparametric proxy for the importance of edges.
For a given tumor type, we computed (1) consensus edge ranks by taking the average of ranks from the TOP6 methods, and (2) consensus edge weights by taking the average of weights again from the TOP6 methods. The consensus ranks served as a nonparametric proxy for our importance levels, while the consensus weights were used in the clustering steps. Comparing consensus edge ranks obtained from the TOP6 methods with those obtained from all 13 methods (ALL13) showed that the TOP6 methods yielded slightly higher AUPR than ALL13 against the Pathway Commons gold standard (S3B Fig, S1C Text). This finding confirmed the use of TOP6 as a superior choice over ALL13.
The number of edges to use for the unsupervised clustering of tumor types was determined in the following way. For a certain threshold, we extracted all edges from a given tumor type that have a consensus edge rank smaller (more significant) than the threshold level. We then formed a matrix of edges by tumor types by combining extracted edges from all 11 tumor types. Next, we computed the PCs constructed as a linear combination of the tumor-type vectors, and inspected the behavior of the percentage of variance explained by the first three PCs as the rank threshold was varied from 25 to 2000. We observed that the sum of the variance percentages from the first three PCs exhibited an inflection point at rank 425, and thus determined 425 as the consensus rank threshold that determined significant and non-significant edges in each tumor type (S4b Fig, S1D Text).
Using the consensus rank threshold of 425, we investigated the natural groupings in the set of 11 tumor types when each tumor type was represented with the consensus edge weights obtained from the TOP6 methods. The number of edges in each tumor type that pass the consensus rank threshold is shown in Table B in S1E Text. The union set of these significant edges from the tested tumor types has 1008 edges. We refer to this union set as the discovery set, and use it to perform PCA and hierarchical clustering of tumor types. The edges in the discovery set and the corresponding weights in the 11 tumor types are given in S1A Table. We note that, among the 187 antibodies in our dataset, all but STAT3_pY705 has at least one interaction in the discovery set (N = 186).
We see in the PCA that PC1 and PC2 jointly separate the 11 tumor types into three groups, and also that PC3 further breaks down one group into two to result in a total of four groups: 1) COAD, READ; 2) LUSC, LUAD, HNSC; 3) GBM, KIRC; and 4) OV, BRCA, BLCA, and UCEC (Fig 4A). These results are concordant with the clusters from hierarchical clustering (Fig 4B dendrogram) and also with the previously defined Pan-Cancer groups in the literature, as we elaborate below.
As for the first group, COAD and READ have previously been shown to cluster together in the Pan-Cancer subtypes defined both by RNA expression[27] and by protein expression[4]. These tumors have also been shown to have common DNA-based drivers (mutations and somatic copy number alterations), and hence have been treated as one disease [2,3,28]. Our finding that COAD and READ have the highest percentage of shared PPIs in this study (Fig 4B heat map) is also in line with these observations. Note that the order of tumor types in the heat map is taken from the dendrogram on the left, and that each cell represents the Jaccard index, i.e. the fraction of the intersection set over the union set of edges from two tumor types.
The tumors in the second group (LUSC, LUAD, and HNSC) have also been previously assigned to a single Pan-Cancer subtype in terms of protein expression[4]. However, RNA expression and somatic copy-number alteration (SCNA) data types have divided these tumor types into two groups: (1) a squamous-like subtype including HNSC and LUSC, and (2) a separate LUAD-enriched group [3,27]. In contrast to this separation where cell histology plays a more important role, both protein expression levels and PPI weights primarily separate these three tumor types based on tissue of origin: (1) lung-derived tumors LUAD and LUSC, and (2) a separate HNSC group (Fig 4B dendrogram and [4]).
Tumors in the third and fourth groups (GBM, KIRC, OV, UCEC, BRCA, and BLCA) can be separated along a continuum in the PC3 dimension (Fig 4A). However, we can consider GBM and KIRC as a separate group as these two tumor types separate from the other four in the unsupervised clustering dendrogram in Fig 4B. GBM and KIRC also cluster most closely among this set of 11 tumor types according to somatic copy-number alterations and protein expression levels [3,4]. However, KIRC also shows an outlier behavior for PPI networks in that it exhibits the lowest fraction of shared PPIs with other tumor types (Fig 4B). GBM, on the other hand, has an outlier property by being on one extreme of the separation along the PC3 dimension. This may reflect the fact that GBM samples arise from glial cells in the brain, a histological origin that shows marked differences from epithelial cells. Indeed, GBM was previously shown to have a distinct cluster comprised of only GBM samples in terms of both RNA and protein expression levels [4,27].
The fourth group contains OV, UCEC, BRCA, and BLCA; the first three of which are categorized as women’s cancers. The proximity of women’s cancers in clustering results may point to female hormones, such as estrogen and progesterone, causing a similar profile of PPI weights. BLCA is most similar to women’s cancers (Fig 4B), but it also is on one extreme of the separation along the PC3 dimension. This is concordant with the previously discovered Pan-Cancer subtypes because BLCA was shown to have the characteristic property of being one of the most diverse tumor types in the TCGA Pan-Cancer dataset. It had samples in 7 major RNA expression subtypes, and histologies in squamous, adenocarcinoma, and other variants in bladder carcinoma [27]. Next, we performed unsupervised clustering and community detection on the 1008 discovery set interactions to investigate patterns both among the interactions and also in the network formed by the interactions.
Unsupervised hierarchical clustering of the 1008 discovery set PPIs shows that these interactions form three main groups (Fig 5): (1) a positive dominant group where interactions generally have positive consensus weight and occurrence in multiple tumor types (mean pan-cancer weight = 0.25, mean pan-cancer recurrence = 4.42, N = 136, recurrence for an individual edge is computed over the binary values in S1B Table), (2) a negative dominant group where interactions generally have negative consensus weight (mean pan-cancer weight = –0.099, mean pan-cancer recurrence = 1.2, N = 133), and (3) a heterogeneous group (mean pan-cancer weight = 0.093, mean pan-cancer recurrence = 1.56, N = 739) which is a mixture of positive and negative, and also recurrent and non-recurrent interactions (S1A Table). In this set of 1008 most significant edges, both the number and the overall weight of negative interactions are smaller with respect to positive interactions. This may indicate either the lower prevalence of mutual exclusivity relationships for in vivo protein-protein interactions, or merely the difficulty of discovering negative PPIs from RPPA data.
We next visualized as networks the positive dominant, negative dominant and heterogeneous groups of interactions in order to gain insight on the related biological processes. However, the number of interactions in the heterogeneous group (N = 739) is too large to allow a clear interpretation of the results. Thus, we investigated whether the complete set of 1008 edges could further be broken down into densely connected modules (with high level of intra-module connectivity, and relatively lower levels of inter-module connectivity). To this end, we employed five different community detection algorithms: (1) fast greedy modularity optimization[29], (2) a spin-glass model from statistical mechanics coupled with simulated annealing for optimization[30], (3) multi-level modularity optimization[31], (4) an information theoretic approach that minimizes the expected description length of a random walker trajectory[32], and (5) random walk-based Walktrap community finding algorithm[33]. In the discovery set network, these algorithms detected 6, 8, 6, 11, and 22 modules respectively with similar and relatively high modularity scores (range 0.41–0.44, modularity due to [29]). Even though the number of detected modules was variable across the methods, we defined a consensus measure to identify the agreement/disagreement between the five predictions. For each antibody pair, the number of methods out of five, i.e. the frequency, of being predicted to be in the same module was utilized as a measure to quantify the level of method concordance. The consensus matrix of frequencies formed this way revealed four robust (Modules 1–4), and two less robust (Modules 5–6) modules among the discovery set interactions (Fig 6, S2A Table for consensus matrix, S2B Table for module membership of antibodies, S1A Table for module membership of interactions).
The four robust modules (1–4) discovered in Fig 6 are also the densely connected ones with per antibody averages of 7.72, 8.24, 9.24, and 7.875 interactions respectively (Table 3). Mapping the positive dominant, negative dominant, and heterogeneous group memberships onto Module 1 reveals that 88% (170/193) of the edges in Module 1 are from the heterogeneous group (Fig 7). The major hubs in this module, N.Cadherin (22 edges), Mre11 (21 edges), and Bid (15 edges), have predominantly heterogeneous-group edges (Table 3). Interactions in this module may play roles in cell cycle, DNA damage repair, apoptosis, hormone and receptor tyrosine kinase (RTK) signaling pathways.
Modules 2 and 4 are distinguishable from the other modules by having zero edges from the negative dominant group, and an abundance of edges from both the positive dominant and heterogeneous groups (Table 3, Fig 7). Interestingly, 91% (30/33) of the antibodies in Module 2 are phosphospecific. On the contrary, only 6.2% (1/16) of the antibodies in Module 4 are phosphospecific. These results raise the question whether significant correlations can only be found between antibodies of the same type (i.e. phosphospecific with each other and non-phosphospecific with each other). To address this question, we analyzed the interactions in Module 1, which is the only module other than Module 2 that contains a subtantial number of phosphospecific antibodies (N = 10). In this module, phosphospecific antibodies have 64 interactions, but only 14 of these are between two phosphospecific antibodies (22%) (S3 Table). This result shows that it is possible to observe significant correlations between phosphospecific and non-phosphospecific antibodies, and suggests that there may be biological differences between Module 1 and Module 2 antibodies that lead to the differences in correlation patterns.
We next compared the biological functions of the phosphospecific antibodies in Module 1 and 2 to investigate potential differences. Processes such as cell cycle, proliferation, RTK signaling, RAS/MAPK signaling were shared between the modules; but Module 2 antibodies were involved in a greater variety of oncogenic pathways such as AKT/mTOR, Wnt, and NFκB signaling. Even though these differences do not rule out technical bias as a reason for the absence of non-phosphospecific antibodies in Module 2, they provide grounds for a biological reason such as the coordinated regulation of signal transduction pathways. An example of a technical bias that could lead an antibody to correlate highly only with another antibody of the same type would be that non-phosphospecific antibodies are expected to bind to both phosphorylated and non-phosphorylated states of a target protein, whereas phosphospecific antibodies only bind to target phosphosites (barring off-target activity).
Module 3 is the densest network among the six, with 9.24 interactions per antibody on average. This module almost exclusively contains non-phosphospecific antibodies (32 out of 34), but has a good representation of edges from all three of positive dominant, negative dominant and heterogeneous groups (Table 3, Fig 7). Akt, Tuberin, and Ku80 antibodies are major hubs (20, 19, and 18 edges respectively) with predominantly positive-dominant and heterogeneous-group edges. There is an absence of phosphospecific antibodies, but due to the cross-reactivity of non-phosphospecific antibodies, this module may be related to several signal transduction pathways (e.g. Akt, mTOR, B.Raf, β-catenin) and DNA double-strand break repair (e.g Ku80, Rad50). Interestingly, this module is the only one that has hubs with predominantly negative-dominant-group edges. Chk1 and PDK1 antibodies have, respectively, 12 negative-dominant-group edges out of 13 (92%), and 9 negative-dominant-group edges out of 11 (82%). One speculation for the underlying cause of the negative edges could be the mutual exclusivity relationship between processes that promote cellular proliferation and those that promote cell cycle arrest or apoptosis. Module 5 and 6 are relatively unstable communities with smaller numbers of intra-module interactions, some of which may play roles in cell cycle (CDK1, Cyclin_B1, Cyclin_E1), translation (eIF4E, 4E.BP1, 4E.BP1_pT70), and apoptosis (Bim, Bcl.2, Bax).
The network visualization of the discovery set edges presents an opportunity to ‘discover’ biologically interesting cancer-related interactions. However, a thorough understanding of the interactions in densely connected modules may still be challenging. To facilitate this ‘interpretation’ and potentially identify the biological processes that each interaction may take part in, we mapped the discovery set interactions to Reactome[21] gene lists. This mapping could not be performed with Pathway Commons because the only unit of analysis in Pathway Commons is interactions, i.e. Pathway Commons does not have previously defined gene lists as in Reactome.
The mapping of inferred interactions to Reactome gene lists involved multiple steps. We first obtained a complete set of Reactome gene lists (N = 1705), and filtered these to keep only the human-specific ones (N = 1669). We then reduced each inferred interaction to the genes corresponding to its interaction partners. If both antibodies corresponded to the same gene, the PPI was left out of the analysis as it would cause a self-interaction at the gene level. We then identified the Reactome gene lists that contained these two interacting genes, and increased their scores by the relevant consensus edge weight. Finally, gene list scores were normalized by the number of matching interactions and also the number of genes in the gene lists to obtain the tumor-type-specific ‘average interaction strengths’ (S4 Table) (Methods). 339 out of 1669 gene lists had a match with at least one of the discovery set interactions in one of the tested tumor types.
The universe of Reactome gene lists is not a flat structure, but a hierarchy. The top level of the hierarchy for human-specific gene lists consists of 24 biological processes according to Reactome Pathway Browser (S7 Fig). We first performed a parent-child analysis for the 339 gene lists in S4 Table. 16 out of 339 were one of the top-level (most general) biological processes, whereas 323 gene lists were child gene lists at different depths of the hierarchy (S5 Table). An analysis of the parents of the 339 gene lists revealed that the top-level ‘events’ signal transduction, cell cycle, and immune system had the greatest number of child gene lists (114, 52, and 46 respectively) that matched to at least one interaction in a tested tumor type (Fig 8A). This analysis also showed that more gene lists with positive ‘average interaction strengths’ were shared across tumor types than those specific for one or two tumor types (Fig 8B). In the Fig 8B heat map, we also tracked the number of significant (consensus rank < 425) interactions that match to each gene list broken down by module or group, and averaged over tumor types (S11 Table). Module 2 and 3 have the highest numbers of interactions that match the shown 339 Reactome gene lists, and these interactions are predominantly from the heterogeneous and the positive dominant groups (Fig 8B).
The 11 tumor types show strong similarities in terms of signal transduction interaction strengths (Fig 9 left panel). The signal transduction gene lists common to all tested tumor types include signaling by RTKs such as fibroblast growth factor (FGF) receptor family, epidermal growth factor (ErbB) receptor family, platelet-derived growth factor (PDGF) receptor family, vascular endothelial growth factor (VEGF) receptor family, insulin receptor family as well as signaling by G-protein-coupled receptors (GPCR) and Wnt, AKT/mTOR, and RAS/MAPK pathways. The gene lists not common across tumor types include the Hippo signaling (specific to LUAD, LUSC, and BRCA) and phospholipase C-related pathways (specific to COAD, READ, UCEC, GBM, KIRC, BLCA). However, the gene lists not common across tumor types are associated with very few matching interactions as indicated by the ‘group’ membership track on the left of the heatmap. It is possible that one interaction matches with multiple similar Reactome gene lists and populates the heatmap. The disease top-level event recapitulates the above signal transduction-related findings as about twenty interactions from Module 2 positive dominant group, and Module 3 heterogeneous group match with these gene lists (Fig 9 middle bottom panel, labeled as ‘signaling in disease’ as the majority of the gene lists are associated with signaling).
All tested tumor types exhibit moderate interaction strengths for cell cycle related gene lists (Fig 9 middle top panel). Most of these interactions are heterogeneous group edges from Module 5 and 6 as indicated by the group and module tracks. Carcinomas of the ovary, uterus, and breast, and adenocarcinoma of the colon have higher interaction strengths compared to other tumor types for several anaphase and metaphase-related gene lists such as those involving anaphase promoting complex (APC/C), its inhibitor Emi1, NIMA family kinases, and nuclear mitotic apparatus (NuMA). Other biological processes common to all tested tumor types are innate and adaptive immune signaling, metabolism, and DNA repair, as expected with transformed cells and the immune cells infiltrating the tumor microenvironment (Fig 9 middle and right panels). Interestingly, interactions that match to ‘metabolism’ gene lists are predominantly positive dominant group edges from Module 6. The ‘immune system’ gene lists match mostly with Module 2 edges that are from the heterogeneous or positive dominant groups. ‘DNA repair’ gene lists match with both Module 2 and 3 edges, albeit predominantly from the heterogeneous group. Other Reactome top-level ‘events’ that match to discovery set interactions include apoptosis, extracellular matrix organization, and cell-cell communication (S8 Fig).
Discovering PPIs in cancerous cells is an important but challenging goal. In this study, we computationally inferred proteomic networks in 11 human cancers using 13 different methods, and presented a performance comparison of the methods accepting a simplified reference network from the Pathway Commons information resource, which is based on experiments and publication digests, as the standard of truth. Pathway Commons is a collection of curated interactions from many different normal and disease conditions (a formal and computable representation of available pathways and interactions). We acknowledge that a complete standard of truth for pathways is not currently available and that our methodology is therefore subject to certain caveats, as discussed below. Despite these caveats, computational inference of PPI networks from measurements of protein levels across a set of conditions are attractive in that they can reduce the hypothesis space of interactions and inform researchers on the potentially active pathways in the experimental model.
Our comparison of the performance of network inference methods indicates that no single method has the best performance in all tumor types, but a group of six methods, including diverse techniques such as correlation, mutual information, and regression, consistently rank highly among the tested methods. These six methods consist of RIDGENET and LASSONET, ridge and lasso regression-based partial correlation methods employing an ℓ2 and ℓ1 penalty respectively; ARACNE-A, ARACNE-M, and CLR, mutual information methods that differ based on their penalty type or the choice of standardization for mutual information; and SPEARMAN CORRELATION, which assesses the strength of the linear relationship between the ranks of the values in two same-length vectors. From a tumor-type perspective, we find that not all tumor types are equally amenable to network discovery with RPPA data. Five tumor types (KIRC, OV, COAD, READ, and BLCA) consistently had lower AUPR predictions by all network inference methods.
In a recent multi-method comparison study for gene network inference, regression-based methods were represented mostly by modifications of the ℓ1-penalized lasso algorithm; however methods involving an ℓ2 penalty, such as ridge regression or elastic net, were not included [35]. Moreover, the ℓ1-penalized methods did not achieve the best overall performance in gene network inference. We find in this study that ℓ2-penalized methods such as ridge regression can outperform the lasso in the inference of proteomic networks. Even though the concurrent execution of feature selection and model fitting may appear to be an attractive property for lasso regression, we recommend performing an unbiased test for both ℓ1 and ℓ2-penalized models in the exploratory phase of a study. It is not guaranteed that the variables selected by the ℓ1 penalty will be the most biologically important ones in the system.
A network of 1008 most significant interactions inferred by high-performing methods reveals that these interactions can be grouped into three. The group termed ‘positive dominant’ contains mostly positive interactions with generally high weights. Nine interactions that exist in at least 10 of the 11 tumor types with very strong consensus weights are also in this group (S1A Table), and potentially occur due to cross-reactivity of the antibodies. The other two groups are termed the ‘negative dominant’ (generally negative weights), and the ‘heterogeneous’ groups. The network of 1008 edges contains four robust densely connected modules, two of which do not include any edges from the negative dominant group (Modules 2 and 4). Strikingly, the ratio of phosphospecific antibodies in one of these two modules is 91% (Module 2). While it is possible that a technical bias may be leading to high correlations between antibodies of the same type, a biological reason such as the coordinated regulation of signal transduction events may also be strongly contributing to the Module 2 interactions between phosphospecific antibodies because phosphospecific and non-phosphosphospecific antibodies may exhibit a high number of interactions as in Module 1. The positive dominant and heteogeneous groups are scattered, albeit unevenly, to the four robust modules. Mapping these 1008 most significant interactions to Reactome gene lists confirms the pan-cancer importance of signal transduction pathways, innate and adaptive immune signaling, cell cycle, metabolism, and DNA repair; and also suggests several gene lists that may be specific to a subset of tumor types.
We observe a paucity of negative dominant group edges that match with Reactome gene lists (Figs 7 and 8). A possible reason may be that negative correlations in our study imply mutual exclusivity, not inhibitory relationships. Reactome gene lists are not designed to group together mutually exclusive proteins unless there is a flow of influence (i.e. activation/inhibition) from one to the other. It is also not implausible that an inhibitory event, such as phosphorylation of protein A by protein B, leads to a positive correlation between A and B as their concentrations may increase or decrease together.
The caveats in our workflow, as shown in Fig 1, concern both the computational inference and the pathway knowledgebase arms of the analysis. In the computational inference arm (Steps 1 and 2), the caveats include questions regarding (1) the quality of RPPA experiments and whether the signal-to-noise ratio in RPPA experiments is high enough to allow the inference of direct interactions, and (2) the reliability of results from computational network inference methods (S1A Text). In the pathway knowledgebase arm (Steps 3–5), the fidelity of pathway models in knowledgebases is limited due to factors including (1) the quality of wet-lab experiments for PPIs such as yeast-2-hybrid [34], (2) missing or inaccurate information in the database due to poor curation, (3) the lack of context information for PPIs, such as cell or tissue type or physiological conditions, and (4) the ascertainment bias in the knowledgebase (primarily incomplete coverage) as discussed in the Introduction. More generally, pathways in knowledgebases such as Pathway Commons are only model descriptions of reality typically summarizing a set of experiments and do not represent an absolutely ‘true’ (and certainly not complete) set of interactions.
In future work, it will be important to assess the predictive power of the inferred PPI networks. For example, it would be useful to evaluate these networks in terms of how much they assist in the understanding of oncogenesis, response to therapy, and design of combination therapies that deal with feedback loops. It is also desirable to incorporate time-dependent readouts from perturbation experiments to be able to build causal models and enhance the predictive power of proteomic networks. An obstacle against building causal models, such as Bayesian networks, with the PANCAN11 RPPA data was the relatively large size of the network (187 nodes, i.e. antibodies) compared with the number of available samples in individual tumor types (between 127 and 747). Probabilistic models such as Bayesian networks require the number of samples to be at least an order of magnitude larger than the number of nodes for a sound estimation of model parameters [36].
The significance of this work extends beyond cancer. Discovering direct, potentially causal interactions between proteins is an opportunity in all areas of molecular biology where proteins are measured in different conditions and where correlations are informative. The methodology presented here can easily be adopted to study interactions in different molecular biology contexts.
The pan-cancer reverse phase protein array (RPPA) dataset was downloaded from The Cancer Proteome Atlas[5] on April 12, 2013. This dataset is denoted as PanCan11 and contains protein expression data for 3467 tumor samples and 187 antibodies, 51 of which are phosphospecific and 136 of which are non-phosphospecific. The 11 tumor types represented in this dataset are bladder urothelial carcinoma (BLCA), breast invasive carcinoma (BRCA), colon adenocarcinoma (COAD), glioblastoma multiforme (GBM), head and neck squamous cell carcinoma (HNSC), kidney renal clear cell carcinoma (KIRC), lung adenocarcinoma (LUAD), lung squamous cell carcinoma (LUSC), ovarian serous cystadenocarcinoma (OVCA), rectum adenocarcinoma (READ), and uterine corpus endometrioid carcinoma (UCEC).
PanCan11 patient samples were profiled with RPPA in different batches, and normalized with replicate-based normalization (RBN). RBN uses replicate samples that are common between batches to adjust antibody means and standard deviations so that the means and standard deviations of the replicates become the same across batches.
Pathway Commons stores pathway information in BioPAX[22] models that contain formal computable representations of diverse events such as biochemical reactions, complex assembly, transport, catalysis, and physical interactions. We queried Pathway Commons with the "prior extraction and reduction algorithm" (PERA)[26] for the proteins and phosphoproteins in the PANCAN11 RPPA dataset.
PERA is a software tool and a protocol that, given a set of observable (phospho- and/or whole) proteins, extracts the direct and indirect relationships between these observables from BioPAX-formatted pathway models[26]. PERA accepts a list of (phospho)proteins identified by their HGNC symbols, phosphorylation sites and their molecular status (one of ‘active’, ‘inactive’, or ‘concentration’) as input and, based on the pathway information provided by the Pathway Commons information resource[20], it produces a binary and directed network. The major advantage of PERA over other similar tools, such as STRING[37] or GeneMania[38], is that it considers not only the name/symbol of a protein but also its phosphorylation sites, enabling finer mapping of entities and pathways.
We downloaded the web ontology (OWL) file for Pathway Commons version 2 on 10/1/2013, and implemented PERA v2.9.1 with the following command:
java -Xmx3g -jar bpp291.jar \
-l 1 –t 3 \
-o output.tsv input.tsv \
pathwaycommons2.owl
The PERA input file (i.e. input.tsv) is provided as S6 Table. The–l value of 1 determines that PERA will output an interaction between two entities only if the distance between them is 1, i.e. there is no intermediary entities. The–t value of 3 determines that the phosphorylation site mismatch tolerance is 3. For instance, if a PERA input includes phosphorylation site S473 for Akt, PERA will consider all interactions in the residue range (470, 476) for this phosphoprotein. The post-processing of the PERA output file included two steps: 1) As the network inference methods employed in this study produce only undirected network predictions, we first converted the directed network in the PERA output to an undirected network. 2) We then removed any existing duplicate and/or self edges before using this network as a gold standard in performance evaluation.
The analysis was performed using the R language[39]. The R functions used to implement the network inference methods are as follows: The cor function in the stats[40] package for PEARSONCOR and SPEARMANCOR; the ggm.estimate.pcor and cor2pcor functions in the GeneNet[41] package for GENENET and SIMPLEPARCOR; the ridge.net, pls.net, and adalasso.net functions in the parcor[42] package for RIDGENET, PLSNET, and LASSONET; the glasso function in the glasso[43] package for GLASSO; the aracne.a, aracne.m, clr, and mrnet functions in the parmigene package[44] for ARACNE-A, ARACNE-M, CLR and MRNET. The ELASTICNET method was implemented as a modification of the adalasso.net function in the parcor package, and is available upon request. Mathematical descriptions of the algorithms used are provided in S1F Text.
Community detection algorithms were implemented with the cluster_fast_greedy, cluster_spinglass, cluster_infomap, cluster_louvain, and cluster_walktrap functions in the igraph[45] package version 1.0.1.
The steps involved in computational network prediction and performance evaluation are discussed in detail in S1E Text. Here we discuss the construction of precision-recall curves. Precision was defined as the fraction of the number of correctly predicted edges (predicted edges that can be found in the gold standard) to the number of all predicted edges. Recall was defined as the fraction of correctly predicted edges to the number of all edges in the gold standard.
The PR curve for a given parameter configuration was constructed by taking the edge list ranked from the most significant to the least, and then iterating over the edges so that we obtained, at each iteration, a cumulative edge set that included all the edges seen up to and including that iteration. For each iteration, we computed the precision-recall value pair for the edge set and placed this value pair on the PR plot. We plotted a separate PR curve for each parameter configuration for the nine methods that required specification of parameter values (all methods except PEARSONCOR, SPEARMANCOR, SIMPLEPARCOR, and GENENET). The PR curve that had the greatest area under the curve (AUPR) between the [0,0.1] recall range (i.e. limited-recall) was identified as the optimal PR curve for that particular method. The optimal parameter values for the limited-recall case are shown in S5 Fig. For methods that did not have user-specified parameters, there was only one PR curve and that was adopted as the optimal PR curve. In the subsequent step, the AUPRs from the optimal PR curves were compared in order to rank the methods and evaluate their performance relative to the gold standard network.
We find that the inferred interactions in various tumor types are a relatively small subset of the gold standard network derived from Pathway Commons (i.e. low recall). Low levels of recall are readily acceptable for satisfactory performance because it is expected that interactions inferred from a single disease (cancer) and a single cancer type will not retrieve all of the interactions in the gold standard. However, it is desirable that, when an algorithm calls an interaction, there is a high probability that this inference is correct, i.e. high levels of precision are essential for nominating a network inference method as competitive.
Three data files were downloaded from the Reactome website http://www.reactome.org/pages/download-data/: 1) Reactome Pathways Gene Set (S7A Table) on 11/11/2015, 2) Complete list of gene lists (ReactomePathwayLabels.txt) on 11/16/2015, and 3) gene list hierarchy relationship (ReactomePathwaysRelation.txt) on 11/16/2015. The first file contained a total of 1705 gene lists. The second file contained gene list labels, unique Reactome identifiers, and species information. The third file contained the unique identifiers of parent gene lists adjacent to those of the child gene lists. The ID in the left column was one level above, in other words a superset of the ID in the right column. Reactome identifiers also include characters to denote the species information. The information in the second and third files was used to filter out non-human gene lists (S7B and S7C Table respectively). After the removal of duplicate entries, the number of human-specific gene lists in file 2 was 1869. The overlap between these 1869 human-specific gene lists and the 1705 gene lists in file 1 was 1669, which was used as a gold standard in the subsequent mapping analysis.
The PERA input in S6 Table lists the gene(s) that correspond to each antibody used in this study. The complete list of these genes including all paralogs is provided in S8 Table, and contains 167 uniqe genes. In mapping the discovery set interactions to Reactome gene lists, each interaction is represented by the gene(s) corresponding to the interaction partners. The pseudocode for this mapping is as follows:
# Constants
N = 11 # Number of tumor types
T = 425 # The threshold for consensus edge rank
R = 1669 # The number of gene lists in the Reactome
# human gold standard
Initialize P # 1669 by 11 matrix that stores the
# ‘average interaction strength’ of each
# Reactome gene list for a given tumor type
for i = 1:N # Tumor types
q <- Number of edges in tumor type i that pass consensus rank threshold T
Initialize M # 1669 by q matrix that maps the most
# significant interactions in tumor type
# i to Reactome gene lists
for j = 1:q # Interactions
1. Identify the corresponding gene(s) for each interaction partner. Let these be set1 and set2.
2. If set1 and set2 has a non-empty intersection, this is a self interaction. Skip to next interaction.
3. Otherwise
for k = 1:R # Reactome gene lists
4. If set1 and set2 both have at least one member in gene list k, this is a match. Assign absolute weight to entry M[k,j]. This weight is the consensus edge weight penalized (divided) by the number of genes in gene list k.
end
end
5. Average the q interactions (only the real-valued ones) for each one of the 1669 gene lists. This is the average interaction strength for a given Reactome gene list in a given tumor type.
6. Insert the vector created in Step 5 as a column in matrix P
end
Networks from the TCGA RPPA or tab-delimited user data can be inferred and visualized with the ProtNet web application at http://www.sanderlab.org/protnet (A tutorial is provided in S1 Protocol).
R scripts used in the analysis are provided in S2 Protocol.
|
10.1371/journal.pbio.0050242 | Variants in a Novel Epidermal Collagen Gene (COL29A1) Are Associated with Atopic Dermatitis | Atopic dermatitis (AD) is a common chronic inflammatory skin disorder and a major manifestation of allergic disease. AD typically presents in early childhood often preceding the onset of an allergic airway disease, such as asthma or hay fever. We previously mapped a susceptibility locus for AD on Chromosome 3q21. To identify the underlying disease gene, we used a dense map of microsatellite markers and single nucleotide polymorphisms, and we detected association with AD. In concordance with the linkage results, we found a maternal transmission pattern. Furthermore, we demonstrated that the same families contribute to linkage and association. We replicated the association and the maternal effect in a large independent family cohort. A common haplotype showed strong association with AD (p = 0.000059). The associated region contained a single gene, COL29A1, which encodes a novel epidermal collagen. COL29A1 shows a specific gene expression pattern with the highest transcript levels in skin, lung, and the gastrointestinal tract, which are the major sites of allergic disease manifestation. Lack of COL29A1 expression in the outer epidermis of AD patients points to a role of collagen XXIX in epidermal integrity and function, the breakdown of which is a clinical hallmark of AD.
| Atopic dermatitis (AD, eczema) is a common chronic inflammatory skin disorder and a major manifestation of allergic disease. Typically, AD first occurs in early childhood, often preceding the onset of allergic airways disease, such as asthma and hay fever. A family history of allergic disorders is the single strongest predictor for AD, showing that genetic factors play a major role in the disease development. We have previously mapped a disease locus for AD on Chromosome 3q21, Now we have used a dense map of microsatellite markers and single nucleotide polymorphisms (SNPs) to find the underlying disease gene. We identified genetic markers in a subregion that showed association with AD, and replicated this finding in a large independent family cohort. The associated region contained a single gene, COL29A1, which encodes a novel collagen. We demonstrate that AD patients lack COL29A1 expression in the outer epidermis, implicating collagen XXIX in epidermal integrity and function. The gene expression pattern of COL29A1 in other organs, including the lung and the gut, suggests that this gene could have a role in a wider spectrum of allergic diseases and may provide a molecular link between AD and respiratory airways disease and food allergies.
| Atopic dermatitis (AD) is a chronic inflammatory skin disease that is characterized by intensely itchy skin lesions. AD is one of the most common chronic diseases in childhood affecting 10%–20% of children in industrialized societies [1], with a steady increase over the past decades [2,3]. Along with asthma and hay fever, AD is commonly associated with an abnormal immune response and the formation of allergy antibodies (IgE) against innocuous environmental allergens.
AD is often the first clinical manifestation of allergic disease. The onset of disease is typically observed during the first two years of life [4]. For the majority of affected children, AD heralds a lifetime of allergic disease. A susceptible child commonly passes a characteristic sequence of transient or persistent disease stages that is known as the “atopic march” which begins with AD and food allergy in the young infant and continues with the development of allergic airways disease later in childhood and adulthood [5]. The close familial and intra-individual association of the allergic disorder strongly suggests shared genetic etiology.
A strong genetic component in allergic disorders has been recognized almost a century ago. Cooke and van der Veer first reported that the relatives of patients are at significantly increased risk of developing allergic disease [6]. Even today, a positive family history for allergic disorders is the single strongest predictor for the development of AD [7]. Additional evidence for the importance of genetic factors in atopic disease comes from twin studies. The concordance rate for AD among monozygotic twins of about 80% far exceeds the concordance rate of 20% observed among dizygotic twins [8,9]. These data clearly indicate that the genetic contribution to the expression of AD is substantial. Furthermore, studies on the vertical transmission of AD and atopic disease show that children are more likely to inherit these disorders if the mother is affected (parent-of-origin effect) [10]. The predominance of maternal inheritance may be due to environmental factors such as uterine milieu or breast feeding, but they may also arise due to genetic mechanisms such as parent-specific gene expression (genomic imprinting) [11].
AD and atopic disorders are regarded as multifactorial conditions, the onset and severity of which are influenced by both genetic and environmental factors. The data are consistent with an immune etiology shared by all allergic diseases and a congenital target organ defect, the penetrance of which is modified by multiple environmental factors during early childhood. The identification of genes underlying AD and allergic disorders has the capacity to define primary physiologic mechanisms, thereby clarifying disease pathogenesis, identifying pathways and targets for therapeutic intervention.
Although several genome-wide linkage screens for AD have been conducted [12–14], there was no substantial overlap between the identified regions of highest linkage, and the underlying genes remained elusive [15]. We have previously mapped a major susceptibility locus for AD on Chromosome 3q21 [16]. Here we report the identification and characterization of a novel epidermal collagen gene as the underlying disease gene.
To narrow the candidate region spanning 12.75 centiMorgans (cM) (13.5 Mb) on Chromosome 3q21 (Figure 1A), 96 additional microsatellite markers at an average distance of 140 kb were genotyped in 199 affected sibling families with AD from the original linkage scan. Linkage analysis yielded a 1-lod support interval of 5.4 Mb between the markers M3CS075 and M3CS233 (Figure 1B). Subsequently, we conducted an association scan of the 5.4-Mb region using 212 single nucleotide polymorphisms (SNPs) at an average distance of 25.5 kb (Table S1). The SNPs were selected primarily to cover known and predicted genes. Because we had observed a strong maternal effect in the linkage study [16], we chose a family-based association analysis that allowed us to search for risk alleles preferentially transmitted from the mother [17]. Two adjacent SNPs, rs5852593 (p = 0.0079) and rs1497309 (p = 0.016), located 36 kb apart, were associated with AD (Table S1 and Figure 1B). To define the critical region, we typed 16 additional SNPs. Eight of these markers showed association with AD and a maternal transmission pattern to affected children, which was consistent with our previous linkage results (Table 1). The strongest association with a single marker was observed for rs4688761 (pall = 0.0016, pmaternal = 0.0006). We selected eight markers spanning 96 kb that were associated with AD and carried nonredundant information based on the linkage disequilibrium (LD) in the region, and we performed haplotype analysis which confirmed the results (transmissions (T)all:non-transmissions (NT)all = 222:168, pall = 0.0076, Tmat:NTmat = 105:68, pmaternal = 0.0070). In addition, we assessed the significance of the difference in maternal-versus-paternal haplotype over transmissions empirically using the permutation procedure for parent-of-origin transmission disequilibrium test (TDT) implemented in PLINK (P_POOmaternal emp = 0.014) [18].
Next, we investigated whether the observed association accounted for the linkage in the region. We used the marker that had shown the strongest association, rs4688761, to identify 73 of the 199 families from the original linkage scan in whom the disease-associated allele had been transmitted to affected offspring. Nonparametric linkage analysis in the 73 associated families yielded significant evidence for linkage (Zall = 4.18 versus 4.31 in the complete cohort), demonstrating that the majority of the linkage signal was attributable to these families. The significance of this finding was assessed by performing nonparametric linkage analysis in 10,000 random selections of 73 families. An empirical significance level was calculated as the proportion of replicates for which the maximum Zall score was equal or greater than that obtained in the actual analysis. The probability of obtaining a Zall score of ≥4.18 by chance in a random selection of 73 families was estimated by 10,000 simulations to be 0.027.
To confirm the association result, we used a large independent replication set consisting of 292 complete nuclear families including 481 children with AD. We genotyped the selected eight SNPs covering an interval of 96 kb that were associated with AD in the discovery dataset. We confirmed the association with AD across all markers with the strongest association with AD observed for marker A36637742 (p = 0.00038), which also showed a significant overtransmission of the maternal allele (p = 0.0013) (Table 2). The association remained significant after correction for multiple testing. For each marker, it was the more common allele that was overtransmitted. Haplotypes were constructed over the region, which confirmed the association with AD and showed that this phenotype is associated with the overtransmission of the most common haplotype (haplotype frequency 65%, Table 3), and of the maternal allele (pmaternal = 0.000025 for AD, P_POOmaternal emp = 0.033). Next, we compared the AD status among the parents: in the discovery cohort, significantly more mothers (n = 63) than fathers (n = 19) suffered from AD (odds ratio [OR] 4.39, 95% confidence interval [CI] 2.51 – 7.68, p = 5.46 × 10−8). Similarly, in the replication cohort, mothers (n = 83) were significantly more frequently affected with AD than fathers (n = 55) (OR 1.71, 95% CI 1.16 – 2.52, p = 8.3 × 10−3). To assess whether the parent-of-origin effect that was observed originated from the discrepancy in AD prevalence among the parents, we compared haplotype transmissions from affected and unaffected mothers. In the discovery cohort, the excess in maternal transmissions was not predominantly attributable to affected mothers (T:UT = 24:21, p = 0.14) compared to unaffected ones (T:UT = 81:47, p = 0.0059). Similarly, affected mothers (T:UT = 25:17, p = 0.14) of the replication cohort did not contribute a larger transmission excess than the unaffected ones (T:UT = 70:29, p = 0.00013). We conclude that the observed maternal overtransmission pattern in both cohorts was not due to the higher AD prevalence among mothers.
A database search within the associated 96-kb interval revealed a single predicted gene, FLJ35880, extending 11.6 kb into the associated region. No other expressed sequence tag was detected. To identify any additional transcripts, we used putative exons predicted with The National Center for Biotechnology Information (NCBI) Modelmaker within and bordering the critical region to perform rapid amplification of cDNA ends (RACE) from human skin mRNA. We thus identified a single transcript of 9226 bp that consisted of 42 exons (Figure 1D) and included all eight exons of FLJ35880. The corresponding gene spanned 139 kb of genomic sequence and completely encompassed the associated region. Pairwise LD measures (D′) of the genotyped markers indicated that the gene was contained within one 170-kb region of increased LD (Figure 2), whereas the neighboring genes, LOC440978 and LOC131873, were located in separate blocks. The LD structure was consistent with the data for the European population in the HapMap database (Figure S1) [19]. The size of the transcript was confirmed by Northern blot hybridization with an FLJ35880-specific probe detecting a single transcript of 9.6 kb in human skin mRNA (unpublished data), which was in good agreement with the RACE experiments.
The open reading frame yielded a protein of 2614 amino acids with an estimated molecular weight of 289.9 kDa. The predicted protein contained a collagenous domain in the central part and was therefore classified as a new member of the collagen superfamily, collagen XXIX. A BLAST search revealed the human collagen VI alpha 3 chain as its closest neighbor (32% identity). Homology with collagen VI alpha 3 was further strengthened by a similar domain architecture consisting of six N-terminal and three C-terminal von Willebrand factor–type A domains (vWAs) flanking a short collagen triple helix (Figure 1E), and an 18–amino acid secretion signal [20]. Sequence analysis of all 42 exons in 46 unrelated children with AD and 2 unaffected individuals revealed 13 common and six rare sequence variations (frequency <2%) predicted to cause nonsynonymous amino acid substitutions (Table S2). Four variants were located within the triple helix repeat of a collagen domain, however none of them changed the first glycine residue in the repeating sequence patterns (Gly-X-Y), which is critical for triple helix formation [21]. In addition, we performed in silico comparisons of the six variants located within a vWA with the crystal structure of the human vWF A3 domain. One variant, V669G, affects an amino acid within a highly conserved stretch of eight amino acids. The mutation of an adjacent amino acid within this conserved region of a vWA in the homologous gene COLVIA1 has been reported to cause the monogenic muscle disorder Bethlem myopathy [22]. This variant, however was rare (allele frequency 1,3%). One additional variant (E455K) is located in a region predicted to be important for integrin–collagen interaction, and another one (M56T) is located in a helix near a rather conserved region that may also affect protein–protein interaction. All coding SNPs were genotyped in the discovery cohort. Four of them showed a positive association with AD that did, however, not account for the observed haplotype association (Table S2).
Gene expression analysis in human tissues revealed a tissue-specific expression pattern of COL29A1. The highest expression was observed in the skin, but also in the lung, small intestine, colon, and testis (Figure 3A). Overall, COL29A1 expression is moderately low compared to more abundant epidermal transcripts such as keratin 10 (unpublished data).
To specify the expression sites in the layers of the skin and to assess the role of collagen XXIX in AD, we performed in situ hybridization using a COL29A1-specific cRNA-probe on skin biopsies of five patients with AD and five healthy patients (controls). In normal skin, COL29A1 was exclusively expressed in the epidermis with the strongest staining in the suprabasal viable layers. In contrast, the skin of patients with AD revealed a striking absence of COL29A1 mRNA staining in the most differentiated upper spinous and granular layers (Figure 3B and Figure S2). No significant difference in the expression of COL29A1 was observed comparing patients with AD to controls (1.28- ± 0.53-fold down-regulation in AD patients, p = 0.41) using quantitative Taqman reverse-transcriptase (RT)-PCR. These results indicate that while differences in mRNA quantity were not detected, AD patients show a distinct abnormal cellular distribution pattern of COL29A1 expression in the differentiated outer epidermis.
We generated a polyclonal antibody to visualize the collagen XXIX protein in the skin of five patients with AD and five normal controls, including four and three of each group, respectively, in whom in situ analysis was performed. Consistent with the in situ findings, we observed collagen XXIX staining in the differentiated suprabasal layers of the epidermis in normal human skin and a remarkable absence of staining in the most differentiated upper spinous and granular layers (Figure 4 and Figure S3).
In a whole-genome linkage scan for AD, we previously identified a susceptibility locus on human Chromosome 3q21. The candidacy of this chromosomal region was further supported by Bradley et al., who mapped a locus for AD severity in close proximity (3q14) in a Swedish population [13]. By positional cloning, we have now identified the disease-causing gene, COL29A1, which encodes a novel epidermal collagen.
The disease gene was located in a two-staged investigation consisting of systematic linkage and association scanning of the region and subsequent confirmation of the association in a large, independent, replication dataset. In the first stage, we genotyped additional microsatellite markers in the candidate region, which narrowed the initial 12.75-cM linkage interval to 5.4 Mb. The association analysis with an average marker distance of 25.5 kb, finally, revealed an association that was confined to a haplotype block of 170 kb, which included a single gene, COL29A1. Pairwise LD measures indicated LD across the entire gene and defined two subblocks of increased LD. The strongest association was observed within the 96-kb subblock encoding the collagenous domain and the C terminus of collagen XXIX. A rapid decay of LD at the borders of the COL29A1 haplotype block and lack of association of the SNPs located within the neighboring genes clearly limited the association to COL29A1. In addition, we demonstrated in the discovery dataset that the families that contributed to the association of SNP rs4688761 with AD also accounted for most of the linkage signal. This finding corroborated that variants in COL29A1 explained the previously reported linkage of AD to 3q21 [16]. Finally, we confirmed association in a large independent family cohort, making COL29A1 the first AD susceptibility gene, to our knowledge, that is identified by positional cloning.
Consistent with the linkage analysis, we found a maternal transmission pattern to affected offspring in both family cohorts. Although the sexes were equally represented among the affected children of both cohorts, we observed a marked maternal preponderance in AD status among the parents in both cohorts. This finding clearly supports the notion of a maternal effect. It may, however, also raise the question whether the maternal overtransmission pattern observed for the COL29A1 haplotype was due to the different prevalence of AD in mothers and fathers. This is unlikely to be the case, because the analytical tools used in this study, nonparametric linkage and TDT, do not take into account the parental phenotype. Furthermore, we showed that the observed maternal effect was not predominantly attributable to transmissions from affected rather than unaffected mothers. Although parent-of-origin effects have not previously been reported for genes from Chromosome 3q21, they have been observed at other loci influencing allergic disease [23]. Tissue-specific imprinting of genes encoding extracellular matrix (ECM) proteins has been reported in the mouse [24], and their disruption has been shown to impair skin structure and function [25]. Interestingly, COL29A1 is expressed in human placenta, an organ of embryonic origin. Apart from classic genomic imprinting mechanisms, maternal effects may be due to an interaction of the child's genotype with the maternal environment during prenatal life.
Sequencing of the exons and promoter region revealed 19 nonsynonymous coding SNPs, six of which were located within a vWA. In silico analysis of these variants revealed only one rare mutation altering a highly conserved amino acid, but all of them may affect protein–protein interaction. None of the nonsynonymous coding SNPs explained the observed association on its own. It has been demonstrated in other complex diseases that multiple independent variants may occur in a single disease gene [26]. It is therefore possible that several variants or combinations thereof which are associated with the most common haplotype of COL29A1 are involved in the disease pathogenesis. The functional influence of the associated variants remains to be determined.
Involvement of COL29A1 in AD is further supported by its tissue- and cell-specific expression pattern. Like COL29A1, a growing number of collagens recently identified show a restricted expression pattern. These are not mainly found in fibrous connective tissue, but rather in the ECM of more specialized tissue structures pointing to a distinctive function of these proteins [27]. Highest COL29A1 expression was observed in the skin, but also in other epithelial tissues such as the lung, small intestine, and colon, which are the main manifestation sites of allergic disorders, including asthma and food allergies. This gene expression pattern might indicate a role of collagen XXIX in a wider spectrum of allergic diseases and suggests a molecular link between AD, respiratory airways disease, and food allergies, which are epidemiologically closely associated [28,29].
In human skin, collagen XXIX was detected throughout the viable layers of the epidermis with an increase toward the differentiated cells of the granular layer. Comparative expression analysis of COL29A1 by in situ hybridization and immunohistochemistry in skin biopsies of patients with AD and healthy controls revealed a distinct lack of COL29A1 mRNA and protein in the outer viable layers of the epidermis. The process of epidermal stratification is tightly regulated by an increasing gradient of extracellular Ca2+ concentration and a specific special and temporal expression pattern of transcriptional regulators [30]. Our findings indicate that the specific cellular milieu acquired during terminal epidermal differentiation affect the regulation or degradation of COL29A1 mRNA in the outer epidermis. However, our findings do not allow us to distinguish between these two processes.
Lack of collagen XXIX in the outer epidermis of AD patients indicates that a defective ECM may give rise to the disease, proposing a new pathomechanism for AD. Collagens are the most abundant ECM proteins in vertebrates and play a crucial role in maintaining tissue integrity. Their importance for tissue function has been highlighted by the wide spectrum of human diseases caused by mutations in collagen genes [31]. Although a large number of collagens in the connective tissue–rich dermis have been characterized, little is known about collagens in the ECM of the epidermis [32,33]. Collagen XXIX belongs to the vWA containing collagens. They form filaments with globular domains containing the vWA motifs, which are involved in protein–ligand interactions for the organization of tissue architecture and cell adhesion [34]. It is therefore conceivable that collagen XXIX plays an important role in keratinocyte cohesion. Lack of collagen XXIX may facilitate antigen penetration through the skin, which may explain the association found between COL29A variants and allergic sensitization, a common feature in AD patients [35]. Recent findings indicate that structural and functional integrity of the epidermis is a key factor in the development of AD [36] and in the disease progression to allergic airways disease [37].
Furthermore, ECM collagens influence the migration of epidermal antigen-presenting Langerhans cells and T cells [38,39] and may thus play an important role in the initiation and maintenance of cutaneous immune responses. In addition, ECM collagens participate in immune regulation by binding to inhibitory immune receptors [40], rendering collagen XXIX an interesting novel susceptibility gene for AD. Impairment of the immune defense function of the skin is a clinical hallmark of AD. Patients with AD show a diminished resistance against microbes resulting in the colonization of nonlesional skin with Staphylococcus aureus in nearly 90% of patients and an increased susceptibility to bacterial and viral skin infections [41]. Our findings led to the identification of collagen XXIX as a novel component of the epidermal ECM and propose a new disease mechanism in the etiology of atopic dermatitis and allergies.
The diagnosis of AD was made according to standard criteria, as previously described [16]. Recruitment was restricted to patients with an age of onset below 2 y and moderate to severe disease expression. Total IgE levels and levels of specific IgE against 12 common environmental allergens were determined using ImmunoCAP (Phadia AB; http://www.phadia.com/). Allergic sensitization was defined as either the presence of specific IgE to at least one allergen (detection limit 0.35 kU/l) or a total serum IgE level elevated above the age-specific norm. The institutional review boards of the participating centers approved the study protocol and informed consent was obtained from all probands or their legal guardians.
One discovery study sample and one replication sample were investigated. All families were of European origin. The discovery data set consisted of 199 complete affected sibling families composed of 427 children with AD that were studied in the original genome scan. [16] The replication set consisted of 292 families including 481 children with AD with an age of onset ≤2 y and moderate to severe disease expression. Among AD patients in the discovery and the replication cohorts, the proportion of boys was 52% and 50.9%, and the proportion of children with allergic sensitization was 74% and 72.1%, respectively.
Punch biopsies of human skin were obtained from six patients with AD and 7 healthy donors aged 24–55 y with written informed consent. Specimens were prepared for in situ hybridization and immunohistochemistry as described below.
In the first stage, fine mapping with microsatellite markers was performed in the discovery dataset. 96 short tandem repeat markers were selected within the interval between D3S1303 (126.07 cM) and D3S1292 (138.82 cM) from the Genethon (http://www.genethon.fr/) and Marshfield (http://research.marshfieldclinic.org/genetics/) databases, or were identified by screening human genome sequence data for short tandem repeats. Fluorescence-based semi-automated genotyping was performed as previously described [16]. Primer sequences, amplification conditions, and allele size are available on request.
We performed an association scan of the 5.4-Mb region using 212 SNPs at an average distance of 25.5 kb. The SNPs were selected from the NCBI database (http://www.ncbi.nlm.nih.gov/projects/SNP/) to cover known and predicted genes. To determine the allele frequency of the polymorphisms, we amplified 600 to 800 bp surrounding each SNP by PCR for resequencing in 31 unrelated Caucasian individuals. Markers with a minor allele frequency (MAF) of >5% were selected for genotyping. To identify functional variants in collagen XXIX, we sequenced all 42 exons including the exon–intron boundaries and 5.1 kb of the promoter region in 46 unrelated patients with AD and two controls. Sequencing was performed on an ABI3730 DNA sequencer (Applied Biosystems; http://www.appliedbiosystems.com) using standard procedures.
We carried out SNP genotyping using TaqMan allelic discrimination, with probes and primers designed and synthesized by the supplier (Applied Biosystems), or by pyrosequencing using PSQ HS 96A (Pyrosequencing AB; http://www.biotagebio.com/). The averages genotyping success rate was 97.9%. All primer and probe sequences are available on request.
We performed linkage analysis of the microsatellite data using Genehunter V. 2.1 [42]. Each SNP was checked for compliance with Hardy-Weinberg equilibrium (HWE) in the parent population using a χ2 test, and those markers that were not in HWE were excluded from the analysis. We calculated pairwise LD between each marker pair using the D′ statistic.
In view of the strong imprinting effect at our locus, we conducted family-based association tests, because this strategy allows us to determine the parental origin of an associated allele. In the affected sibling families, we used the sib_TDT of the ASPEX software that performs a permutation procedure to calculate empirical p-values that are independent of linkage within families [17]. Furthermore, to assess the significance of the maternal effect, we calculated empirical p-values for the difference in maternal versus paternal haplotype transmissions using the parent-of-origin TDT implemented in PLINK [18].
The sib_TDT was also used in the analysis of eight markers in the replication dataset. The significance level of the replication results was assessed empirically. Using all pedigrees and all genetic markers used in the actual analysis, we generated 10,000 replicates using Allegro V1.2c [43] and conducted an association analysis as in the original dataset. An empirical significance level was calculated as the proportion of replicates for which the maximum χ2 score was greater than that obtained in the real dataset. All p-values are two-sided, significance was defined as statistical evidence expected to occur 0.05 times at random in the analysis. For multipoint analysis, we used the FBAT tools package [44] to generate haplotypes and performed family-based association tests for five marker haplotypes using the empirical variance option to adjust for correlation among sibling genotypes. To evaluate parent-of-origin effects in the multipoint analysis, we estimated haplotypes using MERLIN, recoded the haplotypes as alleles, and performed the sib_TDT using ASPEX.
Protein sequences were retrieved from the UniProt (http://www.uniprot.org) and Ensembl (http://www.ensembl.org) databases. The domain architecture of the collagen XXIX protein was retrieved from the NCBI conserved domain search website (http://www.ncbi.nlm.nih.gov:80/Structure/cdd/wrpsb.cgi). The following domains were found in collagen XXIX and analyzed further: cd01472 (vWA_collagen), cd01450 (vWFA_subfamily_ECM), cd01470 (vWA_complement_factors), cd01465 (vWA_subgroup), and Pfam01391 (collagen triple helix repeat). To predict the 3D structure of the vWF protein domains in collagen XXIX, we explored the structure prediction results returned by the web servers GenTHREADER (http://bioinf.cs.ucl.ac.uk/psipred/) and FFAS03 (http://ffas.ljcrf.edu). Based on their very similar predictions, the human vWF A3 domain was chosen as the structural template to analyze the position of COL29A1 SNPs in 3D and to predict their potential effect on protein function.
A 2447-bp sequence from a human testis cDNA library which covered eight exons was the starting point for the characterization of COL29A1. Using rapid amplification of cDNA ends together with the Model Maker of NCBI, we identified a total of 42 exons and determined the transcription start site as well as the 3′ end of the COL29A1 transcript in cDNA from human skin. The complete sequence of the transcript was confirmed by cloning and sequencing of the full-length cDNA. To explore the potential gene function of COL29A1, the protein sequence was predicted (http://us.expasy.org/tools/dna.html), a domain search was performed (http://www.sanger.ac.uk/Software/Pfam/search.shtml) [45], and the presence and location of signalling peptides was analyzed (http://www.cbs.dtu.dk/services/SignalP/) [46].
We examined tissue-specific expression of COL29A1 using oligo(dT)-primed cDNA of 17 different human tissues. cDNA samples of 16 tissues were from the human MTC Panels I and II, which are standardized for the expression of GAPDH (BD Biosciences; http://www.bdbiosciences.com). In addition, human skin poly(A)+ RNA (BD Biosciences) was transcribed into cDNA using the Transcriptor First Strand cDNA Synthesis Kit (Roche Diagnostics; http://www.roche.com). PCR was performed using COL29A1-specific primers 5′-GTTCTAACCAGAATGTATAATCATC (forward) and 5′-TAATTCCCAAGAACATCTCTGGT (reverse), yielding a product of 694 bp, and the GAPDH control primers supplied with the MTC panels.
For in situ hybridization, we generated a plasmid by cloning a COL29A1-specific PCR product amplified from human skin cDNA [5′-ACCTTAGGAGACAGGGTTGCT (forward); 5′-AGTTCCAATCTGGCTTGTGG (reverse)] into the pCRII vector (Invitrogen; http://www.invitrogen.com). We synthesized antisense and sense digoxigenin-labeled riboprobes using the Dig RNA Labeling Kit (Roche Diagnostics). Punch biopsies of human skin were obtained from five AD patients and five healthy donors, immediately fixed in 4% paraformaldehyde for 4 h, cryoprotected in 30% sucrose overnight, and embedded in Tissue-Tek (Sakura; http://www.sakura.com) for cryosectioning. 10 μm cryosections were mounted on slides and dried for 15 min at 50 °C. Sections were postfixed in 4% paraformaldehyde for 7 min at 4 °C and acetylated with 0.25% acetic acid for 10 min. Sections were prehybridized for 3 h and hybridized overnight at 50 °C with digoxigenin-labeled riboprobes. After hybridization, sections were washed twice with 2 × SSC at 53 °C and once with 0.1 × SSC at 58 °C. For detection of the hybridized probe, slides were incubated with BCIP/NBT staining solution (Roche Diagnostics) for 4 d according to the manufacturer's recommendations.
To quantify gene expression in skin specimens, total RNA was isolated from 160 μm cryosections of skin biopsies using the RNeasy Mini Kit (Qiagen; http://www.qiagen.com). RNA was reverse transcribed into cDNA with random hexamer primers using the Transcriptor First Strand cDNA Synthesis Kit (Roche Diagnostics). Taqman real-time PCR was performed with iTaq SYBR Green (BioRad; http://www.biorad.com) and gene-specific PCR products were detected on the ABI PRISM 7900 sequence detection system (Applied Biosystems). All measurements were performed in duplicate. COL29A1 expression was normalized for 18S rRNA expression. Differences in gene expression were calculated using the ΔΔct method and were expressed as fold change. Gene-specific primers were as follows; COL29A1-forward, 5′-CCACCCTCTGGATCATCACT, COL29A1-reverse, 5′-GTTTTCTGTGCCACCGTTCT, KRT10-forward, 5′-CTGAAACCGAGTGCCAGAAT, KRT10-reverse, 5′-GCCTCCGGAACTTCCCTCT, 18S rRNA-forward, 5′-GGATGCGTGCATTTATCAGA, 18S rRNA-reverse, 5′-GATCAGCCCGAGGTTATCTA. The sizes of the PCR products were confirmed by gel electrophoresis and the specificity of the reaction was confirmed by melting curve analysis of the PCR products. For statistical analysis, the unpaired t-test was used.
A polyclonal antibody against human collagen XXIX protein was raised by immunizing rabbits with a collagen XXIX specific peptide (SLGSTRKDDMEELAC, residues 2115–2128) (Eurogentec; http://www.eurogentec.com). The specificity of the antibodies purified by affinity chromatography was tested by comparing their reactivity against recombinant proteins by Western blotting and by blocking experiments.
For immunohistochemistry, freshly isolated skin specimens from five AD patients and five healthy individuals were embedded in Tissue-Tek. Cryosections of 5 μm thickness were prepared and fixed with acetone. Sections were incubated with anti-collagen XXIX antibodies followed by dextran-coupled anti-rabbit antibody, an alkaline phosphatase labelled amplification polymer (DAKO EnVision System; http://www.dako.com) and detection with fuchsin (DAKO). Nuclei were counterstained with Mayer's hematoxylin solution (Sigma-Aldrich; http://www.sigmaaldrich.com). Omission of primary antibody and preincubation with equimolar amount of peptide used for generation of the antibody in rabbits were used as negative controls for parallel sections. The results were consistent among the AD patients on one hand, and among the control biopsies on the other.
The GenBank (http://www.ncbi.nlm.nih.gov/Genbank/index.html) accession numbers for genes and proteins discussed in this paper are: AK093199 (FLJ35880), NM 153264 (FLJ35880), and NP 476507 (human collagen VI alpha 3 chain),
The Protein Databank (http://www.pdb.com) accession number for the human vWF A3 domain is 1atz.
|
10.1371/journal.pcbi.1005482 | Combination therapeutics of Nilotinib and radiation in acute lymphoblastic leukemia as an effective method against drug-resistance | Philadelphia chromosome-positive (Ph+) acute lymphoblastic leukemia (ALL) is characterized by a very poor prognosis and a high likelihood of acquired chemo-resistance. Although tyrosine kinase inhibitor (TKI) therapy has improved clinical outcome, most ALL patients relapse following treatment with TKI due to the development of resistance. We developed an in vitro model of Nilotinib-resistant Ph+ leukemia cells to investigate whether low dose radiation (LDR) in combination with TKI therapy overcome chemo-resistance. Additionally, we developed a mathematical model, parameterized by cell viability experiments under Nilotinib treatment and LDR, to explain the cellular response to combination therapy. The addition of LDR significantly reduced drug resistance both in vitro and in computational model. Decreased expression level of phosphorylated AKT suggests that the combination treatment plays an important role in overcoming resistance through the AKT pathway. Model-predicted cellular responses to the combined therapy provide good agreement with experimental results. Augmentation of LDR and Nilotinib therapy seems to be beneficial to control Ph+ leukemia resistance and the quantitative model can determine optimal dosing schedule to enhance the effectiveness of the combination therapy.
| High likelihood of evolution of resistance to therapy is common in most forms of leukemia. This issue persists for tyrosine kinase inhibitor drug treatments as well as other forms of therapies. In the current work, we suggest a combination therapy where Ph+ acute lymphoblastic leukemic cells are treated with low-dose radiation before chemotherapy (Nilotinib). Our in vitro results of the combined therapy accompanied with a mathematical model shows successful suppression of resistance to Nilotinib. The mathematical model shows a synergistic interaction between Nilotinib and low dose radiation in the chemo dose response function. Beside acute radiation we investigate low dose fractionated therapies with model predicted optimal dosing schedules.
| The persistence of chemo-resistant leukemia-initiating cells in Philadelphia-chromosome positive (Ph+) B-cell Acute Lymphoblastic Leukemia (B-ALL) in the bone marrow is a primary mechanism responsible for disease relapse, following treatment, which occurs in the majority of patients. B-ALL is due, in part, to chromosomal translocations (9;22) that result in the generation of a BCR-ABL fusion protein, which fosters the transformation of immature B cells [1]. BCR-ABL+ (i.e., Ph+) leukemia has a poor prognosis; this is particularly true when matched with deletions in Cdkn2a, the gene encoding the tumor suppressor protein ARF, which occurs frequently in B-ALL [2, 3].
A significant breakthrough in the treatment of Ph+ ALL as well as the treatment of chronic myeloid leukemia (CML is associated with p210 isoform, whereas ALL is associated with p190 isoform) was the development of the tyrosine kinase inhibitor (TKI) Imatinib [1]). This drug, and the more potent second generation drugs Dasatinib and Nilotinib, are able to selectively inhibit the BCR-ABL mutant protein and thus significantly reduce Ph+ cell counts [2, 4]. While TKI therapy has long-term efficacy in the treatment of CML, most ALL patients eventually relapse following treatment with TKI due to the development of resistance [5, 6, 7, 8]. Thus a common treatment protocol for ALL patients is TKI therapy until the first remission [9, 10] followed by stem cell transplantation. However, since stem cell transplantation itself carries many risks to patient survival, the ability to extend the efficacy of TKI therapy in Ph+ ALL patients is of great clinical interest. Combination therapy such as Nilotinib with inhibitors of various other pathways (MEK, AKT, and JNK) showed greater reduction in cell viability and lowered risk of resistance [11]. Ionizing radiation has been used for leukemia disease in limited cases, e.g. i) disease involve in the central nervous system (CNS), potential due to ineffective penetration of chemotherapy to CNS [12], (ii) conditioning regimen with high doses of radiation and chemotherapy prior to stem cell transplantation for patients with high risk of relapse [13].
Taking advantage of leukemia radiosensitivity and the benefit of low dose radiation (LDR) in preserving bone marrow functions, we investigated whether the combination of Nilotinib and low dose radiation will be more effective treatment for BCR-ABL+ (i.e., Ph+) leukemia over Nilotinib alone. Furthermore, to optimize the effectiveness of this combination treatment, we developed a mathematical model, parameterized via cell viability experiments under Nilotinib treatment and radiation exposure, to predict cellular response to the combination therapy. The optimized mathematical model predicts a synergy between LDR and TKI treatment. We propose a combined Nilotinib dose-response function after LDR that accounts for a possible synergistic interaction between LDR and TKI treatment. Model parameters are obtained from in vitro viability measurements in the absence of TKI (Fig 1(a)), with zero LDR (Fig 1(b)) and combination of LDR and TKI for several radiation doses (Fig 1(c)). The model is validated by precise prediction for the drug-dose responses and radiation-dose response to combination LDR and TKI treatment.
It is important to emphasize that our model is focused on the relevance of LDR to prevent small-molecule inhibitor drug resistance. It does not address the efficacy of a successful radiation-drug combination treatment. Answering the question that whether the resistance is pre-existing, selected for or evolves de novo is out the scope of the current work. We do assume a small fraction of pre-existing Nilotinib resistant subtype, as well as possibility to transform into resistant type in the presence of the drug.
Our computational model is a coupled system of ordinary differential equations representing two populations. A Nilotinib sensitive and a Nilotinib resistant population. The use of ordinary differential equations is very common to describe the population dynamics and emergence of a new trait [15, 16] Logistic terms impose limitation on growth of each population, and an additional term allows for conversion from Nilotinib sensitive to Nilotinib resistant phenotype in the presence of non-zero concentration of the drug. More specifically, we assumed the following for the dynamics of the two subpopulations of the sensitive and resistant phenotypes:
The dynamical model that describes the above mechanisms is detailed in the Supplementary Information (SI). We refer to this model as the proliferation-mutation model.
To identify the response of Ph+ ALL cells to Nilotinib treatment, radiation, and the combination of both therapies, we proposed a simple functional form of this dependence and evaluated the fit against a large set of dose response data from experiment. This simple linear dose-response model can be written as follows:
( proliferation rate ) = r 0 - ( Nilotinib dose ) × r 1 + radiation dose × r 2 , ( apoptosis rate ) = d 0 + ( Nilotinib dose ) × d 1 + radiation dose × d 2 , (1)
where the growth rate coefficients ri and di are constants that need to be identified based on the experimental results. It is common to model dose response functions using a Hill function structure (3- or 4-parameter logistic function). The Hill function imposes a low level of drug efficacy at low doses and a saturation of drug efficacy at high doses. A linear response is most accurate to approximate a Hill function dose-response near IC50 does—which is the case here. In the SI section, we show that how the above linear dose-response functions can be derived from a more common Hill -function.
Eq (1) can be rewritten as
r S = r S , 0 − ( r S , 1 + r S , 2 D ) c , r R = r R , 0 − ( r R , 1 + r R , 2 D ) c , d S = d S , 0 + ( d S , 1 + d S , 2 D ) c , d R = d R , 0 + ( d R , 1 + d R , 2 D ) c , (2)
where c and D represent the Nilotinib and radiation doses respectively. The constants rS,0 and rR,0 denote the proliferation rates of sensitive and resistant populations in the absence of therapy, and dS,0 and dR,0 denote the death rates of sensitive and resistant populations in the absence of therapy. The coefficients rS,1, rR,1 represent the dose-response relationship between Nilotinib and proliferation rate of sensitive and resistant cells, respectively. Similarly, the coefficients dS,1, dR,1 describe how Nilotinib impacts the death rate of sensitive and resistant cells, respectively. Lastly, the coefficients rS,2, rR,2, dS,2, dR,2 determine the strength of the radiation-drug interaction on proliferation and death rates sensitive and resistant cells, respectively. These coefficients were fit to experimental data sets studying proliferation and death rates at a variety of Nilotinib and radiation doses. For example, non-negligible positive fitted values of rS/R,2 and dS/R,2 reveal synergistic interaction between the therapies.
We also incorporated an immediate cell-kill term after each radiation dose in accordance with the standard linear- quadratic model [17, 18, 19, 20]. According to the LQ model the effects of radiation cell kill is given by the survival fraction after the radiation exposure
Surviving fraction=exp [ − α × ( radiation dose ) − β × ( radiation dose ) 2 ) ] (3)
where α and β are the radio-sensitivity parameters to be determined from the experimental data. In short, α represents the rate of cell kill due to single tracks of radiation and β represents cell kill due to two independent radiation tracks. The linear quadratic model is widely used due to its excellent agreement with empirical data for a wide range of radiation doses. In SI section we explain how to incorporate the linear quadratic framework for surviving fraction into our mutation-proliferation model.
The above mathematical framework can predict the population fraction (cell viability) at each day given the initial values of Nilotinib and irradiation doses. We fit the model parameters in an iterative fashion. All parameter estimation are obtained by finding optimal parameter set that minimizes the square root distance of solution of Eq. S3 (in SI) with respect to the time series viabilities at days 0, 3, 6, 8, 10. The steps are as follows:
We first measured the time-dependent cell viability of Ph+ ALL cells in vitro in response to Nilotinib, radiation, and combination therapy with both Nilotinib and radiation (Fig 3(a)). For the radiation-only arm, we observed an initial large reduction inviability by day 6 for 2 Gy and day 8 for 4 Gy. Experiments using Nilotinib monotherapy showed an incremental reduction in cell viability over the first 6 days to about 57.2%. Subsequently the cell populations began to develop a resistance to the drug and viabilities began to increase. When used in combination, Nilotinib + radiation induced a more effective initial cell killing and the cancer cell population was controlled at very low numbers (under 10% viability) for the duration of the experiment. In other words, there was nearly no cell population recovery, and resistance to Nilotinib was not developed. Under the Nilotinib + 4 Gy treatment arm, the cell population was entirely eliminated. Thus, the combination therapies were able to not only elicit a larger reduction in cell population, but also to maintain control of low cell viabilities without the development of resistance over a longer time period. The use of Triciribine was able to keep the cell viability as low as Nilotinib+ radiation group (Fig 3(b)).
The radiation dose-response assay revealed an LD50 of 2 Gy in the absence of Nilotinib, and 0.5 Gy in the presence of 18 nM Nilotinib. The Nilotinib dose-response assay revealed an IC50 oof 18nM in the absence of radiation. To investigate how the combined therapy provides the synergistic effect, we evaluated one of the key pathways associated with cell proliferation, and chemoresistance. At day 3 after treatment, p-AKT was slightly increased by adding either Nilotinib or radiation (Fig 4(a)). However, the expression level was gradually decreased at large dose of radiation (6 Gy) with or without Nilotinib (Fig 4(b)). Notably, the combined therapy eventually (day 7 and10) decreased the p-AKT even at 2 Gy (Fig 4(c)).
Using the numerical solutions of the proliferation-mutation model optimized for the best fit with experimental data points, we determined the coefficients in the Nilotinib dose-response function, ri and di, as well as the transformation rate (Table 1). The optimized coefficients indicate an increase in fitness of resistant cells relative to that of sensitive cells in the presence of Nilotinib. (Fitness is defined as the difference between division rate and apoptosis rate in the cell population.) Using a numerical sensitivity analysis, we confirmed that these parameter estimates are robust to perturbations in the initial frequency of resistant cells in the population. Since the experiments demonstrated an initially strong sensitivity of the populations to Nilotinib treatment, we set the frequency of resistant cells to be small. We set this to be 0.1% of the total population for the remainder of investigations. The optimized parameters are reported in Table 2 with the carrying capacity set to K = 4.2 in all cases. These results indicate the presence of synergistic effects between Nilotinib and radiation. In particular, the radiation tyrosine-kinase inhibitor interaction was strongest in the resistant population. This is not surprising as in the presence of Nilotinib the sensitive population is already highly disadvantaged in terms of growth. The value of transformation rate ν in this case is independent of radiation dose and is higher than the respective transformation rate in the Nilotinib-only case suggesting that radiation exposure contributes to the production of new resistant cells.
The results for population fractions (cell viabilities) as a function of time (days) were plotted in Fig 5. The agreement between mathematical model predictions and the in vitro measurements suggests that the proposed radiation-drug interaction term in the linear dose-response is a plausible choice. All the above results of the parameter estimation process are summarized in the Tables 1 and 2.
We next tested predictions of the fitted model with an independent set of experimental results. In particular, we compared model predictions to the experimental results of the dose-response assays for Nilotinib (in absence of radiation) and radiation (at 0 and 18 nM Nilotinib). See Fig 6 for these comparisons. As can be seen, the results are in very good agreement. We then used the model to predict the Nilotinib dose-response curve with 1, 2, 3 and 4 Gy radiation. The results are plotted for 0-22nM of Nilotinib. Note that at higher doses we see a discrepancy with the experiment (25nM 0Gy). This is due to the fact that the linear approximation of the Hill dose-response curve only applies to the initial section of the curve and is expected to break down at higher doses.
The combination therapy above was shown to reduce tumor cell viability, and through computational techniques an optimal schedule of dosing may be approached. Given the potential toxicity of Nilotinib to normal tissues, we first assumed a standard, constant dose of 18 nM Nilotinib. Next, we considered a five-day radiation treatment protocol, where the summed dose of the protocol is constant. As an example, we use a total dose of 2Gy; which, as can be seen in Fig 5, has room for improvement with the combination therapies and does not irradiate the cells to a viability that is not interesting mathematically (4Gy in Fig 5, for example). The aforementioned result will serve as a comparison for our proposed protocol, which will attempt to minimize the total tumor cell viability at day 10. The control parameters are the radiation doses given at days 0, 1, 2, 3 and 4 denoted by Di(i = 1, 2, 3, 4, 5). Placed under the constraint of the total allowable dose, the control parameters were varied by a nonlinear constrained minimization protocol that sought to minimize the tumor cell viability at day 10. Under this minimization protocol, the minimum tumor cell viability was determined by searching the potential dose regimen in the answer space and producing the dosing protocol that best minimized the tumor cell viability at 10 days. This resulted in an optimal dosing schedule of (in Gy) D1 = 0.9371, D2 = 0.5139, D3 = 0.6445, D4 = 0.045, D4 = 0.0064, which front-loads the radiation in the first three days. Notice that the negligible value for D4 indicates that optimal solutions are effectively that of a 4-fraction protocol, with variable doses per fraction. In between any two radiation doses, the cells undergo repopulation. For acute radiation protocols, the linear dose response function predicts a change in proliferation potentials of both sensitive and resistant cells which depends on the total radiation dose at day 0. For a fractionated protocol, we assume the the proliferation potentials of the two cell types between the kth and k+1th radiation doses depends on total radiation dose until fraction k. The model prediction for total cell viabilities versus time for the optimal fractionation protocol is plotted in Fig 7 and compared with a constant 5-fraction radiation protocol (Di = 0.4) and 2Gy acute radiation treatment (D1 = 2Gy, D2 = D3 = D4 = D4 = 0 Gy). At larger time, that is, greater than 10 days, the optimal protocol has the lower cell viability more importantly in a downward trend beyond the 10 day point, whereas the acute treatment begins an upward trend and the constant fractionation protocol does not suppress the cell viability to the levels of the optimal or acute protocols. This suggests the synergistic interactions between the therapies is heightened with larger initial doses of fractionated radiation, and the optimal protocol for long term suppression falls between acute dosing and constant fraction protocols, which results in the front-loading protocol. This dose dependent fractionation may be generalized to various scenarios in the clinical setting as determined by clinical status. Further, the computational model may be extended to consider alternative Nilotinib dosing strategies in combination of the radiation protocol to minimize the cell viability at day 10. Concurrent use of the 5-day optimal fractionation protocol with a varied Nilotinib dosing strategy shows an alternative, though equally efficacious (similar cell viability at 10 day) strategy. This alternative Nilotinib dosing strategy maintains the average daily dose of 18 nM, however, it begins at lower concentration and increases throughout the course of treatment. Specifically, the Nilotinib dose over the first 3 days is 10 nM, followed by 18 nM for 4 days, and finally 26 nM for 3 days, essentially back-loading the Nilotinib onto the irradiated cells. After 10 days, the dose returns to 18 nM daily. This protocol is visualized in Fig 7 as well, where at larger times the new strategy of Nilotinib dosing reduces the cell viability as well as the previous protocol. This is clinically relevant as dose titration is common, if needed, and there is no trade off with treatment efficacy should this dosing schedule be indicated.
Although the clinical outcome of BCR-ABL leukemia has been improved with the advent of TKI therapy, overcoming chemo-resistance has been a major challenge. This study demonstrated that low dose irradiation combined with Nilotinib provided enhanced and prolonged efficacy for leukemic cells in vitro. A companion theoretical model provided good agreement with experimental results with an opportunity for further optimization to enhance treatment efficacy. We explored here the possibility of using low dose radiation as a first line therapy in combination with chemotherapy to enhance the effectiveness controlling BCR-ABL leukemia. There are mainly two reasons for such a combination therapy approach: a) advent of image guided targeted radiation allow more focused dose delivery [21, 22, 23] b) known radiosensitivity of leukemia but not known if low dose radiation with chemotherapy could be an effective alternative to control over the TKI drug resistance. We found that combining low dose of radiation with a commonly used TKI drug could not only substantially reduce cell population, but also maintain low cell viabilities without the development of resistance over a longer time period. To elucidate the mechanism for which low dose radiation provide beneficial effect, we investigated the role of AKT pathway using western blot analysis (Fig 4). Our results revealed that high dose radiation (i.e. 6 Gy) reduced the phosphorylation of AKT. Interestingly, even low dose radiation was adequate to inhibit the phosphorylation of AKT when Nilotinib was also used. To support of this, the cell viability of chemo-resistant cells (with increased p-AKT) was significantly reduced when AKT inhibitor (Triciribine) was used in combination with Nilotinib. These results suggest that radiation can play roles not only in a conditioning regimen before stem cell transplant but also an alternative for an AKT inhibitor. In other words, low dose radiation therapy could be an alternative treatment option for AKT inhibitor in ALL patients. To our best knowledge, this is the first report that demonstrates the promising combined efficacy of Nilotinib and radiation for ALL, with minimum damage to vital organs. Because of low dose with limited toxicity, it can be delivered to the whole body. Since our study was performed in normal cell culture, however, in vivo study is required to take into consideration of microenvironmental factors.
We also constructed a dynamical model to explain the observations and to predict response to additional combination therapy schedules. To tease apart the responses of Ph+ ALL cells to Nilotinib, radiation, and the combination of both therapies, we proposed a simple functional form of the combination dose-response relationship and evaluated the fit against a large set of dose response data. In particular, we proposed a simple Nilotinib dose-response function in which radiation dose may alter the strength of the Nilotinib response in a dose-dependent fashion. Parameter-fitting revealed an optimal parameter set that showed very good agreement with experimental results. Analysis of this optimal parameter set revealed dose-dependent synergistic effects between Nilotinib and radiation response. In particular the radiation tyrosine-kinase inhibitor interaction is strongest in the resistant subpopulation of cells. Our analysis also demonstrated that the model predictions are robust to variation in the initial frequency of resistant cells. We compared fitted model predictions with an independently generated second set of experimental data and found good agreement.
We next utilized the validated model to investigate optimal combination strategies for Nilotinib and radiation in Ph+ ALL. As a simple test, we assumed a standard 18 nM Nilotinib dose and investigated strategies allowing up to 2 Gy over the course of a 5-day treatment. We determined that the optimal therapeutic schedule, given these constraints, spread most of the radiation dose over the first three days. Thus, a ‘sweetspot’ exists between acute radiation protocols and constant treatment protocols. These results suggest a promising direction for investigation of new treatment strategies in Ph+ ALL and for providing an optimal treatment regimen. However, we note that all experiments (and thus parametrization of the model) was done in vitro. Further in vivo studies are needed to determine treatment schedules for the clinic with more accuracy. To summarize, augmentation of LDR-Nilotinib therapy may be beneficial to control Ph+ve leukemia resistance and the model can determine optimal dosing schedule to enhance the effectiveness of the combination therapy.
|
10.1371/journal.pntd.0005214 | Transmission dynamics and elimination potential of zoonotic tuberculosis in morocco | Bovine tuberculosis (BTB) is an endemic zoonosis in Morocco caused by Mycobacterium bovis, which infects many domestic animals and is transmitted to humans through consumption of raw milk or from contact with infected animals. The prevalence of BTB in Moroccan cattle is estimated at 18%, and 33% at the individual and the herd level respectively, but the human M. bovis burden needs further clarification. The current control strategy based on test and slaughter should be improved through local context adaptation taking into account a suitable compensation in order to reduce BTB prevalence in Morocco and decrease the disease burden in humans and animals. We established a simple compartmental deterministic mathematical model for BTB transmission in cattle and humans to provide a general understanding of BTB, in particular regarding transmission to humans. Differential equations were used to model the different pathways between the compartments for cattle and humans. Scenarios of test and slaughter were simulated to determine the effects of varying the proportion of tested animals (p) on the time to elimination of BTB (individual animal prevalence of less than one in a thousand) in cattle and humans. The time to freedom from disease ranged from 75 years for p = 20% to 12 years for p = 100%. For p > 60% the time to elimination was less than 20 years. The cumulated cost was largely stable: for p values higher than 40%, cost ranged from 1.47 to 1.60 billion euros with a time frame of 12 to 32 years to reach freedom from disease. The model simulations also suggest that using a 2mm cut off instead of a 4mm cut off in the Single Intradermal Comparative Cervical Tuberculin skin test (SICCT) would result in cheaper and quicker elimination programs. This analysis informs Moroccan bovine tuberculosis control policy regarding time frame, range of cost and levels of intervention. However, further research is needed to clarify the national human-bovine tuberculosis ratio in Morocco.
| Tuberculosis is a disease of humans and animals which mainly affects the lungs but can also manifest in other organs. A variety of tuberculosis bacteria cause the disease and are usually transmitted through air, i.e. inhalation of aerosols. Bovine tuberculosis (BTB) occurs predominantly among domestic cattle, although wild animals are an important reservoir for transmission. Humans are usually infected with BTB through contaminated dairy products or close contact with cattle. While BTB has been eliminated in cattle and human populations of most high-income countries, it is still a major health threat in low- and middle-income countries. In Morocco, the disease frequently occurs in cattle and poses a health risk for humans. An effective intervention to reduce BTB among domestic cattle and reduce the risk to humans is slaughter of cattle which test positive for the disease. We simulated BTB in the Moroccan cattle and human populations using a disease transmission model. We assessed effects of test and slaughter in regard to elimination of disease in cattle and humans and estimated the associated costs with the model. The time to elimination of disease depended on the number of cattle tested. Our model suggests that the disease might be eliminated in cattle within 32 years if 40% of Moroccan cattle are tested annually and infected individuals are slaughtered or within 13 years if at least 90% of the cattle population are tested. The estimated total costs for the time periods until elimination ranged from 1.55 billion euros for testing and slaughter at 50% to 1.48 billion euros at 90%. These results can be used as a guide for planning BTB control policy in Morocco with regard to time frames and associated costs.
| Bovine tuberculosis (BTB) is a zoonotic bacterial infection caused by Mycobacterium bovis. It belongs to a group of well-known and newer mycobacteria, together with Mycobacterium tuberculosis, all of which derive from a common ancestor forming the Mycobacterium tuberculosis complex (MTBC) [1–4]. M. bovis is capable of infecting a broad range of hosts, including ruminants (predominantly domestic cattle), humans and other primates [5–10]. The wide host range makes BTB highly relevant to conservation projects and difficult to eliminate where wildlife reservoirs are involved, for instance, badgers (Meles meles) in the United Kingdom [11, 12].
Bovine tuberculosis infection in cattle is a chronic disease which first affects the lymph nodes and from weeks to decades later can affect lungs. The disease can also be manifested in other organs such as, mammary tissue, and the gastrointestinal or urinary tract. Since transmission between cattle occurs predominantly through aerosol inhalation [13–17], the transmission rate is increased by risk factors such as high herd density and intensive breeding [18]. Pseudo-vertical transmission from cows to suckling calves through infected milk has been described [19]. Factors like a long survival period for the microbes in manure and soil and close contact between animals, for example around water sources, also contribute to an increased risk of infection [20, 21]. In humans, contaminated dairy products are considered to be the main source of BTB infection, usually resulting in extra-pulmonary infection such as lymphadenitis [22–24]. These patients are missed by thoracic radiographic screening and the resulting diagnostic cascade [25]. Aerosol cattle-to-human transmission can occur during close contact with infected animals, posing an occupational risk, especially for pastoralists and farmers [2, 26]. Infection risks linked to local cultural practices, for instance consumption of fresh blood, are reviewed by Daborn [27]. There is evidence that human patients can transmit BTB to animals, and human to human transmission occurs [28, 29].
There is a bottleneck in detecting human BTB cases because the routine diagnostic protocols were developed for patients with pulmonary tuberculosis, as caused by M. tuberculosis. Tuberculosis (TB) and BTB cannot be distinguished on the basis of clinical symptoms, radiography or histopathology [30]. Glycerol-containing Löwenstein-Jensen medium, the long-time gold standard for TB culture, inhibits the growth of M. bovis, thereby increasing the number of undetected cases [31]. New molecular diagnostic tools, for example spoligotyping, and even whole genome sequencing have been developed for M. bovis detection [32, 33]. Although they require enhanced laboratory infrastructure and personnel training which are not currently available in some developing countries, these new techniques offer promise for epidemiological research, control and adequate treatment, particularly since M. bovis is resistant to pyrazinamide, one of the first-line antibiotics for TB treatment [34].
Morocco is transitioning from extensive pastoralist livestock and dairy production to more intensified production due to increasing demands for dietary protein by a growing human population [35]. The shift in agricultural practice and increased use of high-producing Holstein cattle in place of local breeds may have an impact on BTB epidemiology and contribute to a higher prevalence [36]. The official national control program in Morocco is currently based on a test and slaughter scheme. However, large-scale application remains challenging because testing is not mandatory, and the proposed compensation, ranging from 470 euros for local breeds to 980 euros for improved breeds, is considered lower than market value.
In Morocco, BTB is an endemic zoonosis in livestock. Even though the predominant livestock species in Morocco are sheep and goats, cattle remain of major importance. The most recent national survey, conducted in 2004, showed an individual cattle prevalence of 18% and a herd prevalence of 33% [37]. This prevalence remained similar in the individual level (20%), while the herd prevalence increased (58%) in a 2012 pilot study of 1,200 cattle using the tuberculin skin test [38]. Since 2000, the health risk of tuberculosis in Morocco has been addressed through a national TB program funded by the Ministry of Health in collaboration with the World Health Organization (WHO). In 2014, TB caused 2’800 deaths in Morocco [39], and human tuberculosis had a relatively high incidence, with 36’000 new cases (106 cases per 100,000 inhabitants) [40]. These data do not appear to differentiate between M. tuberculosis and M. bovis infection. A recent meta-analysis estimated the median proportion of human BTB among all TB cases in 13 African countries at 2.8%, with a range of 0–37.7% [41]. National prevalence data from a range of countries worldwide were summarized in a 2014 review; in Mexico up to 13% of all TB cases are reportedly due to BTB, while in the United States it is only 1.4% [42]. In Morocco, Bendadda et al. reported M. bovis prevalence of 17.8% among drug resistant TB isolates from 200 human sputum samples [43].
In the early 20th century, the prevalence in German cattle herds was 90% [44], with 25–80% in other European states and only 2–10% in the US [45]. In most industrialized countries, the health risk and economic loss from M. bovis were considerably reduced or eliminated through strict test-and-slaughter and meat inspection protocols for cattle, along with the implementation of milk pasteurization and financial compensation of farmers [45]. Using a similar control strategy, Switzerland eradicated BTB in 1960 [46]. In most developing countries where the disease is endemic, such measures are not feasible due to financial constraints, particularly for farmer compensation, and inadequate veterinary services [47]. Alternatives for BTB endemic countries to reduce the health and economic risk related to the disease must be sought.
Although cost estimation is difficult due to immense local variability in production parameters and prices [48], the global economic loss due to BTB is thought to be about 3 billion USD annually [49]. In cattle, the disease has a significant economic impact, through an increased death rate and decreased milk and meat production, draft power and fertility [50]. Modelling approaches have been used, mainly for developing countries, to estimate different parameters and factors related to BTB. Previous publications on the economics of BTB focus mainly on the cost of disease and control efforts. Analyses of the profitability of control efforts are very scarce [51]. A study on the economics of BTB in Africa showed higher losses in intensive dairy systems in peri-urban areas of Ethiopia when compared to extensive pastoral production systems in rural areas but did not include the cost to public health [48]. Zinsstag et al presented a simplified framework for a model of animal to human transmission in which transmission between cattle and from cattle to humans is considered [51]. This model allows the simulation of different scenarios over 5–10 years, with and without intervention, where the measurable outcome is prevalence in humans and in cattle. The economic analysis further delineates broader issues such as inter-sectorial contributions from agricultural and public health or private households affected by BTB.
Interventions to reduce health and economic risks, such as those related to BTB, are non-linear processes. Although, statistical analyses of data have been used for many years to analyze different types of health interventions, mathematical modeling represents an alternative approach which provides a broader understanding, especially regarding disease transmission to humans.
Several mathematical BTB models have been developed to study the transmission dynamics and to assess the effectiveness of control measures mostly for wildlife (badgers and possums) but also for cattle [52]. In Italy, a compartmental stochastic model of within and between-farm BTB dynamics in cattle was developed using a monte-Carlo simulation of BTB epidemics following the random introduction of infected individuals in the network. Consequently, slaughterhouse inspection has been found to be the most effective surveillance component [53]. The same methodology was used in Great Britain in order to study BTB transmission dynamics and to assess the currently used control measures [54]. A model composed of two sub-models for buffalo and cattle populations in South Africa showed that BTB infection is only sustained in cattle and buffaloes when all transmission routes are involved [55]. In France, a compartmental stochastic model was used to assess within-herd spread of BTB operating in discrete time in order to design, calibrate and validate a model of spread of M. bovis within a cattle herd. Various herd management practices as well as control programs were parameterized into the model. Therefore, the median effective reproductive ratio was estimated to be 2.2 and 1.7 respectively in beef and dairy herds [56]. This paper presents an ordinary differential equation (ODE) mathematical model of BTB transmission from cattle to humans in order to estimate the disease cost and simulate potential interventions in Moroccan cattle.
Annual data on cattle numbers are estimated using data routinely collected by the Ministry of Agriculture and reported to veterinary services. Our model considered these data from 1995 to 2013. In the transmission model, the cattle population was not stratified by age and sex. We used a tuberculin prevalence of 18% for cattle, as reported in the most recent national survey (2003) [33]. A similar prevalence was noted in a smaller study performed in 2012 in Sidi Kacem, Morocco [38].
A schematic diagram of the model is depicted in Fig 1 and the variables and parameters are described in Tables 1 and 2, respectively.
The cattle population is divided into three mutually exclusive compartments consisting of susceptible cattle (SC), exposed cattle with latent BTB which are positive to the tuberculin test, without showing gross visible lesions (EC) and infected cattle with active BTB showing tuberculosis lesions (IC). Those parameters were estimated based on Ngandolo et al 2009 [57]. The total cattle population (NC) at time t is:
NC(t) = SC(t) + EC(t) + IC(t)
(1)
The human population consists of four mutually exclusive compartments: susceptible humans (SH), exposed humans with latent BTB reacting to the Mantoux test (EH), infected humans with active BTB (IH) and humans recovered from BTB with temporary immunity (RH). The total human population (NH) at time t is:
NH(t) = SH(t) + EH(t) + IH(t) + RH(t)
(2)
The susceptible cattle population (SC(t)) increases through birth (at a rate bC) and decreases through exposure to BTB (at a rate βC) and mortality (at a rate μC). Exposure to BTB is assumed to be frequency dependent. According to Bernues [58], the rate bC decreases by 5% for exposed cattle with latent BTB (EC(t)) and for infected cattle with active BTB (IC(t)), such that:
dSCdt=bCSC+(0.95×bC(EC+IC))−βCICSCNC−μCSC−(1−sp)pSc
(3)
The population of exposed cattle with latent BTB (EC(t)) is generated through infection of susceptible cattle with BTB (at a rate βC) and decreased through the development of active BTB (at a rate αC) and through mortality (at a rate μC). Consequently:
dECdt=βCICSCNC−αCEC−μCEC−se*pEc
(4)
Similarly, the infected cattle population with active BTB (IC(t)) is generated through the development of active BTB among exposed cattle with latent BTB (at a rate αC) and decreased through mortality (at a rate μC):
dICdt=αCEC−μCIC−se*pIc
(5)
For simplicity, no additional mortality rate due to BTB is assumed for the cattle and human populations in the model.
The susceptible human population (SH(t)) increases through birth (at a rate bH) and through recovered humans becoming susceptible again (at a rate δC). The population decreases through exposure to BTB from cattle with active BTB (at a rate βH) and through natural mortality (at a rate μH). For the susceptible human population, the exposure (both direct (aerosol) and indirect (milk) transmission) to BTB from cattle with active BTB is assumed to be frequency dependent and proportional to the number of infected cattle (IC):
dSHdt=bHSH−βHICSHNC−μHSH
(6)
Human to human transmission is assumed to be negligible. The population of exposed humans with latent BTB (EH(t)) is generated through infection of susceptible humans with BTB (at a rate βH) and decreased through the development of active BTB (at a rate αH) and through natural mortality (at a rate μH):
dEHdt=βCICSHNC−αHEH−μHEH
(7)
The infected human population with active BTB (IH(t)) is generated by the development of active BTB among exposed humans with latent BTB (at a rate αH) and decreased by recovery of humans with active BTB due to treatment (at a rate γH) and by natural mortality (at a rate μH):
dIHdt=αHEH−γHIH−μHIH
(8)
The population of humans recovered from BTB and temporarily immune due to treatment is generated through the recovery of humans with active BTB (at a rate γH) and decreased through humans becoming susceptible to BTB again after the end of the prophylactic period of the drugs (at a rate δH) and through natural mortality (at a rate μH), so that:
dRHdt=γHIH−δHRH−μHRH
(9)
A sensitivity analysis of the model recalculated the change of prevalence if individual parameters varied from baseline over 30 years.
Although Morocco has a test and slaughter policy for the control of BTB, it is currently not effectively implemented. The BTB transmission model was used to estimate the effect of the proportion of tested and slaughtered tuberculin positive animals on the duration to reach freedom from disease, achieving an individual animal prevalence of less than one in a thousand tested animals (<1/1000) according to the standards of the World Organization for Animal Health (OIE) [65]. The proportion of tested and slaughtered animals was simulated by removing 10–100% of the exposed (Ec) and infectious cattle (Ic) per year from the herd. The control reproductive number Rc including the test and slaughter intervention as proportion p with a test of sensitivity se was computed as:
Rc=α*β(α+μ+se*p)(μ+se*p)
(18)
The associated costs were estimated in a summaric way, assuming an incremental cost of comparative intradermal or interferon gamma (BOVIGAM) testing of 3 euros per animal. The cost of compensation at 80% of the market value varies from 470 euros for local breeds to 970 euros for improved breeds [66]. For the current estimation of the cost of BTB elimination in Morocco, we used a single value of 500 euros of compensation per slaughtered animal. Models run with and without interventions were simulated using data from 2013 onwards.
The OIE recommended cut-off for SICCT interpretation is 4 mm, however, many studies showed that a severe cut-off of 2mm increased the sensitivity of the test [67, 68], without affecting the specificity compared to the recommended cut-off [69]. Consequently, we decided to consider both 2mm and 4mm cut-offs in the model, and to compare the respective results. The present model considered both options of SICTT cut-off at 2mm and 4mm.
The reproductive ratio of the cattle to cattle transmission of bovine tuberculosis without intervention was 1.325. For the total cost, the birth rate of cattle bc was the most sensitive parameter influencing the dynamics of BTB transmission (Fig 2). High birth rate values lead to an increased cattle population yielding higher costs for elimination. For the time to elimination, the sensitivity of the test was the most sensitive parameter. Low test sensitivity (i.e. with cut-off at 4mm) leads to low detection of infected animals and therefore less culling of infectious cattle, which leads to a longer time to elimination.
The simulation of a test and slaughter intervention led to a decline in BTB prevalence depending on the proportion p of testing (Fig 3). The time to elimination, i.e. the time to reach an individual animal prevalence of less than one in a thousand, ranged from 75 years for p = 20% to 12 years for p = 100%. For values of p > 60%, the time to elimination was below 20 years (Fig 3). The reproductive number decreased rapidly below one with an increasing proportion test and slaughter p (Fig 4). With 60% testing and culling, the prevalence of exposed and active human BTB cases decreased from 3.5 per 1,000.000 to less than 1 in 1,000,000 at the time of freedom from disease after 20 years (Fig 5).
The cost of test and slaughter depends on the percentage p of test and slaughter (Table 5). Lower p results in lower cumulated costs but longer time until elimination. The cumulated cost is remarkably stable for p values higher than 0.2, ranging between 1.47 to 1.87 billion Euros within a time range of 12 to 75 years to reach freedom from disease. The cumulated cost of BTB test and slaughter intervention and the time to elimination were lower using 2mm cut-off of SICTT compared to the 4mm cut-off (Fig 6).
This manuscript presents the first cattle to cattle and cattle to human compartmental deterministic mathematical model. Differential equations were used to describe different pathways within and between compartments at human and animal level. Sensitivity analysis of the model has been used to determine the most sensitive parameter. Additionally, different scenarios of test and culling interventions were simulated by considering different proportions of tested population per year (p).
To our knowledge this is the first model describing cattle to cattle and cattle to human transmission of BTB in Morocco, although an African buffalo-human model has been published [70]. Time series data similar to those for brucellosis in Mongolia [71] are unfortunately not available, but the available data allows for a parameterization under the assumption of endemic stability similar to Ethiopia [72]. The estimated reproductive number of 1.325 is in the range for both low risk areas (R0 = 0.6–1.4) and high risk areas (R0 = 1.3–1.9) reported for the United Kingdom [73] but is lower than the 1.7–2.2 reported by Bekara et al. for France [56]. Using a sensitivity analysis, the birth rate of cattle (bc) was determined to be the most sensitive parameter of the model.
A key challenge in this model was to distinguish between exposed and infected cattle because the diagnostic test utilized was the intradermal tuberculin test. We used the proportion of cattle with active TB (13.5%) among cattle tested positive by tuberculin skin test, as reported by Ngandolo et al [57], to calculate the number of infected cattle. Further microbiological data is required to better describe BTB prevalence in humans in Morocco. Patients treated for active BTB do not completely clear all organisms from their body, with some bacteria persisting in bone marrow [61]. Therefore, in contrast to our model, humans do not become completely susceptible again but technically are subject to re-infection [62]. We argue that this has only a minor impact on total human BTB prevalence, but re-infection should be considered to refine the model.
Our model contains many simplifications. The primary simplification is that of homogeneity: all cattle are not the same. Risk exposure of animals and humans to BTB could change according to sex and age, and we have ignored these differences. Contact between cattle is also not random and is far more likely within herds than between cattle in different herds. Many models have included this heterogeneity in contact patterns within and between herds [52–54], but we ignore it here because of a lack of data on within herd BTB transmission in Moroccan husbandry systems. Furthermore, although deterministic models provide reasonable estimates of mean behaviour in large populations, they cannot provide expected distributions of rare events. Therefore, they may not be appropriate for analysing very low transmission settings (that are necessary before elimination can occur). We circumvent this issue with a rather generous definition of elimination as prevalence of less than 1 in 1000.
BTB prevalence was found to reach less than one per thousand in less than 20 years when the proportion of tested cattle was above 60%. The annual cost for this potential intervention was nearly 77 million euros. Intervention in cattle was found to impact the prevalence of human TB due to M. bovis, which decreased from 5 per 10,000 to 1 per 10,000 after 17 years.
The economic assessment presented here is preliminary, and a detailed cost and cost-effectiveness analysis will be published separately. However, our analysis informs Moroccan bovine tuberculosis control policy on the time horizon, range of cost and optimal levels of intervention. An effective control program will depend on the human resources and technical and logistical capacity of the veterinary services to implement testing and slaughtering of animals. If the proportion of cattle subjected to test and slaughter was greater than 60%, freedom from disease would be reached in less than 20 years. The simulation results suggest that switching from a 4mm cut off to a 2mm cut off would be likely to result in significantly shorter durations of elimination programs and much cheaper elimination campaigns.
Our model simulates the removal of individual animals rather than whole herds. Past experience in Europe has shown that whole herd removal is critical for effective elimination in low prevalence situations [46]. In addition, a herd based model of the Moroccan cattle population could potentially lead to a lower intervention cost, as it is more realistic.
A recalculation of the intervention cost taking into account stratification by breed, sex and age should be undertaken, as it could lead to a different cost estimation of BTB control strategy.
Shortage of human resources should be considered for intervention planning, a maximum of 40% cull rate might be feasible; however, it would be costly in view of current Moroccan economic situation. One may think that test and slaughter implementation would lead to a reduction in cattle population and its by-product. But on the other hand, the increased import of cattle from other countries, with enhanced control measures, could maintain the current population density. In the meantime, as dairy products are provided mostly from highly controlled farms where BTB has a very low prevalence, we could argue that milk production would not be significantly affected.
The WHO includes BTB amongst the seven neglected zoonoses which are perceived to be severe threats to public health [1]. Further molecular epidemiology investigations in Morocco are needed in order to clarify local and national human BTB/TB ratios. To reach this goal, closer collaboration, at the national and international level, between the human and animal health sectors through a One Health approach is highly recommended. Operations in these two sectors remain largely independent in Morocco. Communication must be enhanced to establish a One Health approach, which has proven efficacy in health service delivery and potential for economic savings in zoonosis control [74, 75].
|
10.1371/journal.pntd.0007110 | Evaluation of direct costs associated with alveolar and cystic echinococcosis in Austria | Cystic echinococcosis (CE) is a globally occurring zoonosis, whereas alveolar echinococcosis (AE) is endemic only in certain parts of the Northern Hemisphere. The socioeconomic impact of human echinococcosis has been shown to be considerable in highly endemic regions. However, detailed data on direct healthcare-related costs associated with CE and AE are scarce for high income countries. The aim of this study was to evaluate direct costs of human disease caused by CE and AE in Austria.
Clinical data from a registry maintained at a national reference center for echinococcosis at the Medical University of Vienna were obtained for the years 2012–2014. These data were used in conjunction with epidemiological data from Austria’s national disease reporting system and diagnostic reference laboratory for echinococcosis to assess nationwide costs attributable to CE and AE.
In Austria, total modelled direct costs were 486,598€ (95%CI 341,825€ – 631,372€) per year for CE, and 683,824€ (95%CI 469,161€ - 898,486€) for AE. Median costs per patient with AE from diagnosis until the end of a 10-year follow-up period were 30,832€ (25th– 75th percentile: 23,197€ - 31,220€) and 62,777€ (25th– 75th percentile: 60,806€ - 67,867€) for inoperable and operable patients, respectively. Median costs per patients with CE from diagnosis until end of follow-up after 10 years were 16,253€ (25th– 75th percentile: 8,555€ - 24,832€) and 1,786€ (25th– 75th percentile: 736€ - 2,146€) for patients with active and inactive cyst stages, respectively. The first year after inclusion was the most cost-intense year in the observed period, with hospitalizations and albendazole therapy the main contributors to direct costs.
This study provides detailed information on direct healthcare-related costs associated with CE and AE in Austria, which may reflect trends for other high-income countries. Surgery and albendazole therapy, due to surprisingly high drug prices, were identified as important cost-drivers. These data will be important for cost-effectiveness analyses of possible prevention programs.
| Cystic and alveolar echinococcosis, caused by E. granulosus and E. multilocularis, both occur in humans in Austria. Lesions may develop at any site, with the liver being most frequently affected. Morbidity–especially for alveolar echinococcosis–can be significant and treatment may include major surgery or long-term suppressive medical therapy. The present study was performed to investigate direct healthcare-related costs of cystic and alveolar echinococcosis in Austria based on the data from a clinical registry maintained at the reference center for echinococcosis, the reference laboratory for echinococcosis and data from the national disease reporting system. Annual incidences of AE and CE were estimated at λ = 12 and λ = 40 newly diagnosed cases per year, respectively. Estimated costs due to cystic echinococcosis were 486,598€ (95%CI 341,825€ – 631,372€) per year, and 683,824€ (95%CI 469,161€ - 898,486€) for AE. Major cost drivers were surgical interventions with hospitalizations and drug costs. These data will provide a basis for cost-effectiveness analyses of prevention programs, and highlight possible targets for cost reduction strategies.
| Cystic echinococcosis (CE) is a zoonosis caused by Echinococcous granulosus and is endemic on all continents. In contrast, Echinocococcus multilocularis, the pathogen responsible for alveolar echincoccosis (AE), occurs only in certain areas of the Northern Hemisphere [1]. Both infections occur in humans in Austria. The western provinces of Austria are traditional hotspots of E. multilocularis transmission, with an increase in cases seen recently [2]. The geographic distribution of AE has expanded over the past several decades and now includes the entire country. Autochthonous E. granulosus transmission is rare in Austria, with only occasional locally-acquired CE cases reported in the past 20 years [3]. The majority of CE cases treated in Austrian hospitals are migrants from South-Eastern Europe and the Middle East, residing foremost in Austria’s Eastern provinces.
AE is associated with significant morbidity and mortality if left untreated. Treatment varies with disease stage, but therapeutic options are typically limited to hepatic surgery and albendazole therapy. If surgery is not feasible, palliative management with long-term albendazole treatment is indicated. Ideally, AE cases should receive long-term follow-up with advanced radiological imaging modalities, including positron-emission-tomography/computed tomography (PET/CT) [4]. CE is treated using a cyst stage-specific scheme consisting of surgery, percutaneous interventions, pharmacological treatment, or a watch-and-wait approach [5][6].
Although the socioeconomic impact of echinococcosis has previously been assessed for endemic countries [7], detailed studies on direct costs linked to echinococcosis, and in particular AE, are scarce and even more so for high-income regions.
Consequently, there is a dearth of information about the economic impact of AE and CE in high income regions like the Central European country of Austria, which may guide decision makers about the cost-effectiveness of control measures and of clinical management. The aim of this study was, therefore, to quantify costs associated with the treatment of echinococcosis in Austria from a societal perspective.
Clinical and epidemiological data were collected from a clinical registry of CE and AE patients managed at a reference center for the clinical management of echinococcosis at the General Hospital (AKH) of the Medical University of Vienna in Austria from 2012 to 2014. Patients seeking care for the first time in the given time period and subsequently managed at the reference center were eligible to be included. However, not all echinococcosis patients are registered, referred or managed at this center with other large centers treating patients on an individual basis. Thus, additional epidemiological data on the incidence of echinococcosis in Austria were collected from the Austrian Ministry of Health’s report on zoonoses [8] and the national reference laboratory for the diagnosis of echinococcosis, which is part of the Department of Parasitology at the Medical University of Vienna, and the only reference laboratory for echinococcosis in Austria. This laboratory processes samples from all suspected cases of echinococcosis in Austria. Ethics approval was obtained from the ethics committee of the Medical University of Vienna (#2031/2012).
Frequency data on hospitalizations and interventions were extracted from the clinical registry and patient records. Costs associated with these hospitalizations were obtained from the hospital’s Office for Medical Economics [5]. Data were also collected from patient records on the duration of albendazole drug therapy, including deviations from standard dosing. When complete dosing information was not available, these data were imputed based on published guidelines [5][6]. For outpatient routine laboratory testing (complete blood count and standard chemical parameters), specific diagnostic tests for echinococcosis (ELISAs, Western Blots, PCRs and histological examinations), clinical follow-up visits, diagnostic imaging (CT-scans, magnetic resonance imaging (MRI) scans, PET/CT and PET/MRI), and albendazole treatment, cost estimates were obtained from the Department of Parasitology, the hospital’s outpatient self-payer guide (not freely available), and the Austrian Reimbursement Code (“Erstattungskodex”) for registered drugs [9].
Overall direct costs for incident cases presenting to the AKH in the years 2012, 2013, and 2014 were summed and the proportional contribution of respective items (hospitalizations, imaging, etc.) analyzed per year of follow-up. Patients were classified as suffering from AE or CE, and according to CE-specific disease stage at inclusion into the study. Cost data were assessed for normality and the Wilcoxon matched-pairs signed-ranks test used to compare costs for the year of diagnosis and subsequent two years of follow-up. In order to model nationwide costs, estimates of incidence were based on data from Austria’s diagnostic reference laboratory for echinococcosis in addition to national data provided by the Ministry of Health. Data from these institutions represent the best available estimates of AE and CE incidence in Austria.
A spreadsheet model was developed in Microsoft Excel 2010 for Windows to evaluate national direct costs associated with AE and CE. Parameters were sampled across their distributions using a Monte Carlo approach with 1,000 bootstrap replicates. Annual incidence was modelled using a Poisson distribution with λ = 12 for AE and λ = 40 for CE [2]. Mean patient age was estimated at 60 years for AE and 44 years for CE based on clinical registry data. A year 2013 Austrian life table was used to model yearly survival probability for patients aged 60 and 44 years, respectively [10]. The proportion of patients receiving an intervention (surgery or interventional radiology) was modeled using a binomial distribution based on clinical registry data, with probability = 0.65 for AE and probability = 0.66 for CE. The seemingly comparable proportion of patients requiring an intervention for both AE and CE is explained by the fact that minimally invasive procedures such as PAIR (puncture, aspiration, injection, reaspiration) are part of the estimate for CE. A follow-up time of 10 years was assumed according to the local standard operating procedure. Direct costs for the first 3 years of treatment were obtained based on registry data. For year 4 through 10, yearly costs were estimated to be 90% of the previous year until the end of follow-up, for both AE and CE. In a previous Swiss study on AE [11], a discount factor of 3% was applied. However, this factor is believed to overestimate costs for the population under care in Austria. Therefore, in order to account for monitoring and the long-term use of albendazole, the current study includes a slightly steeper decrease in costs over time. In order to evaluate the impact of this discounting factor, a sensitivity analysis was performed by comparing the results to a model without discounting (i.e., assuming year 3 costs would be applicable for years 4–10) and to a model with 3% discounting per year.
In total, 45 patients (7 patients with AE and 38 with CE) presented for the first time to the center between 2012 and 2014. Of the 38 patients with CE, 8 were first diagnosed in 2012, 17 were first diagnosed in 2013, and 13 were first diagnosed in 2014. Type of infection, anatomical location of cysts, and CE hepatic cyst stage are shown in Table 1. Of the evaluated patients with CE, 16/38 (42%) were male and the mean age was 44 years (median 45 years, range 8 to 81 years) at inclusion. Out of the 7 AE patients, 2 were first diagnosed in 2012, 0 were first diagnosed in 2013, and 5 were first diagnosed in 2014. One (14%) AE patient was male and mean patient age was 60 years (median 62 years, range 27 to 81 years) at inclusion. No deaths occurred in the observed period.
Based on time of follow-up, overall direct, healthcare-related cost associated with CE for the first year after diagnosis at our center was 336,419€ (n = 38 patients), 70,619€ for the second year (n = 25) and 5,335€ (n = 8) for the third year (see Table 2). The most cost-intensive components in the first year of treatment were hospitalizations (66%), followed by albendazole therapy (23%), whereas laboratory procedures (7%), diagnostic imaging (4%), and outpatient visits (1%) only contributed to a small proportion of direct healthcare-related costs. Proportional costs per year of follow-up are depicted in Fig 1. Per patient costs were highest in the year of diagnosis (8,853€) and significantly decreased in the second year of treatment (2,824€) (p = <0.001; n = 25). Cost per patient continued to decline into the third year of treatment (667€). However, there was not a statistically significant difference between year 2 and year 3 (p = 0.161; n = 8). For year 1, a stage-based analysis showed that per-patient costs were higher for active cysts (CE1 to CE3b and extrahepatic active cysts combined) compared to inactive cysts (CE4 and CE5 combined) (see Table 3) (p = 0.001).
Assuming 10% discounting in years 4–10, modelled median costs per patients with CE from diagnosis until end of follow-up after 10 years were 16,253€ (25th– 75th percentile: 8,555€ - 24,832€) and 1,786€ (25th– 75th percentile: 736€ - 2,146€) for patients with active and inactive cyst stages, respectively. Based on an estimated incidence of 40 cases per year, a mean age of 44 years, and a 66% probability of requiring an invasive intervention (surgery or interventional radiology), the modelled yearly direct costs for CE in Austria were 486,598€ (95%CI 341,825€ – 631,372€). For the results of the sensitivity analysis with (a) no discounting factor for years 4–10 and (b) a discounting factor of 3% per year, see Table 4.
Overall direct costs for the 7 AE patients included in this cohort amounted to 132,739€. The largest contributors to first-year costs were hospitalizations, including surgical interventions (79%), and drug therapy with albendazole (12%). Diagnostic imaging (4%), laboratory procedures (4%), and outpatient visits (<1%) contributed only marginally to the direct costs of AE in year one (see Table 5). Most costs occurred in the year of diagnosis and decreased in subsequent years (see Table 6). However, no formal statistical testing on cost differences between years was performed due to the low number of patients in follow-up in years two and three.
Assuming 10% discounting in years 4–10, median modelled direct costs per patients with AE from diagnosis until end of follow-up after 10 years were 30,832€ (25th– 75th percentile: 23,197€ - 31,220€) and 62,777€ (25th– 75th percentile: 60,806€ - 67,867€) for inoperable and operable patients, respectively. Based on an estimated incidence of 12 cases per year, a mean age of 60 years at diagnosis, and a 65% probability of receiving a surgical intervention, the modelled yearly direct costs for AE in Austria were 683,824€ (95%CI 469,161€ - 898,486€). The results of the sensitivity analysis are presented in Table 4. In a sensitivity analysis with (a) no discounting factor for years 4–10 and (b) a discounting factor of 3% per year, the modelled overall costs per year were 838,465€ (95% CI 597,826€ – 1,079,106€) and 781,077€ (95% CI 548,099€ – 1,014,055€), respectively.
This study presents detailed direct costs associated with the diagnosis and treatment of echinococcosis in Austria, which is considered a country with very high human development by the Human Development Index. In contrast to most other economic analyses of echinococcosis, this study was based on a real cohort of patients. Nationwide costs for CE were estimated to cumulate to 486,598€ (95%CI 341,825€ – 631,372€) per year. Surgical interventions with accompanying hospitalizations and therapy with albendazole accounted for the vast majority of costs. Similar studies from high income countries are scarce. Differences in healthcare systems and billing systems, distinct methods and parameters used for the cost models of each publication, economic disparities, inflation and diverging price-levels of each country make direct comparisons almost impossible. However, the high costs of surgical interventions were also noted in a previous study from Italy [12]. Based on an Austrian population of 8.7 million inhabitants, and estimated incidence of 0.46 cases /100,000 persons / year, costs for CE accumulated to an estimated 5,660€ per 100,000 inhabitants per year (95%CI 4,006€ - 7,315€). A study from Italy estimated direct CE-associated costs of 6,398€ per 100,000 inhabitants per year with an estimated incidence between 1.06 and 2.78 cases per 100,000 inhabitants per year [13]. In one study from Spain published in 2005, the overall direct cost of CE in humans was estimated at 603,671€ (95% CI 499,200€ – 662,638€) per year [14]. Assuming a population of 43 million inhabitants, as stated in the paper, this corresponds to 1,404€ per 100,000 inhabitants per year, based on 159 diagnosed cases in 2005 (i.e. 0.36 cases per 100,000 inhabitants).
CE is a rare condition in Austria, with few endemic cases. Disability Adjusted Life Years (DALYS) lost due to CE accounted for only 0.0078% of DALYs lost to infectious and non-infection conditions in Austria as estimated by the Global Burden of Disease Study 2016 data [15]. The costs attributed to CE appear negligible when compared to major drivers of healthcare costs. For example, it was estimated that approximately 1.2 billion € are spent on healthcare costs related to cancer per year in Austria [16]. Nevertheless, morbidity and direct costs per case may be substantial, especially when considering that CE is largely preventable by appropriate livestock management, deworming of dogs, not feeding raw offal to dogs, vaccination of sheep, and good food hygiene.
As expected, costs associated with CE were significantly higher for active stages of the disease compared to inactive ones, and costs were highest in the year of diagnosis. At the center, for evaluation of interventions, advanced and relatively expensive imaging modalities, including MRI and CT scans, were performed in the management of CE to assess potential surgical operability or feasibility of percutaneous interventions [5]. Ultrasound examinations were performed, usually during an outpatient visit and were not billed separately. However, from a perspective of cost-effectiveness, replacing CT- or MRI-scans by ultrasound would only marginally reduce the direct costs of echinococcosis due to the low proportion of costs contributed by imaging in an affluent healthcare system. On the other hand, in low to middle income countries, management of CE is most probably overwhelmingly based on ultrasound, and may constitute a relevant cost-factor.
Interestingly, albendazole consumption was high for both AE and CE in the first year of follow-up when peri-interventional albendazole-therapy or first line pharmacologic therapy (depending on the cyst stage) occur. In the second and third year, however, albendazole consumption was very low in CE patients, but still high in AE patients, some of which need long term pharmacologic therapy, which is unusual in CE. Thus, overall we think that our data on albendazole consumption seem plausible and reflect treatment recommendations.
Interestingly, albendazole, a drug listed by the World Health Organization as an “essential medicine”[17], of which millions of doses are used at very low cost or through donation programs each year in developing countries, was a major driver of healthcare costs in the management of both AE and CE. Costs for albendazole are prohibitively high in some countries of Western Europe and other developed regions [18][19], and could be a primary target to decrease costs associated with the management of echinococcosis. Costs for albendazole are volatile and differ massively even between neighboring countries. In a recent study from Italy, the problem of albendazole drug shortages was also highlighted [20]. This has, to our knowledge, not yet occurred in Austria, but may potentially endanger treatment success. Multiplication of costs for old, unpatented drugs such as albendazole or daraprim following the acquisition of the main supplier by another company has possibly contributed to the high costs [21][22]. However, this is probably a problem requiring a political solution and not a medical or scientific one. Consequently, the contributors of direct costs associated with CE may be different in resource-limited regions of the world, where drug costs are lower and resources are often not available for surgical interventions.
AE is endemic in several medium to high-income countries in the Northern Hemisphere, including Austria. Median costs per patients with AE from diagnosis until the end of a 10-year follow-up period were 30,832€ (25th– 75th percentile: 23,197€ - 31,220€) and 62,777€ (25th– 75th percentile: 60,806€ - 67,867€) for inoperable and operable patients, respectively. This is lower than direct costs of 108,762€ per patient estimated in a Swiss study [11]. One reason for this observation is the different discount factor used (10% in our model vs. 3% in the Swiss study), which partially explains the difference. Secondly, according to the Organization for Economic Co-operation and Development (OECD), the price level index in Switzerland is 1.46 times that of Austria [23], which may account for another part of the observed difference in direct costs between these two neighboring countries. As the treatment of AE consists of major surgery or long-term albendazole therapy, with no percutaneous interventions available, cost per case was higher than for CE. This is highly plausible as CE usually does not require major surgery or long-term suppressive therapy with albendazole, which were found to be the two main drivers of costs.
Although cost per case will differ between countries, the main cost drivers for Western and Central European counties will most likely be similar, with disparities mostly attributable to difference in albendazole drug prices. This cost estimate can also be used to determine cost-effectiveness of control programs, which may include use of praziquantel-baits for E. multilocularis in foxes or systematic screening of at-risk populations [24]. Bait-based programs often involve the distribution of baits via air, and are thus expensive. Although cost estimations for such programs are not available for Austria, it seems unlikely that such interventions would be cost-effective with respect to the results of this study. Cost-effectiveness, as modelled by Hegglin et al., could only be achieved in an area with high population density after decades of campaigning [25].
There are several limitations of the current study. First, the sample size was small particularly for differentiating CE disease-stage specific costs per year in the third year of follow-up, and for AE in the second and third year of follow-up. Thus, modelled costs may be affected by individual outliers. Secondly, patient data were abstracted for a maximum of three years. Therefore, healthcare-related costs linked to echinococcosis occurring later than 3 years after diagnosis were not recorded and had to be modelled. We chose to introduce a discounting factor (10% per year) to approach this problem. This may still be relatively conservative, but is substantially higher than the previously used 3% [11]. In fact, the first year after diagnosis is the main driver of costs with later years of follow-up only contributing a relatively small proportion of costs. No deaths were observed in the study population, but excessive deaths due to echinococcosis that occurred beyond the time period based on real data (i.e. in the modelled period) were not included in the model. Excessive or earlier deaths would result in lower overall costs than modelled. Likewise, costs from patients that needed more than one intervention were included if these occurred in 2012–2014. However, relapses that may occur during the modelled period were not accounted for leading to a possible underestimation of overall costs. Measurement of albendazole-sulphoxide in blood is recommend by some guidelines for AE or in the event of adverse drug effects [5], but this was not performed during the respective period as it was unavailable at our center. Frequent measurements would slightly increase overall costs. Finally, albendazole treatment duration had to be estimated for a few cases based on current treatment guidelines [5][6], which are in line with institutional treatment guidelines.
In summary, this study presents a detailed analysis of direct healthcare-related costs linked to AE and CE in a developed country. This analysis demonstrates that hospitalizations (including surgical interventions) and albendazole treatment are the main drivers of costs for both CE and AE and, therefore, constitute potential targets for cost reduction strategies. Future studies from healthcare systems in countries with other socioeconomic backgrounds could further improve our understanding of drivers of the financial burden of echinococcosis in order to develop future and improve current prevention, treatment, and long-term care strategies.
|
10.1371/journal.pgen.1003794 | Ultra-Sensitive Sequencing Reveals an Age-Related Increase in Somatic Mitochondrial Mutations That Are Inconsistent with Oxidative Damage | Mitochondrial DNA (mtDNA) is believed to be highly vulnerable to age-associated damage and mutagenesis by reactive oxygen species (ROS). However, somatic mtDNA mutations have historically been difficult to study because of technical limitations in accurately quantifying rare mtDNA mutations. We have applied the highly sensitive Duplex Sequencing methodology, which can detect a single mutation among >107 wild type molecules, to sequence mtDNA purified from human brain tissue from both young and old individuals with unprecedented accuracy. We find that the frequency of point mutations increases ∼5-fold over the course of 80 years of life. Overall, the mutation spectra of both groups are comprised predominantly of transition mutations, consistent with misincorporation by DNA polymerase γ or deamination of cytidine and adenosine as the primary mutagenic events in mtDNA. Surprisingly, G→T mutations, considered the hallmark of oxidative damage to DNA, do not significantly increase with age. We observe a non-uniform, age-independent distribution of mutations in mtDNA, with the D-loop exhibiting a significantly higher mutation frequency than the rest of the genome. The coding regions, but not the D-loop, exhibit a pronounced asymmetric accumulation of mutations between the two strands, with G→A and T→C mutations occurring more often on the light strand than the heavy strand. The patterns and biases we observe in our data closely mirror the mutational spectrum which has been reported in studies of human populations and closely related species. Overall our results argue against oxidative damage being a major driver of aging and suggest that replication errors by DNA polymerase γ and/or spontaneous base hydrolysis are responsible for the bulk of accumulating point mutations in mtDNA.
| Owing to their evolutionary history, mitochondria harbor independently replicating genomes. Failure to faithfully transmit the genetic information of mtDNA during replication can lead to the production of dysfunctional electron transport proteins and a subsequent decline in energy production. Cellularly-derived reactive oxygen species (ROS) and environmental agents preferentially damage mtDNA compared to nuclear DNA. However, little is known about the consequences of mtDNA damage for mutagenesis. This lack of knowledge stems, in part, from an absence of methods capable of accurately detecting these mutations throughout the mitochondrial genome. Using a new, highly sensitive DNA sequencing strategy, we find that the frequency of point mutations is 10–100-fold lower than what has been previously reported using less precise means. Moreover, the frequency increases 5-fold over an 80 year lifespan. We also find that it is predominantly transition mutations, rather than mutations commonly associated with oxidative damage to mtDNA, that increase with age. This finding is inconsistent with free radical theories of aging. Finally, the mutagenic patterns and biases we observe in our data are similar to what is seen in population studies of mitochondrial polymorphisms and suggest a common mechanism by which somatic and germline mtDNA mutations arise.
| Mitochondria are the primary source of energy for cells. Owing to their evolutionary history, these organelles harbor a small, independently replicated genome (mtDNA). Human mtDNA encodes two rRNA genes, 13 protein coding genes that are essential components of the electron transport chain (ETC), and a full complement of 22 tRNAs used in translation of the ETC peptides. The escape of electrons from the ETC can lead to the formation of reactive oxygen species (ROS), which are capable of damaging a variety of cellular components, including DNA. Due to its proximity to the ETC, absence of protective histones, and a lack of nucleotide excision or mismatch repair, mtDNA is thought to be especially vulnerable to ROS-mediated damage and the generation of mutations. Failure to faithfully transmit the encoded information during mtDNA replication leads to the production of dysfunctional ETC proteins, leading to the release of more free electrons and ROS in what has been termed ‘the vicious cycle’ [1], [2]. Thus, it is not surprising that mutations in mtDNA have been associated with a decline in energy production, a loss of organismal fitness, an increased propensity for a number of pathological conditions, and aging (reviewed in [3], [4]).
Numerous lines of evidence have suggested mtDNA mutations are involved in the aging process. In particular, ETC activity declines with age [5], [6], and this decrease is coincident with accumulation of mitochondria with large deletions in their mtDNA [7], [8], [9], [10]. Large, kilobase-sized deletions in mtDNA become more prevalent with age in a variety of tissues, including brain [11], heart [12], and skeletal muscle [7]. Furthermore, these large deletions have been shown to increase in frequency in a number of neurodegenerative conditions, including Parkinson's disease [13], [14] and Alzheimer's disease [15]. In addition, DNA damage, predominantly in the form of 8-hydroxy-2′-deoxyguanosine (8-oxo-dG) [16], increases with age in both nuclear and mitochondrial DNA [17], [18], [19], [20]. While the role of mtDNA deletions in aging is well established, the role of point mutations remains controversial [21], [22].
Several previous studies have examined the accumulation of point mutations in human aging and disease [23], [24], [25], [26]. Until very recently, hypotheses that required the observation of rare mutations in mtDNA have been extremely difficult to experimentally validate due to: 1) the lack of genetic tools for introducing reporters or selectable markers into mtDNA; 2) the high background error rate of most DNA sequencing methods [27], [28]; and 3) the sampling limitations of the few available high-sensitivity mutation assays that screen only a tiny subset of the genome [29]. The mitochondrial genome is 16,569 bp, and individual human cells frequently contain hundreds to thousands of molecules of mtDNA; thus, a single human cell typically contains millions of nucleotides of mtDNA sequence. The rate of accumulation of mtDNA mutations has previously been estimated as 6×10−8 mutations per base pair per year [30]. Therefore, reliable study of spontaneous mtDNA mutations requires methodologies that can accurately detect a single mutation among >106 wild-type base-pairs. However, most prior studies of mtDNA mutations and aging have relied upon methods with background error frequencies of 10−3 to 10−4; hence the many reported differences likely reflect changes in mutation clonality or technical artifacts (e.g. due to increases in DNA damage with age) rather than true spontaneous mutations.
Massively parallel sequencing technologies allow mtDNA to be subjected to ‘deep sequencing’ in order to detect rare/sub-clonal mutations on a genome-wide level. However, these new sequencing methods are highly error prone, with artifactual error rates of approximately one spurious mutation per 100 to 1,000 nucleotides sequenced. These high error rates have precluded the study of spontaneous mutations in mtDNA [31]. To circumvent this limitation, we recently developed a new, highly accurate sequencing methodology termed Duplex Sequencing (DS), which has the unique ability to detect a single mutation among >107 sequenced bases [32].
In the study herein, we determined the effect of aging on mtDNA mutation burden by using DS to compare human mtDNA purified from brain tissue of five young individuals (ages <1) and five aged individuals (ages 75–99 years) obtained via rapid autopsy (Table S1). As brain is among the most metabolically active tissues in the human body, we reasoned it to be particularly prone to damage from ROS, and thus, an optimal tissue for comparison between age groups. We assessed the relative frequency, spectrum, and distribution of mtDNA mutations in the two groups. We find that point mutations increase with age, but do so in a non-uniform manner. Furthermore, we find that mutations show a bias in their occurrence with respect to both genome location and strand orientation. The types of mutations we detect are inconsistent with oxidative damage being a major driver of mtDNA mutagenesis.
Duplex Sequencing relies on the concept of molecular tagging and the fact that the two strands of DNA contain complementary information. Fragmented duplex DNA is tagged with adapters bearing a random, yet complementary, double-stranded nucleotide sequence (Fig. 1A). Following ligation, the individually labeled strands are PCR amplified, thus creating sequence “families” that share a common tag sequence, derived from each of the two single parental strands (Fig. 1B). After sequencing, members of each duplicate family are grouped by tag, and a consensus sequence is calculated for each family, creating a single strand consensus sequence (SSCS) (Fig. 1C). This step eliminates random sequencing or PCR errors that occur during library amplification; however, the single-stranded consensus process does not filter out artifactual mutations that are the consequence of first round PCR errors, such as those commonly caused by DNA damage. To remove this latter type of error, the complementary SSCS families derived from the two single-stranded halves of the original DNA duplex are compared to each other (Fig. 1C). The base identity at each position in a read is kept in the final consensus only if the two strands match perfectly at that position. Upon remapping of these duplex consensus sequence (DCS) reads back to the reference genome, any deviations from the reference sequence are considered true mutations. The frequency of mutations in a sampled population of mtDNA is calculated as the number of DCS mutant molecules divided by the number of DCS wild-type molecules observed at any given genomic position.
Point mutations in mtDNA could be the result of maternal inheritance or a de novo mutation event. Maternally inherited mutations or mutations arising during early embryonic development are more likely to be clonal (i.e. the same mutation being present at the same location in most or all mtDNA molecules). Therefore, in order to quantify the frequency of de novo events, we used a clonality cutoff that excluded any positions with variants occurring at a frequency of >1%, and scored each type of mutation only once at each position of the genome. Based on these criteria, the mtDNA from aged individuals show a highly significant ∼5-fold increase in mutational frequency, relative to those obtained from young individuals (Young: 3.7±0.9×10−6 vs. Aged: 1.9±0.2×10−5, p<10−4, two-sample t-test) (Fig. 2A). These mutation frequencies are between one and two orders of magnitude lower than the previously reported values for both young and old individuals using PCR-based methods or conventional next-generation sequencing [24], [33], [34]. This discordance likely stems from artifactual scoring of mutations by these latter methods due to misinsertion of incorrect bases at sites of damage in template DNA during the PCR steps. Duplex Sequencing, in contrast, is unaffected by DNA damage [32].
Inspection of the mutation spectra for both the young and old samples reveals that all samples are significantly biased towards transitions (Fig. 2B). Specifically, the most common mutation type, G→A/C→T, is consistent with either misincorporation by DNA polymerase γ or deamination of cytosine to form uracil, as being the largest mutagenic drivers in mtDNA [35], [36]. The second most common mutation type, T→C/A→G, is consistent with either deamination of adenosine to inosine or a T-dGTP mispairing, the primary base misinsertion mistake made by DNA polymerase γ [37], [38], [39], [40]. Plotting the frequency of each type of mutation as a proportion of total mutations (Fig. 2C) reveals that the relative abundance of each mutation type is the same in young and old, suggesting that the mutagenic pressures that result in the observed spectra are constant throughout the human lifespan.
Surprisingly, comparison of the mutation spectra of the young and old samples reveals a notable absence of the mutational signature of oxidative damage. A number of studies have shown that oxidative damage to DNA accumulates in both the nuclear and mitochondrial genomes as a function of age, as well as several age-associated pathologies [17], [18], [19], . The most frequent alteration produced by oxidative damage is 8-oxo-dG, which, when copied during replication or repair, results in dA substitutions, yielding G→T/C→A transversions [42]. A number of theories of aging invoke ROS-mediated damage to mtDNA as being a major driver of the aging phenotype (reviewed in [43] and [44]). A key prediction for these theories is that the frequency of G→T/C→A mutations would be expected to increase with time. We failed to find either a preponderance of G→T/C→A substitutions or a proportionally greater increase with age in this type of mutation relative to other types, despite a span of >80 years between our sequenced sample groups (Fig. 2C).
Our data indicate that point mutations increase with age and that these mutations are inconsistent with oxidative damage being a primary driver of mutagenesis; we next assessed whether these mutations lead to alterations in the protein coding sequence. We find that in the aged samples, 78.3% of mutations are non-synonymous. The incidence of non-synonymous mutations is close to the expected value of 75.7% for mtDNA that would occur if non-synonymous and synonymous mutations occur randomly. In contrast, only 62.9% of mutations are non-synonymous in the young samples. The reduced mutation load observed in the young samples is consistent with that negative intergenerational selection against such mutations and that this selection is relieved during development and could play a role in the aging phenotype.
However, the existence of a high load of non-synonymous mutations does not necessarily mean that the coding changes lead to functional protein alteration. To examine this possibility, we compared the predicted “pathogenicity” of all non-synonymous mutations in both the young and aged samples using MutPred [45], a software package that calculates the likelihood of a mutation being deleterious based on a number of criteria, including protein structure, the presence of functional protein motifs, evolutionary conservation, and amino acid composition bias. A score between zero and one is assigned to each mutation, with a higher score denoting a higher likelihood of being deleterious. Based on this analysis (Fig. S1), the predicted pathogenicity of mutations, indeed, increases with age (p<0.02, Wilcoxon Rank Sum analysis), suggesting that mutations acquired during aging may have functional consequences for the electron transport chain. A similar increase in predicted deleterious mutations was also observed using the SWIFT software package (data not shown). The increase in predicted pathogenicity is consistent with mutations causing coding changes occurring randomly and argues against a mechanism by which point mutations are selected against by the cell. Similar finding in clonally expanded mutations were recently reported in colon tissue show a similar increase in predicted pathogenic mutation in mtDNA [46].
The mitochondrial genome can be divided into three different regions: 1) protein coding genes, 2) RNA coding genes (consisting of both rRNA and tRNA), and 3) non-coding/regulatory regions including the origin of replication known as the D-loop. Phylogenetic analysis of both human and other mammalian lineages has shown that population level single nucleotide variants (SNVs) tend to cluster in a number of ‘hotspots’ in the mitochondrial genome, most notably in Hypervariable Regions I and II of the D-loop [47], [48], [49]. We sought to determine if the distribution of non-clonal mutations within the mtDNA of individuals exhibited a uniform distribution or if certain regions of the genome similarly show variations in mutation frequency. Comparison of the mutation frequencies of the RNA coding genes to the protein coding genes yielded no significant differences in either the young or old samples (p = 0.15, two-tailed t-test).
In contrast, we observed a significant increase in mutation frequency of the D-loop (bp 16024-576) relative to the coding regions (bp 577–16023) in both young (D-loop: 1.5±0.6×10−5 vs. coding region: 2.9±0.7×10−6, p<0.01, two-tailed t-test) and aged (D-loop: 5.7±1.5×10−5 vs. coding region: 1.65±0.2×10−5, p<0.01, two-tailed t-test) samples, suggesting that the D-loop is a mutagenic hotspot. However comparing the relative increase in the mutation frequency of the D-loop between the young and old sample groups (3.8±1.6-fold increase) to the relative increase seen between the two sample groups in the non-D-loop regions (5.6±2.0 fold increase) shows no difference. This finding is inconsistent with the idea that the D-loop accumulates significantly more mutations during aging than the rest of the mitochondrial genome. Spectrum analysis shows a similar predominance of transition mutations in both the D-loop and coding regions of the genome (Fig. 3A), with no significant difference in the relative abundance of the different mutation types (Fig. 3B). Taken together, our data suggest that the mutagenic processes of mtDNA are largely uniform across the genome.
The human mitochondrial genome has a significant bias in the cytosine/guanine composition between the two strands. Specifically, the light strand (L-strand), which is the coding strand for only nine genes, contains about three-fold more cytosine than guanine, whereas the heavy strand (H-strand) codes for the remaining 28 genes and has the opposite composition bias. Human population studies, as well as the comparative analysis of evolutionarily related species, have shown a bias towards the occurrence of G→A and T→C SNPs of the L-strand [50], [51], [52], [53]. These population-level compositional biases are hypothesized to be due to an asymmetric accumulation of mutations between the two strands of mtDNA in the germline; however, to date, the biases have not been observed at the sub-clonal/random level within individuals. To examine this, we compared the frequency of reciprocal mutations occurring on the L-strand (i.e. G→A on the L-strand vs. C→T mutations on the L-strand). By definition, mutations cause complementary sequence changes on both strands of a DNA molecule. Therefore, if a bias does not exist in the orientation of specific mutations towards a particular strand, then the frequency of reciprocal mutations on the same strand would be expected to be equal. Alternatively, the presence of a strand orientation bias would manifest itself in the form of a particular type of mutation occurring more frequently than its reciprocal mutation.
We find that the majority of the human mitochondrial genome shows a significant strand orientation bias in the occurrence of transitions, whereas transversions show no apparent asymmetry (Fig. 4A). Specifically, in young samples, G→A/C→T mutations are more likely to occur when the dG base is present on the L-strand and the dC base is in the H-strand, respectively. This pattern is even more pronounced in aged individuals, consistent with this bias being due to ongoing mutagenic process and not the result of maternal inheritance. In addition to the G→A/C→T bias, the aged samples also exhibit a strand orientation bias in the occurrence of T→C/A→G, where dT is more likely to be mutated to a dC when it is located on the L-strand than on the H-strand. Interestingly, this bias, which appears uniformly throughout most of the mtDNA, is uniquely absent in the D-loop region (Fig. 4B). Thus, both the spectrum and strand orientation asymmetry of somatic mtDNA mutation accumulation recapitulates what has been previously recognized in population studies.
The accumulation of somatic mutations in mtDNA has frequently been hypothesized to drive the aging process and its associated pathologies, including neurodegeneration, cancer, and atrophy (reviewed in [54]). The underlying mechanisms by which these mutations occur and accumulate have been the subject of intense study, but remain incompletely defined. One of the major limitations has been the lack of methodologies with sufficient sensitivity to detect rare mutations among a much larger population of wild-type molecules. We recently developed a robust next-generation sequencing methodology, termed Duplex Sequencing, which is able to detect a single point mutation among >107 sequenced bases [32] and has now enabled us to precisely characterize the genome-wide frequency, spectrum, and distribution of somatic mtDNA mutations in aging human brain with unprecedented accuracy.
Our data show a significant increase in the load of point mutations as a function of human age, with absolute frequencies 10–100 fold lower than what has been typically reported in the literature using less sensitive assays. Recent work using the Random Mutation Capture assay has reported an age associated increase in mtDNA point mutation frequencies in mice and Drosophila that are on par with the values that we have determined here; however, these studies were limited to only a very small region of the genome [22] (Leo Pallanck-submitted). Of particular interest, despite a ∼1000-fold difference in lifespan, the increase in mutation load with age appears to be highly consistent among multiple species. This surprising finding suggests that the underlying mechanisms behind the age-dependent accumulation of point mutations in mtDNA are conserved between humans, flies, and mice and merit more detailed comparison.
Oxidative damage to DNA, most notably in the form of 8-oxo-dG, has long been believed to be a primary driver of mutagenesis in both nuclear and mitochondrial DNA [42], [55], [56]. However, our results do not support this hypothesis. In our data, the relative proportion of G→T/C→A mutations is quite low in the young samples examined and, importantly, does not show a disproportionate increase with age relative to other types of mutations. Other recent reports, which used less sensitive methods to detect intermediate frequency sub-clonal mutations, have similarly failed to detect this classic signature of oxidative damage to DNA. For instance, one conventional deep sequencing analysis of aged mice reported no significant burden of G→T/C→A transversions [57]. Even more surprising is the observation that a transgenic mouse strain deficient for both MutY and OGG1, which are the two primary enzymes responsible for repairing 8-oxo-dG, do not exhibit an increase in mtDNA mutations [58].
Comparison of the spectrum of our reported data (Fig. 2B) to that of the clonal SNV's in our data (i.e. mutations present at >90%), as well as those reported in the Mitomap database [59], reveals an identical bias towards transitions with a minimal number of G→T/C→A transversions (Fig. S2). Indeed, a similar propensity towards transitions has been noted in numerous animal phylogenies [60]. This consistency in mutational pattern suggests that the mutagenic processes that cause the accumulation of mutations in somatic tissue are also responsible for clonal population variants arising in the maternal germline.
In addition to 8-oxo-dG, ROS can also cause a number of other mutagenic lesions, including thymine glycol and deamination of cytidine and adenosine, all of which can induce transition mutations [37], [61], [62]. Our data clearly show an excess of transitions relative to transversions, which could be consistent with oxidatively induced deamination events becoming fixed as mutations. It is well established that ROS production and oxidative damage increase with age [17], [18], [19], [20]. Yet, if oxidative damage were the main driver for deamination events, this model would predict that the relative proportion of transitions should be disproportionately higher in aged individuals, which is not the case with our data. While we cannot conclusively rule out a role for ROS in inducing transition mutations, the excess of transitions could additionally be explained by spontaneous deamination of either cytidine or adenosine, especially in single-stranded replication intermediates, or base misincorporation events by DNA polymerase γ during genome replication, which has a known propensity for transition mutations [36], [39], [40]. Overall, the absence of key mutagenic signatures of oxidative damage argues against ROS being the major driver of mutagenesis in mtDNA in normal aging brain.
Several explanations may account for why we observed few 8-oxo-dG associated mutations in mtDNA despite an extensive literature showing that 8-oxo-dG levels increase with age. First, rapid DNA repair may remove 8-oxo-dG prior to genome replication and this repair capacity may increase with age [63], thereby keeping 8-oxo-dG mutagenesis to a minimum. Secondly, mitochondrial quality control pathways may simply eliminate mitochondria with damaged mtDNA. Consistent with this idea is the observation that oxidatively damaged mtDNA rapidly disappears from cells treated with H2O2 [64]. In addition, the cellular levels of parkin, a major component of the quality control pathway involved in mitochondrial turnover, increase under conditions of high oxidative stress [65]. Finally, DNA polymerase γ itself may actively discriminate against incorporating 8-oxo-dG [66]. Regardless of the mechanism, our data suggest that cells have evolved one or more strategies to effectively deal with the challenge of replicating mtDNA in the highly damaging environment of the mitochondria.
We find a striking excess of mutations in the D-loop in young individuals; however, D-loop mutations accumulate during aging at the same rate as other parts of the mitochondrial genome, consistent with the D-loop not being inherently error prone. Instead, our data suggest that young individuals are born with a higher aggregate mutation burden in the D-loop relative to the rest of the genome, thus preserving the D-loop's disproportionate mutation load as mutations accumulate during life. The higher proportion of low-level/random mutations in the D-loop also offers a likely explanation as to why population level SNVs tend to cluster in a number of ‘hotspots’ in the D-loop.
The disproportionate mutation load of the D-loop at birth provides a likely explanation as to why the D-loop has long been considered prone to mutagenesis. Due to the D-loop's higher aggregate burden, the previously used assays were only sensitive enough to detect mutations located in the D-loop. Thus, even though point mutations increase uniformly throughout mtDNA with age, the mutation load in the non-D-loop portions of the genome would likely be below the sensitivity of the previously available assays, thus giving the appearance that the D-loop's mutations increased with age.
Human population studies have previously identified a bias in the occurrence of G→A and T→C SNPs of the L-strand, as has comparison of human mtDNA sequences with those of evolutionary related species [50], [51], [52]. This imbalance has been hypothesized to be the result of preferential deamination of cytidine and adenosine on the H-strand; however, because this strand orientation bias has only been previously observed at the clonal population scale, the influence of natural selection could not be discounted [50], [52]. Our data demonstrate that this bias originates from mutagenesis of mtDNA through a process that is continuous throughout life; however, the source of this bias is unclear. One possibility is that this bias arises during mtDNA replication. Replication of mtDNA is thought to occur in an asymmetric manner [67], [68]. In this model, replication of the H-strand (G-rich strand of mtDNA) begins at the H-strand origin, using the L-strand (C-rich strand of mtDNA) as its template. During this time, the parental H-strand remains in a single-stranded state until the synthesis of the L-strand is initiated at the L-strand origin. Several studies previously documenting a strand orientation bias of clonal SNVs at the population level hypothesized that the bias is due to an increased rate of spontaneous hydrolytic deamination of cytidine and adenosine on the H-strand spending a greater amount of time in an unprotected single-stranded state during replication [50], [51], [52], [69]. Indeed, there is in vitro evidence that replication past a deaminated dC (i.e. dU) by DNA polymerase γ is responsible for the observed asymmetry. Zheng et al. used DNA polymerase γ to amplify small regions of homoduplex mtDNA via PCR and observed the emergence of an excess of G→A/C→T mutations in these reactions [40]. Their experimental conditions involved extended heating of the DNA, which is known to increase the rate of deamination of single-stranded DNA. While our results do not, on their own, provide direct proof that an asymmetry in deamination of single-stranded replication intermediates is responsible for the observed inter-strand mutational skew, our data are consistent the findings and conclusions in Zheng et al., thus providing strong support for this model [40]. Furthermore, the observation that a similar strand bias exists in the absence of intergenerational selection argues that it is due to an ongoing mutagenic process within the cell and that the bias observed at the population level is likely due to the same process.
In summary, Duplex Sequencing is a powerful technology for high-sensitivity study of mtDNA mutagenesis that has enabled us to uncover a number of mutagenic patterns and biases in human brain tissue that have not been previously observed at the somatic level. Taken together, these mutation patterns argue against oxidative DNA damage being a major driver of aging and suggest that replication errors by DNA polymerase γ and/or spontaneous base hydrolysis are responsible for the bulk of point mutations that accumulate in mtDNA. In light of this finding, the central role for oxidative damage to mtDNA in theories of human aging and disease merits re-evaluation.
Mitochondria were isolated from the pre-frontal cortex of either young (<1 yr) or old (>75 yr) brain tissue with no known brain pathologies (Table S1). Approximately 500–800 mg of brain tissue was sub-divided into smaller 100–150 mg pieces, with the mitochondria from each piece purified independently from the other pieces of the same tissue sample. Each tissue fraction was dounced in 5 mL of homogenization buffer (0.32 M sucrose, 1 mM EDTA, 10 mM Tris-HCl, pH 7.8) with no more than five strokes, as excessive douncing increases the contamination of nuclear DNA. The homogenate was centrifuged at 1000 g for 20 min at 4°C. The supernatant was transferred to a new centrifuge tube and centrifuged at 1000 g for 10 min at 4°C. Samples that exhibited a pellet at the bottom of the tube after the second centrifugation step were centrifuged a third time using the same condition. The supernatant was removed and centrifuged at 13,000 g for 30 min at 4°C to give a crude mitochondrial pellet. Each crude pellet was resuspended in Mito-DNase buffer (0.3 M sucrose, 10 mM MgCl2, 0.15% BSA (w/v), 20 mM Tris-HCl, pH 7.5) and DNase added to a concentration of 0.01 mg/mL and incubated at 37°C for 1.5 hrs. This step helps remove contaminating nuclear DNA while not affecting mtDNA protected within intact mitochondria. Mitochondria were then re-pelleted at 13,000 g for 30 min and the supernatant discarded. The mitochondrial pellets were washed two times by resuspending the pellets in 1 mL of Mito-DNase buffer followed by centrifugation at 13,000 g for 30 min. Finally, the enriched mitochondrial pellets were resuspended in 500 µL Lysis Buffer (150 mM NaCl, 20 mM EDTA, 1% SDS (w/v), 10 mM Tris-HCl, pH 7.8, 0.2 mg/mL Proteinase K, 0.01 mg/mL RNase) and incubated at 56°C for 1 hr. The mtDNA was extracted using a standard phenol∶chloroform approach. The relative purity of each mtDNA prep from nuclear DNA was determined by qPCR using the following primers sets: nDNA: fwd 5′- TTGCCAGACCATGGGATTGTCTCA, rev 5′-TTCCTACCGAACGAGGACTCCAAA; mtDNA: fwd 5′-ACAGTTTATGTAGCTTACCTC, rev 5′-TTGCTGCGTGCTTGATGCTTG. DNA fractions from the same tissue samples exhibiting a ΔC(t) ≥17.5 cycles (corresponding to ∼50% mtDNA by mass) were pooled and used for library preparation and sequencing. Typical yields were between 100–500 ng of highly pure mtDNA.
We carried out synthesis and preparation of the Duplex Tag-labeled adapters and sequencing library preparation as previously described with the following minor mtDNA specific modifications [32]: 1) 100–500 ng of mtDNA was sheared using the Covaris AFA system with a duty cycle of 10%, intensity of 5, cycles/burst 100, time 20 seconds×5, temperature of 4°C; 2) prior to adapter ligation, the DNA was quantified and a 40∶1 molar excess of adapters was used for the ligation step; 3) after adapter ligation and clean up, the library was re-quantified and ∼37 fmol of product was amplified by PCR for a total of 20 cycles. The resulting libraries were subjected to sequencing on an Illumina HiSeq 2000/2500 platform using 101 bp, paired-end reads.
This study involved the use of human brain tissue obtained from autopsy. The tissues were collected at University of Washington under the direction of the Alzheimer's Research Center and Seattle Children's hospital under their approved IRB protocols. Our research did not fall under human subjects requirements due to the samples being anonymous.
|
10.1371/journal.ppat.1006158 | Proline Metabolism is Essential for Trypanosoma brucei brucei Survival in the Tsetse Vector | Adaptation to different nutritional environments is essential for life cycle completion by all Trypanosoma brucei sub-species. In the tsetse fly vector, L-proline is among the most abundant amino acids and is mainly used by the fly for lactation and to fuel flight muscle. The procyclic (insect) stage of T. b. brucei uses L-proline as its main carbon source, relying on an efficient catabolic pathway to convert it to glutamate, and then to succinate, acetate and alanine as the main secreted end products. Here we investigated the essentiality of an undisrupted proline catabolic pathway in T. b. brucei by studying mitochondrial Δ1-pyrroline-5-carboxylate dehydrogenase (TbP5CDH), which catalyzes the irreversible conversion of gamma-glutamate semialdehyde (γGS) into L-glutamate and NADH. In addition, we provided evidence for the absence of a functional proline biosynthetic pathway. TbP5CDH expression is developmentally regulated in the insect stages of the parasite, but absent in bloodstream forms grown in vitro. RNAi down-regulation of TbP5CDH severely affected the growth of procyclic trypanosomes in vitro in the absence of glucose, and altered the metabolic flux when proline was the sole carbon source. Furthermore, TbP5CDH knocked-down cells exhibited alterations in the mitochondrial inner membrane potential (ΔΨm), respiratory control ratio and ATP production. Also, changes in the proline-glutamate oxidative capacity slightly affected the surface expression of the major surface glycoprotein EP-procyclin. In the tsetse, TbP5CDH knocked-down cells were impaired and thus unable to colonize the fly’s midgut, probably due to the lack of glucose between bloodmeals. Altogether, our data show that the regulated expression of the proline metabolism pathway in T. b. brucei allows this parasite to adapt to the nutritional environment of the tsetse midgut.
| Bloodsucking insects play a major role in the transmission of pathogens that cause major tropical diseases. Their capacity to transmit these diseases is directly associated with the availability and turnover of energy sources. Proline is the main readily-mobilizable fuel of the tsetse fly, which is the vector of sub-species of Trypanosoma brucei parasites that cause human sleeping sickness and are partly responsible for animal trypanosomiasis (Nagana disease) in sub-Saharan Africa. Once trypanosomes are ingested from an infected host by the tsetse, the parasites encounter an environment that is poor in glucose (as it is rapidly metabolized by the fly) but rich in proline, which then becomes the main carbon source once the parasite differentiates into the first insect (procyclic) stage. In this work, we provide evidence on the essentiality of T. b. brucei proline catabolism for procyclic survival within the tsetse’s digestive tract, as this organism is unable to synthesize this amino acid and strictly depends on the proline provided by the fly. We also show that parasites deficient in TbP5CDH, a mitochondrial enzyme involved in the proline degradative pathway, failed to proliferate in vitro, showed a diminished respiratory capacity, and showed compromised maintenance of energy levels and metabolic flux when proline was offered as the main carbon source. Thus, the integrity of the trypanosome proline degradation pathway is needed to maintain essential functions related to parasite bioenergetics, replication and infectivity within the insect host. Our observations answer a long-standing question on the role of parasite proline metabolism in tsetse-trypanosome interplay.
| The study of the metabolic interactions between parasites and insect vectors is critical to understanding their biology and evolution, as well as to aid the design of control strategies that aim to prevent transmission of vector-borne pathogens. Parasites of the Trypanosoma brucei sub-species cause sleeping sickness and Nagana disease in sub-Saharan Africa, and are exclusively transmitted by tsetse (Glossina spp.) flies [1–3]. When T. b. brucei bloodstream forms (BSF) are ingested by a fly, the replicative ‘slender’ trypanosomes rapidly die within the insect midgut (MG), whereas the pre-adapted ‘stumpy’ trypanosomes differentiate into the procyclic form (PF) within 24h [4]. Establishment of a trypanosome infection in the tsetse MG involves parasite colonization of the ectoperitrophic space (a cavity between the peritrophic matrix and the gut epithelium) and subsequent migration to the proventriculus (PV) [5], where the parasite is confined and further differentiates [6]. After multiple morphological and biochemical changes (reviewed in [7, 8]), the parasites then migrate to the salivary glands (SG), where they remain attached to the epithelial cells as epimastigotes ([9] and reviewed in [7]). After colonizing the SG, epimastigotes differentiate into infectious metacyclic forms, which are then released into the fly’s saliva and transmitted to another vertebrate host during a subsequent feed [4].
Unlike most Dipterans, tsetse flies do not store carbohydrates for ATP production [10]. Furthermore, glucose does not seem to constitute a relevant source of energy, is rapidly metabolized (~1h) after the bloodmeal is ingested, and is also found in low amounts in the fluids of these insects [11]. The use of minute amounts of glucose seems to be restricted to the production of other metabolites, such as non-essential amino acids in anabolism-requiring situations, e.g. pregnancy [12]. Thus, tsetse flies are adapted to efficiently metabolize amino acids and, more specifically, to catabolize proline to accomplish ATP biosynthesis [13, 14], a characteristic that is associated to obligatory blood feeding dipterans [15]. Additionally, proline is important in lactation, it is the metabolite that energetically supports the flight process and it is preferentially utilized by sarcomeres (flight muscle cells), yielding alanine as the main product. In this context, proline is a critical metabolite for tsetse biology [16].
Amino acid metabolism requires a robust transamination network that allows the transfer of amino groups (-NH2) to different acceptors, mainly ketoacids. In the specific case of glutamate, -NH2 is preferentially transferred to pyruvate, and yields alanine and oxoglutarate, which are the main intermediate products of proline catabolism. In tsetse flies, alanine is produced from proline by muscle cells and is further delivered into the hemolymph, which is then taken up into the fat body cells, for proline production [17]. This newly synthesized proline is, in turn, delivered to the hemolymph and taken up by flight muscle cells [13, 18]. This cycle allows the continuous supply of proline to flight muscles by keeping high proline levels in the hemolymph, which fuels insect flight [19].
During the T. b. brucei life cycle, the parasite goes through a deep metabolic reprogramming; this process allows the parasite to optimize its nutritional requirements according to the available metabolic resources in each environment. This is the case when trypanosomes transit from glucose-rich environment (in the bloodstream of the mammal) to one rich in amino acids (tsetse midgut), which requires a profound metabolic switch (reviewed in [4, 20]). Among the amino acids catabolized, L-proline plays a major role in the bioenergetics of trypanosomes [21–24]. In particular, the procyclic stage of T. b. brucei uses L-proline as a major carbon and energy source [23], which is actively taken up [25] and catabolized inside the mitochondrion into succinate, alanine and acetate with production of intermediate metabolites, reduced cofactors and ATP [26, 27]. Conversion of proline into glutamate is mediated by two enzymatic steps and one non-enzymatic step. First, proline is oxidized into Δ1-pyrroline-5-carboxylate (P5C) by a FAD-dependent proline dehydrogenase (TbProDH) [EC 1.5.99.8] [23]. Second, the cyclic P5C ring is spontaneously opened through a non-enzymatic reaction to produce glutamate-γ-semialdehyde (γGS). Third, the carbonyl moiety of γGS is further oxidized to glutamic acid by a P5C dehydrogenase (TbP5CDH) [EC 1.5.1.12] with a concomitant reduction of NAD(P)+ into NAD(P)H [28]. Unlike Trypanosoma cruzi, there are no genomic or biochemical data supporting the existence of a proline biosynthetic pathway in T. b. brucei [29], which suggests it is auxotrophic for this amino acid. Moreover, in PFs it was reported that proline degradation is downregulated in the presence of glucose [24], and the importance of Ca2+ regulation of TbProDH activity in the energy metabolism of trypanosome insect stages was recently suggested [30]. Collectively, both proline oxidation to glutamate and further oxidation through a part of the tricarboxylic acid cycle (TCA) are able to produce reduced equivalents, as well as fuel oxidative phosphorylation, and thus contribute to fulfilling the parasite’s energy requirements [31].
The relevance of proline metabolism for both T. b. brucei and the tsetse led us to address the long-standing question on the role of this amino acid in the parasite´s ability to infect flies. While the importance of TbProDH to the parasite’s biology has previously been studied, little is known on the specific role of TbP5CDH, besides its participation in the complete oxidation of proline. In this work we addressed this issue by studying the role of TbP5CDH in the bioenergetics of T. b. brucei as well as its importance during a tsetse infection. Our data show that in the absence of glucose, T. b. brucei PFs rely on the proline provided by the fly and on a fully functional proline catabolic pathway to successfully survive within the tsetse midgut.
In order to understand the role(s) of TbP5CDH in T. b. brucei biology, we first characterized its expression during the in vitro growth of both procyclic cultured forms (PCFs) and BSFs. Parasites were cultured in complete SDM79 and HMI9 media, respectively, and their growth followed up for 72h (although the analyses were made at 24 and 48h depending on the different parasite doubling times; Fig 1A). To analyze the expression profile of TbP5CDH and its influence on proline metabolism, TbP5CDH mRNA and protein levels were determined by qPCR and western blot, respectively. While both the mRNA and protein levels remained almost constant over time in PCFs, no TbP5CDH protein was detected in BSFs (Fig 1B and 1C). This indicates that, at least in vitro, expression of this enzyme is tightly regulated between different trypanosome stages. This observation is consistent with previous data showing that proline catabolism seems to be repressed in T. b. brucei BSFs [32]. We then investigated whether TbP5CDH expression is developmentally regulated during tsetse infection by isolating parasites from different infected organs; i.e. MG, PV and SG. TbP5CDH mRNA was detected in parasites collected from the PV and MG but not from SG-derived forms (Fig 1D). No significant changes in the expression levels were observed between PV and MG forms, but there was a strong reduction (60-fold change, p<0.05) in mRNA levels in SG forms. Notably, it was not possible to examine TbP5CDH protein expression by western blotting due to strong cross-reactivity with the Glossina P5CDH protein. Collectively, these results suggest that both PV and MG trypanosome forms express the proline-oxidizing pathway, which would be necessary to fulfill the energy requirements for cell proliferation, although the enzyme is downregulated as the infection progresses towards the SGs.
To determine the subcellular location of TbP5CDH, T. brucei PCFs were submitted to digitonin fractionation and the enzyme was detected by western blotting. As shown in Fig 2A, TbP5CDH was released together with the mitochondrial markers TbASCT and TbProDH, while the cytosolic marker enolase was released at much lower digitonin concentrations (20 μg compared to 350 μg of digitonin mg-1 of protein) [33]. Under these assay conditions, we also detected TbProDH but at low amounts, which is consistent with its possible association with the mitochondrial inner membrane (Fig 2A) [23]. Furthermore, immunofluorescence of fixed PCFs showed co-localization of TbProDH and TbP5CDH (Fig 2B), thus confirming the results obtained by digitonin fractionation.
To determine the importance of TbP5CDH in the bioenergetics of trypanosomes, we downregulated its expression by RNAi using a tetracycline-inducible system [34]. After 72h of tetracycline-induction (RNAiTbP5CDH tet+), no TbP5CDH was detected by western blotting (Fig 3A). However, when we assayed its enzymatic activity, we observed ~16% remaining activity compared to non-induced cells (RNAiTbP5CDH tet-) (Fig 3B). No changes in the levels of TbP5CDH were observed in wt cells supplemented or not with tetracycline (wt tet-/+), which showed that addition of this antibiotic had no direct effect on TbP5CDH expression (Fig 3A and 3B).
As previously shown, wt PCFs are able to replicate in standard SDM79 supplemented (or not) with glucose (SDM79 and SDM79 glc-, respectively) [23]. In standard SDM79, glucose is the preferred carbon source for PCFs, whereas in the absence of glucose, the parasites mainly use proline as a carbon source and for ATP production. In the case of TbP5CDH, the enzyme was essential when proline was the major carbon source. However, the phenotype was not lethal most likely because of the remaining enzymatic activity in the RNAiTbP5CDH cell line (Fig 3C and 3D). These findings prompted us to evaluate the main mitochondrial functions (i.e. ΔΨm, O2 consumption rates and ATP levels) in RNAiTbP5CDH cells energized with proline.
In digitonin-permeabilized cells, downregulation of TbP5CDH caused a diminished capacity to retain the mitochondrial dye safranin and to respond to the addition of ADP compared to non-induced cells. This profile reflects a partial depolarization of mitochondria from RNAiTbP5CDH tet+ cells when proline is the electron source for the oxidative phosphorylation (OxPHOS) process (Fig 4A). No changes were observed for the same parameters when succinate was used as a mitochondrial substrate (S1A and S1B Fig). In addition, ADP failed to induce the proton flux into the matrix space through the Fo/F1 ATP synthase complex and did not decrease ΔΨm to the same levels shown by non-induced cells. Moreover, addition of oligomycin, an inhibitor of ATP synthase, also resulted in a slight increase in ΔΨm, and reestablished the resting levels, which were significantly lower than control. This is likely due to the diminished electron flux from proline degradation to the respiratory complexes in RNAiTbP5CDH tet+ parasites, which seem to be insufficient to sustain physiological levels of OxPHOS (Fig 4A). Interestingly, addition of Ca2+ to these mitochondrial preparations did not affect the ΔΨm of wt and RNAiTbP5CDH cells, which suggests that variations in the electrochemical potential using proline are due exclusively to mitochondrial electron transfer chain (mt-ETC) capacity rather than mitochondrial Ca2+ influx (Fig 4B). Observations made at the ΔΨm level are consistent with the diminished ability of the mutant cell line to consume O2 when proline and ADP were present at high concentrations (respiratory state 3), and the high respiration rates are limited by respiratory chain activity [35]. Moreover, the maximal oxygen reduction capacity was dramatically affected in the RNAiTbP5CDH tet+ cells when FCCP (which collapses the mitochondrial membrane potential) was added to the mitochondrial preparations (Fig 4C, Table 1), and the respiratory control ratio significantly decreased to 1.44 ± 0.02 (Table 1). When succinate was used as the respiratory substrate in control and RNAiTbP5CDH tet-/+ parasites, no differences in ΔΨm or O2 consumption rates were observed (S1A–S1C Fig). The ATP levels in parasites cultivated in either SDM79 and SDM79 glc- media were also determined. As expected, the absence of TbP5CDH did not affect ATP levels when glucose was present (Fig 4D). Conversely, when ATP synthesis relied on proline oxidation (cells grown in SDM79-glc-), the capacity of RNAiTbP5CDH tet+ cells to produce ATP was diminished (Fig 4D).
Given that the T. b. brucei genome does not appear to contain genes that encode putative P5C/γGS metabolizing enzymes (with the exception of TbP5CDH), it is assumed that the proline-glutamate pathway has no branches. On this basis, it is expected that TbP5CDH-knocked down cells would produce elevated quantities of intracellular P5C, which has been described as a toxic metabolite in several cell types [36, 37]. Thus, the deleterious effect observed in TbP5CDH knockdown cells could be due not only to a diminished efficiency in ATP synthesis but also due to P5C accumulation. To evaluate this, RNAiTbP5CDH tet-/+ parasites were incubated in vitro under different metabolic conditions (i.e. PBS supplemented with L-proline, glucose, proline plus glucose, or P5C/γGS), and their viability was assessed over a 3h period. Controls consisted of RNAiTbP5CDH tet-/+ cells incubated with either SDM79 (100% viability) or PBS (which yielded a 3% viability compared to cells incubated in SDM79 alone). PCFs incubated in the presence of either proline or proline plus glucose showed a viability of 65% versus SDM79-treated cells, and no significant differences were found for these treatments between induced or not-induced cells (Fig 5A). The addition of P5C/γGS to the RNAiTbP5CDH tet- cells resulted in almost the same viability as proline treatment (50%). Notably, incubation of RNAiTbP5CDH tet+ cells with P5C/γGS reduced their viability by more than 90% (Fig 5A). In addition, non-induced and RNAi-induced procyclics were treated with proline or P5C for 1 or 3h, and P5C toxicity was indicated based on loss of plasma membrane integrity. Only in RNAiTbP5CDH tet+ cells P5C but not proline treatment resulted in 15% and 57% of Propidium Iodide (PI)-positive cells after 1h and 3h challenge, respectively (Fig 5B). These data were compatible with observed mitochondrial and morphological alterations (Fig 5C). Interestingly, in spite of its deleterious effect, P5C was able to support MitoTracker accumulation (a process that is dependent on the mitochondrial inner membrane potential) and to maintain higher ATP levels in wt or RNAiTbP5CDH tet-, when compared to RNAiTbP5CDH tet+ cells. These results, along with previous published evidence [36, 37], suggest that i) P5C is able to reach the mitochondrial matrix; ii) the only metabolic fate for P5C/γGS is to be oxidized to glutamate via TbP5CDH; and iii) the intracellular accumulation of P5C/γGS has a detrimental effect on PCFs viability.
The increased susceptibility of RNAiTbP5CDH tet+ when exogenous P5C/γGS is added is indicative of the inability of T. brucei PCFs to reduce it to proline. Thus, we then evaluated whether proline biosynthesis from glutamate or from P5C could happen in T. b. brucei. To address this question, parasites were grown in defined media supplemented or not with proline. When PCFs were grown in either complete SDM79 or SDM79 glc- media no differences were found in the cells doubling time (19.3 ± 1.1 h and 20.3 ± 1.4 h, respectively). After proline deprivation of the media (SDM79 pro- glc-), PCFs showed a delay in doubling time (48.4 ± 6h) (Fig 6A). This diminished capability for proliferation under proline-depleted media strongly suggests that T. brucei is auxotrophic for this amino acid. Furthermore, when the T. b. brucei genome was interrogated for putative genes that encode P5C-synthase (P5CS; converts glutamate into P5C/γGS0) and P5C-reductase (reduces P5C/γGS into proline), using T. cruzi sequences as queries, only a protein sequence with 65% similarity with T. cruzi P5CR was found (TritrypDB accession number: Tb927.7.2440). No significant hits were found for P5CS (TritrypDB accession number: TCSYLVIO_005298). The presence of a putative P5CR ortholog in T. b. brucei prompted us to evaluate its enzymatic activity by measuring the reduction of P5C to proline in PCF cell-free extracts. The enzymatic test for P5CR revealed activities of 8.6 ± 0.5 versus 60 ± 9 nmol NADPH/min/mg of protein in T. b. brucei PCF and T. cruzi epimastigote cell-free extracts, respectively (Fig 6B). Furthermore, P5CS protein was not detected in T. b. brucei lysate using antibodies raised against its T. cruzi ortholog (Fig 6C). To evaluate the possible occurrence of a proline biosynthetic pathway in T. b. brucei PCFs, the levels of this amino acid were measured in proline–deprived parasites (after 1h incubation in PBS). The cells were then incubated with different substrates that would restore proline levels, i.e. via uptake (proline), reductive biosynthesis (P5C/γGS, glutamate, glutamine), or through the connection between the urea cycle and proline-glutamate pathway (arginine or alanine) as occur in other organisms. Collectively, the demonstration that the only metabolite capable of restoring the normal intracellular levels of proline in PCFs after starvation was proline (Fig 6D) and the lack of genetic and biochemical evidence for a proline biosynthetic pathway in T. b. brucei further corroborate its auxotrophic nature for this amino acid.
As no proline biosynthetic pathway or ornithine transaminase activity could be evidenced in T. b. brucei, TbP5CDH should be the only enzyme capable of metabolizing intra-mitochondrial P5C in these cells. In order to unambiguously evaluate the occurrence of this enzymatic activity we kinetically characterized TbP5CDH from PCF lysates. Our data revealed that these cells were able to reduce NAD+ upon the addition of P5C in a concentration dependent manner with apparent KM values of 92.7 ± 14 μM and 0.38 ± 0.04 mM for its substrate (P5C/γGS) and cofactor (NAD+), respectively, and Vmax values of 0.15 ± 0.01 and 0.19 ± 0.01 μmol/min/mg of protein for P5C and NAD+, respectively (S2 Fig).
To further determine the metabolic perturbations caused by downregulation of TbP5CDH, end products excreted from the catabolism of proline and [U-13C]-glucose were analyzed by proton-NMR spectroscopy. We used a previously-developed metabolite profiling assay based on the ability of proton NMR spectroscopy to distinguish 13C-enriched from 12C molecules [38]. Cells were incubated in PBS with equal amounts (4 mM) of non-enriched proline and of [U-13C]-glucose in order to perform a quantitative analysis of proline-derived and glucose-derived acetate production by proton NMR. For instance, [13C]-acetate derived from metabolism of [U-13C]-glucose (annotated 13C in Fig 7) is represented by two doublets, with chemical shifts at around 2.0 ppm and 1.75 ppm, respectively, while the central resonance (1.88 ppm) corresponds to proline-derived [12C]-acetate. As expected, the amounts of [U-13C]-glucose-derived end products (13C-enriched succinate, acetate and alanine) are similar in the RNAiTbP5CDH tet+ mutant and wt cells (2081 versus 2057 nmol excreted/h/mg of proteins), whereas the amounts of excreted end products from proline degradation (non-enriched succinate, acetate and alanine) were 2.2-reduced in the RNAiTbP5CDH tet+ cell line (Fig 7 and Table 2). The remaining production of end products excreted from proline metabolism (44% compared to wt cells) was probably due to a 16% residual TbP5CDH activity in the tetracycline-induced RNAiTbP5CDH mutant. Notably, reduction of succinate and acetate production from proline is compensated by an increased production of these molecules from glycolysis (Fig 7, Table 2). Such flux redistribution towards glucose-derived acetate production was also previously observed in the threonine dehydrogenase procyclic mutant incubated with threonine and [U-13C]-glucose [38]. Altogether these metabolic data demonstrate that TbP5CDH is involved in the proline degradation pathway of procyclic trypanosomes.
After observing differences in the expression levels of TbP5CDH during parasite development in the fly (Fig 1D), we then analyzed its essentiality for parasite survival in the tsetse midgut. Flies were infected with a bloodmeal supplemented with either wt or RNAiTbP5CDH PCFs, which were either previously induced or not with tet. At 9 days post-infection (dpi), the flies were dissected and midgut infections were determined. Flies fed with either wt or RNAiTbP5CDH tet- cells had infection rates of 82% (Fig 8A, S3 Fig). Furthermore, there were no differences in the number of parasites in the midguts of wt tet-, wt tet+ or RNAiTbP5CDH tet- infected flies (Fig 8A, S3 Fig). However, after downregulation of TbP5CDH, the midgut infection rates dropped significantly to 58% (p<0.01) and, importantly, only a few parasites were visible (Fig 8A). Furthermore, under normal TbP5CDH expression (i.e. wt tet-, wt tet+ or RNAiTbP5CDH tet-), the infected midguts had a much higher number of parasites (>1000 cells per field) compared to flies infected with RNAiTbP5CDH tet+ cells (≤10 cells per field) (Fig 8A, S3 Fig). Parasites were probably present in the latter group due to residual expression of TbP5CDH and/or to the transient utilization of glucose present in subsequent bloodmeals. Altogether, these data demonstrate that TbP5CDH activity, a key enzyme in the parasite proline metabolism pathway, is crucial for trypanosome survival within the tsetse fly midgut.
EP- and GPEET-procyclins are the most abundant GPI-anchored surface glycoproteins on the surface of T. b. brucei PCFs. The C-terminus of all EP-isoforms contains abundant (up to 30) repeats of glutamate (E) and proline (P) dipeptides [55]. Likewise, GPEET-procyclin is also rich in E and P because of its 5–6 GPEET C-terminal repeats. We investigated whether alterations in the proline-glutamate oxidative flux interfere with the expression of all procyclin isoforms. Western blotting analysis showed a slightly decreased in EP-procyclin expression in RNAiTbP5CDH tet+ cells compared to wt tet-, wt tet+ (Fig 8B). Interestingly, perturbations in the number of EP-positive cells were found after four days of RNAi-induction for TbP5CDH. Two different cell populations were observed, which were named as EP-pop1 and EP-pop2 (Fig 8B, right panel). The EP-pop1 displayed similar values of fluorescence intensity versus controls (wt tet-/+ or RNAiTbP5CDH tet- (Fig 8C), whereas the EP-pop2 population showed a 10-fold reduction in the mean of fluorescence. However, when the repertoire of procyclins was analyzed by MALDI-TOF (S4A–S4D Fig) mainly EP1-2 and EP3 isoforms (containing 25 and 22 EP repeats, respectively [55]) were detected in either induced or non-induced cells. This suggests that although the overall expression of EP-procyclins appears to be slightly compromised when the proline metabolism pathway is altered (Fig 8C), these cells do not seem to compensate the slight EP deficit by re-expressing GPEET-procyclin.
Once T. b. brucei blood forms are ingested by the tsetse, differentiation of stumpy trypanosomes to procyclics is triggered by a combination of at least two key factors that modulate parasite gene expression [39], i.e. a drastic decrease in temperature and the presence of specific metabolites inside the fly’s gut. Many developmental changes allow parasites to adapt to the midgut’s hostile environment, including the expression of a procyclin coat, which helps to protect the parasite surface against tsetse midgut proteases [40], and development of a functional mitochondrion for energy production. In this work, we biochemically and genetically characterize TbP5CDH, an essential mitochondrial enzyme involved in L-proline catabolism. Trypanosomes deficient in the expression of TbP5CDH failed to proliferate in vitro in the absence of glucose (which mimics the tsetse midgut environment) and showed compromised mitochondrial activity. Thus, the integrity of the proline degradation pathway in T. b. brucei is needed to maintain essential functions related to parasite bioenergetics, replication and infectivity within the insect host. We further demonstrated that T. b. brucei is unable to produce L-proline; instead it utilizes the proline available in the tsetse midgut. Collectively, our observations confirm the long-standing suggestion that proline metabolism in T. b. brucei is essential for in vivo energy production, thus ensuring the viability of infection within tsetse fly.
Some Dipterans (including the genus Glossina) are well adapted to use amino acids for energy production. In fact, due to the scarce carbohydrate reserves in tsetse, glycolytic activity is negligible within this insect [11, 41]. Three characteristics make proline a readily mobilizable energy source in tsetse: i) its highly reduced state, which is related to its high yield in terms of metabolic energy production (i.e. 5-fold more efficient than carbohydrates); ii) its high solubility (allowing its transport in high concentrations, thus permitting an efficient distribution in the entire fly body); and iii) its low nitrogen content limiting the amount of energy required for nitrogen detoxification (reviewed in [42]). Einar Bursell concluded that "proline constitutes the only effective substrate for flight metabolism" for species in which both sexes are obligatory blood-feeders (i.e. Glossina spp.) [15]. In fact, proline represents ~4% of the total amino acid content in the tsetse hemolymph and is efficiently burnt during the flight process [43, 44]. It is first oxidized to glutamate, and further converted into oxoglutarate by either an alanine transaminase or a glutamate dehydrogenase (Fig 9, left panel). Consequently, flight time in tsetse (which is limited to about three minutes) is likely to be determined by the amount of proline available in the hemolymph at the outset [43].
On the other hand, the synthesis of proline from alanine in tsetse takes place in the fat body. It is a complex process that comprises an alanine-glyoxylate transaminase, a pyruvate dehydrogenase and part of the TCA cycle. Part of the oxoglutarate produced is converted into glutamate in the same transamination reaction in which a new alanine molecule is converted into pyruvate to feed again the TCA cycle (Fig 9, right panel) [17, 18, 45]. Thus, there is a strong interdependence between proline/alanine metabolism between the fat body and flight muscles of tsetse, which are metabolically connected through the hemolymph. Notably, this metabolic system does not work at the steady state: the release of CO2 by the flight muscles creates a deficit of carbon. This deficit is possibly compensated by using Acetyl-CoA from the β-oxidation of lipids in the fat body for proline de novo biosynthesis [17].
During a trypanosome infection, parasite colonization of different tsetse organs may alter the fly’s proline-alanine cycle. Such an alteration would not only have an impact on the activity of flight muscles, but also affects tsetse reproduction [45]. The crosstalk in the utilization of proline in trypanosome-infected flies becomes even more complex with the dependency on Wigglesworthia glossinidia (the obligate tsetse bacterial symbiont) for the production of vitamin B6, which is essential for activity of alanine-glyoxylate aminotransferase (involved in proline regeneration in fat body) (reviewed in [16]).
It has previously been shown that TbProDH is essential when parasites are grown in the presence of proline as the main energy source (SDM79 glc-) [23]. In the present work, a different phenotype was observed: TbP5CDH knockdown cells did not die in vitro. However, the essentiality of TbP5CDH for survival in the fly was evident by the low midgut infection phenotype in knockdown cells. This discrepancy between the in vitro phenotypes could be due to the residual activity (~16%) of TbP5CDH in the tetracycline-induced RNAiTbP5CDH mutant, which would be able to maintain a low but significant metabolic flux thus allowing proline oxidation in the TCA cycle/ETC. Alternatively, the activity of TbProDH, a FAD dependent enzyme, might also be able to directly transfer electrons to the ubiquinone pool at the ETC (similarly to succinate dehydrogenase), as already proposed for T. cruzi [21]. Both possibilities, individually or combined, could explain parasite survival by partially fulfilling the energy requirements of these cells. In T. b. brucei, proline conversion into P5C/γGS produces FADH2, which can transfer 2e- to the UQ pool with further reduction of cytochromes [21]. γGS is then converted into glutamate to produce NADH. In addition, four additional reactions downstream to the proline-glutamate conversion produce NADH in T. b. brucei (reviewed in [29]). T. b. brucei expresses a mitochondrial NADH: ubiquinone oxidoreductase (which is rotenone insensitive), uses FMN as cofactor, transfers one e- to the ubiquinone (UQ) pool and can reduce O2 to O2-• anion [46]. This enzyme is likely to be involved in the reoxidation of NADH, thus reducing UQ and driving proton pumping at level of C-III and C-IV in the mt-ETC [47]. Then, both proline oxidation steps generate reducing equivalents that feed the OxPHOS, thus driving the ATP synthesis through the FoF1/ATP synthase. The intramitochondrial glutamate produced from proline can either be: i) deaminated into oxoglutarate or ii) transaminated to pyruvate forming alanine. Oxoglutarate can be converted into succinyl-CoA and then into succinate, constituting two points of ATP generation, by substrate level phosphorylation (at succinyl-CoA synthetase level) and OxPHOS (via succinate dehydrogenase complex), respectively [48]. Succinate can also be excreted as an end product. In the absence of glucose, alanine is also excreted by T. b. brucei PCFs as the end product of proline degradation. It may be possible that tsetse also utilizes trypanosome-excreted alanine for further conversion into proline, especially in highly infected tissues.
An intriguing phenotype evidenced in our RNAiTbP5CDH cell line, was the cell toxicity displayed when exogenous P5C/γGS was added to tet-induced parasites. In most eukaryotes, P5C/γGS can be synthesized by proline oxidation, glutamate reduction or by loss of the -NH2 group at the δ-carbon of ornithine through an ornithine transamination reaction. In turn, P5C/γGS can be decreased by its oxidation to glutamate by P5CDH, its reduction to proline, or by its amination to form ornithine [49]. Thus, the amount of free P5C/γGS mainly results from the balance between all these enzymatic activities. T. b. brucei lacks a functional urea cycle, which eliminates any connection between this pathway and P5C/γGS [50]. Furthermore, our results show that neither relevant enzymatic activities related to proline biosynthesis nor genes encoding the putative enzymes for these pathways are present in the T. b. brucei genome. In addition, there was a cytotoxic effect for externally added P5C/γGS to RNAiTbP5CDH tet+, which supports the oxidation to glutamate as the only fate for this metabolite in PCFs. It should be noted that these results differ to those obtained when TbP5CDH-knockdown cells were treated with proline (resulting in the intracellular accumulation of P5C/γGS due to the TbProDH activity), which indicated the cells remained viable (although non-replicative). This was also consistent with the viability shown by cells with increased levels of P5C/γGS by overexpressing a mitochondrial carrier (TbMCP14) [51]. Altogether our results showed that, unlike T. cruzi –whose energetic metabolism also relies on proline consumption [21, 28]-, T. b. brucei PCF is auxotrophic for proline, this being an essential metabolite and a main carbon source during this stage. As a consequence, P5C/γGS levels depend exclusively on the balance between its formation from proline oxidation and its depletion by oxidation to glutamate. In addition, we confirmed that proline deprivation dramatically affects cell proliferation as previously suggested [52]. Altogether, our data provide evidence that T. b. brucei has a strict requirement of a complete proline to glutamate oxidation pathway to successfully colonize tsetse midguts.
The relevance of EP-procyclin expression for the successful development of T. b. brucei within the fly has been widely reported [53, 54]. Expression of procyclins (GPEET and EP) varies according to the parasite stages in the tsetse [40]. There seems to be a correlation between the expression of such molecules in midgut forms and elevated mitochondrial activities [55]. More specifically, GPEET expression (normally in early stages) can be reactivated in late forms when mitochondrial activities such as the ASCT cycle or alternative oxidase are inhibited [55]. It was also stated that glycolytic activity, disrupted by RNAi-silencing of the trypanosome hexokinase gene, produces a switch in the surface expression from EP- to GPEET-procyclin [56]. We observed herein that alterations in the proline-glutamate pathway slightly affects the levels of EP-procyclin expressed at the surface. However, this alteration did not induce a change in the type of procyclins these cells expressed and no evidence of GPEET re-expression was observed. Given the specificity of the anti-EP mAb 247 for the glu-pro dipeptides [57], it is likely that the alterations in EP-procyclin expression after down-regulation of TbP5CDH, could be simply due to a reduction in the overall levels of intracellular glutamate available for making the C-terminus glu-pro repeats. Furthermore, it is unlikely that such a small reduction in the surface expression of EP-procyclins accounts for the inability of RNAiTbP5CDH tet+ cells to colonize the tsetse midgut, although EP-procyclin-null trypanosomes are less efficient in establishing midgut infections [53, 54]. Thus, these results further confirm that the fly phenotype observed in knocked down cells appears to be mainly a direct consequence of an interrupted proline metabolism pathway in these parasites.
Animal experiments in this work were performed in accordance with the local ethical approval requirements of the Liverpool School of Tropical Medicine and the UK Home Office Animal (Scientific Procedures) Act (1986) under license number 40/2958.
BSF of T. b. brucei TSW196 strain [58], which is a fully fly-transmissible, was used for gene expression studies and proline determination in infected flies. BSF T. b. brucei from 2T1 strain (kindly provided by David Horn, University of Dundee, UK) were cultured in HMI-9 medium supplemented with 15% (v/v) FCS (Gibco) at 37°C and 5% CO2 [59]. The initial cell density was 5x104 cells/ml, which was sub-cultured each 48h. Parasite densities were determined by cell counting using a hemocytometer. PCFs of T. b. brucei Lister 427 (29–13 clone) (T7-RNAp+ NEO+ TET+ HYG+), which expresses the T7 RNA polymerase under the control of the tetracycline (tet) promoter, was cultured in vitro in SDM79 media (Gibco) supplemented with GlutaMAX (Gibco), 7.5 μg/ml hemin and 10% (v/v) heat-inactivated fetal calf serum (FCS) [60, 61] at 26°C. For RNAi experiments, parasites were grown in SDM79 media supplemented with 25 μg/ml G418 (G) and 12.5 μg/ml hygromycin (H) as indicated [62]. To grow PCFs in defined medium, we used SDM79 base media without sodium bicarbonate, glucose, glutamine, glutamate, proline, pyruvate, threonine and acetate (SDM79-CGGGPPTA) (PAA laboratories, Pasching, Austria) and then supplemented, except for glucose (SDM79 glc-) or proline (SDM79 glc- pro-) as indicated [23]. In both cases, the preparation was supplemented with 10% (v/v) tet-free FCS (Clontech Laboratories) and an excess of 50 mM N-acetyl-D-glucosamine (GlcNAc) to inhibit uptake of glucose presented in serum (about 1.5 mM) [63]. The initial cell density was 106 cells/ml and sub-culturing was done every 72h [60].
Glossina morsitans morsitans flies were maintained in a laboratory colony at the Liverpool School of Tropical Medicine (LSTM) at 26°C and 65–70% relative humidity. Teneral (12-24h post-emergence) flies were fed on sterile defibrinated horse blood (TCS Biosciences Ltd., Buckingham, UK).
Fly-derived trypanosomes were isolated from the MG, PV and SG of infected flies, as described. Parasites were resuspended in SDM79 medium and midgut debris were removed by filtration through cytometer-tubes filter (Becton Dickinson). Cells were harvested by centrifugation (2,000 g for 10 min at 4°C), washed twice with cold PBS (137 mM NaCl, 2.7 mM KCl, 8 mM Na2HPO4, 1.5 mM KH2PO4 adjusted to pH 7.3), counted and stored at -80°C until RNA or protein analysis. Total RNA extractions from fly-derived parasites (~5x106 cells) were performed with TriZol reagent (Sigma) following standard procedures [64]. Then, 300 ng total-RNA were used for cDNA synthesis with oligo (dT)20 and SuperScript III Reverse Transcriptase (RT) (Invitrogen). Resultant cDNA samples were diluted (1:4) in nuclease-free water (NFW) for use in quantitative RT-PCR (qPCR). Based on DNA-sequences for TbGAPDH and TbP5CDH (TritrypDB accession numbers: Tb927.6.4300 and Tb427.10.3210, respectively), specific primers were designed (S1 File for oligonucleotides sequences). qPCR reactions were performed in 96-wells (Stratagene, Agilent Technologies, La Jolla, TX, USA) using 3.2 pg of each primer, 5 μl fast SYBR green master mix (Applied Biosystems, Life Technologies, CA) and 5 μl of cDNA samples to a final volume of 20 μl per well. Reactions were run in a Mx3000P qPCR-system (Stratagene) followed by a dissociation curve. Samples from naïve tissues were also used to verify primer specificity.
The pZJM vector, which contains a cloning site between two opposing T7 promoters, was used to silence TbP5CDH expression (Tb427.10.3210) [34]. A 5ʹ DNA fragment (480 bp) corresponding to TbP5CDH was amplified by conventional PCR using specific primers (see S1 File for oligonucleotides sequences), cloned into the pZJM vector (pZJM/RNAiTbP5CDH) and the resulting construct was confirmed by sequencing. Plasmid preparation was done using the QIAGEN plasmid Maxi Kit according to the manufacturer’s instructions (QIAGEN). For transfections, 10 μg pZJM/RNAiTbP5CDH was linearized by digestion with the restriction endonuclease NotI (Thermo Scientific), precipitated by standard procedures and dissolved in NFW. PCF trypanosomes (2x107 cells maintained in mid-log phase in SDM79 H/G medium) were transformed using a Nucleofector transfection system II/2b, following the manufacturer’s instructions (Lonza). Parasites were seeded into 24-well plates (<10 cells/well) and cloned by limiting dilution in SDM79 H/G supplemented with 2.5 μg/ml phleomycin as a selection marker. The obtained parasite lineages were referred to as wt (parental Lister 427 29–13 strain) or RNAiTbP5CDH, as the RNAi-TbP5CDH cell line, in the presence or absence of tetracycline (tet-/+). RNAi was induced by adding 0.5 μg/ml tetracycline disodium salt (freshly dissolved in PBS) to the selective media (at 26°C).
Teneral flies were infected with bloodmeal preparations that contained either wt or RNAiTbP5CDH parasites. Briefly, non-induced (tet-) or tetracycline-added (tet+) parasites were added to sterile horse blood at a density of 5x105 parasites/ml. RNAi induction was maintained by adding 25 μg/ml tetracycline to the bloodmeal, and 24h after receiving an infectious blood meal, the flies were sorted and only fed flies were used. After nine days, flies were dissected and the number and intensity of infected midguts was determined by microscopy. A score was attributed to each infection as previously described [65].
Enzymatic determinations for both P5C reduction to proline or P5C oxidation to glutamate were performed. The substrate of TbP5CDH, a racemic mixture of DL-Δ1-pyrroline-5-carboxylate (DL-P5C) and its ring-open form gamma-glutamate semialdehyde (γGS), was synthesized from peroxidation with NaIO4 (Sigma), and eluted in acidic medium (1 M HCl) as previously described [66]. The steady-state activity for TbP5CDH was measured in cell-free homogenates from PCFs, as previously described for T. cruzi [28]. The TbP5CDH reaction mixture contained: 0.3 mM P5C/γGS (freshly prepared), 1 mM nicotinamide adenine nucleotide disodium salt (NAD+) and 90 mM potassium phosphate buffer pH 7.2, made up to a final volume of 3 ml with distilled water. The reaction was started after adding 200 μg cell-free homogenates from PCFs and the linear rate was determined by following the increase in absorbance (λ340nm) over 5 mins at 28°C with constant stirring. A blank without substrate (P5C/γGS) was used as a control. Readings of samples and controls were made in parallel in a double-beam Thermo Evolution 300 spectrophotometer (Thermo Scientific). Kinetic parameters for P5C and the cofactor of TbP5CDH were also determined in PCF homogenates. Substrate dependence was assayed by varying the P5C/γGS concentrations over the range of 20–600 μM (freshly prepared) and 1 mM of NAD+ as saturating concentration. Cofactor dependence was assayed by varying the NAD+ concentrations over the range of 0.01–2.5 mM and 600 μM P5C/γGS as saturating concentration. The P5C-reductase reaction mixture contained: 500 μM P5C/γGS (freshly prepared), 50 μM NADPH and 100 mM Tris-HCl pH 7.0, and was made up to a final volume of 1 ml with distilled water. The reaction was started by adding different concentrations of PCF homogenates. The linear rate was determined by following the decrease in absorbance (λ340nm) over 3 min at 28°C with constant stirring. P5C-reductase enzymatic activity determinations from T. cruzi homogenates were used as controls under the same conditions.
Parasites (Lister 427 29–13 strain) were incubated in PBS (for 1h) to diminish the intracellular pool of free proline. Parasites were then incubated for 40 min in the presence of different carbon sources and cofactors (S2 File for detailed mix composition) to determine which combination was able to restore the intracellular proline levels. Additional treatments consisted of parasite incubation with PBS supplemented with 5 mM L-proline (positive control) or non-supplemented PBS (negative control). Parasites were then washed with cold PBS and centrifuged (3,000 g for 5 min at 4°C). Pellets were resuspended in 100 μl lysis buffer (100 mM Tris-HCl pH 8.1, 0.25 M sorbitol, 1 mM EDTA, 1% (v/v) Triton X-100, 1 mM phenylmethanesulfonylfluoride (PMSF), 4 μg/ml aprotinin, 10 μg/ml tosyl-L-lysyl-chloromethane hydrochloride (TLCK) and 10 μM E-64), and submitted to two cycles of snap freezing in liquid nitrogen thawing. Crude extracts were clarified by centrifugation (15,000 g for 15 min at 4°C) and 100 μl supernatant was mixed (in a separate reaction) with 1 volume 20% (w/v) trichloroacetic acid for deproteinization. Samples were precipitated by centrifugation (20,000 g for 30 min at 4°C) and 200 μl of the resultant supernatants were used for the Bates assay, as described elsewhere [67].
Non-induced and RNAi-induced (tet-/+) PCFs from wt and RNAiTbP5CDH cell lines were grown for three days in SDM79 media at 26°C. Then, parasites were harvested by centrifugation and resuspended in either PBS or PBS supplemented with 5 mM L-proline, 1.5 mM P5C/γGS and 5 mM D-glucose, or with 5 mM L-proline + 5 mM D-glucose, and further incubated for 4h at 26°C. Cell viability was evaluated after incubation with 3-(4,5- dimethylthiazol-2-yl)-2,5-diphenyltetrazolium bromide (MTT) [68]. Results were obtained from three biological replicates (n = 3). Comparisons between non-induced and RNAi-induced cells were calculated using the one-way ANOVA test in GraphPad Prism v5.0a for Mac OS X (GraphPad Software, USA).
PCFs (Lister 427 strain, 29–13 clone) cultivated at the late logarithmic phase of growth (3x109 total cells) in SDM79 medium were harvested by centrifugation [1,000 g for 10 min at room temperature (RT°)] and washed twice with PBS buffer. Total protein concentration was determined by the Bradford method [69] and the final pellet was resuspended in STEN buffer (250 mM sucrose, 25 mM Tris-HCl pH 7.4, 1 mM EDTA, 150 mM NaCl, 1 mM DTT and sigma-protease inhibitor mixture) adjusted to a final concentration of 1 mg/protein in 200 μl. Cells were treated with variable concentrations of digitonin (dissolved in STEN + dimethylformamide 40 mg/ml) in a final volume of 300 μl for each treatment, incubated for 4 min at 25°C and centrifuged (2 min at max speed), as previously described [70]. Supernatants corresponding to solubilized fractions were mixed with 1x SDS Laemmli buffer and analyzed by western blotting.
The presence of TbP5CDH, TbP5C-synthase, acetate:succinyl-CoA transferase (ASCT), enolase, TbProDH and EP-procyclins was determined by antibody detection in parasite homogenates. Briefly, parasites were harvested as described above and resuspended in lysis buffer that contained: 20 mM Tris-HCl pH 7.9, 1 mM EDTA pH 8.0, 0.25 M sucrose, 50 mM NaCl, 5% (v/v) glycerol, 1% (v/v) Triton X-100, 1 mM PMSF, 10 μg/ml aprotinin and 10 μg/ml leupeptin. Samples were chilled on ice (for 40 min) and clarified by centrifugation (15,000 g for 15 min at 4°C). Protein concentration was determined by the Bradford method using bovine serum albumin (BSA) as a standard [69]. Samples were submitted to protein electrophoresis (SDS-PAGE) and an equal amount of protein (30 μg) was loaded per lane. Proteins were transferred into 0.2 μm PVDF membranes (Amersham, GE, Life Sciences), blocked with PBS buffer plus 0.3% (v/v) Tween-20 (PBST) supplemented with 5% (w/v) skimmed milk powder and probed (16h at 4°C) against specific sera. The enzyme TbP5CDH was probed with a polyclonal specific serum (1:4,000) raised against its T. cruzi ortholog (TcP5CDH, TritrypDB accession number: Tc00.1047053510943.50) [28]. The enzyme P5C-synthase was probed with a polyclonal serum (1:3,000) produced in mouse against its close species T. cruzi, (TcP5CS, access code: TCSYLVIO_005298) exactly as previously described [28]. For digitonin assays, extracted fractions were probed with rabbit polyclonal antibodies against T. brucei ASCT (1:1,000), enolase (1:10,000), PPKD (1:1,000) and ProDH (1:500). EP-procyclins were probed with the monoclonal mAb-247 (1:1,500), which recognizes the EP-repeats of T. brucei procyclins (generous gift from Dr Terry W. Pearson, University of Victoria, Canada) [57]. As loading controls, two different polyclonal antisera were used: the mouse anti-TcGAPDH (1:3,000) and anti-HSP60 (access code: Tb427.10.6400) (1:2,000), dissolved in PBST-skim milk. Membranes were washed three times and incubated with goat anti-mouse IgG horseradish peroxidase (Sigma) diluted in PBST (1:50,000). Developing was done by using SuperSignal West Pico Chemiluminescent ECL substrate (Thermo Scientific) following the manufacturer’s instructions.
PCFs were cultured up to mid-exponential growth phase in SDM79. After this, parasites were washed with Voorheis’s modified PBS buffer (vPBS: 137 mM NaCl, 3 mM KCl, 16 mM Na2HPO4, 3 mM KH2PO4, 46 mM sucrose, 10 mM glucose) and harvested by centrifugation (850 g for 10 min at 4°C). Fixation, permeabilization and blocking were performed on poly-lysine coated glass slides, as previously described [71]. For antibody staining, polyclonal antisera produced against TbProDH (1:200) and TcP5CDH (1:250) [28] were dissolved in vPBS containing 20% (v/v) FBS and incubated for 2h at room temperature. Slides were washed five times with PBS and then incubated with AlexaFluor488-coupled goat anti-mouse IgG (Invitrogen) secondary antibody (1:600) plus TexasRed-X conjugated goat anti-mouse IgG (H+L) (Invitrogen) (1:400) for 1h. DNA staining was performed by adding 10 μg/ml of Hoechst probe (Invitrogen) and incubated for 5 min. Next, 2μl Fluoromount-G (GE, Healthcare) was added and a cover slip was mounted. Trypanosomes were visualized in a Leica DMi8 fluorescence microscope (Leica Microsystems) under an apochromatic 40x magnification lens. Image overlaying was done in imageJ software (NIH, Bethesda, MA, USA).
Wild type and RNAiTbP5CDH (tet-/+) PCFs (5x106 parasites) grown (3 days) in complete SDM79, as described above, were harvested (2,000g for 10 min at 4°C), washed twice with cold PBS, resuspended in 500 μl fixing solution [2% (v/v) formaldehyde and 0.05% (v/v) glutaraldehyde in PBS] and incubated for 20 min, as described before [72]. After fixation, parasites were washed twice and blocked in 200 μl PBS plus 2% (w/v) BSA (PBS-BSA) for 1h. Then, the cells were incubated with 200 μl monoclonal anti-EP procyclin solution (mAb 247 diluted 1:500 in PBS-BSA) for 2h [57]. After three washings with PBS, cells were incubated with a secondary antibody solution that contained goat anti-mouse IgG AlexaFluor-488 (Invitrogen) (1:1,000 in PBS-BSA) for 1h and were protected from light. Flow cytometry analysis was performed in a FACSCalibur flow cytometer (Becton Dickenson). FACS-acquired data were normalized using the unstained cells and only secondary antibodies provided as controls in the FlowJo v10 software (Tree Star, Inc.).
To analyze the mitochondrial functions in wt and RNAiTbP5CDH PCF cells, three parameters were taken into account: mitochondrial inner membrane potential (ΔΨm), control of respiration and total ATP levels. After culture, cells were prepared as follow: parasites were harvested by centrifugation (1,000 g for 7 min at RT°) and dissolved in buffer A with glucose (BAG: 116 mM NaCl, 5.4 mM KCl, 0.8 mM MgSO4, 50 mM HEPES-KOH, pH 7.2 and 5.5 mM D-glucose), as previously described [30]. Final densities were adjusted to 109 parasites/ml in BAG and kept on ice until further use. Parasite aliquots of 50 μl (5x107 cells) of each group were used for measurements. The ΔΨm determinations was made spectrofluorometrically in parasites dissolved in cell respiration medium (CRM: 125 mM sucrose, 65 mM KCl, 10 mM HEPES-KOH pH 7.2, 1 mM MgCl2, 2.5 mM potassium phosphate) supplemented with 5 mM L-proline, 10 μM EGTA, 20% (w/v) non-fatty acids BSA (NFA-BSA) (Sigma) and 10 μM of the safranin-o dye (Sigma), as previously described [73]. Changes in the fluorescence were recorded on a Hitachi 2500 spectrofluorometer (λexi496nm, λemi586 nm) at 28°C under constant stirring. Oxygen consumption was determined using a high-resolution oxygraph (O2k, OROBOROS Instruments, Innsbruck, Austria), under constant stirring in a 2.1 ml final volume at 28°C. The reaction buffer was supplemented with NFA-BSA and EGTA as mentioned above. Assays were initiated by adding 5x107 parasites to the oxygraph chamber. After adding the cells to the tightly closed oxygen-chamber, preparations were supplemented with 5 mM succinate or 5 mM proline, as indicated in each experiment. In order to measure parameters at mitochondrial levels, parasite suspensions were further permeabilized by adding 40 μM digitonin. Data were recorded using DatLab software (O2k, OROBOROS). In both measurements, additions of uncoupler or respiratory complex inhibitors were made as detailed in each experiment. ATP levels were determined using a luciferase bioluminescence assay (Sigma) according to the manufacturer’s indications. Briefly, the cells were harvested by centrifugation (2,000 g for 10 min at 4°C), washed twice with cold PBS and resuspended in the kit lysis buffer according to manufacturer´s instructions (Sigma). The intracellular ATP contents were extrapolated from a standard curve with known concentrations of ATP disodium salt. Results were obtained from four separate biological replicates (n = 3). Statistical analysis was performed using a one-way ANOVA test in GraphPad Prism v5.0a for Mac OS X (GraphPad Software, USA).
PCFs use of glucose and proline as carbon sources was evaluated by nuclear magnetic resonance (proton-NMR) for the excreted end-products. Wt and RNAiTbP5CDH (tet-/+) PCFs (106 parasites/ml) were grown in complete SDM79 medium for 72h. Then, parasites were harvested by centrifugation (1,300 g for 10 min at 4°C) and washed twice with PBS. Then, 5x108 parasites were transferred to 5 ml PBS supplemented or not with 4 mM L-proline + 4 mM D-[U-13C]-glucose. After 6h incubation at 26°C, cell suspensions were centrifuged and supernatants were submitted to NMR analysis, after adding 50 μl maleate (20 mM) as an internal reference to a 500 μl aliquot of the collected supernatant. 1H-NMR spectra were performed at 125.77 MHz on a Bruker DPX500 spectrometer equipped with a 5 mm broadband probe head. Measurements were recorded at 25°C with an ERETIC (Electronic REference To access In vivo Concentrations) method, which provides an electronically-synthesized reference signal. Acquisition conditions were as follows: 90° flip angle, 5000 Hz spectral width, 32 K memory size and 9.3 s total recycle time. Measurements were performed with 256 scans for a total time of almost 40 min. Before each experiment, the phase of the ERETIC peak was precisely adjusted. Protons linked to acetate carbon C2 generate by 1H-NMR five resonances, a single peak (unenriched acetate) flanked by two doublets ([13C]-acetate).
|
10.1371/journal.pgen.1003432 | Rod Monochromacy and the Coevolution of Cetacean Retinal Opsins | Cetaceans have a long history of commitment to a fully aquatic lifestyle that extends back to the Eocene. Extant species have evolved a spectacular array of adaptations in conjunction with their deployment into a diverse array of aquatic habitats. Sensory systems are among those that have experienced radical transformations in the evolutionary history of this clade. In the case of vision, previous studies have demonstrated important changes in the genes encoding rod opsin (RH1), short-wavelength sensitive opsin 1 (SWS1), and long-wavelength sensitive opsin (LWS) in selected cetaceans, but have not examined the full complement of opsin genes across the complete range of cetacean families. Here, we report protein-coding sequences for RH1 and both color opsin genes (SWS1, LWS) from representatives of all extant cetacean families. We examine competing hypotheses pertaining to the timing of blue shifts in RH1 relative to SWS1 inactivation in the early history of Cetacea, and we test the hypothesis that some cetaceans are rod monochomats. Molecular evolutionary analyses contradict the “coastal” hypothesis, wherein SWS1 was pseudogenized in the common ancestor of Cetacea, and instead suggest that RH1 was blue-shifted in the common ancestor of Cetacea before SWS1 was independently knocked out in baleen whales (Mysticeti) and in toothed whales (Odontoceti). Further, molecular evidence implies that LWS was inactivated convergently on at least five occasions in Cetacea: (1) Balaenidae (bowhead and right whales), (2) Balaenopteroidea (rorquals plus gray whale), (3) Mesoplodon bidens (Sowerby's beaked whale), (4) Physeter macrocephalus (giant sperm whale), and (5) Kogia breviceps (pygmy sperm whale). All of these cetaceans are known to dive to depths of at least 100 m where the underwater light field is dim and dominated by blue light. The knockout of both SWS1 and LWS in multiple cetacean lineages renders these taxa rod monochromats, a condition previously unknown among mammalian species.
| The emergence of Cetacea (whales, dolphins, porpoises) represents a profound transition in the history of life. Living cetaceans have evolved a spectacular array of adaptations in association with their return to aquatic habitats. Aquatic environments impose challenging constraints on sensory systems, including vision, and the cetacean eye exhibits both anatomical and molecular specializations that enhance underwater sight. Most mammals have one photopigment (RH1) for dim-light vision and two photopigments (long wavelength-sensitive opsin [LWS], short wavelength-sensitive opsin [SWS1]) for daytime, color vision. By contrast, cetaceans have an inactivated copy of the gene that encodes SWS1. Here, we show that LWS is also inactivated in several cetacean lineages including the giant sperm whale, Sowerby's beaked whale, and balaenopteroids (rorquals plus gray whale). These cetaceans dive to depths of at least 100 meters where the underwater light field is dominated by dim, blue light. The knockout of both cone pigments renders these taxa rod monochromats, a condition that is previously unknown among mammalian species. Rod opsin remains functional in these taxa and is blue-shifted to increase its sensitivity to the available blue light that occurs in deep water conditions. These results further elucidate the molecular blueprint of modern cetacean species.
| Cetacea [dolphins, porpoises, and whales] represents a remarkable example of aquatic specialization within Mammalia [1]. With their return to river and marine environments, the ancestors of modern toothed cetaceans (Odontoceti) and baleen whales (Mysticeti) underwent extensive modifications that included the evolution of novel structures [e.g., baleen plates, tail flukes], major anatomical rearrangements (e.g., telescoping of the skull, development of fore-flippers), the loss or reduction of typical mammalian traits (e.g., olfactory structures, hair, hindlimbs), and associated behavioral changes (echolocation, filter-feeding, deep-diving) [2]–[4]. At the genetic level this restructuring includes evidence of positive selection in loci related to high-frequency audition [5]–[7], brain size [8], [9], and flipper development [10], as well as degradation of genes related to olfaction [11]–[13], taste [14], tooth enamel formation [15]–[17], and vomeronasal chemoreception [18].
In the case of vision, aquatic environments impose challenging constraints, and the cetacean eye exhibits both morphological and molecular specializations that enhance underwater sight [19]. Possible morphological adaptations include an extensive reflective tapetum lucidum, a spherical lens with high refractive power, a relatively large cornea, and a rod-dominated retina, all of which enhance visual capabilities under dim light conditions [20], [21]. At the molecular level, most mammals have dichromatic color vision based on presence of three visual pigments, each of which is a G protein-coupled receptor that consists of an opsin protein moiety linked via a Schiff base to a retinal chromophore [22]. The three opsins that characterize most mammals include a rod opsin (RH1) and two cone opsins, short wavelength-sensitive opsin (SWS1) and long wavelength-sensitive opsin (LWS). Rods mainly function in dim light conditions (scotopic/night vision) whereas cones require more light (photopic vision) and are necessary with color vision. By contrast with most other mammals, all cetaceans that have been investigated are thought to be L-cone monochromats that possess an inactivated copy of SWS1 and two functional opsins, RH1 and LWS, which are expressed in rod and L-cone cells of the retina, respectively [19], [23], [24].
Griebel and Peichl [19] and Peichl [24] suggested that retinal S-cones, which express SWS1 and are sensitive to blue wavelengths, were lost during an early, coastal period of cetacean evolution. Near-shore waters commonly have an underwater light spectrum that is red shifted owing to the absorption of blue light by organic and inorganic debris, and the loss of ‘jobless’ S-cones may have constituted an economical advantage in this environment by simplifying retinal and cortical visual information processing [19]. There are no inactivating frameshift mutations in SWS1 that are shared by all odontocetes and mysticetes [23], but Griebel and Peichl [19] suggested that an unidentified genetic change, possibly in the promoter region, thwarted expression of the SWS1 protein in the common ancestor of crown Cetacea. Following the knockout of SWS1, crown cetacean lineages that independently conquered the open ocean were forced to shift λmax [the wavelength of maximal absorption] of RH1 and LWS to bluer wavelengths because SWS1 had previously been inactivated [19], [24]. By contrast, Bischoff et al. [25] offered an alternative scenario in which RH1 was blue shifted in the common ancestor of Cetacea. Specifically, Bischoff et al. [25] speculated that the ancestral cetacean RH1 possessed 83Asn, 292Ser, and 299Ala at three key tuning sites, as in the deep-diving giant sperm whale [Physeter macrocephalus], but stopped short of using explicit methods to reconstruct the ancestral RH1 sequence of Cetacea. If RH1 was blue-shifted in the common ancestor of Cetacea, then SWS1 may have been inactivated independently in mysticetes and odontocetes, perhaps due to the inefficiency at S-cones at photon capture in dim light conditions [22].
Another intriguing hypothesis posits rod monochromacy, as opposed to L-cone monochromacy, in at least some cetaceans. McFarland [20] suggested that some cetaceans are probably rod monochromats in which S-cones, L-cones, and their associated opsin genes [SWS1 and LWS, respectively] are lacking, so that vision is based entirely on rods and the rod opsin gene RH1. Immunocytochemical studies have failed to support this hypothesis and instead demonstrated the presence of L-cones in representative odontocete species belonging to the families Delphinidae and Phocoenidae [26]. More recently, Fasick et al. [21] reported the first partial L-cone opsin sequence (LWS) of a mysticete, Eubalaena glacialis (Atlantic right whale), and suggested that the L-cone opsin in this taxon is blue shifted, as are the L-cone opsins of representative odontocetes [27]. A potential shortcoming of this study is that Fasick et al. [21] only sequenced exons 3 and 5 of the E. glacialis LWS gene. In addition, LWS sequences have not been characterized from several other cetacean families including the deep-diving Ziphiidae, Physeteridae, and Kogiidae. Thus, McFarland's [20] suggestion that some cetaceans are rod monochromats remains to be tested by a more complete sampling of opsin gene sequences from a broader array of species.
Here, we report complete or nearly complete protein-coding sequences for all three opsin genes (RH1, SWS1, LWS) from representatives of all extant families of Cetacea and the cetacean sister group, Hippopotamidae. Previous studies have characterized the evolutionary patterns of individual cetacean opsins in isolation, but have not yet integrated information from all three retinal opsin genes (RH1, LWS, SWS1) into a single, comprehensive analysis. We utilized selection intensity estimates, ancestral sequence reconstructions, shifts in spectral tuning, and shared missense/frameshift mutations to infer the complex history of opsin evolution in Cetacea. Our reconstructions suggest that RH1 was blue-shifted in the common ancestor of Cetacea prior to the independent inactivation of SWS1 on the stem mysticete and odontocete branches. LWS, in turn, was pseudogenized convergently in five different cetacean lineages [right whale plus bowhead, rorquals plus gray whale, Sowerby's beaked whale, giant sperm whale, pygmy sperm whale], all of which are deep divers that feed on bioluminescent organisms. The tandem inactivation of SWS1 and LWS in these taxa presumably renders them rod monochromats, a condition that was previously unknown within Mammalia.
Maximum likelihood trees based on SWS1 exons plus introns, SWS1 exons, RH1 exons, and LWS exons are shown in Figures S1, S2, S3, S4. With a few exceptions, clades with high bootstrap support percentages (>90) on individual gene trees are in agreement with the species tree in Figure 1. All of the gene trees recovered Cetancodonta [Cetacea + Hippopotamidae], Cetacea, Mysticeti, Balaenidae, Balaenopteroidea, Physeteroidea, Ziphiidae, Iniidae + Pontoporiidae, Phocoenidae, Delphinidae, Delphinoidea, and Iniidae + Pontoporiidae + Delphinoidea [Delphinida]. Odontoceti was only recovered in the SWS1 analyses, but conflicting nodes in the RH1 and LWS trees had low bootstrap support values (≤53%).
Inactivating mutations (frameshift indels, premature stop codons, disrupted intron splice sites, amino acid replacement at the Schiff's base counterion site) were apparent for all cetacean species in the SWS1 alignment (Table 1), but were lacking in SWS1 from the semiaquatic outgroup species, Hippopotamus amphibius (Figure 1). Although the SWS1 genes of all cetacean species show evidence of mutational decay, no inactivating mutations map to the last common ancestral branch of Cetacea (Figure 1, node 26 to node 27). Instead, different molecular lesions define various sublineages of Cetacea (Table 1). An amino acid replacement (E113G; bovine RH1 numbering) at the Schiff's base counterion site that is thought to disrupt opsin-chromophore binding [23] optimizes to the stem branch of Odontoceti (Figure 1; node 27 to node 35), and a four base-pair frameshift deletion was derived on the stem branch of Mysticeti (Figure 1; node 27 to node 28). These independent inactivating mutations imply that SWS1 was pseudogenized convergently in the two major subclades of Cetacea (Figure 1, Figure S5).
Estimates of ω (dN/dS) on different branches of the cetacean tree are consistent with parallel knockouts of SWS1 in Odontoceti and in Mysticeti. The ω estimate for SWS1 on the stem Cetacea branch (Figure 1, node 26 to node 27), just prior to the two inferred inactivating mutations, suggests a pattern of purifying selection based on analyses with two different codon frequency models (ω = 0.31, 0.35). Likewise, a signature of strong purifying selection (ω = 0.16, 0.17) was inferred on the stem Odontoceti branch (node 27 to node 35) (Table 2). Neutrality is predicted on the stem odontocete branch if SWS1 had previously been inactivated on the stem cetacean branch [19], but statistical tests rejected this hypothesis (Table 2). The ω estimate (0.75, 0.83) for the stem mysticete branch (node 27 to node 28), in turn, is only slightly lower than expected for complete neutrality (ω = 1.0) and suggests that pseudogenization occurred very early on this branch. Estimates of ω for crown odontocete + crown mysticete branches are in agreement with expectations for neutrality (Table 2), and in conjunction with numerous frameshift indels within these clades (Table 1) imply a release from selective constraints after the occurrence of inactivating mutations on the stem odontocete and stem mysticete branches (Figure 1).
No inactivating mutations (frameshift indels, splice site mutations) were apparent in the RH1 alignment, implying that RH1 is functional in all of the species that were surveyed (Figure 1).
Ancestral amino acid sequences at key tuning sites (83, 292, 299) in Cetacea [21], [25] are shown in Table 3 for internal nodes with inferred blue or red shifts in λmax. Amino acid changes from DAS to NSS on the stem cetacean branch [node 26 to node 27] resulted in an inferred blue shift from 501 to 484 nm. Additional blue shifts (484 to 479 nm) are inferred in Caperea, in stem Physeteroidea (node 35 to node 36), and in stem Ziphiidae (node 38 to node 39) based on an amino acid changes at site 299 (NSS to NSA) that occurred independently in these three lineages. Seven red shifts were reconstructed in Cetacea, including three in Mysticeti (stem Balaenidae [node 28 to node 29], Megaptera, Eschrichtius) and four in Odontoceti (stem Iniidae + Pontoporiidae [node 42 to node 43], Pontoporia, stem Monodontoidea [node 44 to node 45], Delphinapterus) (Table 3).
Among ten other amino acid sites that have been linked to spectral tuning in vertebrates [28], eight (sites 96, 102, 122, 183, 253, 261, 289, 317) are invariant among the cetaceans and the hippopotamid that were included in our taxon sampling, site 194 exhibits four amino acid replacements within Cetacea, and site 195 shows an amino acid replacement [L to P] on the stem Cetacea branch and four replacements within Cetacea.
Analyses with Codeml rejected site models 2 and 8, which add an extra category for positively selected sites, in favor of models 1 and 8a, respectively. By contrast, branch-site analyses with two different codon frequency models (CF) provided statistically significant support for a bin of five positively selected sites (7, 83, 123, 266, 292) on branches with λmax changes (CF2: P = 0.00036, ω = 5.53; CF3: P = 0.00018, ω = 6.43). Three of the five positively selected sites (83, 266, 292) have probabilities >0.95 of membership in this bin.
Inactivating mutations are apparent in LWS sequences from ten cetacean species (Figure 1, Figure S6, Table 1). Reconstructions of ancestral sequences imply eight frameshift indels and three splice site disruptions within Cetacea, with convergent inactivation of LWS on the following five branches (Figure 1): Physeter macrocephalus, Kogia breviceps, Mesoplodon bidens, stem Balaenidae (node 28 to node 29), and stem Balaenopteroidea (node 30 to node 31). All of these separate knockouts of LWS postdate prior inactivations of SWS1 and therefore result in rod monochromacy (Figure 1).
Estimates of ω throughout the species tree generally are consistent with multiple, independent knockouts of LWS within Cetacea. Branches reconstructed as functional for LWS exhibit a strong signature of purifying selection (ω = 0.09, 0.10). By contrast, ω estimates on “transitional” branches [16], where inactivating mutations in LWS were reconstructed, generally show elevated rates of nonsynonymous substitution (Physeter: ω = 0.38, 0.41, Kogia: ω = 0.73, 0.78, stem balaenopteroid branch: ω = 0.34, 0.37). Exceptions are the short transitional branches for stem Balaenidae (1.5 to 1.7 inferred substitutions, ω = 0.0001) and Mesoplodon (2.4 to 3.1 inferred substitutions, ω = 0.13, 0.19). Branches within crown Balaenopteroidea (node 31 and descendant branches) plus crown Balaenidae (node 29 and descendant branches), which are interpreted as pseudogenic based on the prior occurrence of inactivating mutations, have an ω value based on two codon models (0.69, 0.70) that does not deviate significantly from neutral expectations (ω = 1.00) based on χ2-tests.
Reconstructions of ancestral amino acid sequences at five key tuning sites (amino acids 180, 197, 277, 285, 308) [21], [25] are shown in Table 3 for branches with inferred shifts in λmax. For the five tuning sites, AHYTA (λmax = 552 nm) is the inferred ancestral condition for Cetancodonta (node 26) and for the last common ancestor of extant cetaceans (node 27). Three changes at LWS tuning sites were reconstructed within Cetacea. Parallel changes from AHYTA to AHYTS on the stem Mysticeti branch [node 27 to node 28] and on the stem Delphinoidea branch (node 42 to node 44) imply blue shifts from 552 nm to 522–531 nm. A change from AHYTA to AHYTP was reconstructed on the terminal Inia branch, but the functional effect of A308P is unknown (Table 3).
Analyses with Codeml rejected site models 2 and 8, which add an extra category for positively selected sites, in favor of models 1 and 8a, respectively. Similarly, positively selected sites were not identified in branch-site analyses.
Here, we assembled complete or nearly complete protein-coding sequences for RH1, SWS1, and LWS for representatives of all extant cetacean families. These sequences, in combination with molecular evolutionary analyses, permit a detailed, synthetic reconstruction of opsin evolution in Cetacea (Figure 1).
Recent phylogenetic hypotheses imply that the aquatic ancestry of Cetacea extends back to its last common ancestor with the semi-aquatic Hippopotamidae in the early Eocene, >50 Ma [4], [29], [30]. Whales and hippos share a variety of “aquatic” specializations including sparse hair, loss of sebaceous glands, and the ability to birth and nurse underwater [4], [31], [32], but these features traditionally have been interpreted as parallel evolutionary derivations in these two lineages. Given the hypothesis that the common ancestor of cetaceans and hippos was aquatic/semi-aquatic (Figure 1, node 50 to node 26), shared mutations in opsin genes that enhance vision in aquatic environments might be expected in whales and hippos. ML reconstructions of ancestral opsin sequences imply only two replacement substitutions (LWS: E41D; RH1: L216M) on the stem lineage. These replacements are not at key tuning sites and fail to provide compelling evidence for an aquatic shift in opsin properties in the common ancestor of hippos and whales.
Following divergence from Hippopotamidae, the unique evolutionary history of Cetacea began on the stem cetacean branch (Figure 1, node 26 to node 27). The fossil record indicates that the stem cetacean lineage was marked by a profound transition in anatomy from primitive semi-aquatic forms to obligately aquatic taxa with vestigial hindlimbs [3], [33]–[35]. Ancestral reconstructions imply that stem cetaceans retained dichromatic color vision with functional SWS1, LWS, and RH1 as in Hippopotamus and more distantly related artiodactyls; a blue shift in RH1 also occurred on the stem cetacean branch (Figure 1). Specifically, the amino acid array at three key tuning sites (83, 292, 299) [21], [25] changed from DAS to NSS, with an inferred λmax shift from 501 to 484 nm. Our ML reconstruction supports Bischoff et al.'s [25] hypothesis that RH1 was blue shifted on the stem cetacean branch, but contradicts their assertion that the ancestral cetacean expressed the amino acids NSA as in deep-diving physeteroids.
In addition to replacements at sites 83 and 292, a change at tuning site 195 (K to T) of RH1 occurred on the stem cetacean branch. This change from a polar amino acid to a positively charged residue has been retained in the deep-diving physeteroids (giant sperm whale, pygmy sperm whale). The inferred shift in λmax that results from a K to T replacement at this site, if any, remains to be investigated with mutagenesis studies. Unlike tuning sites 83, 292, and 299, that are situated in transmembrane regions of RH1 and are in close proximity to the chromophore, site 195 is positioned in the luminal face of RH1 [36]. The nature of long distance interactions between this amino acid site and the chromophore are unknown [36].
The basal split in Cetacea defines the separation of Odontoceti from Mysticeti, and also marks the evolution of profound changes in anatomy/feeding strategy in both clades [4], [37]. Echolocation capabilities and degradation of olfactory structures were derived on the stem odontocete branch (Figure 1, node 27 to node 35), whereas the transition to bulk filter feeding with a keratinous baleen sieve evolved on the stem mysticete branch (Figure 1, node 27 to node 28). These divergent specializations represent changes in feeding style that would be expected to impact demands on visual systems.
Following the blue shift in RH1 on the stem cetacean branch, SWS1 was inactivated independently in stem odontocetes and in stem mysticetes, coincident with the evolution of divergent specializations in these two clades (Figure 1). Two lines of evidence support this reconstruction and argue against an earlier knockout of SWS1 in the common ancestor of Cetacea. First, comprehensive sequencing of SWS1 exons and introns revealed no shared inactivating mutations common to all extant cetaceans. Odontocetes have a common missense mutation at the Schiff's base counterion site (E113G) that disrupts opsin-chromophore binding [23]. Mysticetes, in turn, share a 4-bp frameshift mutation in exon 1 of SWS1 that results in a premature stop codon. Frameshift indels in the same position occur in several odontocetes, but these deletions are most parsimoniously reconstructed as convergent between Mysticeti and multiple odontocete subclades (Text S1). Several mutations that disrupt intron boundaries were identified, but in all cases these substitutions map to branches within Odontoceti or within Mysticeti. Second, dN/dS values on the stem odontocete and stem mysticete branches should indicate an absence of selective constraints if SWS1 was inactivated earlier in the common ancestor of Cetacea. Estimates of dN/dS (0.75, 0.83) for the stem mysticete branch are consistent with neutral evolution (dN/dS = 1.00), but neutrality was rejected given the low dN/dS estimates (0.16, 0.17) for the stem odontocete branch, indicative of purifying selection and thus functionality after the split between Odontoceti and Mysticeti (Figure 1, node 27; Table 2).
Together, our reconstructions for the evolution of RH1 and SWS1 contradict the coastal knockout hypothesis [19], [24]. This scenario postulates that SWS1 was inactivated during an early amphibious phase of cetacean history when semi-aquatic whales occupied coastal waters that absorbed blue light, and that RH1 was subsequently blue shifted in crown cetaceans that moved to open ocean environments dominated by blue light. The coastal knockout hypothesis requires an as yet undiscovered inactivating mutation in SWS1, perhaps in the promoter region of this gene or at splice sites [23]. Instead, our results fit the hypothesis that RH1 was blue shifted in the common ancestor of Cetacea, and that SWS1 was convergently knocked out in Odontoceti and in Mysticeti after cetaceans had invaded open ocean habitats. This is perhaps surprising given that SWS1 is well suited to detect the blue light that dominates the open ocean. However, the relative scarcity of S cones in the mammalian retina, which diminishes the efficiency of photon capture under dim light conditions, may have predisposed S-cones to eventual loss through relaxed selection [22]. By contrast, rods are much more efficient at photon capture under dim light conditions because of their higher density in the mammalian retina and their integration with large, sparsely distributed ganglion cells that sum photon detection over huge receptive fields [38]. The preferential retention of a functional copy of LWS rather than SWS1 in some cetaceans may reflect the higher density of L-cones and their greater impact on visual acuity [22].
In addition to the blue shift of RH1 in the ancestral cetacean branch, several amino acid replacements in RH1 imply further adjustments in λmax (Figure 1). These shifts are generally consistent with the photic environments that are occupied by different cetacean species [20], [25]. Among these changes are blue shifts in deep-diving physeteroids [sperm whales] and in ziphiids [beaked whales], a red shift in the common ancestor of Inia and Pontoporia, both of which are found in shallow water environments, and red shifts in several mysticetes [e.g., Eschrichtius, Megaptera]. Bischoff et al. [25] suggested that the red-shifted pigments that occur in some mysticetes are better adapted to relatively shallower foraging environments than the ancestral mysticete pigment. The blue shifts in Physeteroidea and Ziphiidae occured in parallel and in both cases involve amino acid replacements [serine to alanine] at tuning site 299 [21], [25]. Sperm whales and beaked whales rank among the deepest diving mammals and specialize on a cephalopod-rich diet [39]. Several phylogenetic studies of anatomical evidence grouped these suction feeding species, presumably based on convergent character states related to their deep diving habits [40], [41], but most recent work indicates that ziphiids are more closely related to dolphins and porpoises than to physeteroids [37], [42], [43].
Nozawa et al. [44] suggested that Yang's [45] codeml program is not useful for identifying adaptive sites in visual pigments. Our results support Nozawa et al.'s [44] finding that site analyses fail to identify adaptive changes in visual pigments. However, branch-site tests identified five codons in RH1 that have evolved under positive selection on branches with inferred changes in λmax. The ω value for the five positively selected sites is well above one (5.53–6.43), and supports the hypothesis that changes affecting λmax in cetacean RH1 proteins are adaptive. The failure of site analyses to detect positively selected sites in RH1 may be a consequence of mixing positive selection on foreground branches with purifying selection on background branches. Nozawa et al. [44] criticized branch-site tests [45]–[47] for their proclivity to generate false positive results based on simulations, but Yang et al. [48] correctly noted that false positives only occurred in 32/14,000 cases, which is much lower than the nominal significance level (5%) and demonstrates that the branch-site test is conservative. Among the positively selected sites, two (83, 292) are known tuning sites that in part are the basis for inferring changes in λmax (Figure 1). Changes at site 83 may also be important in dim-light conditions because the amino acid at this position affects the rate at which photoreceptor cells generate electrical signals [49]. The other three sites (7, 123, 266) have not been predicted to affect λmax. Site 7 occurs in the extracellular domain, site 123 occurs in transmembrane helix III, and site 266 occurs in transmembrane helix 6 [50]. The functional consequences of mutations at these amino acid positions in cetacean RH1 sequences remain unknown, although conformational changes associated with transmembrane domains III and VI of G protein-coupled receptors may be important in receptor activation [51].
Changes in LWS spectral sensitivity coincide with deployment of cetaceans to diverse aquatic habitats (Figure 1). A blue shift in LWS in stem mysticetes co-occurs with an SWS1 frameshift mutation on the same branch, although the sequence of these events is unclear. An additional LWS blue shift in λmax maps to the common ancestor of Delphinoidea [dolphins, porpoises, beluga], but the most striking feature of LWS evolution in Cetacea is the convergent knockout of this gene in five different lineages: Balaenopteroidea (rorquals and gray whale), Balaenidae (bowhead and right whale), Mesoplodon bidens (Sowerby's beaked whale), Physeter macrocephalus (giant sperm whale), and Kogia breviceps (pygmy sperm whale) (Figure 1). Given that SWS1 is also debilitated in each of these species (Figure 1), the genetic data imply that these taxa are rod monochromats. This iterated degeneration of cetacean LWS was not apparent in earlier studies because complete protein-coding LWS sequences had been generated for only a few cetacean species [21].
Historically, the pure rod retina has been proposed as the “extreme” adaptation to low light levels [52]. Walls [52] and McFarland [20] suggested the possibility of rod monochromacy in at least some cetaceans. More generally, early studies on retinal anatomy hinted at this condition in a variety of nocturnal and aquatic mammalian species with rod dominated retinas, including night monkeys, lemurs, tarsiers, chinchillas, seals, and bats [52]–[57]. Recent work has shown that representative cetaceans are instead L-cone monochromats and retain a functional copy of LWS [21], [26]. Similarly, primates, rodents, pinnipeds, and bats that were previously hypothesized to be rod monochromats are now known to be L-cone monochromats with functional LWS or even cone dichromats with functional LWS and SWS1 [22], [58]–[61]. The present survey of cetacean opsins, which documents pseudogenization of both SWS1 and LWS in multiple cetacean lineages, vindicates McFarland's [20] hypothesis that some cetaceans are rod monochromats (Figure 1). To our knowledge these are the only known examples of rod monochromacy in Mammalia or even Amniota. The observation that five independent derivations of mammalian rod monochromacy are all clustered within Cetacea is striking, and suggests that one or more features of cetacean biology have been pivotal in driving the degenerative pattern of opsin evolution in this aquatic clade.
The naked mole rat (Heterocephalus glaber) is the only other mammal, aside from the cetacean species characterized here, that is known to lack a functional copy of LWS, but H. glaber retains an intact SWS1 and is interpreted as a cone monochromat [62]. This condition contrasts with other mammalian cone monochromats [pinnipeds, dolphins, porpoises, some procyonids, some rodents, some bats] that combine a pseudogenic SWS1 with a functional copy of LWS [22], [23], [28], [60], [63]–[71]. It has been suggested that cetacean cone monochromats [e.g. bottlenose dolphin, Tursiops truncatus] can distinguish colors, possibly via interactions between LWS and RH1 [38], [72], but any vestiges of color vision presumably have been lost in the various rod monochromatic cetacean species documented here (Figure 1).
Among other vertebrates, rod monochromatic taxa are rare and to our knowledge have only been documented in bony and cartilaginous fishes [73]–[81], caecilians [22], [82], and the cave salamander Proteus anguinus [83], although presumed rod monochromacy based entirely on immunocytochemistry, microscopy or spectral analysis does not preclude the possibility that other minor visual pigment classes exist [77], [81], [83]. Most of the rod monochromatic fish species inhabit the deep sea or are nocturnal; caecilians are generally fossorial and/or nocturnal with poorly developed eyes; and the cave salamander Proteus lives in a virtually light-free environment.
The phylogenetic evidence for multiple, independent knockouts of both SWS1 and LWS within Cetacea raises the question of why convergent pseudogenization and rod monochromacy evolved in this clade but not in other mammalian groups. All rod monochromatic cetacean species that were genetically characterized in our survey are capable of diving to depths that exceed 100 m, with sperm and beaked whales ranking among the deepest diving mammals [39], [84]–[88]. The selective pressures on mammalian retinal opsins in deep-water habitats are drastically different from those on land. In the open marine environment, the electromagnetic radiation of visible light is weakened with depth due to absorption and scattering [89]. In the mesopelagic zone (150–1000 m), down-welling sunlight becomes more monochromatic and the spectrum shifts towards shorter, bluer wavelengths with depth [81], [82]. Below 1000 m (bathypelagic zone), there is no down-welling sunlight and localized bluish bioluminescence becomes the predominant source of light [90].
A rod-dominated retina is advantageous in dim light conditions [38]. Therefore, the convergent pseudogenization of SWS1 and LWS in multiple cetacean lineages may be an adaptation to deep-water habitats and/or feeding at night on bioluminescent invertebrate prey. Cone opsins have a higher rate of thermal activation [i.e., dark noise] than RH1 [91] and may interfere with rod sensitivity under scotopic conditions. Thus, combined SWS1 and LWS pseudogenization may have increased RH1 sensitivity in physeteroids and ziphiids that feed in the mesopelagic and bathypelagic zones. Echolocation is a key specialization that has enabled odontocete taxa such as these to forage at night and at great depths on individual prey items, in particular cephalopods [92]; rod monochromatism may be an additional adaptive feature that has enabled predation at depth. Balaenopteroid and balaenid mysticetes are not known to feed in the bathypelagic zone, do not echolocate, and instead batch filter aggregations of small prey items. However, baleen whales do feed at night and much of their diet is composed of bioluminescent prey including krill [93], [94]. The ability to take advantage of this huge resource offers a compelling selective driver on the evolution of visual systems in Mysticeti, and the detection of schools of tiny prey at night would seem to be problematic without echolocation. The reliance of various mysticete species on RH1 might represent one solution for improved night vision given that rods are more useful than cones for contrast detection and hence picking out schools of prey from the background. Along these lines, the parallel pseudogenization of both cone opsins in Cetacea (Figure 1) could be the result of natural selection favoring an all-rod retina, in which case cone opsins were either selected against because of interference with RH1, or were rendered ‘jobless’ by the elimination of cones and released from selective constraints on color vision in this aquatic clade [22].
The emergence of Cetacea represents a profound macroevolutionary transition that entails comprehensive remodeling at both the genetic and morphological levels [4]. Our results elucidate key events in the evolutionary history of cetacean opsins, including an initial blue shift of RH1 in stem Cetacea, parallel knockouts of SWS1 in Odontoceti and Mysticeti, and five independent inactivations of LWS in deep-diving cetacean lineages. As correctly surmised by McFarland [20], some cetaceans are rod monochromats and have evolved eyes that are highly specialized for dim-light vision.
Previously published RH1, SWS1, and LWS sequences for Cetacea were combined with new sequences that were generated through PCR and dideoxy sequencing. We targeted complete coding regions of all three opsin genes for representatives of Hippopotamidae and all extant cetacean families (Text S2). RH1, LWS, and SWS1 sequences for additional cetartiodactyl families [Bovidae, Cervidae, Suidae, Camelidae] were assembled from Ensembl, Pre-Ensembl, and NCBI based on availability with minor augmentation by new sequences (Table S1, Text S2).
Aligned sequences for Bos taurus, Sus scrofa, Tursiops truncatus, Vicugna pacos, and available GenBank sequences (Table S1) were used to design PCR primers for SWS1, RH1, and LWS. SWS1 (exons 1–4; partial exon 5; introns 1–4) was amplified in five overlapping segments. PCR primers for RH1 (exons 1–5) and LWS (exons 1–6) were positioned in the flanking intronic regions of each exon (see Text S2 for additional details on PCR reactions). Accession numbers for new cetartiodactyl sequences are KC676796–KC677023 (Table S1). Primer sequences are provided in Table S2.
Sequences were aligned manually using Se-Al [95]. The virtual mRNA alignment lengths were 1014 base pairs (bp) for SWS1, 1092 bp for LWS, and 1044 bp for RH1. The complete alignment for SWS1, including exons and introns, was 4163 bp. All alignments for phylogenetic and PAML analyses, along with alignments for non-overlapping PCR amplicons (exons plus partial introns for LWS and RH1), are provided in Text S3 in nexus format. Phylogenetic analyses were performed with RAxML 7.2.7 [96] and the GTR + Γ model of sequence evolution. Additional details are provided in Text S2.
Opsin alignments were manually inspected for putative inactivating mutations, including substitutions that result in stop codons, changes at intron splice donor/acceptor sites, and frameshift indels. We also examined SWS1 sequences for a missense mutation at Schiff's counterion site (E113G; bovine RH1 numbering) that disrupts opsin-chromophore binding [23].
Ancestral DNA sequences for SWS1, LWS, and RH1 were reconstructed with the Baseml program implemented in PAML 4.4b [45]. We used the REV model and a composite species tree based on McGowen et al. [42] for cetaceans and Gatesy [97] for all other cetartiodactyls. Frameshift mutations and other indels were optimized with Fitch parsimony, as implemented in Mesquite [98].
Spectral tuning in RH1 is influenced by at least 13 amino acid sites [28], although replacements at only three of these sites (83, 292, 299) fully explain the absorbance difference between cow RH1 (Bos taurus, λmax = 500 nm) and bottlenose dolphin RH1 (Tursiops truncatus, λmax = 488 nm) [99]. These replacements are D83N, A292S, and A299S. Different combinations of ancestral and derived amino acids at these three sites have been tested in mutagenic studies of Bos RH1 to explain the various λmax values that occur in other cetaceans [21], [25]. For LWS, Yokoyama [36] suggested a “five-sites” rule whereby λmax values between 510 and 560 in vertebrates can be fully explained by amino acid changes S180A, H197Y, Y277F, T285A, A308S and their interactions. Here, we follow Fasick et al. [21] and Bischoff et al. [25] and provide λmax estimates for newly determined RH1 and LWS sequences based on directly determined λmax values from expressed RH1 and LWS pigments that possess identical amino acids at the same key sites for each of these opsins. It will be important in future studies to perform direct measurements of λmax on reconstructions of ancestral RH1 sequences. Even without these experiments, empirical measurements on a diverse array of opsins from cetacean species and Bos taurus (wild type and mutagenesis variants) provide a strong foundation for inferring λmax values in ancestral cetacean sequences [21], [25].
The Codeml program in PAML 4.4b [45] was used to estimate the ratio (ω) of the non-synonymous substitution rate [dN] to the synonymous substitution rate (dS) at individual sites (RH1, LWS) and on branches (SWS1, LWS). We also performed branch-site analyses [45]–[47], [100] on RH1 and LWS sequences. In both cases, branches with predicted changes in λmax (Figure 1) of the relevant opsin were assigned to the foreground, and all other branches were assigned to the background. We used a composite species tree for all cetartiodactyl taxa as detailed above. Statistical tests of neutrality [complete absence of functional constraints] for branches and sets of branches were executed as in Meredith et al. [16]. See Text S2 for details.
|
10.1371/journal.pntd.0007164 | A multi-country study of the economic burden of dengue fever based on patient-specific field surveys in Burkina Faso, Kenya, and Cambodia | Dengue fever is a rapidly growing public health problem in many parts of the tropics and sub-tropics in the world. While there are existing studies on the economic burden of dengue fever in some of dengue-endemic countries, cost components are often not standardized, making cross-country comparisons challenging. Furthermore, no such studies have been available in Africa.
A patient-specific survey questionnaire was developed and applied in Burkina Faso, Kenya, and Cambodia in a standardized format. Multiple interviews were carried out in order to capture the entire cost incurred during the period of dengue illness. Both private (patient’s out-of-pocket) and public (non-private) expenditure were accessed to understand how the economic burden of dengue is distributed between private and non-private payers.
A substantial number of dengue-confirmed patients were identified in all three countries: 414 in Burkina Faso, 149 in Kenya, and 254 in Cambodia. The average cost of illness for dengue fever was $26 (95% CI $23-$29) and $134 (95% CI $119-$152) per inpatient in Burkina Faso and Cambodia, respectively. In the case of outpatients, the average economic burden per episode was $13 (95% CI $23-$29) in Burkina Faso and $23 (95% CI $19-$28) in Kenya. Compared to Cambodia, public contributions were trivial in Burkina Faso and Kenya, reflecting that a majority of medical costs had to be directly borne by patients in the two countries.
The cost of illness for dengue fever is significant in the three countries. In particular, the current study sheds light on the potential economic burden of the disease in Burkina Faso and Kenya where existing evidence is sparse in the context of dengue fever, and underscores the need to achieve Universal Health Coverage. Given the availability of the current (CYD-TDV) and second-generation dengue vaccines in the near future, our study outcomes can be used to guide decision makers in setting health policy priorities.
| Dengue fever is a major public health concern in many parts of South-East Asia and South America. In addition to countries where dengue has been highly prevalent for many years, there is a growing concern on the undocumented burden of dengue in Africa. Following the successful execution of the first-round economic burden study in Vietnam, Thailand, and Colombia by the Dengue Vaccine Initiative, the second-round economic burden study was implemented in Burkina Faso, Kenya and Cambodia using the same standardized methodology. In particular, the second-round study targeted GAVI eligible countries for future vaccine introductions and included two African countries where the burden of dengue was relatively unknown. Our study outcomes show that the economic burden of dengue fever is significant in all three countries. The dengue vaccination era began in 2016 with the first dengue vaccine (CYD-TDV) although its public use should be carefully determined due to the safety concerns related to the vaccine. Considering that there are other second-generation dengue vaccines in development, the current study outcomes provide an important step to estimate the economic benefits of vaccination in the three countries.
| Dengue fever is a vector-borne disease and transmitted by Aedes mosquitoes [1–3]. Dengue immunity and population biology are complex [4,5]. There are four serotypes, which are antigenically distinct viruses but interact with each other. It is known that infection with one serotype provides life-long protection against that specific serotype, but a subsequent heterotypic infection may lead to favorable (short-term cross protection) or detrimental (the development of more severe illness) outcomes due to a high degree of antigenic cross-reactivity [5–7]. Despite continuous efforts to disentangle the complexity of the disease, it is still not clear how all four serotypes interact with each other in terms of cross protection, antibody dependent enhancement (ADE), and the duration of the serotype interactions [8,9].
The complex nature of the disease also imposes difficulties on the development of safe and effective dengue vaccines. A first live-attenuated, tetravalent dengue vaccine (Dengvaxia, CYD-TDV) became commercially available in 2016, but the safety concerns related to the vaccine have created a wide range of controversial debates [10–13]. Such challenges for developing safe and effective dengue vaccines have been a part of the reasons why there has been relatively less attention paid to the health-economic aspect of the disease.
As previously mentioned by Lee et al. [14], a relatively small number of empirical economic burden studies of dengue are available. Some of the existing studies relied on secondary data sources and extrapolated to other countries given a lack of field-based datasets. While Suaya et al. and Lee et al. conducted the economic burden study of dengue fever in a multi-country setting based on primary data sources using standardized methods [14,15], many of other studies applied different study designs and methodologies, making it difficult to make proper comparisons across countries. There is no doubt that all of the existing studies have contributed to informing the importance of the economic burden of dengue fever, but it is also true that more field-based studies with standardized methods are essential to better understand the economic burden of dengue fever in many of known and unknown dengue-endemic countries.
In order to fill the existing knowledge gaps, the Dengue Vaccine Initiative (DVI) implemented multi-disciplinary cost-of-illness (COI) field surveys in six countries in collaboration with research partners. As a first round of the project, the economic burden study had been carried out in Vietnam, Thailand, and Colombia from 2012 to 2015, and the final study outcome was recently published [14]. Following the successful execution of the first-round surveys, the DVI expanded the COI field studies to three additional countries: Burkina Faso, Kenya, and Cambodia. Considering that the first dengue vaccine was about to be available when the second-round countries were being selected, GAVI eligibility was also taken into account for vaccination in the future. In particular, the second-round field surveys included two countries in Africa where dengue burden is relatively unknown compared to other tropical and sub-tropical countries in South Asia and Latin America.
Understanding the accurate economic burden of a disease is one of the important steps to grasp a full scope of vaccination benefits from the societal perspective. As previously shown, the range of the total COI for dengue fever plays a critical role in determining the threshold costs for which dengue vaccination would be effective [16]. Considering that there are several second-generation vaccine candidates which are currently in phase 3 trials, the current economic burden study would contribute to filling the knowledge gaps in Burkina Faso, Kenya, and Cambodia where healthcare resources are limited, and healthcare budget may be highly constrained considering several competing health problems.
Ideally, the COI study would be conducted in an area where dengue transmission is prevalent. This may not be an issue for many of South Asian countries such as Cambodia where the prevalence of dengue fever has been known for many years. However, appropriate site selection was a challenge in Burkina Faso and Kenya because there was a lack of information regarding dengue fever during the initial phase of the study. Thus, study sites in these two countries were selected based on their likelihood of supporting dengue transmission using outbreaks and case reports in the literature in Africa. The full description is available in the study by Lim et al. [17].
Table 1 shows information on study sites. Similar to the first round COI study, the current COI study was embedded into ongoing dengue epidemiological field studies. In Burkina Faso, five centers for health and social advancement (CSPS) located in Ouagadougou were selected. The CSPS are the first-level health system facilities which provide basic healthcare and medical resources for local populations. The CSPS consists of three units: Expanded Program of Immunization (EPI), gynecology and obstetrics, and general medicine. The facility has examination rooms, inpatient wards (where patients can stay up to three days), and staff offices. In Kenya where health facilities are categorized into six levels, three health facilities were selected in Mombasa: Ganjoni, Coast Provincial General Hospital (CPGH), and Tudor. Ganjoni is a community-level health service provider (level 1) and mainly provides services for outpatients with limited health services. Tudor sub-county hospital is a district-level healthcare provider (level 3). The facility has 14 beds and provides various services including inpatient department care, family planning, anti-retroviral therapy, as well as home-based care. CPGH is the second largest governmental hospital and serves as a tertiary referral center (level 5). The hospital has about 700 beds with approximately 800 staff members. In Cambodia, the COI study was implemented in two provincial-level and two district-level health facilities in three provinces: Kampong-Cham, Tbong-Khmum, and Kampot. The district- and provincial-level health facilities in Kampong-Cham have 30 beds with 32 staff and 260 beds with approximately 250 health workers, respectively. The health facility in Tbong-Khmum, which was separated out from Kampong Cham province, is a district-level facility with 90 beds and 59 staff. Another provincial-level facility in Kampot province is staffed with 95 health workers and maintains 155 beds.
As one of the main goals of the DVI study was to estimate the economic burden of dengue fever by using a standardized method across the sites, the current COI surveys were implemented following the similar methodologies applied to the first-round countries. The detailed study design and overall structure were fully described by Lee et al. [14].
Briefly, patients experiencing fever for less than 7 days were recruited for the fever surveillance, and rapid tests (NS1, IgM/IgG) were implemented. Due to the slow caseload during the first-half of the study period in Cambodia, the polymerase chain reaction (PCR) was additionally used to meet the desired sample size. Among those who were positive on any of the test results and consented to the study, the economic burden survey was carried out. Multiple interviews were conducted up to three times depending upon the duration of illness in order to capture the entire costs of the current dengue illness: costs spent before the study enrollment visit, during the study enrollment visit, and after the enrollment visit (S1 Text). The target sample size was estimated to be approximately 150 patients for each study site. The simple random sampling method was applied [18], where an acceptable difference between a true population and a sample estimate was assumed to be 0.2 with the 95% confidence interval statistic (= 1.96). The coefficient of variation was obtained from an existing multi-country study [15].
The COI survey included three major cost components: direct medical costs (DMC), direct non-medical costs (DNMC), and indirect costs (IC). DMC consists of consultation fees, medication, laboratory tests, and all other costs which are directly related to the medical treatment of the current dengue illness. Patients were asked how much money they spent for medical services that they received, and whether they had to bear all of the expenditure directly or were covered by any external supports such as private/public insurance, government subsidies, or non-governmental aids. In order to capture the full spectrum of the DMC, hospital bill records were also accessed to understand how the DMC burden was distributed between private and non-private payers. DNMC includes all expenditure spent for food, lodging, and transportation for a patient as well as the patient’s accompanies.
IC takes account of the costs of productivity loss (i.e. wage loss, missing school days), substitute laborers, and caretakers. In order to estimate productivity loss, the self-reported daily wage loss was asked for patients who make earnings. For students who do not earn any wages, the government expenditure per primary student expressed in 2015 USD (2014 USD in Cambodia due to data availability) was used to convert their productivity loss into monetary value. If a patient was neither a wage-worker nor a student (i.e. unpaid housework), the minimum wage of each country was applied. While the government expenditure per primary student is useful for comparing average spending on one student between countries, the use of this indicator may underestimate their productivity loss as the indicator does not include household contributions [14].
In addition to productivity loss, patients were also asked whether they had hired any substitute laborers or caretakers during their illness. If yes, a series of questions related to the duration and payments of having substitute laborers and/or caretakers were asked. In case that patients did not pay anything for having them (i.e. household members), the opportunity costs of substitute laborers/caretakers were estimated by taking into account the daily payments for doing their usual activities which they would have done otherwise. It should be noted that the questionnaire was carefully designed in order to avoid any duplication of the costs. In other words, patient’s productivity loss was not double counted when combining patient’s wage loss with substitute laborer(s)’ costs. The detailed study design which avoids the duplication of productivity loss estimation was fully addressed by Lee et al [14].
In many circumstances, costs may not be the same as charges for various reasons as indicated by previous studies [14,19,20]. While estimating economic burden using hospital charge information reflects better on patients’ direct burden, adjusting hospital charges by the ratio of cost-to-charge (RCC) is another way to understand the overall societal cost of a disease [19,20]. As previously defined, RCC was estimated by dividing the overall annual hospital costs by the total hospital revenue of the hospital [14]. In Burkina Faso and Kenya, the RCC was estimated for each study facility by accessing the annual financial reports which provide comprehensive revenue sources and expenditure types of the health facilities: external funding, additional services, labor costs (staff salary, welfare), material costs, and capital assets, etc. On the other hand, the study team was not able to access the full scope of financial reports for the health facilities in Cambodia due to logistical issues. The study facilities in Cambodia charged patients a package (uniformed) price, which covers a range of medical services such as consultation and medication. Given the limitations, the medical service utilization form was additionally implemented to collect unit costs and quantities of medical services in Cambodia. Other health facility costs such as staff salary, materials, and electricity, etc. were separately obtained. The overall economic burden was presented from both the hospital-charge and societal-cost perspectives. Taking into account the skewed distribution of cost data in general [21,22], bootstrapping was conducted to generate a 95% confidence interval with the percentile method (2.5th and 97.5th percentiles of the distribution) [8]. All estimates were expressed in 2016 USD using the official exchange rate from the World Bank, as well as the purchasing power parity (PPP).
Given that dengue burden is relatively unknown in Burkina Faso and Kenya, we investigated the understanding of the disease in the general public. Thus, the dengue perception score was constructed by combining the following factors: whether a respondent is aware of (1) how dengue is transmitted, (2) proper ways to get treated when infected, and (3) best ways to avoid dengue. The perception score was ranged between 1 and 3 where a higher number indicates more knowledge on dengue fever. In addition, respondents were also asked about their monthly household income, and the total out-of-pocket expenditure of dengue fever was estimated as a proportion of household monthly income to understand the extent of the direct economic burden borne by patients due to dengue infection. Households were categorized into three income groups based on percentiles of monthly household income reported by respondents: low-income group (income≤25%), middle-income group (25%<income≤75%), and high-income group (income>75%). Respondents who did not report their monthly income were categorized into the three income groups by comparing their levels of household-assets with the ones for the respondents who reported income [14].
The cost-of-illness studies were approved by the Institutional Review Boards (IRB) of the International Vaccine Institute, as well as by the ethical review committees of host country institutions: the IRB of the Centre Hospitalier de l’Universitede Montreal (CRCHUM) in Canada and the National Health Ethical Committee in Burkina Faso, KEMRI Scientific and Ethical Review Unit and the Ethical Review Committee of CPGH in Kenya, and the National Ethics Committee for Health Research (NECHR) in Cambodia. All patients who were enrolled into the COI studies completed the written informed consent form. For minors under the age of 18 years old, their parents or guardians were asked to provide consent on behalf of their children.
Table 2 summarizes descriptive statistics. The total number of patients enrolled in the study was the highest in Burkina Faso (n = 414) due to the dengue outbreak occurred during the study period in Ouagadougou. In Cambodia, all dengue-probable cases were automatically hospitalized, thus there was no outpatient enrolled. On the other hand, inpatients were not included because of logistical issues in Kenya. The average number of sick days ranged from 6 to 9 days. Patients tended to have more caretakers than substitute laborers during their illness. While on average, a majority of inpatients were completely unable to perform their usual activities during their illness in Cambodia, patients in Burkina Faso and Kenya were at least partially able to carry out their usual activities during the half of the total sick period. The mean age of patients was lower in Cambodia compared to that in Burkina Faso and Kenya, which in turn, results in the higher proportion of patients studying in Cambodia (see S1 Table for additional information). The average monthly household income was higher in Burkina Faso than in Kenya and Cambodia. Among respondents, dengue vector control activities were more common in Burkina Faso and Cambodia than in Kenya. Types of health facilities where patients visited before and after the study enrollment are further summarized in S1 Fig.
Dengue awareness was quantified by constructing the dengue perception score as shown in Fig 1. As dengue has been prevalent for many years in Cambodia, over 95% of the respondents were well aware of the disease in Cambodia. It is interesting to see that a majority of the respondents fell into the highest category of the perception score in Burkina Faso although the percentage is lower than that of Cambodia. This high perception score observed in Burkina Faso may have been due in part to the dengue outbreaks occurred during and before the study period [23]. In contrast, less than 50% of the respondents scored the highest number in Kenya, reflecting that dengue was a relatively unknown disease to the general public compared to the other two countries. The low-level perception score in Kenya might be related to the fewer number of respondents who conducted vector control activities as shown in Table 1.
Fig 2 demonstrates the proportions of the economic burden by cost component, as well as by expenditure payer. In Fig 2(A), while IC is the biggest burden for patients followed by DNMC and DMC in Cambodia, DMC accounts for the highest proportion of the patient’s private (out-of-pocket) burden in Burkina Faso and Kenya. Fig 2(B) compares the percentage contributions between patient’s private expenditure and public expenditure (i.e. insurance schemes, governmental subsidies, or other NGO aids, etc.). It is clear to see that compared to Cambodia, public contribution to the overall DMC is trivial in Burkina Faso and Kenya, meaning that the most of DMC burden has to be directly borne by patients. This finding is consistent with challenges on achieving Universal Health Coverage (UHC) in Africa [24,25].
The average economic burden of dengue fever is shown in Table 3. The total cost per dengue illness episode for inpatients converted by the official exchange rate is $26 and $134 in Burkina Faso and Cambodia, respectively. For outpatients, the average economic burden per dengue illness episode is estimated to be $13 in Burkina Faso and $23 in Kenya. After taking into account both private and public expenditure, the DMC component appears to be the biggest burden among the three major cost items in Burkina Faso and Kenya, whereas IC still remains the most significant contributor for the overall burden in Cambodia. The average cost per day ranges from $2 for outpatient in Burkina Faso to $15 for inpatient in Cambodia. The economic burden of dengue fever was also presented after adjusting the costs by the RCCs. While the total cost per dengue illness episode went up after the adjustment in Burkina Faso and Cambodia, this was the opposite in Kenya. The estimate in Cambodia shows the biggest change between the official exchange rate and the PPP conversion factor. It is worth noting that the WHO-CHOICE project shows cost per bed day, as well as cost per outpatient visit by hospital level [26]. While direct comparisons may not be appropriate due to different cost components, study designs, and target diseases, the total cost per day shown in the current study may be considered as a similar cost category.
Fig 3 demonstrates the average economic burden of dengue fever by age group. In Burkina Faso and Cambodia, the average cost increases from the younger age group to the older age group. On the other hand, the economic burden is higher for the youngest age group than for the other older age groups in Kenya. The high cost in the youngest group in Kenya was mainly derived from the additional private facility visit where patients paid much higher fees for medical services.
The patient’s private expenditure was estimated as a proportion of household’s monthly income and shown in Fig 4. For all three countries, the proportion of the private economic burden of dengue fever directly borne by patients was the highest in the low income group and decreased as moving towards the high income group. By country, the proportion of the private burden appeared to be relatively more significant in Cambodia compared to Burkina Faso and Kenya. In particular, the average direct expenditure due to dengue infection could be more than household’s monthly earning in the low income group in Cambodia.
The current study reports the most up-to-date estimates of the economic burden in Burkina Faso, Kenya, and Cambodia. In particular, having reviewed existing economic burden studies of dengue fever, our study is the first to understand the economic burden of dengue fever in Burkina Faso and Kenya based on primary data sources [17,27]. The study outcomes showed that the total economic burden of dengue fever is not trivial in all three countries. For inpatients, the average total cost per episode of dengue illness after the RCC adjustment was $26 in Burkina Faso and $134 in Cambodia. In the case of outpatients, the average cost per dengue episode was estimated to be $13 and $23 in Burkina Faso and Kenya, respectively.
Given that dengue has been prevalent for many years in Cambodia, several economic burden studies for dengue were previously done in this country. Four studies were identified at the time of this research [15,28–30]. Out of four, two studies estimated dengue cost-of-illness based on primary data sources [15,30]. Huy et al. reported $40 for inpatient which is lower than our estimate even after the inflation adjustment [30]. This is because Huy et al. only took into account private expenditure, whereas the current study included both private and public payments (i.e. health equity funds). On the other hand, Suaya at el. estimated the average cost of $115 per inpatient which is similar to the RCC-adjusted cost of the current study after the inflation adjustment [15]. Compared to the first round COI countries, the RCC adjusted total cost in Cambodia is similar to that in Thailand ($181) but lower than the costs in Vietnam ($213) and Colombia ($239). It is interesting to observe that the cost per inpatient converted using the PPP conversion factor in Cambodia is higher than those in Thailand and Colombia. Considering that purchasing power and parity is designed to equalize the purchasing power among different currencies, the economic burden of dengue in Cambodia is as significant as other dengue-endemic countries after taking into account differences in cost of living.
Overall, the total cost of illness for dengue fever was higher in Cambodia than in Burkina Faso and Kenya. In particular, the average cost per inpatient was much higher in Cambodia than in Burkina Faso although the average household income in Cambodia was lower than that in Burkina Faso. This was due to the following reasons: (1) the duration of illness was longer in Cambodia than in Burkina Faso, (2) while only 23% of the enrolled patients had sought medical care prior to coming to our study facilities in Burkina Faso, over 80% of the inpatients in Cambodia had done so increasing the overall spending, and (3) not many patients (approximately 30%) had caretakers during their illness in Burkina Faso, whereas all inpatients had caretakers in Cambodia, contributing to the significant increase in IC.
Nonetheless, the economic burden of dengue fever in the two African countries is not insignificant compared to the economic cost of malaria. Albeit by different methods, Beogo et al. reported $15.2 as the average cost of malaria in Burkina Faso [31]. Sicuri et al. estimated the economic costs of malaria in children in selected sub-Saharan countries and reported $11.2 for uncomplicated malaria and $51.9 for hospitalized malaria episodes in Kenya [32].
Some areas of uncertainty deserve attention. Despite the efforts to obtain financial reports from all four study facilities in Cambodia, the study team was not able to collect the financial report from one hospital out of four health facilities due to logistical issues. Thus, the RCC from the other health facility at the same level was applied assuming that the financial structure at the same level would not be substantially different. Nonetheless, additional information was obtained by implementing the medical service utilization form in Cambodia, and the bias was minimized. Similar to the first round COI study, the current COI study sites were limited to the areas where epidemiologic surveillance studies were carried out, thus caution must be exercised when interpreting the estimates beyond the study communities. In Burkina Faso, there was a dengue outbreak during the study period. This may have influenced healthcare practice in the health facilities, as well as health seeking behavior, particularly for children. However, the estimates in Burkina Faso may also be meaningful to understand the economic burden of dengue fever during an epidemic period. In Kenya, the study team tried to cover as many units within CPGH as possible but was unable to include inpatients due to logistical issues. Capital assets were not included in Cambodia and not depreciated in Burkina Faso and Kenya due to the lack of available information, thus the societal costs might be conservative in Cambodia and overestimated in the other two countries.
The standardized COI study was implemented in Burkina Faso, Kenya, and Cambodia. The selected study outcomes were presented in a similar way to the first-round COI study in order to facilitate comparisons across all six sites. In particular, the study findings clearly showed that the economic burden of dengue fever is significant not only in Cambodia but also in the two African countries. Given that the burden of dengue fever is relatively unknown in Africa, and that an increasing number of non-malaria fever patients have been reported [33,34], future research is urgently needed to have a better understanding of dengue disease burden in this region. For example, during the site selection period prior to implementing the current economic burden study in Kenya, health clinicians had repeatedly reported an increasing number of non-malaria fever patients during the mosquito season and were keen to understand the potential causes of the fever cases.
The first live attenuated, tetravalent dengue vaccine called Dengvaxia (CYD-TDV) became available in 2016. In addition, there are several second-generation vaccine candidates in the pipeline. Considering the broader availability of dengue vaccines in the future, it is critical to understand the societal benefits of vaccination and to develop sustainable financing plans taking into account competing health problems in the three countries. Along with more detailed epidemiological data (i.e. incidence rates) and evidence on the long-term behavior of a vaccine, the economic burden outcomes presented in the current study can be used to estimate more accurate vaccination benefits when conducting cost-effectiveness analyses of dengue vaccine interventions in the three countries in the future.
|
10.1371/journal.pcbi.1000320 | On the Growth of Scientific Knowledge: Yeast Biology as a Case Study | The tempo and mode of human knowledge expansion is an enduring yet poorly understood topic. Through a temporal network analysis of three decades of discoveries of protein interactions and genetic interactions in baker's yeast, we show that the growth of scientific knowledge is exponential over time and that important subjects tend to be studied earlier. However, expansions of different domains of knowledge are highly heterogeneous and episodic such that the temporal turnover of knowledge hubs is much greater than expected by chance. Familiar subjects are preferentially studied over new subjects, leading to a reduced pace of innovation. While research is increasingly done in teams, the number of discoveries per researcher is greater in smaller teams. These findings reveal collective human behaviors in scientific research and help design better strategies in future knowledge exploration.
| It is of great interest to understand the patterns and mechanisms of scientific knowledge growth, but such studies have been hampered by the lack of ideal cases in which the structure of the knowledge is known, the knowledge is quantifiable, and the process of knowledge discovery is well understood and documented. The biological knowledge about a species is in part described by its protein interaction network and genetic interaction network. Here, we conduct a temporal meta-analysis of three decades of discoveries of protein interactions and genetic interactions in baker's yeast to reveal the tempo and mode of the growth of yeast biology. We show that the growth is exponential over time and that important subjects tend to be studied earlier. However, expansions of different domains of knowledge are highly heterogeneous and episodic such that the temporal turnover of knowledge hubs is much greater than that expected by chance. Familiar subjects are preferentially studied over new subjects, leading to a reduced pace of innovation. While research is increasingly done in teams, the number of discoveries per researcher is greater in smaller teams. These findings reveal collective human behaviors in scientific research and help design better strategies in future knowledge exploration.
| Scientific knowledge refers to the body of facts and principles that are known in a given field. Modern civilization is built on the knowledge that humans have acquired about the world they live in, and the future of the human species and society critically depends on further accumulation of scientific knowledge. Patterns and mechanisms of human knowledge growth are jointly determined by the intrinsic structure of knowledge and human behaviors in knowledge exploration. Although such behaviors are of interest to many scientists including philosophers [1],[2], sociologists [3], anthropologists [4], economists [5], physicists [6], and psychologists [7], they are poorly studied, due primarily to the lack of ideal cases in which (i) the structure of the knowledge is known, (ii) the knowledge is quantifiable, and (iii) the process of knowledge discovery is well understood and documented.
As biologists, we notice that the above three requirements are all met for biological knowledge of the baker's yeast Saccharomyces cerevisiae. Knowledge can be described largely as relationships among a set of subjects. Over the past three decades, scientists have substantively deepened their understanding of yeast biology through the study of interactions among its ∼6000 genes [8]. By the end of 2007, over 73,000 yeast gene-gene interactions had been discovered and documented in ∼5,400 publications authored by 11,238 researchers (see Materials and Methods). Much of the structure of the knowledge about yeast biology can be described as a gene-gene interaction network, where the unit of knowledge is an interaction. Scientific publications record the approximate date of each relevant discovery, as well as the methodology used. As a case study, we here analyze the temporal growth of the known yeast gene-gene interactions to understand the tempo and mode of scientific knowledge expansion.
Gene-gene interactions are separated into two types: genetic interactions (GIs) and protein-protein interactions (PPIs) [9]. Two genes are said to interact genetically if the effect of one gene on a trait is masked or enhanced by the other. Two genes are said to have a PPI if their protein products physically bind to each other stably or transiently. The data we considered contain 37,809 PPIs among 4,913 genes and 35,231 GIs among 3,743 genes, respectively (see Materials and Methods). Because of the difference in the nature of PPIs and GIs, we study the yeast PPI and GI networks separately.
The PPI data were published from year-1982 to 2007, spanning 26 years, while the GI data were published from year-1977 to 2007, spanning 31 years (see Materials and Methods). The number of new interactions discovered per year increased approximately exponentially over time (Figure 1), and there is no apparent sign of slowing of this exponential growth at present. The exponential growth can be attributed to the increased number of studies per year and/or the enhanced productivity per study over time (Figure 2). P(k), the probability that a study discovers k novel interactions, is proportional to k−r, where r = 1.79 and 1.84 for PPIs and GIs, respectively, indicating that the per-study productivity roughly follows a power-law distribution (Figure 3 and Figure S1). We also observed that the number of co-authors per study increased over time (Figure 4), reflecting a general trend of increased collaboration in scientific research [10],[11]. Increase of productivity per author over time is not significant for PPIs, but significant for GIs (Figure S2). However, within virtually every year, per-author productivity is strongly negatively correlated with the number of co-authors of the study (Figure 5A and Table S1), suggesting that small research teams are more efficient than large teams at all times. Considering the possibility that researchers of small teams may publish fewer papers than those of large teams, we calculated accumulated productivity per-author in a five-year window. Again, authors of small teams consistently outperform those of large teams (Table S2) and this result remains qualitatively unchanged even when we consider the accumulated productivity of only those researchers who served at least once as the last author of a study in a five-year window (Table S3). However, the negative correlation between the productivity of a researcher and his/her mean team size appears to be weakening over the years (Figure 5B and Tables S1, S2 and S3).
The ∼6000 yeast genes have been individually deleted to examine their functional importance, which is defined by the amount of reduction in the fitness of yeast caused by each deletion [12]. We traced the first year of appearance (birth year) of each gene in the PPI and GI networks, and found that genes appearing earlier in the networks (old genes) are more important than those appearing later (young genes) (Figure 6). One possible explanation of this phenomenon is that a gene's importance arises from the sheer number of its interactions [13]–[15]; if each interaction has the same probability of discovery, highly interactive genes are incorporated into the knowledge network earlier simply because they have more interactions. However, we found that old genes are more important than young genes even when the number of now known interactions per gene is controlled for (Spearman's partial correlation coefficient ρ = 0.13, P = 1.8×10−17 for the PPI network; ρ = 0.10, P = 5.3×10−9 for the GI network; Table 1). This result remains unchanged when we further control for the level of gene expression (Table 1). Thus, important genes are studied earlier not simply because of their large numbers of interactions, but also because of their phenotypic importance that is beyond what is predicted from their numbers of interactions.
During the growth of the yeast biological knowledge network, a new interaction can introduce zero, one, or two genes into the network. Generally speaking, follow-up studies tend to discover interactions involving “pre-existing” genes while novel studies tend to discover interactions between previously “uncatalogued” genes [16]. We separately simulated the growths of yeast PPI and GI networks by randomizing the birth years of all interactions while conserving the number of new interactions discovered each year. Interestingly, the growth of gene number in the real networks lags behind the random expectation for many years (Figure 7), suggesting that, compared with the random process, actual researchers tend to focus on finding properties of known genes rather than those of new genes. We conducted 1000 simulations of random growth and found that the number of genes is 655.1±10 at 1995, the mid-point of PPI network growth, and this number is 676.1±14.6 for GI network at its mid-point of growth. Both numbers are significantly (P<0.001) larger than the observed numbers (390 for PPI network and 454 for GI network) in real growth. We also observed that the real growth pattern relative to the random pattern was reversed in recent years. However, this reserve is due to the fixation of total numbers of genes and interactions at year-2007 and does not suggest that the tendency of “novelty-aversion” has been reversed in research. The “novelty-aversion” phenomenon may arise from a high cost of novelty-seeking research and/or a high reward (or desire) for studying previously discovered genes [17]. As a consequence, the cohesiveness of the actual knowledge network is higher than that of a randomly growing network during the early years of yeast research (Figure S3).
Many complex networks are naturally divided into communities or modules, such that interactions within modules are much denser than those between modules [18]. The temporal PPI and GI data allow us to study the relative growths of different modules in a knowledge network compared to random growths. We identified 12 and 16 modules from the present-day PPI and GI networks, respectively [15] (see Materials and Methods). We transformed the network growth information into module growths by assigning one unit for every involved gene of a new interaction to the module that the gene belongs to. We then measured the deviation of the growth of each module from its expectation under homogenous growth, for each temporal PPI or GI network. Interestingly, although the network growth was contributed simultaneously by multiple modules in many years, the among-module heterogeneity in growth is striking, compared to random growths (Figure 8). For example, 4.7% of the PPI network growth was contributed by module #12 in year-2000, but this number becomes 70.8% in year-2007. The fluctuation index measured by mean Euclidean distance (see Materials and Methods) among these distributions is 0.40 and 0.42 for PPI and GI networks, respectively. Both are significantly larger than the expectations from simulated random growths of PPI (0.26±0.03) and GI (0.18±0.02) networks (P<0.001; Figure 9). This heterogeneous and episodic growth also leads to among-module variation in the maturation process of modules (Figure 10).
One wonders whether the observed heterogeneous and episodic growth of PPI and GI modules is owing to some recent large-scale studies that focused on genes involved in specific cellular functions; PPIs and GIs discovered from such studies are expected to be localized to certain knowledge modules rather than evenly distributed among all modules. To examine the effect of large-scale studies, we separately examined the network growth before and after year-1999. In the pre-1999 years, there was only 1 paper reporting >50 PPIs and 8 papers each reporting 20–50 PPIs, among the 919 papers on PPIs. Similarly, in this period, there were only 5 papers each reporting 20–50 GIs, among 1633 papers on GIs. In the post-1999 years, there were many large-scale studies. However, heterogeneous episodic growth of modules is found in both periods (Table S4). Thus, our observation is not simply a result of recent large-scale studies of specific cellular functions.
The heterogeneous and episodic growth of knowledge modules has an important consequence. Like many complex networks [19], connectivity is highly variable among nodes in the yeast PPI and GI networks. Most genes have one or a few interactions while a small fraction of genes have a very large number of interactions (Figure S4). Highly connected nodes (hubs) are known to be of both structural and functional importance to a network [13],[14],[19] (see also Table 1). Therefore, recognizing true hubs earlier would speed up the study of the network structure and function. However, hubs in today's network may not be hubs in the previous year's network and it is important to examine how stable hubs are during network growth. We arbitrarily define hubs in a given year as genes whose total connectivities in a network are among the top 10% of all available genes within the network at that time (only temporal networks with at least 50 genes are considered). We examined hub turnover in each year by computing the proportion of temporal hubs that become non-hubs in the following year. For both the PPI and GI networks, hub turnover rates are usually high (Figure 11). Surprisingly, hub stability did not increase with the growth of the network. For example, 32.5% of year-2006 GI hubs became non-hubs in 2007, and the corresponding number was 15.5% for year-2006 PPI hubs. This suggests that under the current mode of knowledge growth, it is difficult to predict true hubs before completion of network growth. By contrast, in the simulated random network growth, there is a trend of reduction in hub turnover over time. For example, in the GI network the turnover rate became <10% after year-1997 and <1% between year-2006 and 2007. The birth of temporal hubs appears to be strongly associated with heterogeneous expansions of modules (Figure 12).
The heterogeneous and episodic growth of network modules, and the related rapid hub turnover, are likely caused by a high reward (e.g., high-profile publications or large grants) for or biased interest in studying certain topics at certain times. For example, when a human disease-associated gene is identified, its yeast ortholog could be subject to intense studies immediately. Human syntaxin 8 was cloned in 1999 [20] and characterized as a member of the t-SNARE (target soluble N-ethylmaleimide sensitive factor attachment protein receptor) superfamily involved in vesicular trafficking and docking, a critical cellular process implicated in many human diseases [21]–[23]. Soon after the discovery, its yeast ortholog YAL014C was investigated and its 5 PPIs were identified by two studies in 2000 [24] and 2002 [25], respectively.
In addition, different parts of a knowledge network are more likely to be discovered by different technologies that are invented at different times (Figure 13). For instance, in discovering PPIs, affinity approaches [26] tend to identify stable protein complexes while yeast two-hybrid assays [27] find dynamic interactions well. To further demonstrate this point, we directly compared two genome-wide studies that used either yeast two-hybrid assays [28] or affinity approaches [29] to discover PPIs. The across-module PPI distributions of the two studies are significantly different (Table S5). These results illustrate the importance of employing diverse approaches in knowledge exploration.
Although the PPI and GI networks analyzed here are still growing, they have been studied for ∼30 years and have encompassed most yeast genes. Thus, they serve as relatively good representations of the true and complete networks. For example, it is believed that we have already discovered ∼50% of all yeast PPIs [30]. Nevertheless, it is possible that we may have omitted some discoveries, although the BioGRID database, from which our data are acquired, is based on extensive literature searches [31]. To evaluate the potential effect of such omissions, we randomly excluded 10% of studies and repeated our analyses, and found that all major conclusions hold (data not shown). It should also be pointed out that, although the unbiased random network growth was based on the year-2007 networks, all principles should be applicable to the final true and complete networks.
The exponential growth shown in Figure 1 and the assumption that ∼50% of all PPIs in yeast have been identified predict that almost all yeast PPIs will have been discovered by year-2009, if the fraction of false positive discoveries does not increase with the rate of discovery. However, it is fully expected that both the current and future PPI and GI networks contain false interactions. Because false understanding exists in any type of knowledge, it will be interesting to study how false interactions affect the discoveries of true interactions. Unfortunately, BioGRID contains no information about previously reported interactions that are later dismissed. In fact, it is extremely difficult to falsify a previously reported interaction, because (i) the falsification requires one to test an interaction with exactly the same technique and condition as used in the initial experiment that discovered the interaction, and (ii) such falsification is by definition negative evidence for the existence of the interaction and therefore could be subject to other interpretations. Thus, at present it is difficult to evaluate how false interactions affect the growth of yeast biology.
In this work, we considered only the knowledge of the presence of an interaction and ignored detailed knowledge such as the strength of the interaction, the conditions under which the interaction occurs, and the biochemical or genetic basis of the interaction. It is difficult to analyze these types of knowledge at present because their structures are unclear. Paradigm shifts have been emphasized as an important mode of knowledge growth [2]. In the history of yeast research, the publication of the yeast genome sequence in 1996 [8] is widely thought to have triggered a paradigm shift from gene-based studies to genomic studies. However, such a shift in research scale and approach did not cause apparent changes in either the speed or pattern of discovery of new PPIs and GIs. Further analysis may reveal subtle signals of the paradigm shift that escaped our gross analysis. After all, our work represents just one step towards quantitative understanding of the tempo and mode of knowledge growth in the framework of network theories. Although the generality of our findings requires further evaluation, the lessons learned from this case study may help develop strategies for efficient knowledge exploration in the future.
Yeast protein-protein interaction data and genetic interaction data were downloaded from BioGRID (http://www.thebiogrid.org). The publication year and author information for each interaction were extracted from NCBI (http://www.ncbi.nlm.nih.gov) using the PUBMED ID provided by BioGRID. Because we are interested in discoveries of new interactions, interactions that were reported in previous years were excluded. When a new interaction is reported by two or more publications of the same year, one of these publications was randomly chosen for further analyses. We measured the importance of a gene by the reduction in fitness of the yeast strain (i.e., growth rate) in rich medium (YPD) when the gene is deleted. The fitness data were downloaded from http://www-deletion.stanford.edu/YDPM/YDPM_index.html. The expression levels of yeast genes are measured at mid-log phase of growth and obtained from a previous study [32]. Authors with identical names were not differentiated. Although this practice necessarily introduced errors, it should not affect our results, because authors with common names and rare names are not expected to behave differently in research (e.g., they should participate in large teams with equal probabilities).
Random network growth was simulated by randomizing the birth year of each interaction while keeping the number of newly discovered interactions unchanged for each year. Network modules were identified using simulated annealing, which has been shown to perform better than other module-separating algorithms [15]. The parameters used were: iteration factor = 0.1, cooling factor = 0.9, and final temperature = 10−20. For the PPI network, the giant component contains 99.72% of all genes and 99.98% of all interactions. The corresponding numbers are 98.18% and 99.89%, respectively, for the GI network. Relative growths of all modules in each year form a vector. The Euclidean distance between vectors of two consecutive years is then computed. The fluctuation index of a network is defined as the mean of Euclidean distances of all consecutive years. We transformed the network growth information into module growths by assigning one unit for every involved gene of a new interaction to the module that the gene belongs to. To measure the deviation of the actual growth of a module in a given year from the expected homogenous growth, we calculated a transformed chi-squares value, , where Oi is the observed growth of module i in a given year and Ei is the expected (homogenous) growth given the total growth of the network in the year and the relative size of module i in year-2007. , where O is the total number of interactions discovered in a given year and Si is the relative size measured by the sum of node degrees of module i to the entire network in year-2007. In short, for each year, the deviations from homogenous growth were calculated across modules.
|
10.1371/journal.pcbi.1000415 | Exploring the Free Energy Landscape: From Dynamics to Networks and Back | Knowledge of the Free Energy Landscape topology is the essential key to understanding many biochemical processes. The determination of the conformers of a protein and their basins of attraction takes a central role for studying molecular isomerization reactions. In this work, we present a novel framework to unveil the features of a Free Energy Landscape answering questions such as how many meta-stable conformers there are, what the hierarchical relationship among them is, or what the structure and kinetics of the transition paths are. Exploring the landscape by molecular dynamics simulations, the microscopic data of the trajectory are encoded into a Conformational Markov Network. The structure of this graph reveals the regions of the conformational space corresponding to the basins of attraction. In addition, handling the Conformational Markov Network, relevant kinetic magnitudes as dwell times and rate constants, or hierarchical relationships among basins, completes the global picture of the landscape. We show the power of the analysis studying a toy model of a funnel-like potential and computing efficiently the conformers of a short peptide, dialanine, paving the way to a systematic study of the Free Energy Landscape in large peptides.
| A complete description of complex polymers, such as proteins, includes information about their structure and their dynamics. In particular it is of utmost importance to answer the following questions: What are the structural conformations possible? Is there any relevant hierarchy among these conformers? What are the transition paths between them? These and other questions can be addressed by analyzing in an efficient way the Free Energy Landscape of the system. With this knowledge, several problems about biomolecular reactions (such as enzymatic activity, protein folding, protein deposition diseases, etc.) can be tackled. In this article we show how to efficiently describe the Free Energy Landscape for small and large peptides. By mapping the trajectories of molecular dynamics simulations into a graph (the Conformational Markov Network) and unveiling its structural organization, we obtain a coarse grained description of the protein dynamics across the Free Energy Landscape in terms of the relevant kinetic magnitudes of the system. Therefore, we show the way to bridge the gap between the microscopic dynamics and the macroscopic kinetics by means of a mesoscopic description of the associated Conformational Markov Network. Along this path the compromise between the physical nature of the process and the magnitudes that characterize the network is carefully kept to assure the reliability of the results shown.
| Polymers and, more specifically, proteins, show complex behavior at the cellular system level, e.g. in protein-protein interaction networks [1], and also at the individual level, where proteins show a large degree of multistability: a single protein can fold in different conformational states [2]–[4]. As a complex system [5],[6], the dynamics of a protein cannot be understood by studying its parts in isolation, instead, the system must be analyzed as a whole. Tools able to represent and handle the information of the entire picture of a complex system are thus necessary.
Complex network theory [7],[8] has proved to be a powerful tool used in seemingly different biologically-related fields such as the study of metabolic reactions, ecological and food webs, genetic regulatory systems and the study of protein dynamics [7]. In this latter context, diverse studies have analyzed the conformational space of polymers and proteins making use of network representations [9]–[12], where nodes account of polymer conformations. Additionally, some studies have tried to determine the common and general properties of these conformational networks [13],[14] looking at magnitudes such as clustering coefficient, cyclomatic number, connectivity, etc. Recently, trying to decompose the network in modules corresponding to the free energy basins, the use of community algorithms over these conformational networks have been proposed [15]. Although this approach has opened a promising path for the analysis of Free Energy Landscapes (FEL), the community based description of the network leads to multiple characterizations of the FEL and thus it is difficult to establish a clear map from the communities found to the basins of the FEL.
A similar approach, commonly used to analyze the complex dynamics, is the construction of Markovian models. Markovian state models let us treat the information of one or several trajectories of molecular dynamics (MD) as a set of conformations with certain transition probabilities among them [9],[16],[17]. Therefore, the time-continuous trajectory turns into a transition matrix, offering global observables as relaxation times and modes. In [16]–[18] the use of Markovian models is proposed with the aim of detecting FEL meta-stable states. However, the above approaches to analyze FELs of peptides involves extremely large computational cost: either general community algorithms or large transition matrices.
Finally, other strategies to characterize the FEL that have successfully helped to understand the physics of biopolymers, are based on the study of the Potential Energy Surface (PES) [3], [4], [19]–[21]. The classical transition-state theory [22] allows us to project the behavior of the system at certain temperature from the knowledge of the minima and transition states of the PES. This approach entails some feasible approximations, such as harmonic approximation to the PES, limit of high damping, assumption of high barriers, etc. These approximations could be avoided working directly from the MD data.
In this article we make a novel study of the FEL capturing its mesoscopic structure and hence characterizing conformational states and the transitions between them. Inspired by the approaches presented in [12],[15] and [16],[17], we translate a dynamical trajectory obtained by MD simulations into a Conformational Markov Network. We show how to efficiently handle the graph to obtain, through its topology, the main features of the landscape: conformers and their basins of attraction, dwell times, rate constants between the conformational states detected and a coarse-grained picture of the FEL. The framework is shown and validated analyzing a synthetic funnel-like potential. After this, the terminally blocked alanine peptide (Ace-Ala-Nme) is studied unveiling the main characteristics of its FEL.
In this section we show the round way of the FEL analysis: the map of microscopic data of a MD into a Conformational Markov Network (CMN) and, by unveiling its mesoscopic structure, the description of the FEL structure in terms of macroscopic observables.
First, we encode a trajectory of a stochastic MD simulation into a network: the CMN. This map will allow us to use the tools introduced henceforth to analyze a specific dynamics of complex systems such as biopolymers.
Up to now, we have illustrated the conversion of molecular dynamics data into a graph (the CMN). Now, we show how to efficiently obtain the thermo-statistical data from the mesoscopic description of the CMN.
The alanine dipeptide, or terminally blocked alanine peptide (Ace-Ala-Nme, Figure 3A), is the most simple “biological molecule” that exhibits the common features shown by larger biomolecules. Despite of its simplicity, this system has more than one long-life conformational state with different transition pathways. Since the first attempt by Rossky and Karplus [37] to model this dipeptide solvated, this system has been widely studied in theoretical works [38]–[41]. The alanine dipeptide has been also the appropriate molecule to test tools to explore the FEL [15],[16],[42] and, specifically, to study reaction coordinates [41],[43].
The alanine dipeptide has two slow degrees of freedom, the rotatable bonds () and () (see Figure 3A). The FEL projected onto these dihedral angles let us identify the conformational states that characterize the geometry of biopolymers, namely: alpha helix right-handed (), alpha helix left-handed (), beta strands (, ), etc. The number of local minima in the (, ) space depends on the effective potential model used to simulate the system. Up to date, electronic structure calculations have identified a total of nine different conformers [44]. Regarding MD simulations different conformational states have been observed: (i) using classical force fields with explicit solvent up to six conformers are detected [16],[38],[39], (ii) at least four stable states by using implicit solvent [15],[38],[40], and (iii) two stable conformers in vacuum conditions [38],[41]. On the other hand, since the angles and seem appropriate to distinguish the metastable states, the kinetics between them is not accurately described with this choice of reaction coordinates, the solvent coordinates and/or other internal degrees of freedom must be taken into account [41],[43].
We have used SSD algorithm to detect the local minima and their corresponding basins for this molecule in the space. For this purpose, a Langevin MD simulation of 250 ns has been performed at a temperature of 400 K (see Text S3 for further details). Additionally the CMN has been built dividing the Ramachandran plot into cells of surface 9°×9° (40×40) and taking dialanine conformations at time intervals of . The resulting CMN have a total of nodes and directed links.
The algorithm applied to the CMN network reveals 6 basins. Figure 3B shows the resulting network where nodes belonging to the same basin take the same color. Bringing back this information to the Ramachandran map, these 6 sets of nodes define 6 regions represented in Figure 3C. To better illustrate this division, other representation, where each region has a different color, is shown. By comparing with previous studies on this molecule, we identify the regions in orange, red, yellow and pink with conformers , , and respectively [38],[45]. Besides, region green corresponds to conformer and the blue one to [16],[45]. Remarkably, the basin (one of the less populated state) has been visited 1155 times with a mean stay time of 4.20 ps.
We now look at the coarse-grained picture of the FEL by describing the properties of the 6 basins detected. The different weights of the basins are related to the free energy of the corresponding conformational macro-states. In Table 1 these energy differences are shown, taking the most populated basin as the energy reference . The lowest free energy basins correspond to configurations with φ≤0° (, , and ), whereas the two other conformers located in the region φ≥0° have the highest free energy but the largest dwell time. Moreover, we have also analyzed the trapping efficiency of each basin by reporting the mean escape time () as well in Table 1.
The FEL can be represented as a dendogram, see Figure 4, where the hierarchical map of the conformers based on Free Energy gives at first glance a global picture of the landscape. Remarkably, the conformer , despite of having one of the highest free energy, looks like the metastable state with longest life. This result is supported by the values of Mean Escape Time shown in Table 1.
The alanine dipeptide has been also studied because of its “fast” isomerization and a “slow” transition . Our coarse-grained picture of the FEL also allows us to extract information about these transitions. In Table 2 we show some of the relevant characteristic transition times from a basin a to an adjacent basin b, . [The whole data is shown in the Text S3.] Transitions between basins with the same sign of are remarkably faster (e.g and ). While slow transitions are observed for those hops crossing the line φ = 0° ( and ), showing them as rare events. Instead, the alanine dipeptide finds more easily paths to go to φ≥0° conformers through and by crossing φ = 180°.
To round off the description of the FEL, the dendogram corresponding to the temporal hierarchy is shown in Figure 5. From the figure, it becomes clear that the behavior of the dialanine depends on the time scale used for its observation. Again, the same two different sets of conformers are distinguished from this hierarchy. Additionally, the global minimum conformer is reached in around 100 ps from any basin.
Finally, the magnitudes computed here for the alanine dipeptide would allow to construct a first-order kinetic model of 6 coupled differential equations as Eq. (6) (assuming equilibrium intra-basin). This model contains the same information as the kinetic model by Chekmarev et al. for the irreversible transfer of population from [40].
Hierarchical landscapes characterize the dynamical behavior of proteins, which in turn depends on the relation between the topology of the basins, their transitions paths and the kinetics over energy barriers. The CMN analysis of trajectories generated by MD simulations is a powerful tool to explore complex FELs.
In this article, we have proposed how to deal with a CMN to unveil the structure of the FEL in a straightforward way and with a remarkable efficiency. The analysis presented here is based on the physical concept of basin of attraction, making possible the study of the conformational structure of peptides and the complete characterization of its kinetics. Note that this has been done without the estimation of the volume of each conformational macro-state in the coordinates space and without the ‘a priori’ knowledge of the saddle points or the transition paths from a local minimum to another.
On the other hand, the framework introduced in the article provides us with a quantitative description of the dialanine's FEL, coming up directly from a MD dynamics at certain temperature. The peptide explores its landscape building the corresponding CMN and the success of extracting the relevant information is up to the ability of dealing with it. Neither the FE basins were defined by the unique criterion of clustering conformations with a geometrical distance [46], nor the rate constants were projected from the potential energy surface [19],[20]. Moreover, the conformers and their properties were computed from the MD with the only limitation of the discretization of time and space.
Although we have applied the method to low dimensional landscapes, we expect that high dimensional systems could be also studied, by the combination of this technique with the usual methods to reduce the effective degrees of freedom (like principal component analysis or essential dynamics). In conclusion, the large amount of information obtained by working with the CMN, its potential application to any peptide with a large number of monomers, and the possibility of performing the analysis on top of CMN constructed via several short MD simulations [47], make the approach presented here a promising way to describe the FEL of a protein.
|
10.1371/journal.pgen.1006042 | Heat Shock Protein Beta-1 Modifies Anterior to Posterior Purkinje Cell Vulnerability in a Mouse Model of Niemann-Pick Type C Disease | Selective neuronal vulnerability is characteristic of most degenerative disorders of the CNS, yet mechanisms underlying this phenomenon remain poorly characterized. Many forms of cerebellar degeneration exhibit an anterior-to-posterior gradient of Purkinje cell loss including Niemann-Pick type C1 (NPC) disease, a lysosomal storage disorder characterized by progressive neurological deficits that often begin in childhood. Here, we sought to identify candidate genes underlying vulnerability of Purkinje cells in anterior cerebellar lobules using data freely available in the Allen Brain Atlas. This approach led to the identification of 16 candidate neuroprotective or susceptibility genes. We demonstrate that one candidate gene, heat shock protein beta-1 (HSPB1), promoted neuronal survival in cellular models of NPC disease through a mechanism that involved inhibition of apoptosis. Additionally, we show that over-expression of wild type HSPB1 or a phosphomimetic mutant in NPC mice slowed the progression of motor impairment and diminished cerebellar Purkinje cell loss. We confirmed the modulatory effect of Hspb1 on Purkinje cell degeneration in vivo, as knockdown by Hspb1 shRNA significantly enhanced neuron loss. These results suggest that strategies to promote HSPB1 activity may slow the rate of cerebellar degeneration in NPC disease and highlight the use of bioinformatics tools to uncover pathways leading to neuronal protection in neurodegenerative disorders.
| Niemann-Pick type C1 (NPC) disease is an autosomal recessive lipid storage disorder for which there is no effective treatment. Patients develop a clinically heterogeneous phenotype that typically includes childhood onset neurodegeneration and early death. Mice with loss of function mutations in the Npc1 gene model many aspects of the human disease, including cerebellar degeneration that results in marked ataxia. Cerebellar Purkinje cells in mutant mice exhibit striking selective vulnerability, with neuron loss in anterior lobules and preservation in posterior lobules. As this anterior to posterior gradient is reproduced following cell autonomous deletion of Npc1 and is also observed in other forms of cerebellar degeneration, we hypothesized that it is mediated by differential gene expression. To test this notion, we probed the Allen Brain Atlas to identify 16 candidate neuroprotective or susceptibility genes. We confirmed that one of these genes, encoding the small heat shock protein Hspb1, promotes survival in cell culture models of NPC disease. Moreover, we found that modulating Hspb1 expression in NPC mice promoted (following over-expression) or diminished (following knock-down) Purkinje cell survival, confirming its neuroprotective activity. We suggest that this approach may be similarly used in other diseases to uncover pathways that modify selective neuronal vulnerability.
| Selective vulnerability of specific neuronal populations is a well characterized, though often perplexing feature of many neurodegenerative diseases [1]. Most commonly, these disorders are initiated by a uniform stress to the entire CNS, such as a genetic mutation, toxic insult, or aging. However, only a subset of neurons respond to these stressors by degenerating, while others remain resistant and apparently maintain their normal function [2]. Although this phenomenon is widely observed, the underlying mechanisms remain poorly understood. Notably, the factors regulating neuronal vulnerability represent attractive therapeutic targets, with the potential to convert susceptible neuronal populations into ones that are disease resistant.
One particularly striking example of selective vulnerability is the degeneration of cerebellar Purkinje cells [3]. Purkinje cells represent the sole output of the cerebellar cortex. Loss of Purkinje cells, therefore, leads to significant deficits of motor coordination, including ataxia and tremors. Despite the apparent similarity of Purkinje cells in their morphology, connectivity, and electrophysiological properties, many cerebellar disorders affect Purkinje cells in a non-uniform way, leading to a distinct spatiotemporal pattern of loss that is reproducible not only between cases of a single disease, but across many otherwise unrelated diseases and injuries. One common pattern reveals a strong resistance of Purkinje cells in lobule X to degeneration, contrasted with the exquisite sensitivity of the anterior zone (lobules II-V), and moderate susceptibility of the intermediate (lobules VI-VII) and posterior zones (lobule VIII and rostral aspect of lobule IX). Superimposed onto this anterior-to-posterior gradient is often a pattern of parasagittal stripes in which differential vulnerability is also observed [3]. Diseases displaying the classic anterior-to-posterior gradient may arise from genetic mutations, including spinocerebellar ataxias type 1 [4] and 6 [5], late infantile neuronal ceroid lipofuscinosis [6], saposin C deficiency, a rare cause of Gaucher Disease [7], ataxia telangiectasia [8], and Niemann-Pick disease types A/B [9] and C [10]; sporadic disorders, including multiple system atrophy [11] and chronic epilepsy [12]; toxins, including alcohol [13], cytosine arabinoside [14], methotrexate [15]; hypoxia/ischemia [16, 17]; paraneoplastic syndromes [18]; and even normal aging [19]. This pattern is also seen in many spontaneous mouse mutants, including pcd [20], leaner [21], toppler [22], robotic [23], shaker [24], and lurcher [25]; or targeted mutants, such as saposin D knockout [26], prion protein knockout [27], and over-expression of the prion protein related gene doppel [28]. The fact that such a diverse array of insults leads to the same pattern of Purkinje cell death suggests that selective vulnerability of Purkinje cell subpopulations arises not from the initiating event of the disease process, but instead from differential regulation of cellular survival or death pathways in response to these injuries. We hypothesize that the identification of pathways responsible for this phenomenon will yield therapeutic targets broadly applicable to this large class of cerebellar disorders.
As a model for patterned Purkinje cell loss, we have studied murine Niemann-Pick type C1 disease (NPC). NPC is caused by mutations in the genes encoding NPC1 or NPC2 proteins, which are thought to act cooperatively in the efflux of cholesterol from late endosomes (LE) and lysosomes (LY) [29–31]. The consequence of these mutations is the accumulation of cholesterol and glycosphingolipids in the LE/LY compartment, leading to neurodegeneration by mechanisms that are not yet understood [32]. We previously demonstrated that conditional deletion of Npc1 in Purkinje cells leads to cell autonomous degeneration that recapitulates the spatiotemporal pattern of cell loss observed in mice with germline Npc1 deletion [33]. Further, because Purkinje cell death does not cause early mortality in these mice, we were able to follow Purkinje cell survival beyond the typical lifespan of NPC mice. During this period, the population of surviving Purkinje cells in lobule X remained stable, while neurodegeneration continued to progress in lobules II-IX, thus highlighting the strong resistance of these cells to degeneration. Given the cell autonomous nature of Purkinje cell loss in NPC, we hypothesized that this selective vulnerability arises from intrinsic biological differences that are driven by differential gene expression. To test this notion, here we used a bioinformatics approach to identify genes that are differentially expressed between disease-resistant and vulnerable Purkinje cell populations. To test the biological function of these differentially expressed genes, we used in vitro and in vivo model of NPC and characterized the ability of one of these candidate genes to protect neurons from degeneration.
Using mice containing a conditional null allele of the Npc1 gene, we found that gene deletion in Purkinje cells recapitulates the spatiotemporal pattern of neuron loss observed in mice with global germline deletion of Npc1 (Fig 1A) [10, 33]. The population of surviving Purkinje cells is located within posterior lobules of the cerebellar midline, while age-dependent progressive Purkinje cell loss is observed in anterior lobules (Fig 1B) [33]. We hypothesized that differential gene expression underlies this selective neuronal vulnerability. To search for genes differentially expressed between Purkinje cell subpopulations, we utilized the Allen Brain Atlas (Fig 1C). This resource contains quantitative three-dimensional expression data derived from in situ hybridizations for greater than 20,000 genes in the adult C57BL6/J mouse brain [34]. The complete gene expression dataset was downloaded and used to construct a single expression matrix with spatial coordinates and gene identifiers arrayed on separate axes. This strategy allowed us to treat the data for each location in the brain analogously to a single microarray experiment. The coordinates corresponding to cerebellar lobule X, the location of the most resistant Purkinje cells, and lobules II and III, the most highly vulnerable, were defined as regions of interest (Fig 1B). For analysis, all coordinates falling within one region of interest were treated as replicate microarray experiments. We then used bioinformatics tools developed for microarray analysis to query the Allen Brain Atlas dataset. Differential gene expression between lobules was determined by t-test and Significance Analysis of Microarrays (SAM) [35], followed by manual curation of in situ hybridization images. Manual curation was required to remove false positives created by expression in non-Purkinje cell types and technical artifacts in the archived images.
Initial analysis revealed 234 differentially expressed genes, of which 185 were more highly expressed in lobules II and III and 49 were more highly expressed in lobule X. We next sought to prioritize this list to identify testable candidates with putative roles in promoting or preventing neurodegeneration. The Allen Brain Atlas data, being derived from in situ hybridizations, presented a challenge in this regard, as expression levels were regarded as semi-quantitative. Further, because expression data within each z plane came from the same hybridization experiment, they were not considered statistically independent samples. For these reasons, we were unable to rank the gene list by either the magnitude of differential expression or the degree of significance. Instead, we prioritized genes whose expression differences were most robust and tightly correlated with Purkinje cell survival in midline cerebellar sections. To accomplish this, we only included genes whose expression was undetectable in one region of interest, and whose expression matched or was the inverse of the survival pattern in 20 week old Npc1 flox/-;Pcp2-Cre mice: strong in lobule X, patchy throughout the intermediate and posterior zones, with additional sparing in the caudal aspect of lobule IX and a region spanning the caudal aspect of lobule VI and rostral lobule VII (Fig 1A). This yielded sixteen candidate neuroprotective or susceptibility genes (Fig 2A, Table 1); in situ hybridization images from the Allen Brain Atlas for the candidate genes highly expressed in regions of cell survival are shown in S1 Fig.
We analyzed the functions of these candidate genes and their human orthologs by querying their gene ontology (GO) annotations using AmiGO [36]. The GO Term Enrichment tool revealed significant over-representation (p<0.01) for GO terms containing Prkca, Prkcd, and Plcxd2, members of the phospholipase C—protein kinase C signal transduction cascade, suggesting that this pathway is differentially regulated between regions of interest. AmiGO was also used to query the complete list of GO Biological Process terms associated with candidate genes. In support of our hypothesis that the differentially expressed genes would include regulators of cellular survival and death decisions, 5 genes were associated with cell death related annotations, including “cell death” (GO:0008219, Dbc1 and Hspb1), “apoptosis” (GO:0006915, Pde1b and Prkcd), “negative regulation of apoptosis” (GO:0043066, Hspb1), and “induction of apoptosis by intracellular signals” (GO:0008629, Prkca). Furthermore, the gene product of Sgpp2, sphingosine 1-phosphate phosphatase 2, is likely involved in the regulation of apoptosis as well due to its hydrolysis of sphingosine 1-phosphate [37], a lipid second messenger that is a negative regulator of apoptosis [38]. Finally, we performed an analysis of cellular component annotations to determine the subcellular localization of the protein products of candidate genes (Fig 2B). The vast majority of gene products are localized outside of the endosome-lysosome system, further suggesting that selective vulnerability of Purkinje cell populations arises not from the primary site of pathogenesis in NPC disease, but from responses to cellular stress that take place elsewhere.
We next sought to directly test the extent to which candidate genes influence cell survival in models of NPC disease. For this initial analysis, we chose to study one gene that was over-expressed by lobule X Purkinje cells, heat shock protein beta 1 (Hspb1). This gene has been linked previously to neurodegeneration, as mutations in human HSPB1 cause some cases of Charcot-Marie-Tooth disease and distal hereditary motor neuropathy [39]. Additionally, HSPB1 regulates multiple events that influence neuronal viability, including stability of the actin cytoskeleton, protein folding, reactive oxygen species (ROS), and apoptosis [40], and its robust expression has been documented in surviving Purkinje cells from Npc1-/- mice [10].
We initially sought to confirm that Hspb1 expression in mutant mice with active disease matched the pattern predicted by the Allen Brain Atlas. Strong expression of Hspb1 was detected in lobule X Purkinje cells of Npc1 flox/-;Pcp2-Cre mice at 7 weeks of age, prior to the significant Purkinje cell degeneration (Fig 3A). In contrast, Hspb1 was undetectable in the more susceptible Purkinje cells of lobules II and III. To determine whether HSPB1 functions as an inhibitor of cell death pathways in NPC cell models, we knocked down its expression using siRNA. We initially treated HeLa cells with U18666A, a small molecule which induces lipid trafficking defects similar to those seen in NPC disease by binding to NPC1 and inhibiting cholesterol export [41, 42]. Knockdown of HSPB1 in U18666A-treated cells, but not in vehicle controls, led to a significant increase of caspase activity (Fig 3B). Likewise, HSPB1 knockdown in NPC patient fibroblasts significantly increased the percentage of cells with chromatin condensation, while HSPB1 knockdown had no effect on control fibroblasts (Fig 3C). These results are consistent with a model in which HSPB1 prevents the induction of cell death in response to the intracellular lipid trafficking defects caused by NPC1 deficiency.
To initially test the role of HSPB1 in the survival of neurons, the cell type critical for NPC disease neuropathology [33, 43, 44], we utilized a neuronal culture model. Primary cortical neurons treated with U18666A develop filipin-positive lipid inclusions and progressive degeneration, and have been used previously to model NPC disease [45, 46]. Neurons treated with U18666A demonstrated progressive degeneration, and exogenous over-expression of HSPB1 almost completely prevented this death (Fig 3D). To probe the mechanism of this effect, we took advantage of the fact that serine phosphorylation is critical for HSPB1-mediated protection against neuronal damage in vitro and in vivo [47]. Mutation of these residues to alanine (non-phosphorylatable) or aspartate/glutamate (phosphomimetic) has been widely used to study phosphorylation state-dependent properties of HSPB1 [40]. Transduction of U18666A-treated neurons with the phosphomimetic HSPB1-3E recapitulated the neuroprotective effects of wild-type HSPB1, while non-phosphorylatable HSPB1-3A was inactive (Fig 3D). We conclude that the neuroprotective effects of HSPB1 in NPC cell models are mediated by the phosphorylated species.
We next sought to determine whether HSPB1 over-expression impacts Purkinje cell survival and motor impairment in NPC mice. To accomplish this, we generated mice deficient in Npc1 only in Purkinje cells by using a previously characterized conditional null allele [33]. Cre recombinase expression driven by the Pcp2 promoter initiated around postnatal day 6 and was present in all Purkinje cells by postnatal days 14–21 [48]. Therefore, this strategy enabled post-developmental as well as cell-type restricted deletion of Npc1. Expression of the hemagglutinin (HA)-tagged human HSPB1 cDNA transgene was driven by the chicken β-actin promoter and cytomegalovirus enhancer. These transgenic mice express exogenous HSPB1 in brain, spinal cord, heart, muscle, liver, kidney, lung, and pancreas, and exhibit normal reproductive patterns, longevity and behavior [49]. We determined the behavioral effect of HSPB1 over-expression on Npc1 deficiency by measuring the time to traverse a balance beam. Purkinje cell specific null mutants (Npc1 flox/-;Pcp2-Cre), but not littermate controls (Npc1 flox/+;Pcp2-Cre), displayed a progressive, age-dependent behavioral impairment beginning at 10 weeks (Fig 4A), consistent with our previous study [33]. HSPB1 over-expression significantly rescued motor performance in mice at 10 and 15 weeks of age (Fig 4A). Previous work has demonstrated that this motor task is a sensitive measure of Purkinje cell loss in Npc1 deficient mice [33]. To determine the extent to which HSPB1 over-expression improved neuron survival, we examined the density of Purkinje cells in the cerebellar midline of mice at 11 weeks. This analysis revealed that HSPB1 over-expression significantly rescued Purkinje cell density in posterior (lobules VIII-X) but not anterior cerebellar lobules (Fig 4B and 4C–4J). Purkinje cell rescue in posterior lobules was confirmed by immunofluorescence staining for calbindin, a marker of Purkinje cells (Fig 4G versus 4I). This rescue was associated with the expression of HA-tagged HSPB1 transgene (Fig 4H versus 4J). Transgene expression was also noted in anterior lobules, suggesting that HSPB1 over-expression alone was insufficient to account for effects on neuron survival. The HSPB1 transgene did not alter the accumulation of ubiquitinated proteins or filipin-positive unesterified cholesterol in Purkinje cells of posterior lobules (S2 Fig). We conclude that exogenous HSPB1 protects Purkinje cells in posterior lobules and delays the onset of behavioral impairment, without altering the aberrant accumulation of proteins or cholesterol.
To further explore the basis of the beneficial effects of HSPB1 on select Purkinje cell subpopulations, we first evaluated whether the transgene was uniformly expressed. HA staining of the cerebellar midline confirmed diffuse reactivity of Purkinje cells in 7 week old Npc1 flox/-;Pcp2-Cre, HSPB1 mice (Fig 5A). We next considered the possibility that HSPB1 was differentially activated in cerebellar lobules. Because phosphorylation of HSPB1 influences its ability to promote neuronal survival in vitro (Fig 3E), we examined HSPB1 phosphorylation state in Purkinje cells using phospho-HSPB1 [pS15] immunofluorescence. Strikingly, only Purkinje cells in posterior lobules were positive for phospho-HSPB1 (Fig 5B) despite the fact that the transgene was diffusely expressed (Fig 5A). Intriguingly, our expression analysis identified restricted expression of the HSPB1 kinase PKCδ [50–52] to Purkinje cells in the posterior lobules (Fig 2A, Table 1), a finding that was confirmed by immunofluorescence staining (Fig 5C). Taken together, these data indicated that phosphorylation of HSPB1 was tightly associated with Purkinje cell rescue in animals expressing the transgene.
We sought to additionally explore the functional importance of HSPB1 phosphorylation in mediating cell survival in models of NPC. Prior studies have shown that PKCδ phosphorylates HSBP1 at Ser-15 and Ser-86 to reduce apoptosis [50–52], suggesting that these two proteins may act together to promote cell survival. To determine whether this pathway was active in cellular models of NPC, we knocked down the expression of PKCδ with targeted siRNA and then treated cells with U18666A. We found that diminished PKCδ expression significantly increased the sensitivity of cells to U18666A-mediated toxicity (Fig 6A and 6B), similar to the effect of HSPB1 gene knockdown (Fig 3B and 3C). To evaluate in vivo activity of the phosphorylated form of HSBP1, we used an adeno-associated virus serotype 2 (AAV2) vector to over-express phosphomimetic HSPB1-3E. Transgene and control viral vectors were injected into the deep cerebellar nuclei of Npc1 flox/-;Pcp2-Cre mice at 6 weeks and animals were examined four weeks post-infection. Gene delivery as visualized with the 6x-myc tag was strong and consistent in the central and posterior lobules of the cerebellar midline. Quantification of Purkinje cell density confirmed a significant rescue in the central lobules VI and VII, as well as in the posterior lobule VIII, of mice expressing HSPB1-3E compared to controls (Fig 6C and 6D and S3 Fig). As Purkinje cell survival was not significantly rescued in these central lobules by transgenic expression of wild type HSPB1 (Fig 4B), we conclude that the phosphorylated form of HSPB1 was active in promoting Purkinje cell survival in the NPC cerebellum.
Our over-expression studies demonstrated that HSPB1 delays motor impairment and Purkinje cell loss in posterior cerebellar lobules. We next sought to determine the effects of Hspb1 knockdown in the NPC mouse cerebellum. The feasibility of this approach was supported by prior work demonstrating that Hspb1 null mice are viable and fertile, without obvious morphological abnormalities [53]. To accomplish gene knockdown, we used an AAV2 vector to produce a short hairpin RNA (shRNA) driven by the U6 promoter. Hspb1 shRNA was cloned into an AAV2 shuttle plasmid (pFBAAV/mU6mcsCMVeGFP). To initially confirm knockdown efficiency, NIH3T3 cells were transfected to express non-targeted (NT) or Hspb1 shRNA, heat shocked, and analyzed by western blot (Fig 7A). These targeted and control shRNA clones were then used for virus generation, and injected into the deep cerebellar nuclei of Npc1 flox/-;Pcp2-Cre mice at 7 weeks. Animals were examined six weeks post-infection. At this time point, calbindin staining for Purkinje cells was markedly diminished in the posterior cerebellar lobules of mice receiving Hspb1 shRNA (Fig 7B). We confirmed viral transduction of remaining Purkinje neurons by GFP staining and assessed knockdown efficiency by Hspb1 staining. We observed diffuse GFP reactivity of Purkinje cells in mice expressing NT and Hspb1 shRNA, whereas Hspb1 staining was specifically diminished by Hspb1 shRNA (Fig 7C). Quantification of Purkinje cell density confirmed a significant exacerbation of neuron loss in central and posterior cerebellar lobules (lobules VII-IX) of mice expressing Hspb1 shRNA (Fig 7D). Furthermore, although Purkinje cell density was not altered in lobule X, Hspb1 knockdown significantly diminished soma size (Fig 7E). These data indicate that Hspb1 knockdown exacerbates Purkinje cell degeneration due to NPC1 deficiency.
Many progressive neurological diseases are characterized by the selective vulnerability of neuronal populations, yet mechanisms underlying this phenomenon remain poorly characterized. Here, we sought to identify potential modifier genes that influence the susceptibility of neurons to disease. Using NPC disease as a model for the study of selective neuronal vulnerability, we demonstrate that one of the candidate genes we identified, HSPB1, promotes neuronal survival in cellular model systems through a mechanism that likely involves phosphorylation-dependent inhibition of apoptosis. Additionally, we show that HSPB1 over-expression in vivo slows the progression of motor impairment and diminishes cerebellar Purkinje cell loss. The neuroprotection from Npc1 deficiency afforded by HSPB1 over-expression in mice was associated with HSPB1 phosphorylation and expression of the kinase PKCδ. We confirmed the modulatory effect of Hspb1 on Purkinje cell degeneration in vivo, as knockdown by Hspb1 shRNA significantly enhanced neuron loss. This effect of Hspb1 gene knockdown was particularly robust, resulting in Purkinje cell degeneration in posterior lobules (VII-IX) that approached the severity observed in anterior cerebellar lobules. Although diminished Hspb1 expression did not trigger Purkinje neuron loss in lobule X, we observed a significant decrease in soma size, a compensatory change reported in other degenerative ataxias that influences membrane excitability [54]. These results highlight the use of bioinformatics tools to uncover pathways leading to neuronal protection in neurodegenerative disorders.
HSPB1 is a multifunctional protein with documented roles in actin stability, protein folding, oxidative damage, and apoptosis [40]. Interestingly, HSPB1 is a direct inhibitor of apoptosis at multiple levels, through binding and sequestration of cytochrome c [55] and caspase-3 [56], and inhibition of Bax activation [57] and DAXX signaling [58]. The phosphorylation state required for most of these activities is unknown, with the exception of DAXX inhibition, which requires phosphorylated HSPB1 [58]. Recently, phosphomimetic mutants of HSPB1 were shown to protect against a broad array of apoptosis-inducing stimuli, while non-phosphorylatable mutants showed no protection against some stimuli and only mild protection against others, suggesting that anti-apoptotic activities of HSPB1 are primarily attributable to the phosphorylated species [59]. Although both phosphorylated and dephosphorylated HSPB1 have chaperone activity [60, 61] and prevent oxidative damage [62], it is less likely that these functions play a primary role in exerting beneficial effects in NPC models. This conclusion is based on the tight association that we observed between HSPB1 phosphorylation and neuroprotection, and the finding that neuronal rescue is not associated with diminished accumulation of ubiquitinated proteins. Instead, we favor a model in which HSPB1 acts through an anti-apoptotic mechanism. Our findings suggest that strategies that promote HSPB1 expression or phosphorylation may diminish the rate of cerebellar degeneration in NPC disease.
Interestingly, we also identified the expression of PKCδ in disease-resistant Purkinje cells. While this kinase has been shown to phosphorylate Hspb1 at Ser-15 and Ser-86 to reduce apoptosis [50–52], it is possible that other kinases also contribute to the regulation of Hspb1 activity in the cerebellum. To initially explore this possibility, we examined the expression of kinases that have been previously reported to phosphorylate HSPB1. This includes protein kinase D (PKD) [63], mitogen-activated protein kinase-activated protein kinase-2 and 3 (MAPKAPK2/3) [64, 65] and p38 mitogen-activated protein kinase (p38 MAPK) [66, 67]. According to information in the Allen Brain Atlas, PKD1, PKD2, and PKD3 and MAPKAPK3 are not expressed in the cerebellum, while MAPKAPK2 and p38 MAPK are expressed by all Purkinje neurons. The association between restricted expression of PKCδ, the occurrence Hspb1 phosphorylation, and the pattern of Purkinje neuron survival prompts us to favor PKCδ as an important regulator of the survival benefits mediated by Hspb1. Moreover, we note that the gene encoding a phosphatidylinositol-specific phospholipase C, Plcxd2, is also expressed by resistant Purkinje neurons in the posterior cerebellar lobules (Table 1). This observation raises the possibility that both the regulatory components and effectors of this pro-survival pathway are preferentially expressed by the subset of disease-resistant Purkinje neurons.
Our identification of candidate disease modifying genes relied on in situ hybridization data available in the Allen Brain Atlas. Published studies have similarly mined data from this public database to uncover biologically important gene expression variation [68, 69]. For guidance in our study, we looked to tools developed for the analysis of microarray data, where studies of differential gene expression are commonplace. Several caveats exist when applying our strategy to Allen Brain Atlas data. First, this method is heavily dependent upon manual curation as standard statistical tests yielded high false positive rates. These were variably due to signals generated by other cell types that fell within or adjacent to the region of interest, or artifacts and noise on the in situ hybridization images. Second, while the majority of differentially expressed genes were identified by both t-test and SAM, others were found only by one method. Therefore, it was necessary to combine the use of both approaches, and it remains possible that some differentially expressed genes were not discovered by either. To streamline future studies, a more robust method for working with Allen Brain Atlas data may need to be developed. Despite these technical limitations, our study provides proof of concept for the use of Allen Brain Atlas data to identify therapeutic targets in neurodegenerative diseases.
All animal procedures were approved by the University of Michigan Committee on the Use and Care of Animals (protocol number PRO00006114). Euthanasia of mice was performed by anesthesia overdose followed by induction of bilateral pneumothorax or removal of vital organs.
Antibodies used in this study were anti-GAPDH (Santa Cruz, sc-25578), anti-calbindin (Sigma-Aldrich, c2724), anti-HSPB1 (abcam, ab5579), anti-Hspb1 (Enzo Life Sciences, SPA-801), anti-HSPB1 phospho-Ser 15 (Novus, NBP1-60864), anti-PKCδ (Fisher, BDB610397), anti-hemagglutinin (HA) (Covance, 16B12), anti-GFP (Novus, NB 100–1770), anti-Hsp90 (Santa Cruz, sc-7947) and anti-ubiquitin (Dako, Z0458).
Mice containing the Npc1 floxed (exon 9) [33] and null alleles [70], and transgenic mice expressing the Cre transgene driven by the Pcp2 promoter [71] were generated and genotyped as described previously. Transgenic mice over-expressing hemagglutinin tagged human HSPB1 were from Dr. Jacqueline de Belleroche (Imperial College, London, UK) [49]. All lines were backcrossed to C57BL6/J for ≥10 generations. All animal procedures were approved by the University of Michigan Committee on the Use and Care of Animals.
All cell lines were cultured at 37°C with 5% CO2. HeLa cells were maintained in DMEM (Gibco, 11965–092) supplemented with 10% FBS, 1X penicillin, streptomycin, and glutamine (Gibco, 10378–016). Human skin fibroblasts GM03123 from an NPC patient and GM08399 from an age and sex matched control (Coriell Cell Repositories) were maintained in MEM (Gibco, 10370–021) supplemented with 15% FBS, 1X penicillin, streptomycin, and glutamine (Gibco). To manipulate HSPB1 expression, cells were transfected with ON-TARGETplus SMART pool human HSPB1 or non-targeting control (Dharmacon). HeLa cells were transfected using the DharmaFECT reagent (Dharmacon), according to the manufacturer’s instructions. Fibroblasts were transfected by electroporation with the Lonza Nucleofector normal human dermal fibroblast kit. To reduce PKCδ expression, HeLa cells were transfected with ON-TARGET plus SMART pool PKCδ siRNA (Dharmacon, L-003524-00-0005) or ON-TARGET plus non-targeting pool (Dharmacon, D-001810-10-05), using TransIT-X2 (Mirus).
The Expression Energy Volume for each gene in the Allen Mouse Brain Atlas was downloaded via the Allen Brain Atlas API [34]. These data were then reorganized into a single expression matrix and filtered to include locations corresponding to the regions of interest, cerebellar lobules X and II/III, and extending laterally 1400 microns from the midline. This data matrix was then loaded into TM4 MultiExperiment Viewer software [72], in which differential expression between regions of interest was determined by Student’s t-test and Significance Analysis of Microarrays (SAM) [35]. The top 1000 genes returned by each method were manually verified by direct inspection of in situ hybridization data on the Allen Brain Atlas website in midline and several adjacent sagittal sections. Criteria for validation were (1) expression present in the Purkinje cell layer in at least one region of interest, and (2) differential expression between regions of interest.
Cells were lysed in RIPA buffer (Thermo Scientific) containing Complete protease inhibitor (Roche) and Halt phosphatase inhibitor (Thermo Scientific). Samples were electrophoresed through a 10% SDS-PAGE gel, and then transferred to nitrocellulose membranes (BioRad) using a semidry transfer apparatus. Primary antibodies were anti-HSPB1 (1:1000), anti-Hsp90 (add dilution) and anti-GAPDH (1:5000). HRP-conjugated secondary antibodies were from BioRad. Blots were developed using ECL (Thermo Scientific) or TMA-6 (Lumigen) chemilluminescent reagents, following manufacturers’ protocols.
Total RNA was isolated from HeLa cells using TRIzol (Invitrogen). cDNA was synthesized using the High Capacity cDNA Archive Kit (Applied Biosystems). Quantitative real-time PCR was performed on 100 ng of cDNA in triplicate, using primers and probes for PKCδ (cat # 4453320) and 18S rRNA (Applied Biosystems). Threshold cycle (Ct) values were determined using an ABI Prism 7900HT Sequence Detection System. Relative expression values were normalized to 18S rRNA.
Caspase-3 activity in HeLa cells was determined by assaying DEVDase activity in cell lysates using the ApoTarget caspase 3 / CPP32 fluorimetric protease assay kit (Biosource) according to the manufacturer’s instructions. Fluorescence was measured using a SpectraMax Gemini EM plate reader (Molecular Devices). NPC fibroblasts were stained with Hoechst (Immunocytochemistry Technologies). Cells were counted in five randomly selected fields per transfection at 200x magnification and scored for chromatin condensation. The viability of primary mouse cortical neurons and HeLa cells was determined by XTT assay (Cell Proliferation Kit II, Roche). XTT reagent and activation reagent were mixed at a ratio of 50:1 and added to cultures. After incubating for 4 hrs at 37°C, absorbance at 490 nm and 650 nm was measured using a SpectraMax Gemini EM plate reader (Molecular Devices).
Cortices from P0 C57BL6/J mouse pups were dissected free of meninges, minced, and then dissociated and cultured as described previously [73]. Neurons were plated in poly-D-lysine (Millipore) treated 96-well plates at a density of 6x104 cells per well. Cytosine arabinoside (Sigma) was added to the culture media the following day at a final concentration of 5 M to prevent glial growth. U18666A was added at 2.5 μg/ml at 7 div to induce lipid storage.
A lentiviral expression clone of human HSPB1 with a C-terminal FLAG tag was obtained from Genecopoeia. Serine-to-alanine and serine-to-glutamate mutations were introduced at serines 15, 78, and 82 using the QuikChange Lightning Multi Site-Directed Mutagenesis kit (Stratagene). Wild type HSPB1, HSPB1-3A, HSPB1-3E, and empty vector plasmids were packaged into feline immunodeficiency virus (FIV) vectors by the Iowa Vector Core. Viral infection of cultured primary neurons was performed at 10 MOI, followed by a 75% media change four hours after infection. For in vivo gene over-expression, HSPB1-3E with a 6x-myc tag was cloned into pFBAAV/CMVmcspA. For gene knock-down, Hspb1 shRNA was designed and cloned by the Iowa Vector Core. The target region in the Hspb1 sequence was analyzed using siSPOTR and potential miRNA target sequences of 21 nucleotides were identified based on low GC content and other factors, as described [74]. Five potential target sequences were cloned in pFBAAV/mU6mcsCMVeGFP. Knockdown efficiency was tested in NIH3T3 cells. The most efficient plasmid was used in producing AAV2/1mU6miHspb1-CMVeGFP or AAV2/1CMVHSPB1-3E triple transfection virus. Non-targeted virus, AAV2/1mU6-miSafe-CMV eGFP, was used as a control. Before injection, virus was dialyzed at 4°C for 3hrs against 7,000 MWCO Slide -A-Lyzer mini-dialysis units (Thermo Scientific) in a custom buffer formulation distributed through the Gene Transfer Vector Core in University of Iowa.
Stereotaxic administration of AAV2 was performed on 7 week-old Npc1 flox/-, Pcp2-Cre mice placed under anesthesia using a mixture of O2 and isoflurane (dosage 4% for induction, 1.5% maintenance). Mice received bilateral intracerebellar injections (either one or two sites/hemisphere) of virus. For each injection, ~1.4 x 1012 vg/ml of virus (4 μl) was delivered to the medial or lateral cerebellar nucleus at an infusion rate of 0.5 μl/min using a 10-μl Hamilton syringe (BD). One min after the infusion was completed, the micropipette was retracted 0.3 mm and allowed to remain in place for 4 min prior to complete removal from the mouse brain. When two injections sites per hemisphere were used, anterior–posterior coordinates were calculated separately for medial and lateral injection into each cerebellar hemisphere. The coordinates for the medial injection were -6.4 mm anterior-posterior, ±1.3 mm medial-lateral and 1.9 mm dorsal-ventral as measured from bregma. The coordinates for the lateral injection were -6.0 mm anterior-posterior, ±2.0 mm medial-lateral and 2.2 mm dorsal-ventral as measured from bregma. When a single injection per hemisphere was used, the coordinates for the injection were -6.2 mm anterior-posterior, ±0.9 mm medial-lateral and 2.2 mm dorsal-ventral as measured from bregma.
5 μm sections from brains embedded in paraffin were deparaffinized with xylenes and ethanol. Sections were boiled in 10 mM sodium citrate, pH 6, for 10 min for antigen retrieval. After washing with water, sections were blocked with 5% goat serum and 1% BSA in PBS for 1 hr and then incubated in primary antibody (calbindin 1:500, PKCδ 1:50, Hspb1 1:100, HA 1:200, phospho-Hspb1 1:50, GFP 1:100, ubiquitin 1:200) diluted in 1.5% blocking solution overnight at 4°C. Sections were subsequently incubated in secondary antibodies conjugated to Alexa Fluor 594 or 488 for 2 hrs and mounted with mounting medium including DAPI (Vector Lab, H-1200). Images were captured on an Olympus FluoView 500 Confocal microscope.
Motor function was measured by balance beam test. Mice at 4 weeks of age were trained on three consecutive days to cross a 6 mm wide beam suspended at 50 cm. Mice were then tested in triplicate at 5, 10, 15 and 20 weeks of age. Data are reported as average time to traverse the beam, allowing a maximum of 25 sec and scoring falls as 25 sec.
Purkinje cell density was quantified in midline sagittal sections stained with hematoxylin and eosin or calbindin staining. Purkinje cells were recognized as large cells with amphophilic cytoplasm, large nuclei with open chromatin and prominent nucleoli that were located between the molecular and granular layers or as calbindin positive cells. The number of cells was normalized to the length of the Purkinje layer, as measured by NIH ImageJ software. For analysis of Purkinje cell soma size, calbindin staining was used to define the cell soma. The cell soma was selected and measured by NIH ImageJ, and pixel size was converted to μm2 using the scale bar as a calibration standard.
Mouse brain tissue embedded in Optimal Cutting Temperature (OCT) compound (Tissue-Tek) was sectioned at 10 μm in midline. The sections were rinsed with PBS, and fixed with 4% paraformaldehyde for 30 min. After washing with PBS, the sections were incubated with 1.5 mg/ml glycine for 10 min, washed with PBS and stained with 0.05 mg/ml filipin and 10% FBS in PBS for 2 hrs at room temperature. Filipin images were captured with the UV filter set on an Olympus FluoView 500 confocal microscope. Representative images are from one of three mice per genotype.
Statistical significance was assessed by unpaired Student’s t test (for comparison of two means) or ANOVA (for comparison of more than two means). The Newman-Keuls post hoc test was performed to carry out pairwise comparisons of group means if ANOVA rejected the null hypothesis. Statistical analyses were performed using the software package Prism 6.02 (GraphPad Software). P values less than 0.05 were considered significant. Statistical analysis of gene expression data was performed using TM4 MultiExperiment Viewer software [72]. For these calculations, statistical significance was determined using Student’s t-test with Bonferroni correction for multiple comparisons and Significance Analysis of Microarrays (SAM) [35].
|
10.1371/journal.ppat.1005777 | Altered Memory Circulating T Follicular Helper-B Cell Interaction in Early Acute HIV Infection | The RV254 cohort of HIV-infected very early acute (4thG stage 1 and 2) (stage 1/2) and late acute (4thG stage 3) (stage 3) individuals was used to study T helper- B cell responses in acute HIV infection and the impact of early antiretroviral treatment (ART) on T and B cell function. To investigate this, the function of circulating T follicular helper cells (cTfh) from this cohort was examined, and cTfh and memory B cell populations were phenotyped. Impaired cTfh cell function was observed in individuals treated in stage 3 when compared to stage 1/2. The cTfh/B cell cocultures showed lower B cell survival and IgG secretion at stage 3 compared to stage 1/2. This coincided with lower IL-10 and increased RANTES and TNF-α suggesting a role for inflammation in altering cTfh and B cell responses. Elevated plasma viral load in stage 3 was found to correlate with decreased cTfh-mediated B cell IgG production indicating a role for increased viremia in cTfh impairment and dysfunctional humoral response. Phenotypic perturbations were also evident in the mature B cell compartment, most notably a decrease in resting memory B cells in stage 3 compared to stage 1/2, coinciding with higher viremia. Our coculture assay also suggested that intrinsic memory B cell defects could contribute to the impaired response despite at a lower level. Overall, cTfh-mediated B cell responses are significantly altered in stage 3 compared to stage 1/2, coinciding with increased inflammation and a reduction in memory B cells. These data suggest that early ART for acutely HIV infected individuals could prevent immune dysregulation while preserving cTfh function and B cell memory.
| The HIV-specific T cell memory response diminishes rapidly even after the initiation of anti-retroviral treatment (ART), and there is no control of viral rebound if treatment is interrupted. Restoration or preservation of memory T cells or B cells with treatment, to allow for control of virus replication after ART is stopped, is therefore very important. CD4+ T cells, in particular T follicular helper (Tfh) cells, have a major role in mediating antiviral immunity by providing help to B cells, which is key to a strong and efficient anti-HIV antibody response. The unique RV254 cohort provided the best setting to analyze immune responses during very early acute HIV, as the study was able to enroll individuals that were infected for less than 2 weeks and initiated treatment approximately 1–2 days after recruitment. We aimed to study the capacity of memory circulating Tfh (cTfh) cells to promote B cell help in acute HIV infection, and found the interaction to be dysfunctional in the later stage compared to the very early stages, accompanied by increased levels of proinflammatory cytokines and a reduction in regulatory cytokines. High levels of plasma viremia correlated with low cTfh-mediated B cell antibody production in later stage acute individuals; and memory B cells were significantly decreased but could be restored with ART, compared to chronically infected individuals, who could not normalize this compartment compared to healthy individuals. Overall, we show that the cTfh- B cell interaction and B cell memory compartment is altered in late stage acute infection, mainly attributed to an increase in inflammation and skewing of the response away from helper to proinflammatory. Identifying individuals for treatment in the earliest stages of acute infection, prior to immune damage, could preserve cTfh function and the anti-HIV B cell response, therefore reducing the chances of viral rebound upon the cessation of ART.
| The progressive nature of immune dysfunction in HIV infected individuals has implied that early treatment could play a critical role in reducing immune defects and in preserving T and B cell memory responses against HIV infection [1–3]. Despite the fact that antiretroviral treatment (ART) has been pivotal in reducing the viral burden in persons infected with HIV, the concurrent decline in the HIV-specific T and B cell memory response poses a great risk, as treatment interruption can lead to a loss in the control of viremia [4–7]. The need to identify immune parameters that are associated with preservation of the memory response during HIV infection is therefore important to providing clues for therapies going forward.
Even with ART, low level HIV replication in lymphoid tissues has been shown to maintain a state of chronic immune activation [8]. B cell hyperactivation, a hallmark feature of HIV infection, is characteristically evidenced by elevated serum immunoglobulin [9, 10], and also includes changes within the circulating B cell compartment, some of which cannot be reversed by ART as evident in chronic individuals [11]. These changes include an increase in their activation, proliferation, immature and terminal differentiation markers [12–16], as well as a reduction in CD27+ memory cells [17–19].
CD4 T follicular helper cells (Tfh) are specialized in providing help to B cells and support B cell maturation and differentiation to long-lived plasma cells in the germinal center (GC) [20]. It therefore follows that if an efficient HIV-specific B cell response is to be achieved, Tfh function must be preserved. With access to human lymphoid tissue limited, there has been an increased interest in the study of CD4+CXCR5+ T cells in the blood known as memory circulating Tfh (cTfh), that are also very efficient at inducing B cell differentiation and providing B cell help; and much progress has been made to characterize and understand their biology [6, 21–23]. We have previously shown that germinal center Tfh cells in HIV positive lymph nodes are dysregulated and that the B cell response is severely reduced compared to those from HIV negative lymph nodes [24]. We have also recently shown that HIV-associated microenvironment can affect the differentiation and phenotype of cTfh cells, and that these cells from chronic aviremic individuals treated in very late stages (3–6 months after transmission) are dysfunctional in providing B cell help compared to elite controllers or healthy individuals [6]. It is therefore plausible that individuals who undergo very early ART could preserve their CD4+ Tfh function, and that phenotypic perturbations of T and B cell populations that are characteristic of acute HIV infection are better resolved in very early treated individuals. As the HIV-associated microenvironment likely affects the Tfh program [6], it is also possible that in very early acute HIV-infected individuals, there is less immune activation and therefore fewer inflammatory signals, which would reduce the risk for adverse effects on Tfh phenotype and function.
The unique RV254/SEARCH010 Thai cohort is strategic in that HIV infected participants are recruited during the earliest stage of acute infection and receive ART almost immediately [25, 26]. This population of individuals offers the best setting to accurately analyze the benefits of early ART on immune preservation and function. Using cTfh cells as surrogate B-cell helper T cells, we examined T-helper mediated function in stage 1/2 (4thG stages 1 and 2 grouped) (early acute; median average 12 and 17 days respectively from history of exposure) and stage 3 (4thG stage 3) (late acute; median average 18 days from history of exposure) [27]; and investigated the phenotype and immune activation status of cTfh and memory B cells before and post ART. We hypothesized that cTfh-mediated B cell help in untreated late acute individuals will be reduced compared to very early acute individuals and lead to an increased proinflammatory environment.
We initially examined whether the frequency and phenotype of cTfh cells (CXCR5+CXCR3-) in peripheral blood from stage 1/2 are similar to those from stage 3 at week 0. Similar frequencies of cTfh cells were observed in all stages (P = 0.62, S1A and S1B Fig). In addition, there was no significant difference in the expression of inhibitory (PD-1) (P = 0.65), costimulatory (ICOS) (P = 0.85) or activation markers (CD25, CD38) (P = 0.27 and P = 0.82 respectively) when we compared stage 1/2 to stage 3 (S1C–S1F Fig). These results suggest that there are no apparent differences in the phenotype of cTfh cells between the two groups. Additionally we did not observe any differences in the induction of CD38 expression on cTfh cells (P = 0.55) between the two stages after SEB stimulation (S1G Fig). Thus we studied the function of cTfh cells between the two groups in the context of interaction with memory B cells.
To investigate cTfh function at different stages very early in acute infection, coculture assays of sorted cTfh cells and autologous-sorted resting memory B cells (CD21+CD27+) were performed. Cocultures of cTfh cells from stage 3 elicited significantly less help to B cells compared to cocultures of cTfh cells from stage 1/2 individuals, as evidenced by reduced levels of total IgG (P = 0.01, Fig 1A), HIV-specific IgG (P = 0.02, Fig 1B); and reduced number of absolute B cells in the coculture at day 7 (P = 0.002, Fig 1C). Similar numbers of cTfh cells survived through the end of the 7-day coculture in both stage 1/2 and stage 3 (P = 0.41, Fig 1D). The data indicated that the blunted antibody response in the coculture of stage 3 was independent of T cell survival and points to a defective T/B cell interaction. Nonetheless, these results signify preservation of the cTfh- dependent B cell and HIV- specific IgG response in the earliest acute stages of HIV infection compared to the later stages. This is further demonstrated by a positive correlation between absolute B cell numbers and HIV-specific IgG production in coculture (P = 0.046, R = 0.5473; Fig 1E). There were no significant differences observed in HIV-specific IgG (P = 0.60, Fig 1F) production from cocultures of less efficient B-helper T cells (CXCR5+CXCR3+) suggesting that the defect in stage 3 is specific for cTfh/B cell interaction. Of note, although CXCR5+CXCR3- cTfh cells are mostly enriched in a subset that provides efficient B cell help, the CXCR5+CXCR3+ T cell subset can also provide help, but to a lesser degree (S2 Fig), hence they are a more appropriate control population compared to CXCR5- non-cTfh cells.
To evaluate whether our observations on reduced B cell responses in stage 3 cocultures may be related to a skewed cTfh program, we analyzed cocultures of cTfh cells from stage 1/2 and 3 for cytokine production. In the cTfh cocultures of stage 3 individuals, there was a significant increase in the Th1 cell chemotactic cytokine RANTES [28] (P = 0.04, Fig 2A) and a trend towards an increase in the acute phase proinflammatory cytokine TNF-α [29] (P = 0.07, Fig 2B), compared to stage 1/2 individuals. Concurrently, levels of the regulatory cytokine IL-10 [30] were significantly reduced in stage 3 individuals when compared to those of stage 1/2 (P = 0.01 Fig 2C). We also observed an apparent increase in levels of the proinflammatory cytokines IL-1β (P = 0.299), IL-6 (P = 0.142), IFN-γ (P = 0.110) and MIP-1α (P = 0.343) in stage 3 compared to stage 1/2, although this did not reach statistical significance (S3A–S3D Fig). Importantly, we observed a negative correlation between TNF-α and HIV-specific antibody levels in the coculture supernatants (P = 0.003, R = -0.715; Fig 2D) suggesting a negative impact of TNF-α on cTfh/B cell interaction. On the other hand, we observed a trend towards a positive correlation between IL-10 and HIV specific antibody levels in the coculture supernatants (P = 0.062, R = 0.480; Fig 2E) suggesting a positive impact of IL-10 on cTfh/B cell help. This is not surprising as IL-10 is a known potent growth and differentiation factor for B cells and can be secreted by Tfh cells [31, 32]. Overall, these results support the notion that delayed ART could cause alterations in cTfh program towards a Th1 profile as in stage 3 which could affect cTfh-dependent antibody output and phenotype.
We found that plasma viral load was significantly lower in stage 1/2 compared to stage 3 (P< 0.0001, Fig 3A). This suggested that elevated levels of plasma viral load could be associated with cTfh cell mediated B cell defects. To ascertain this association, we measured the correlation between plasma viral load in both stages and antibody levels in the cocultures of cTfh and memory B cells. Interestingly, we found a significant negative correlation between plasma viral load and total IgG in the coculture supernatants (P = 0.039, R = -0.543; Fig 3B). This data provides evidence that elevated viral load can skew cTfh function towards a non-helper program. Similarly, we observed a significant negative correlation between plasma viral load and HIV-specific IgG antibody levels (P = 0.047, R = -0.525; Fig 3C). As expected, a negative correlation trend was observed between plasma viral load and absolute number of B cells in the coculture (P = 0.097, R = -0.484; Fig 3D). Importantly we found a significant negative correlation between viral load and IL-10 concentration in the supernatants of the coculture assay (P = 0.001, Fig 4E). On the other hand, a trend towards a positive correlation was observed between viral load and RANTES (P = 0.086, R = 0.461; Fig 3F) and TNF-α (P = 0.097, R = 0.446; Fig 3G) supernatant levels. Overall these results indicate that prolonged plasma viral load, as in seen in stage 3, could be responsible for blunting cTfh mediated B cell help.
We determined whether elevated plasma viral load in stage 3 was associated with a hyperactivated antibody response and altered B cell differentiation. Although individuals in both groups had similar levels of plasma total IgG from the time of ART initiation and throughout treatment (Fig 4A), stage 1/2 individuals had significantly less HIV-specific IgG from as early as day 5 of ART compared to stage 3 (P = 0.003, Fig 4B), indicating increased B cell hyperactivation in late acute infection that could be partly due to antigen overload. It is possible that this hyper activation of B cells occurs in extrafollicular areas as damage to lymph nodes occurs very early in acute infection as has been previously shown [33]. This could contribute to hypergammaglobulinemia which could directly affect cTfh/B cell interactions as we observed in stage 3 (Fig 1A and 1B). In the HIV viremic setting, plasmablasts are markers of uncontrolled proliferation and exhaustion and have been shown phenotypically and functionally to be increased and to secrete antibodies ex vivo [9, 13, 34]. Frequencies of plasmablasts were significantly increased in stage 3 individuals prior to ART initiation, compared to stage 1/2 individuals (P = 0.014, Fig 4C and S4A Fig); and that this increase in percentages of plasmablasts correlate positively with an increase in viral production (P = 0.002, R = 0.757, Fig 4D). This suggests that the enhanced frequency of plasmablasts and the HIV-specific antibody response, could be in part due to the enhanced inflammatory microenvironment at stage 3 when compared to stage 1/2.
HIV infection leads to early and progressive depletion of peripheral CD27+ resting memory B cells [11, 18]. This loss occurs from the onset of acute infection [33, 35] and persists even after prolonged ART [11]. We studied whether resting memory B cells (S4B Fig) are affected at different stages of acute HIV infection. We observed lower frequencies of resting memory B cells in stage 3 individuals compared to stage 1/2 (P = 0.003, Fig 5A), and this decrease correlated with an increase in plasma viral load (P = 0.004, R = -0.495, Fig 5B).
Importantly, administration of ART at stage 1/2 and at stage 3 preserved the frequency of resting memory B cells (similar frequencies to HIV negative subjects) (stage 1/2, P = 0.210 and stage 3, P = 0.135; Fig 5C). Whereas, initiation of ART much later at the chronic stage (months after transmission), was not able to adequately restore the resting memory B cell compartment (p = 0.068, Fig 5C). It is worth noting that the frequency of resting memory B cells at week 0 of stage 1/2 is similar to that of healthy individuals (P = 0.406); however the frequencies at week 0 of stage 3 are significantly reduced compared to healthy subjects (P = 0.0004) and is similar to that from chronic (P>0.999). In addition, we did not see significant differences between the stages, prior to ART, in the frequency of total CD19+ B cells, IgD+ resting naïve B cells (CD21+CD27-), activated memory B cells (CD21-CD27+) or exhausted tissue-like B cells (CD21-CD27-) between stage 1/2 and stage 3 (Fig 5D–5G).
In this study, we found that the cTfh-B cell interaction is significantly altered in later stage acute HIV infection. Supernatants from these cocultures showed that total and HIV-specific IgG from stage 3 individuals not yet on ART are significantly reduced compared to stage 1/2 individuals. We further found that the reduced levels of HIV-specific antibodies coincided with reduced B cell numbers in the coculture supernatants. Our previously published reports have demonstrated that the HIV microenvironment can negatively impact Tfh mediated B cell help [6, 24]. Specifically, we showed cTfh cells obtained from HIV infected individuals who were treated relatively late (months after transmission) displayed a Tfh1 inflammatory phenotype that is not capable of providing help to B cells. We therefore sought to investigate how early this impairment occurs and examine the underlying mechanisms using the unique RV254 cohort of acute HIV-infected subjects. This work has showed that cTfh cells are impaired in their ability to provide B cell help as early as stage 3 (~18 days) of acute HIV infection. The reduction in absolute numbers of B cells after coculture of cells from stage 3 individuals correlated with increased HV-specific IgG secretion, which provides further evidence that the B cell response, as evidenced by antibody production, is directly linked to cTfh-mediated help. These results indicate that the cTfh-mediated B cells response is better preserved at the earliest stage of HIV infection and that identifying individuals very early on during HIV infection has the potential to preserve cTfh-dependent B cell responses.
To further characterize the dysregulation in cTfh-B cell interaction, we performed a coculture assay where we replaced cTfh cells from stage 3 (defective) with cTfh from healthy individuals (effective) in coculture with memory B cells (obtained from stage 3). We found that replacing cTfh showed signs of increased total and HIV-specific antibody levels (S5A and S5B Fig), as well as increased IL-10 and reduced RANTES levels (S5C and S5D Fig). This does suggest that reducing inflammation and hence reversing impairment of the cTfh-mediated B cell response can potentially be achieved by normalization of cTfh function.
In order to further discern the contribution of memory B cells and determine the extent of B cell impairment prior to coculture with cTfh, we measured memory B cell proliferation and antibody production following polyclonal stimulation with CpG oligonucleotide. Our results indicated that similar proliferation profiles between the two stages and apparent non-significant increase in total IgG production in stage 1/2 when compared to stage 3 (S6A and S6B Fig). Overall, this suggested some level of intrinsic memory B cell defect that could contribute to impaired antibody response in stage 3. Of note, we observed similar expression of the survival molecule Bcl-2 on memory B cells in both stages (S6C Fig). Overall our results suggest that the impairment of the cTfh-B cell interaction is primarily due to impaired interaction of cTfh and memory B cells, however we cannot overlook the contribution of intrinsic memory B cell defect to this phenomenon.
Additionally, we found that cocultures of cTfh cells obtained from patients at stage 3 are polarized towards a proinflammatory phenotype. Our results imply that this skewing of cTfh cells towards a non-cTfh program is associated with elevated levels of viral load observed at stage 3 when compared to stage 1/2. This has been confirmed by the significant correlation between viral load at week 0 and the impaired cTfh function as measured by the production of total and HIV-specific antibodies in culture supernatants. It is not clear from our study, however, whether this impairment is a direct effect of the virus or indirectly through viral-induced microenvironmental changes. Previously published data has suggested elevated proinflammatory cytokine levels including RANTES and TNF-α during chronic HIV infection [28, 29]. This is corroborated by findings from our own studies showing that cTfh from chronic individuals can be polarized to a more Th1 profile through the production of IL-2, TNF-α and IFN-γ [6]. cTfh cells can be further phenotyped into Tfh1, Tfh2 and Tfh17 subsets that provide help to B cells at varying levels and are characterized depending on their expression of CXCR3 and CCR6 [21, 36–38]. The impact that proinflammatory cytokines have on cTfh polarization in this setting would be of great interest.
The regulatory cytokine IL-10 has been shown to be important in B cell proliferation and maintenance [30–32, 39–41], as well as promoting Tfh-dependent responses in HIV and in other disease models [23, 24, 42]. In fact, lack of IL-10 production by CD4+ T helper cells in X-linked lymphoproliferative disease (XLP) patients impaired the development of B cell responses [43]. The increase in IL-10 and decreased proinflammatory cytokines, as well as the positive correlation between IL-10 and coculture HIV-specific IgG we observed in stage 1/2 individuals compared to stage 3, suggests that IL-10 could not only be important in regulating the early acute cytokine environment during HIV infection, but also in promoting cTfh-dependent B cell responses.
Rapid initiation of ART for maintaining low viremia and preserving immunity is a key objective of the RV254 study. We determined that stage 1/2 individuals had significantly lower viral load than stage 3 and consequently less plasma HIV-specific IgG and terminally differentiated B cells compared to stage 3 individuals. We know one of the characteristic effects of increased viremia on the B cell response is hypergammaglobulinemia [9, 10]. In addition, it has not been fully established whether cTfh interactions with memory B cells represent exactly what happens in the follicular microenvironment. However, based on our results, it could be that the hypergammaglobulinemia we observe in the plasma in stage 3 is partly due to antigen overload and hyper activation of the B cells which could occur in the extrafollicular areas of lymphoid tissues as damage to the lymph node has been shown to occur very early in acute infection [33]. This may play a greater role in driving B cell hyperactivation in stage 3. Overall, these results confirm a role for increased viremia in not only increasing plasma antibody responses but also mediating inflammation and skewing immune memory in acute HIV infection, resulting in impaired T cell- B cell interactions.
Resting memory B cells play a critical role in eliciting strong humoral immunity and were significantly lower in stage 3 than in stage 1/2. Interestingly, the frequencies of these resting memory B cells in both stage 1/2 and stage 3, but not chronic individuals were restored to levels similar to what is observed in healthy subjects. These results strongly suggest that intervention at the earliest stages of infection can maintain the integrity and survival of resting memory B cells, which are significantly capable of strengthening humoral defenses and very important in preventing HIV disease progression. This is similar to other studies comparing the effects of early ART versus initiation at the chronic phase, where the resting memory B cell subset was better restored with early treatment [44, 45].
As we investigate what occurs in the periphery, we are cognizant that this environment may not necessarily represent what occurs in GCs in secondary lymphoid tissue. Using cTfh and memory B cells as surrogates allows us to interpret what may be going on generally and these results will help is to refine our future investigations. However, based on our previous work [6, 24] and the results in this study, we propose the notion that cTfh-mediated B cell interaction in early acute infection is compromised in an inflammatory microenvironment. The exact mechanism of dysregulation is still not entirely clear, as both cell types appear to be affected during acute infection, so further investigation is needed. However, the idea to initiate ART very early in acute infection is validated as we observe that initiation of ART in the first week of infection leads to an environment with less inflammation. We hypothesize that this may prevent irreversible deleterious changes to the immune system, and may preserve immune function.
Of note, there has been pioneering but not extensive work done on the timing of ART for controlling viral rebound after treatment interruption. This study as well as work done by others, has shown that the timing of ART is important to preserve T cell-dependent B cell responses [44, 45] and mucosal T cell responses [46], all of which provide more insight into how preservation of memory could be important in overcoming viral rebound after ART interruption. Other studies have concluded there is no evidence for post-treatment control after ART interruption, even in the earliest participant group investigated (8.6 weeks from the estimated time of infection) [47]. Their observations however support what we believe occurs in acute HIV infection, and highlights the significant differences between participants our studies, in that initiating ART at 8.6 weeks from the estimated time of HIV infection is too late to preserve memory and hence prevent viral rebound after treatment interruption. As we were able to show in our current study, there is preservation of humoral function and memory in individuals that started ART from as early as 12 days from the history of exposure, with significant perturbations observed in individuals starting treatment at 18 days from the history of exposure. We therefore conclude that initiating ART at the earliest stages of HIV infection is key to preserving immune memory, and could prove most effective in controlling viral replication after treatment interruption.
The RV254/SEARCH 010 study is an ongoing study in Bangkok, Thailand (clinicaltrials.gov NCT00796146), recruiting participants at the earliest stages of acute HIV infection. Peripheral blood mononuclear cells (PBMCs) from participants in the study cohort were used for these experiments (S1 and S2 Tables). Individuals with non-reactive HIV IgG were enrolled as previously described [48, 49], and later reclassified into stages 1, 2 or 3 using the 4th generation immunoassay (4thG IA) staging system [27]. Definitions for the stages were stage 1 (HIV RNA+, 4thG IA-, 3rd generation (3rdG) IA-, stage 2 (HIV RNA+, 4thG IA+, 3rdG IA-), and stage 3 (HIV RNA+, 4thG IA+, 3rdG IA+, Western Blot negative or indeterminate). Participants subsequently initiated ART at a median (IQR) of 2 (1–2) days from enrollment. Individuals from stage 1 and stage 2 were grouped together to represent stage 1/2, and compared to stage 3; as evaluating circulating Tfh function indicated that stage 2 clustered more similarly with stage 1 in the context of B cell help compared to stage 3. PBMCs from chronically HIV-infected (CHI) and HIV negative (HIV-) participants were also obtained for analysis from other protocols (SEARCH 011 and RV304/SEARCH 013 respectively) (S3 and S4 Tables).
The RV254/SEARCH010 study (NCT00796146) was approved by the Institutional Review Boards (IRBs) of Chulalongkorn University, Walter Reed Army Institute of Research (WRAIR), University of California at San Francisco (UCSF), Yale University, University of Hawaii (UH), University of Texas Medical Branch at Galveston, University of Sydney, and Centre Hospitalier de l’Université de Montréal Le Comité d’Ethique de la Recherche. The RV304/SEARCH013 study (NCT01397669) was approved by the IRBs of Chulalongkorn University, Walter Reed Army Institute of Research (WRAIR), University of California at San Francisco (UCSF), and Yale University. The SEARCH011 study (NCT00782808) was approved by the IRBs of Chulalongkorn University, University of California at San Francisco (UCSF), and University of Hawaii (UH). Initiation of ART was voluntary and done as part of enrollment. All participants gave written informed consent for all the studies detailed above.
PBMCs from HIV-infected patients were incubated with fluorochrome-conjugated antibodies for at least 15–20 minutes at 4°C or on ice, protected from light. Samples were sorted on a BD™ FACSAria II for: cTfh cells defined as CD3+CD4+CD45RA-CXCR5+CXCR3-, less efficient helper T cells denoted CXCR5+CXCR3+ and defined as CD3+CD4+CD45RA-CXCR5+CXCR3+; and resting memory B cells defined as CD3-CD19+CD10-CD27+. Sorted cTfh or CXCR5+CXCR3+ T cells were placed in coculture with sorted memory B cells at an equal ratio with 100ng/mL of staphylococcal enterotoxin B (SEB) (Toxin Technology) in RPMI 1640 (Corning) with 10% fetal bovine serum (Access Biologicals) and 1% penicillin/streptomycin (Gibco). SEB was required to induce T cell activation and promote T/B crosstalk [21]. Of note, total IgG measured in T-B cultures without SEB had very low levels (S2 Fig). Supernatant and cells were harvested at day 7 for subsequent analysis. The absolute number of cells were counted by flow cytometry with Flow-count Fluorospheres (Beckman Coulter) using the protocol recommended by the manufacturer.
PBMCs from HIV-infected or HIV-negative patients were incubated with fluorochrome-conjugated antibodies for at least 15–20 minutes at 4°C or on ice, protected from light. The following fluorochrome-conjugated anti-human antibodies were used: CD3 (HIT3α), CD4 (RPA-T4), CD10 (HI10a), CD20 (2H7), CD25 (BC96), CD27 (O323), CD38 (HIT2), PD-1 (EH12.2H7), and CXCR3 (G025H7) were all from BioLegend. CD19 (HIB19), CXCR5 (RF8 B2), ICOS (DX29) and CD21 (B-ly4) were from BD Biosciences and CD45RA (2H4LDH11LDB9) from Beckman Coulter. LIVE/DEAD Fixable Dead Cell Stain (Life Technologies) was used to gate on live cells; and in some cases both LIVE/DEAD stain and Annexin V (BD Biosciences) were used. Samples were acquired on a BD LSR II.
Total and HIV-specific IgG was measured by ELISA on culture supernatant and plasma as previously described [6]. Total IgG was detected by coating 96-well Immulon 2HB plates (Thermo Fisher Scientific) with anti-human monoclonal IgG (Mabtech, clone MT91/145) at a concentration of 1μg/mL in phosphate buffered saline (PBS) overnight at 4°C. The next day plates were washed 5 times (unless otherwise stated) with wash buffer (PBS + 0.05% Tween 20), and subsequently left to block with wash buffer for 1 hour at room temperature. Plates were then washed before the addition of sample and IgG standards at different dilutions, for 1 hour at room temperature. HIV Env-specific antibody responses were detected by coating 96-well high-binding half-area plates (Greiner Bio-One) with 1μg/mL recombinant HIV-1 envelope protein (ProSpec) in PBS and incubating overnight at 4°C. The next day plates were washed 5 times (unless otherwise stated) with wash buffer (PBS + 0.05% Tween 20), and subsequently left to block with wash buffer for 1 hour at room temperature. Plates were then washed before the addition of samples at different dilutions for 2 hours at room temperature. For the standard curve, human HIV immunoglobulin was used, and was obtained through the NIH AIDS Research and Reference Reagent Program, Division of AIDS, NIAD, NIH (Cat no. 3957) via Dr. Luiz Barbosa, NABI and National Heart, Lung and Blood Institute. For both assays, following sample incubation and washing, the plates were left to incubate with 1μg/mL of anti-human IgG-biotin (Mabtech, clone MT78/145) for 1 hour at room temperature. The wash step was repeated and the plates incubated with streptavidin-HRP (Mabtech) for 1 hour at room temperature. An extra wash was added to the last wash step before adding 100μL of TMB substrate (Sigma-Aldrich) to each well until a color change was observed. The reaction was stopped by the addition of 50μL of 1M H3PO4. The OD values were read at 450nm using a spectrophotometer (SpectraMax Plus, Molecular Devices).
IL-1β, IL-6, IL-10, IFN-γ, MIP-1α, RANTES and TNF-α were measured in 7 day culture supernatant from T and B cell coculture experiments using the Bio-Plex Pro™ Human Cytokine Luminex Kit (Bio-Rad) exactly according the manufacturer’s instructions. Samples were added neat to the plate and cytokine standards were supplied by the manufacturer. The samples were acquired using a Bioplex-200 system and the data analyzed on the BioPlex Manager Software (Bio-Rad).
All data were analyzed using Graphpad Prism v10. Unpaired Student’s t test (Mann Whitney) was used when comparing two groups. The unpaired multiple t test and non-parametric one-way ANOVA (Dunn’s) test was used when comparing more than two groups to each other. Comparisons by correlations were analyzed using nonparametric spearman correlation. P values of less than 0.05 were considered statistically significant.
|
10.1371/journal.ppat.1005991 | A Trematode Parasite Derived Growth Factor Binds and Exerts Influences on Host Immune Functions via Host Cytokine Receptor Complexes | The trematode Fasciola hepatica is responsible for chronic zoonotic infection globally. Despite causing a potent T-helper 2 response, it is believed that potent immunomodulation is responsible for rendering this host reactive non-protective host response thereby allowing the parasite to remain long-lived. We have previously identified a growth factor, FhTLM, belonging to the TGF superfamily can have developmental effects on the parasite. Herein we demonstrate that FhTLM can exert influence over host immune functions in a host receptor specific fashion. FhTLM can bind to receptor members of the Transforming Growth Factor (TGF) superfamily, with a greater affinity for TGF-β RII. Upon ligation FhTLM initiates the Smad2/3 pathway resulting in phenotypic changes in both fibroblasts and macrophages. The formation of fibroblast CFUs is reduced when cells are cultured with FhTLM, as a result of TGF-β RI kinase activity. In parallel the wound closure response of fibroblasts is also delayed in the presence of FhTLM. When stimulated with FhTLM blood monocyte derived macrophages adopt an alternative or regulatory phenotype. They express high levels interleukin (IL)-10 and arginase-1 while displaying low levels of IL-12 and nitric oxide. Moreover they also undergo significant upregulation of the inhibitory receptor PD-L1 and the mannose receptor. Use of RNAi demonstrates that this effect is dependent on TGF-β RII and mRNA knock-down leads to a loss of IL-10 and PD-L1. Finally, we demonstrate that FhTLM aids newly excysted juveniles (NEJs) in their evasion of antibody-dependent cell cytotoxicity (ADCC) by reducing the NO response of macrophages—again dependent on TGF-β RI kinase. FhTLM displays restricted expression to the F. hepatica gut resident NEJ stages. The altered fibroblast responses would suggest a role for dampened tissue repair responses in facilitating parasite migration. Furthermore, the adoption of a regulatory macrophage phenotype would allow for a reduced effector response targeting juvenile parasites which we demonstrate extends to an abrogation of the ADCC response. Thus suggesting that FhTLM is a stage specific evasion molecule that utilises host cytokine receptors. These findings are the first to clearly demonstrate the interaction of a helminth cytokine with a host receptor complex resulting in immune modifications that facilitate the non-protective chronic immune response which is characteristic of F. hepatica infection.
| Parasitic worms, helminths, can cause long-lived chronic infection in many hosts that they infection. The liver fluke, Fasciola hepatica, is one such parasite causing global infection of both humans and animals. F. hepatica exerts an influence over the immune system such that it avoids effector mechanisms and prevents the development of effective immunity. Here we characterise a molecule—FhTLM—derived from juvenile parasites that is similar to the regulatory cytokine TGF-β. We show that FhTLM will bind to host TGF-β receptors with a reduced affinity when compared with mammalian TGF-β. Despite this FhTLM can induce Smad2/3 signalling in host leukocytes, which is key to initiating gene transcription. Phenotypically FhTLM causes fibroblasts to slow their growth and replication response resulting in slower wound healing. Importantly FhTLM induces a macrophage phenotype that resembles a regulatory macrophage phenotype identified in other species undergoing helminth infection. Finally we Our work highlights the potential of FhTLM to play important roles in controlling host immunity when initially infected with juvenile parasites, thereby preventing the development of effective immunity.
| The trematode Fasciola hepatica is capable of establishing chronic infection in multiple hosts that can last for many years. In terms of animal infections F. hepatica is highly prevalent within sheep, cattle and goats throughout temperate regions of the globe with varying levels of infection reported from 30%-70%[1]. This problem is compounded by a growing degree of resistance against what has been the drug of choice for combating infection, triclabendazole [2]. F. hepatica however is not solely an animal problem as it also has growing implications for human health with large endemic foci of infection within South America and the Middle East [3]. Crucially, there has now been a reported case of triclabendazole resistant parasites causing human infection [4]. As such F. hepatica has been added to the list of emerging zoonotic diseases [5]. In response to infection an extremely polarised T-helper (Th) 2 response characterised by high levels of IgG1, IL-4 and eosinophilia. Despite the magnitude of this response naturally infected animals fail to develop immunity [6] and efforts at experimental vaccination have thus far demonstrated that a mixed Th2/Th1 profile is required to achieve a reduction in parasite burdens, egg outputs and liver damage [7]. As infection progresses the Th2 response switches to a characteristic regulatory response—this is denoted by high levels IL-10 and TGF-β and suppression of antigen specific T-cells [8].
Much research has identified the nature and mode of action of parasite derived immunomodulators. Chief amongst these is the aforementioned cathepsin L1 which has also been shown to suppress Th1 cytokine secretion. Cathepsin L1 has also been shown to cleave CD4 from the surface of lymphocytes [9] and prevent antibody-dependent cell cytotoxicity (ADCC) from killing newly excysted juvenile (NEJ) parasites [10]; one of the few mechanisms known to kill NEJs. More recently Donnelly and colleagues have shown multiple modes of immunosuppression involving a family of helminth defence molecules, similar to host defence peptides that are capable of suppressing macrophage activation and B-cell cytokine responses [11]; providing evidence parasite for mimicry.
Another family of proteins which has demonstrated conservation between host and multiple parasites is the Transforming Growth Factor (TGF) superfamily [12]. Members of this protein superfamily have been shown to predominantly play roles in body patterning, optimal sexual development and reproductive success. Members of this family have been found in free-living and parasitic worms including Schistosoma mansoni [13], S. japonicium [14], Ancylostoma caninum [15] and Caenorhabditis elegans [16]. We have recently demonstrated that the F. hepatica contains a TGF-like molecule which we termed FhTLM [17]. Previously we have shown that the expression of FhTLM was restricted to the NEJ stage. Further to this we have provided evidence that recombinant FhTLM could enhance motility and survival of NEJs while increasing the rate at which eggs underwent embryonation. Members of the TGF superfamily have been ascribed many roles including within leukocytes. TGF-β is a requirement for the development of both Th17 and Treg cell subsets with the levels IL-6 playing a crucial role in dictating the fate of these cells. TGF-β secreted from Treg cells or other sources can have an anti-proliferative effect on T-cells after TCR stimulation. Macrophage responses to TGF-β are broadly known to result in anti-tumouricidal [18] and anti-inflammatory [19] macrophages. Recently, the effect of TGF-β has been shown to direct the effects of myeloid suppressor cells in Nippostrongylus brasiliensis infection thus controlling Th2 immunity within the lung [20]. TGF-β is also known to be crucial to development of fibrosis, this response can be serve to promote pathology via hypertension during S. mansoni infection [21]. Studies on Heligmosomoides polygyrus suggests that there is a protein(s) which can bind the mammalian TGF-receptor complex and initiate a Smad2/3 signalling program [22]. This H. polygyrus derived antigen was found to upregulate FoxP3 expression within naïve T-cells, directly generating Tregs; a finding that explains the protective effect of H. polygyrus in lung inflammation [23].
Given the above and our own findings with regard to FhTLM we sought to determine if FhTLM could directly interact with the native receptor complex and initiate a phenotype. To begin this process we determined if FhTLM could generate a response signal using a luciferase reporter cell line and physically bind host TGF-β RI and RII. FhTLM initiates a Smad3 signal and can alter the responses of fibroblasts in a TGF-β RI kinase dependent fashion. Moreover when used to activate macrophages the response to FhTLM and the resultant phenotype resembled a regulatory macrophage rather than the helminth-linked alternatively activated macrophage; with high levels of IL-10 and PD-L1 and moderate arginase-1 activity. These processes occurred in a tgf-βRII dependent fashion as demonstrated by siRNA knockdown. Finally, the FhTLM macrophage phenotype was incapable of killing NEJ parasites by ADCC demonstrating that this stage specific parasite protein might elicit non-protective responses from resident cells within the intestinal phase of infection.
We have recently shown that the trematode parasite Fasciola hepatica contains three ligand members of the transforming growth factor superfamily [17]. These include two bone morphogenic proteins (BMPs) and an activin-like molecule which we have terms F. hepatica TGF-like molecule, FhTLM. We have demonstrated a restricted pattern of expression within the parasite with the highest level of expression within the newly excysted juvenile that emerges within the intestine of hosts. To determine if FhTLM is a bioactive molecule similar to TGF-β we assessed its activity in a reporter assay [24]. A dose response analysis suggests that FhTLM can indeed generate a positive luciferase signal and when compared to a TGF-β1 standard curve it would suggest that FhTLM has a lower degree of bioactivity in this assay when compared with mammalian equivalents (Fig 1A). We were also able to demonstrate a similar activity within crude parasite homogenate (LFH) which required higher concentrations to induce comparable responses (S1 Fig). Initial attempts to use a monoclonal antibody to inhibit FhTLM activity did not demonstrate inhibitory capacity. However, a polyclonal anti-sera raised with broad specificity was found to reduce the activity of FhTLM in the same bioassay in a dose dependent manner (Fig 1B). To ensure the effects of FhTLM were dependent on ligand-receptor based interactions we sought to determine if FhTLM could bind either of the bovine TGF-β RI or RII. We cloned and expressed fusion proteins comprised of the bovine TGF-β RI and RII extracellular domain fused to the human IgG1 Fc domain (S1 Table). Using these proteins we performed a comparison between the binding of these fusion proteins to FhTLM and human TGF-β1 (Fig 1C & 1D –note differences in Y-axis scale). Using both fusion proteins we could confirm that human TGF-β1 could bind the bovine receptors RI and RII. More interestingly we also confirmed that FhTLM could cause the binding of both fusion proteins with a greater affinity for TGF-β RII-Fc, which is similar to the reported affinity of TGF-β RII with TGF-β1 elsewhere [25, 26]. Final confirmation of the greater affinity of TGF-β RII with FhTLM was confirmed by repeating the above assays but with the inclusion of increasing concentrations of KSCN after initial addition of fusion proteins to the plate. A greater concentration of KSCN was required to disassociate the interaction between FhTLM with TGF-β RII-Fc when compared with TGF-β RI-Fc (Fig 1E). Moreover when we tested the affinity of FhTLM for either TGF-β RII-Fc or TGF-β RI-Fc in a competition assay we were able to demonstrate that FhTLM was only moderately able to reduce the binding of free TGF-β RII-Fc or TGF-β RI-Fc to immobilised TGF-β1 reducing binding to TGF-β RII-Fc and TGF-β RI-Fc by 46% and 42%, respectively. In comparison TGF-β was able to reduce binding of FhTLM to TGF-β RII-Fc and TGF-β RII-Fc by 50% and 22%, respectively, as comparable doses (Fig 1F & 1G).
The TGF-β receptors are G-protein coupled receptors and after ligand binding heterodimers of TGF-β RI and RII are formed. The resultant phosphorylation of this receptor complex triggers movement of the signalling proteins phosphorylated (p)Smad2/3 into the nucleus where gene transcription is initiated [27]. To determine if FhTLM was capable of driving Smad2/3 signalling after engagement with the receptor complex we used primary bovine peripheral blood mononuclear cells (PBMCs) in a stimulation assay to measure the extent of co-localisation of pSmad2/3 with the nucleus. Given the differences we noted above between activity of recombinant FhTLM in our luciferase assay and the binding data determined from our receptor fusion protein assays we conducted a dose response curve for bovine macrophages using IL-10 as our readout to determine the optimal working concentration (S2 Fig). PBMCs were stimulated for between 3 and 4hrs, fixed and stained with DAPI (Fig 2A top row), anti-pSmad2/3-FITC (Fig 2A middle row) and the percentage of co-localisation (or pSmad2/3 positive cells) was determined (Fig 2A bottom row). TGFβ clearly drives pSmad2/3 signalling in this setting with 28% (±4.1%) of cells being pSmad2/3 positive at 3hrs post-stimulation with a small, but not significant decrease, at 4hrs to 22.4% (±5.3%). Interestingly when we initially examined cells stimulated for 3hrs with FhTLM we found only 10.7% (±1.9%) of cells positive which was not significantly increased when compared to our controls [6.8%(±2.5%)]. However, when we extended our analysis to 4hrs we found that FhTLM induced pSmad2/3 in 15.8% (±2.9%) of cells which was significantly different when compared to controls [5.4% (±3.4%)]. A recent analysis of the bovine il10 promoter within our lab has indicated a role for GATA1 in driving il10 expression. Furthermore GATA1 has been in implicated in anti-helminth immunity in previous studies using Nippostrongylus brasiliensis [28] and S. mansoni [29]. We subjected PBMCs to a 4hr stimulation as above and then determined the levels of GATA1 co-localisation and found that both TGF-β (11%±1.3%) and FhTLM (6.1%±1.4) could induce significantly more GATA1 co-localisation in PBMCs when compared to controls (4.9%±1.1). Our results clearly demonstrate that FhTLM can act to induce both direct and indirect transcription factors in primary host PBMCs.
Characterisation of TGFβ demonstrated a profound an anti-proliferative and developmental effect on multiple cell types [30–35]. In an effort to ascribe a phenotype to the effects of FhTLM we performed CFU assays using the NIH 3T3 fibroblast line. Cells were seeded at a density of 6 cells/petri dish and incubated for 10 days. This was done in the presence of TGF-β or increasing concentrations of FhTLM (2.5, 25, 200 ng/mL). CFUs that formed were counted and our results clearly show a significant decrease in number of CFUs that formed when higher doses of FhTLM were used (Fig 3A). 25ng/mL of FhTLM was sufficient to reduce the number of CFUs to comparable level as seen on those incubated with TGF-β (128.3±6.173 vs. 129.7 ±19.10). To further determine the effects of FhTLM on fibroblast activity we performed in vitro scratch assay/wound closure experiments. Confluent cells were scratched and imaged before incubation for 24hrs with TGF-β or FhTLM. After 24hrs the wounds were imaged and total area determined (Fig 3B). Expressing this area as % wound closure we found that both TGF-β and FhTLM significantly reduced wound closure (P<0.01) [Ctrl = 67.77%±12.17 vs. TGF = 47.95±4.906 vs. FhTLM = 44.47±7.235]. Decreased arginase-1 has been previously shown to correlate with delayed wound resolution [36]. We determined the levels of arginase-1 in wounded cultures 24hr after incubation (Fig 3C). We confirmed that in cultures treated with TGFβ or FhTLM the levels of arginase-1 were decreased suggesting an ability of FhTLM in altering arginase-1 activity. We next using chemical inhibition to block the kinase activity of TGF-β RI to determine if the effects of FhTLM are indeed TGF-β dependent and specific. As can be seen in Fig 3B when cells were co-cultured with the inhibitor SB- 431542 [37] during formation of fibroblast CFUs the effect of both TGF-β and FhTLM were abolished. These results suggest that FhTLM can both bind and initiate a specific signal via the TGF-β receptor complex.
Having demonstrated a role for FhTLM in modulating cell growth and regulating the arginase levels of these cells we sought to determine if the effects of FhTLM on arginase levels could be extended to macrophages. The classical and alternative pathways for macrophage activity can be broadly defined in terms metabolism of L-arginine either using iNOS or arginase following stimulation with LPS/IFN-γ or IL-4, respectively [38]. We produced bovine macrophages using purified blood derived CD14+ monocytes; these were then stimulated with IL-4, LPS, TGF-β or FhTLM. Our initial analysis confirmed that IL-4 and LPS act to induce arginase or NO, respectively (Fig 4A & 4B). While FhTLM can induce a slight increase, similar to that seen in response to induced TGF-β, in the levels of arginase-1 this was not significant when compared with IL-4 but was significantly different when compared with the control. Similarly LPS induced NO but IL-4, TGF-β or FhTLM induced marginal levels in comparison. To further determine if FhTLM could alter the phenotype of macrophages we measured IL-10 and IL-12. Only LPS stimulation induced significantly more IL-12 when compared to controls (Fig 4D). However, both IL-4, TGF-β and FhTLM induced IL-10, raising levels significantly above controls (Fig 4C). While TGF-β tended towards higher levels of IL-10 induction when compared with L-4 this was not significant. Reports of a regulatory macrophage phenotype in helminth infection or in response to helminth products suggest that this cell population is distinct from alternatively activated macrophages [39, 40] and in some cases it would appear to be independent of arginase-1 [39]. These reports suggest that upregulation of mannose receptor (MR) and PD-L1 serve as proxy markers for these cells. We determined MR levels in stimulated cells by immunofluorescence or PD-L1 levels of qPCR (Fig 4E & 4F). FhTLM significantly upregulated MR expression compared to controls and the number of cells becoming MR+ after stimulation was comparable to IL-4 treatment. However TGF-β, in comparison to both IL-4 and FhTLM, was induced a higher number of MR+ cells [~60% vs 20%] (Fig 4E). When we examined PD-L1 expression only FhTLM and TGF-β were able to induced PD-L1 above levels seen in controls, again with TGF-β inducing higher levels of PD-L1 compared to FhTLM [30 fold change vs 10 fold change] (Fig 4F).
To determine what host factor confers specificity on interaction of FhTLM with bovine macrophages we employed siRNA directed against the tgf-βRII. Primary bovine macrophages were as standard, however at point prior to normal cytokine stimulation cells were transfected with target siRNA or with scrambled siRNA, thereafter we measured changes in tgf-βRII levels over a 72hr time period. Our results show we could reliably suppress tgf-βRII mRNA levels up to 24hrs post transfection (Fig 5A). We then proceeded to stimulate macrophages 6 hours after knock-down, with FhTLM or TGF-β. Our results show that absence of tgf-βRII mRNA at the time of cytokine treatment results in a reduced levels of PD-L1 being upregulated in response to both FhTLM and TGF-β with knock-down reducing PD-L1 upregulation by 47% and 90%, respectively (Fig 5B). While in the case of IL-10 induction, measured 54 hours post-transfection and 48hrs post-stimulation, we same similar reductions in the levels of IL-10 in response to FhTLM and TGF-β (Fig 5C). These results strongly support the conclusion that the effects of FhTLM are dependent on the host cytokine receptor complex TGF-β RI and RII, especially when taken in together with our findings above showing that ALK inhibition also negated the effects of FhTLM.
We have previously shown that FhTLM is selectively expressed within the newly excysted juvenile (NEJs) stage of F. hepatica infection [17]. NEJs are thought to be resident within the intestine for only a number of hours before entering the peritoneal cavity and continuing on their migration to the liver. Within the intestine, multiple type-2 effector responses could be active including antibody-dependent cell cytotoxicity (ADCC). ADCC is one of the few mechanisms shown to actively kill F. hepatica and has previously been shown to target NEJs, both in vitro and in vivo; moreover it has previously been shown to be a target of parasite immune evasion mechanisms [10, 41, 42]. We sought to determine if FhTLM altered the macrophage component of this process to benefit parasite survival. Using naïve donor macrophages we incubated cells and NEJs in the presence of either immune or naïve sera. Thereafter viable parasites were counted by visual inspection, as can be seen in Fig 6A the presence of macrophages plus immune sera, but not non-immune sera, resulted in the death of NEJs and was accompanied by induction of NO (Fig 6B). When macrophages were incubated with TGF-β or FhTLM prior to addition of NEJs and sera a different outcome was recorded. As can be seen in Fig 6A both TGF-β and FhTLM reduced the capacity of immune sera to induce ADCC-mediated death of NEJs, this was also accompanied by a loss of NO production (Fig 6B).
We next determined if the effect of FhTLM was dependent on the TGF-β RI, via kinase activity, pre-incubation of cells with the inhibitor SB-431542. Pre-incubation of macrophages with inhibitor prior to TGF-β or FhTLM reversed the negative effect on cells and rescued the ADCC response to NEJs in the presence of immune sera in comparison to cells pre-incubated with vehicle only (Fig 6C). To accompany this we also found that the NO response was restored in both TGF-β and FhTLM cultures in cells pre-incubated with inhibitor but not vehicle only (Fig 6D). Thus our findings demonstrate that FhTLM alters the host macrophage phenotype, via TGF-β RI and RII, to evade ADCC killing of NEJs.
Multiple studies have shown that chronic infection with F. hepatica can be long lived and accompanied by parasite-specific and non-specific immunosuppression [8, 43, 44]. As the host progresses from a Th2 type response to a more regulatory response it is assumed that secretion of IL-10 and TGF-β increases as a result of either T-cell phenotypic changes or the expansion of T-regulatory cells. The parallel expansion of both Th2 and Treg populations has previously been demonstrated in S. mansoni [45] and recent work in murine models of F. hepatica have also demonstrated that infected mice generate a FoxP3+ population of cells as infection progresses [46]; however the relevance of this to ruminant immune responses remains to be determined. All of these mechanisms would appear to be driven, or at executed, by the host as a balance to minimise immunopathology. Here we demonstrate that F. hepatica can utilise a host-exogenous cytokine/growth factor, FhTLM—previously shown to have developmental functions, to direct the host immune system. Indeed this mechanism fits with previous patterns identified whereby for optimal host and parasite survival a balance of immune effector mechanisms must be maintained, allowing parasite survival while avoiding immunopathology [47].
Our data demonstrates that FhTLM is capable of directly engaging TGF-β RI and RII in an ELISA format which makes use of fusion proteins of RI and RII. This data helps to explain the activation by FhTLM of the luciferase reporter. Indeed there are prior reports of parasite derived molecules driving activation in this assay system previously [22]. What is apparent from these data is that FhTLM has a) a higher affinity for TGF-β RII over RI and b) has a lower binding capacity for either receptor when compared with human TGF-β1; this is further evidenced by our competition data. This effect is seen again in the results of our luciferase assay whereby FhTLM was needed in the ng/mL range to generate signals equivalent to those seen in TGF-β in the pg/mL range. The higher affinity of TGF-β for TGF-β RI and RII is also seen in mammalian systems [25, 26] and highlights the conservation between the two ligands despite being phylogenetically distinct, demonstrating the close association amongst the parasite and host. To ensure that the immunomodulatory capacity of FhTLM was not due to binding but not initiating signalling from the receptor complex we examined the canonical intracellular signalling molecule Smad2/3. Using immunofluorescence we can see that not only does FhTLM drive p-Smad2/3 translocation to the nucleus but also does so at slower rate when compared to mammalian TGF-β, again in line with our findings from our binding experiments. It took 4hrs of stimulation with FhTLM to drive a p-Smad2/3 signal distinct from background when compared with the higher translocation rate and shorter time period required in response to TGF-β.
TGF-β is pleiotropic in terms of its effects being implicated in developmental processes, anti-proliferative in a context dependent fashion, responsible for fibrosis, and key to differentiation of two distinct CD4+ T-helper subsets. TGF-β is known to be anti-proliferative in terms of fibroblasts [30, 34, 35] and we confirm this finding here and also demonstrate that FhTLM can cause a similar response which is also dependent on TGF-β RI kinase activity. FhTLM reduced the number of CFUs formed to a similar rate of TGF-β over a 10 day period. Likewise when included as a growth factor in in vitro wound assays we found that FhTLM, like TGF-β, reduced the rate of wound closure. The role of TGF-β in wound responses is still disputed with some reports finding a positive or negative role dependent on the phase of wound resolution in which it is examined. A recent study however determined that arginase-1 was crucial for healing in murine model wounding [36] and here we found that in parallel with reducing wound closure FhTLM also reduced the cellular levels of arginase-1. Campbell et al [36] found the reduction in arginase-1 levels also resulted in a reduction in pro-inflammatory cell recruitment, including macrophages. The benefits in delayed wound healing during a parasite infection are not apparent however given the migratory nature of F. hepatica infection, it could be speculated that reducing the rate at which wounds or migratory paths caused by the parasite are healed may confer a benefit to the parasite. The parasite excysts within the intestine and migrates into the peritoneal cavity where it gains access to liver before moving to bile ducts [48]. As the parasites migrate through the intestine they formed a cavity around themselves which would require healing post migration. The delay in healing may increase the rate of successful migrating NEJs; it is already knwon that NEJs secrete proteases to digest surrounding tissue to facilitate their movement [49].
Given the context specific effects of TGF-β we sought to determine its effects on other cell types. Studies suggest that helminth infection [39] and a recombinant helminth immunomodulator [40] can induce a macrophage phenotype that is distinct from the alternatively activated phenotype that is normally associated with helminth infection [50]. Our data demonstrated a subtle yet significant rise in arginase-1 levels following FhTLM, in contrast to our results in the fibroblast experiments, and no increase in NO and in comparison to the strongly polarising effect that IL-4 has on these readouts the results were not striking and more akin to the response to TGF-β. This pattern concurs with the findings of Smith et al.,[39] who found a helminth elicited macrophage population could protect from colitis but in an arginase-1 dependent manner. We next examined the cytokine profile, IL-10 and IL-12, of these cells we found a more pronounced effect of FhTLM. FhTLM could upregulate IL-10 while also suppressing the expression of IL-12, this has been a reported feature of regulatory macrophages for some time [51]; again this pattern of responses being more similar to TGF-β than IL-4. Moreover mRNA expression of PD-L1 and the number of mannose receptor positive cells were significantly upregulated in FhTLM or TGF-β treated macrophages only. During infection with Taenia crassiceps PD-L1 has been shown to suppress T-cell responses and neutralisation of PD-L1 on macrophages from infected mice abrogated their suppressive capacity [52]. Recently, an Ancanthocheilonema viteae derived immunomodulatory was shown to induce macrophages with high levels of expression of IL-10 and PD-L1 capable of reducing signs of colitis in mice after cell transfer [40]. The mannose receptor (CD206) has previously been shown to be upregulated on regulatory macrophages from a variety of settings [53, 54] including controlling their role in regulating inflammatory cytokine release in Pneumocystis infection [55] and endotoxin-induced lung injury [56]. We found the effects of FhTLM on PD-L1 and IL-10 to be dependent on tgf-βRII expression, as use of siRNA resulted in a loss of their expression following stimulation. Given the effects of FhTLM on macrophages and the restricted expression of FhTLM to NEJs within the parasite itself; we sought to determine the effects of FhTLM on a NEJ targeting effector mechanism—ADCC. ADCC has been shown to kill NEJs when using cells from cattle [41], rats [42], and mice [10] but not sheep [57].This is thought to be as a result of a lack of NO generation in sheep macrophages. Here we demonstrate that bovine macrophages, in the presence of immune sera, kill NEJs and release NO into the supernatant. Moreover the culture of macrophages with FhTLM or TGF-β prior to this assay effectively removed the killing phenotype and reduced the levels of NO. When these assays were repeated in the presence of the TGF-β RI kinase inhibitor the ADCC effect was rescued and parasites were rendered non-viable and NO levels were restored, again showing the effects of FhTLM to be TGF-β receptor dependent.
The expression of TGF-β homologues within helminth parasites has been previously identified [58] however this, to our knowledge, is the first full description of the suppressive effect of a recombinant helminth TGF-β homologue on its host immune system. Our findings indicate a role for FhTLM in the modulation of host macrophages to avoid a well-recognised mechanism of killing ADCC. The complete function of FhTLM during infection has yet to be explored but the on-going development of stable gene silencing techniques in F. hepatica will make this achievable [59]. A loss-of-function approach would be the best method to approach this subject, however there exists a number of hurdles, metacercariae have yet to be successfully treated with RNAi and as such NEJs treated with RNAi would need to be transplanted into the intestines of suitable hosts. The macrophage response to surgery has been shown to tend towards alternative activation, thus attempting to analyse the immune phenotype in such circumstances may prove difficult. A second complicating factor are the parasite-intrinsic effects of FhTLM [17], knock-out of FhTLM may yield a near-lethal or lethal phenotype, for reasons unrelated to host immunity, again complicating the analysis. A system for conditional gene targeting within the parasite metacercariae would best allow for natural infection and thus a faithful analysis of the resulting immune response; however these tools do not yet exist. Implementation of this technology will aid us in answering unresolved questions surrounding exact timing of FhTLM expression within the intestine of hosts, the full range of target cells and whether the effects of FhTLM are confined both physically and temporally confined to the intestine.
We have previously described the cloning and expression of FhTLM [17]. A pET28-based construct (Novagen) was used to express a 6XHis-Tagged protein in BL21 E. coli (Novagen) using kanamycin and chloramphenicol to select for transformed bacteria. Recombinant protein was purified using a Nickel resin column (Sigma-Aldrich). Recombinant proteins were subject to two rounds of phase separation prior to use [60]. To generate the receptor-fusion proteins the bovine TGFβRI extracellular domain sequence from nucleotide (nt) +88 to +331 and TGFβRII extracellular domain sequence from nt +139 to +453 relative to the translation initiation site (+1) were PCR amplified using specific primers. Using a forward primer with a NCOI restriction site incorporated and reverse primer with a Bg1II restriction site incorporated for TGFβRI and forward primer with an EcoRI restriction site incorporated and reverse primer with a Bg1II restriction site incorporated for generation of TGFβRII (See S1 Table). The amplified extracellular domains of TGFβRI was sub-cloned into the NCOI and Bg1II and TGFβRII ED into EcoR1 and Bg1II multi cloning site of the pFUSE-hIgG1-Fc2 vector (InvivoGen, UK) respectively. Following ligation and confirmation of insertion plasmids were used to chemically transform E. coli DH5α cells. The transformed cells were grown on LB agar plates supplemented with 50μg/ml Zeocine (InvivoGen, UK) at 37°C overnight. Plasmid was purified and used to transfect mammalian HEK-293 cells (Invivogen UK) maintained in DMEM (Sigma Aldrich) supplemented with 10% FCS (Sigma Aldrich), 100 μg/ml penicillin, 100 μg/ml streptomycin and grown to 80% confluency. Transfection with recombinant plasmids was carried out using jetPRIME DNA and siRNA transfection reagents (Polyplus-transfection, USA) as per manufacturer’s instructions. The day before transfection cells were seeded into 6 well culture plate at 2x105 cells /well, DMEM medium were added to final volume of 2ml per well and incubated at 37°C overnight 5% CO2. 2 μg of plasmid was diluted into 200 μl of jetPRIME buffer and 4 μl of jetPRIME were added to each well for transfection. The transfection medium were replaced with complete DMEM medium after 4 hrs and incubated at 37°C for 72 hrs. A positive control GFP reporter plasmid, Pc-gfp-c2 (Clontech), was used in parallel to confirm transformation. After 72 hrs the positive controls was assessed under an inverted microscope (LEICA DMIL). Positive cells were cloned under limiting dilution conditions, using zeocine (InvivoGen) as a selective.
96-well plates was coated with 50 μl of hTGF-β1 or FhTLM, concentration indicated on figures, at room temperature overnight. The plate was washed three times with 0.05% Tween/PBS. Additional protein binding site were blocked by adding 200 μl of 4% BSA-PBS and incubatedfor 1 hr at room temperature. TGFβ-RI and RII Fc fusion proteins were used at 1.25μg/mL and supernatant from non-transfected HEK were used as negative control and added to the wells of the plate and incubated for 1 hr at room temperature. HRP-conjugated Anti-human IgG1 Ab (HP6070, Life Technology) at a concentration of 5 μg/ml. Colour was developed with TMB substrate and the reaction was stopped with 1% HCL; absorbance was measured at 450nm using micro-plate reader (LT-4000, Labtech, UK).
To estimate the avidity of the FhTLM interaction with Fc fusion proteins of the bTGFβ-RIED and RIIED, potassium thiocyanate (KSCN) was introduced into the ELISA to disrupt the binding between the bovine receptors and the recombinant FhTLM protein. ELISA was performed as stated above using FhTLM as coating antigen at concentration of (500 ng/ml). After addition and incubation with (1.25 μg/ml) of Fc fusion proteins, different concentrations (1, 2, 3, 4, 5 and 0 M) of KSCN were added to each well and incubated for 1 hr at RT. Thereafter the binding and optical densities were measured as above. To determine the extent of competition between FhTLM and TGF-β in the context of binding to receptor fusion proteins, FhTLM or TGF-β were coated on plates at 400ng/mL. After this the opposite increasing concentrations of competing protein were incubated with the Receptor-Fc fusion in solution at 37°C for 1hr, then added to the plate and the ELISA proceeded as above.
Mink Lung epithelial cells (MLECs—a gift from Prof D Rifkin, New York University) were maintained in T25 flask (Sarstedt) containing 5 ml of Dulbecco’s modified Eagle’s Media (DMEM) (Sigma-Aldrich) supplemented with 10% of heat inactivated fetal calf serum (Sigma-Aldrich), penicillin (100 U/ml), streptomycin (100 U/ml), L-glutamine and 200 μg/ml, G418 (Sigma-Aldrich). For use in luciferase measurements cells were used at a concentration of 1.6x106 cells/ml. The suspension was plated in 96 well tissue culture plates (Sarstedt) 100 μl/well. The culture plate was incubated at 37°C, 5% CO2 overnight for optimal cell attachment. Proteins including a TGF-β standard curve were added to cells in DMEM with 0.1% BSA. Luciferase was measured the luciferase Assay (Promega) on a BMG luminometer as per Abe et al [24].
Whole blood was collected under terminal exsanguination from healthy donor animals under a Home Office regulated schedule 1 procedure. CD14+ cells were isolated and cultured as before [61] IL-4 was used at 20ng/mL, LPS was used at 100μg/mL while FhTLM was either used at the indicated dose or 200ng/mL. Cells were stimulated for 6hr for RNA isolation or 48hrs for collection of supernatants and cell lysates.
RNAi knock-down of tgfβRII (NM_001159566.1) was conducted using the methods of Jensen et al [62] Briefly, siRNA oligos were designed by Sigma Aldrich and used at a final concentration of 100nM. Macrophages were cultured to maturity in 48 well plates at a density of 1 x 105 cells and after 10 days were transfected with JetPrime media was replaced after 24hrs. Cells were tested for knock-down beginning at a further 24hrs after media change this using PCR primers designed against tgfβRII;Forward primer 5’ -GGACTATGAGCCTCCGTTCG- 3’ and reverse primer 5’–GGTTCCAGGAAGCATCGTCA- 3’. Alternatively, once an optimum time post-transfection was selected– 12hrs—stimulation was conducted.
The fibroblast cell line NIH 3T3 (A gift from Dr Janet Daly University of Nottingham) was routinely maintained and to conduct the scratch assay published methods were used. Briefly, 3x105 cells were seeded into 6 well plates and incubated overnight. To scratch the monolayer, a linear scratch was made to the fibroblasts from the top of the well to the bottom using a 20μl pipette tip at time 0; plates were then incubated with the indicated proteins for 24hrs. Images of scratches were obtained using an inverted light microscope set to 5 x magnification. Lecia imaging software (leicra microsystems LTB Milton Keynes UK) was used to acquire digital images. ImageJ, was used to analyse the images (version 1.49v from National Institutes of Health, USA). Scratch areas were measured and compared by a blinded operator independent to the culture treatments for each well.
To conduct the CFU assay 6 cells/well were seeded in a 6-well plate, thereafter proteins were added and plates incubated for 10 days. Plates were stained with 0.5% crystal violet and imaged as above. CFUs were determined per well by an operator blinded to treatments before data analysis. In some experiments cells were co-cultured with SB-431542 a TGF-β RI kinase inhibitor (Tocris) and TGF-β or FhTLM with inhibitor at a final concentration of 5μM. Inhibitor stocks were prepared in DMSO and vehicle controls were prepared using an appropriate comparative dilution of DMSO.
For immunofluorescence staining cells were isolated and stimulated as above but grown on coverslips (Corning). Following stimulation plates were centrifuged at 300xg for 10 min following incubation, to collect cells on the cover slips. Medium were removed and cells were washed three times with 1XPBS. PBMCs were then fixed by in 4% paraformaldehyde for 15 min at room temperature and washed with 1XPBS. Afterward, cells were permeabilized for 10 min at room temperature with 0.5% Triton X-100 in PBS and washed with PBS. Cells were incubated for 1 hr at room temperature with a 1:100 dilution of primary polyconal rabbit anti-GATA1 (Santa Cruz sc-13053) or goat anti-pSmad2/3 (Santa Cruz sc-11769) or mouse anti-mannose receptor (ThermoFisher 2G11). After extensive washing with PBS, the cells were incubated for 1 h at room temperature in the dark with 1:1000 dilution of secondary anti-IgG-FITC. Slides were then washed with PBS, mounted with Vectashield (Vector Labs Ltd., UK) containing DAPI staining reagent. Images were captured using Leica DM500B microscope and DFC 350FX camera (Leica Microsystem Ltd., UK) using X40 and X63 magnification.
ELISAs or paired antibodies were used to detect cytokines were conducted as per manufacturer’s instructions the kits were as follows; IL-10 (CSB-E12917B) was purchased from Cusabio; IFN-γ (ESS0026B) from ThermoScientific; and IL-12 paired antibodies (CC301 –capture and CC326 –detection) were from AbD Serotec. Nitric oxide was measured using a Griess Reagent Kit (Promega) and arginase levels were determined using the method of [63]. To measure PD-L1 via qPCR RNA was isolated from cells 6hrs after stimulation and converted to cDNA using the GoScript Reverse Transcription Kit (Promega). Primers and conditions used are as previously reported [64].
F. hepatica metacercariare were obtained from the University of Liverpool clonal strain FhepLiv and NEJs excysted as previously described [17]. Parasites were rested for 4hrs prior to use in the ADCC assay which was conducted as previously described by Piedrafita et al [57] and Van Milligen et al [65]. Briefly, macrophages were cultured as described above and mixed with rested NEJs in wells of a 48 well plate containing 40 NEJs and 2 x 105 macrophages per well. Serum was collected from three cattle prior to infection and 13 weeks post infection, with 250 metacercariae of F. hepatica (Kind gift from Dr Divya Sachdev University of Nottingham). Sera was added to wells at a final concentration of 15% in a total volume of 250μl/well. Plates were incubated for 48hrs and NEJs were observed for viability by monitoring motility, absence of defined intestinal structures and exclusion of trypan blue. Parasites were only classified as non-viable if motility and intestinal structure were absent/not visible and trypan blue was taken up in the tegument. NO was measured in the same supernatant using the methods above. In some experiments, macrophages were pre-incubated with the inhibitor SB-431542 prior to FhTLM or TGF-β stimulation as described above.
Data was entered into Prism 6.01 (Graphpad) for statistical analysis. Data was analysed using a 1-way Anova with post-test comparison using a Tukey correction. Apart from data in Fig 6 which was analysed using 2-way anova to determine the effect of serum type and macrophage stimulation. P values <0.05 were taken as significant and individual P values are listed in figure legends.
|
10.1371/journal.ppat.1003482 | Trichomonas vaginalis Exosomes Deliver Cargo to Host Cells and Mediate Host∶Parasite Interactions | Trichomonas vaginalis is a common sexually transmitted parasite that colonizes the human urogential tract where it remains extracellular and adheres to epithelial cells. Infections range from asymptomatic to highly inflammatory, depending on the host and the parasite strain. Here, we use a combination of methodologies including cell fractionation, immunofluorescence and electron microscopy, RNA, proteomic and cytokine analyses and cell adherence assays to examine pathogenic properties of T. vaginalis. We have found that T.vaginalis produces and secretes microvesicles with physical and biochemical properties similar to mammalian exosomes. The parasite-derived exosomes are characterized by the presence of RNA and core, conserved exosomal proteins as well as parasite-specific proteins. We demonstrate that T. vaginalis exosomes fuse with and deliver their contents to host cells and modulate host cell immune responses. Moreover, exosomes from highly adherent parasite strains increase the adherence of poorly adherent parasites to vaginal and prostate epithelial cells. In contrast, exosomes from poorly adherent strains had no measurable effect on parasite adherence. Exosomes from parasite strains that preferentially bind prostate cells increased binding of parasites to these cells relative to vaginal cells. In addition to establishing that parasite exosomes act to modulate host∶parasite interactions, these studies are the first to reveal a potential role for exosomes in promoting parasite∶parasite communication and host cell colonization.
| Trichomoniasis, the most common non-viral sexually transmitted disease worldwide, infects over 275 million people annually. Infection results from the colonization of the human urogenital tract by the parasite Trichomonas vaginalis. To establish and maintain infection the parasite adheres to host cells, a process that is poorly understood. Here, we show that T. vaginalis secretes small vesicles called exosomes that are capable of fusing with and delivering their contents to host cells. Parasite exosomes were found to induce changes in the host cell and to mediate the interaction of T. vaginalis with host by increasing the adherence of the parasite to host cells. Exosomes have been primarily studied in mammalian cells where they have been shown to mediate intercellular communication and have been implicated in processes including development, antigen presentation and cancer metastasis. Our data extend the function of exosomes to mediating host∶parasite interactions, cellular communication between two species and promoting colonization of an extracellular parasite. Research on T. vaginalis exosomes holds the potential for developing applications that would allow exosomes to be used in detecting and diagnosing trichomoniasis and for targeting drugs to the site of infection.
| The parasite Trichomonas vaginalis causes the most common non-viral sexually transmitted infection, with an estimated 275 million people infected each year worldwide [1]. Disease manifestations may include vaginitis, cervicitis, urethritis, pelvic inflammatory disease, and adverse birth outcomes [2]. T.vaginalis infection may also increase risk of HIV transmission and the incidence and severity of cervical and prostate cancer [3]–[5]. Despite the need to define key pathogenic properties of the parasite in order to prevent and control the infection, little is known about parasite or host factors involved in pathogenesis [6], [7]
As an extracellular parasite residing in the urogenital tract, T. vaginalis must adhere to epithelial cells as an initial step towards colonizing the host and establishing infection. Although several families of membrane proteins and secreted proteases have been proposed to play roles in host cell attachment [7]–[10] only three T.vaginalis surface molecules have been shown to be involved in attachment of the parasite to host epithelial cells. The best studied is an abundant lipoglycan (TvLG) [11]–[13] that binds to galectin-1, the only host cell receptor described for T.vaginalis [14]. Two related surface proteins of unknown function are known to increase the attachment of T.vaginalis to host cells when expressed in the parasite [15].
A recent analysis of the surface membrane proteome of T.vaginalis revealed that at least three tetraspanin (Tsp) proteins of the nine found in the genome are present on the parasite's surface [15], [16]. Tsps are involved in a wide variety of activities in mammalian cells including attachment, fusion, motility, migration, and proliferation [17]. Of the 33 human tetraspanins, a small subset including CD63, CD9, CD81, CD82 are constitutive components of exosomes and various other tetraspanins may be present depending on cell of origin and cellular environment [18]. Tsps are also present in all examined mammalian exosomes and, as such, are routinely used as markers for these small secreted extracellular vesicles [19], [20].
Exosomes are 30–100 nm membrane-bound vesicles derived from endocytic compartments that are secreted into the extracellular milieu. Studies of mammalian cells have established that exosomes package specific cargo used for intercellular communication [21], immune modulation and surveillance and the metastasis of diverse tumor cells [18], [22]–[24]. Roles in antigen presentation, the delivery of surface receptors and the transfer of RNA to recipient cells have also been described for exosomes [25]–[27].
Exosomes have recently been shown to be released by pathogens or mammalian cells infected with pathogens [22], [28]. However, to date only a handful of non-mammalian cell types: the fungi Histoplasma, Cryptococcus, Paracoccidiodes, the nematode C.elegans and the parasite Leishmania [29]–[31] have been shown to produce exosomes.
Here we report that T.vaginalis secretes exosomes with physical characteristics and protein components similar to mammalian exosomes. Our analyses demonstrate that parasite exosomes mediate both host∶parasite and parasite∶parasite interactions and play a role in the attachment of the parasite to host epithelial cells. T. vaginalis exosomes are also shown to fuse with and deliver their contents to host cells thereby modulating host cell immune response. These studies are the first to indicate a role for exosomes in promoting host cell colonization and parasite∶parasite communication.
As previously published in the surface proteome, three tetraspanin (Tsp) membrane proteins were identified [15], [16]. Examination of exogenously expressed, hemagglutinin (HA) tagged Tvag_019180 (Tsp1) revealed it is mainly on the plasma membrane (Fig. 1A) [15]. However, when parasites expressing Tsp1-HA are exposed for an hour or longer to ectocervical cells (Ects) the protein accumulates in large vesicular bodies within the parasite (Fig. 1B). These structures are reminiscent of mammalian multivesicular bodies that give rise to secreted exosomes. As mammalian Tsps are enriched in exosomes [19], [26], these data raised the possibility that T. vaginalis secretes Tsp-containing vesicles. As shown in Fig. 1C, this membrane protein is secreted, consistent with the secretion of exosomes by T. vaginalis.
To test whether T. vaginalis produces exosomes, vesicles were isolated from parasite growth media through a series of ultracentrifugation steps, similar to that described for isolating mammalian exosomes [32]. Examination of the preparation by electron microscopy (EM) revealed cup-shaped vesicles of ∼50–100 nm (Fig. 2A), similar in size and shape to mammalian exosomes [33]. To determine if the vesicles had the density reported for mammalian exosomes, vesicles containing a hemagglutinin (HA) tagged Tsp1 (Tsp1-HA) for tracking purposes were floated on a linear sucrose density gradient. The Tsp1-HA-tagged vesicles were found to have densities of 1.03–1.25 g/cm3 (Fig. 2B), similar to that reported for mammalian exosomes (∼1.1–1.2 g/cm3, [34]). The higher MW band in Fig. 2B is most likely a homodimer of Tsp1 as tetraspanin proteins are known to form dimers [35]. To determine vesicle size variation, we used nanoparticle tracking analysis (Nanosight, Costa Mesa, CA) to directly examine millions of vesicles. This analysis showed that the size of vesicles peak with a mean diameter of 95 nm and 83.3% are between 50–150 nm in size (Fig. 2C and 2D).
We then determined whether these vesicles contain RNA using an Agilent 2000 Bioanalyzer, as mammalian exosomes have been reported to deliver miRNAs and mRNAs to recipient cells [26]. A heterogenous population of small RNAs ranging in size from between 25 and 200 nt were found (Fig. 3A and3B). Taken together, the size, morphology, density, and presence of Tsp1 and RNA indicates that T. vaginalis produces and excretes exosomes.
SDS-PAGE followed by silver staining of proteins of exosomes and whole cell lysates normalized by protein concentration indicates an enrichment of specific proteins in exosomes (Fig. 4A). To define the proteins packaged in adherent T. vaginalis B7RC2 strain exosomes and compare their contents with exosomes from other eukaryotes, we determined their protein composition using MudPIT-based proteomic mass spectrometry. Proteins with two or more identified peptides that were found in at least three of seven MudPIT analyses were included in the exosome proteome and revealed a total of 215 proteins (Table S1). These inclusion parameters are conservative compared with other exosomal proteome analyses wherein proteins identified in two of ten experiments or proteins with one peptide in a single experiment were included [36], [37].
When compared with the compiled list of common mammalian exosome proteins in ExoCarta [38], we found that T. vaginalis exosomes contained orthologs of approximately 73% of mammalian exosome proteomes and 39.5% are orthologous to Leishmania exosomal proteins [39]. The extensive overlap with mammalian exosomal proteins was surprising as roughly 2/3 of T. vaginalis genes have no mammalian orthologs [40]. Shared proteins represent 60 core conserved exosome protein/protein families such as tetraspanins, Alix, Rabs, Hsp70, subunits of heterotrimeric G proteins and TcTP [25] as well as hypothetical proteins identified by BLAST analyses as similar to mammalian exosomal proteins. Identified proteins were sorted into functional groups by BLAST analyses and genome annotation and assigned a predicted function (Fig. 4B). Fourteen percent are signaling proteins, 14% are metabolic enzymes, 13% are cytoskeletal proteins, 8% are involved in transport and 6% are vacuolar proteins. The remaining proteins include 32 hypothetical proteins (15% of total) of unknown function as well as proteins involved in protein folding and other cellular activities (Fig. 4B and Table S1). It is notable that one protein in the exosome proteome Tvag_452120 (TvG402) was previously found to localize to in large vesicular structures within T.vaginalis [41]. Additionally, 24.6% were previously found in the T.vaginalis surface proteome [15]). Nine percent have predicted transmembrane domains and 3.7% have predicted signal peptides as annotated in the TrichDB database (Figure 4C). It should be noted, however, that T. vaginalis membrane proteins often appear to lack conventional, identifiable N-terminal signal peptides or transmembrane domains [15], making it difficult to predict whether many exosomal proteins are membrane associated or soluble. Nevertheless, the high degree of conservation between mammalian and T. vaginalis exosome proteomes firmly establishes that T. vaginalis secretes exosomes.
Novel proteins that may have a role in T. vaginalis pathogenesis are also present in the exosome proteome. Interestingly we identified surface proteins (Tvag_240680, BspA family) and proteases (Tvag_224980, metallopeptidase) thought to be involved in pathogenesis [6], [40], [42], [43]. For example, Tvag_340570 is related to two surface proteins involved in parasite attachment which are significantly more abundant in highly adherent versus poorly adherent parasites [15]. These adhesion related surface proteins are part of a ∼150 member family, 30% of which have EST evidence for expression (www.trichdb.org) and 18% of which were found in the surface proteome [15]. Orthologs of virulence proteins characterized in other parasites (Tvag_371800, GP63-like) and those potentially involved in host immune regulation (Tvag_137880, peptidyl proyl isomerase A also known as cyclophilin A) are also notable. Many of the identified exosomal proteins are members of very large gene families [40]; however, only 1 or 2 members are packaged in the exosome, indicating a specificity in expression and/or packaging of specific proteins. For example, there are 911 putative BspA family proteins, ∼30% of which have EST, RT-PCR, or microarray data to support expression [43] and while 11 were found in the surface proteome only 1 was identified in the exosome proteome. Additionally there are ∼700 proteases encoded in the genome, of which ∼120 are annotated as metallopeptidases with 60% having EST evidence for expression and only 3 are found in the exosome proteome. Similarly, ∼90 GP63-like proteases are annotated in the genome, 33% have been shown to be expressed, 16 were found in the surface proteome [15], while only 1 was identified in the exosome proteome. Future studies on the differential expression of proteins packaged into exosomes by T.vaginalis strains with varying virulent phenotypes should help identify exosomal proteins involved in pathogenesis..
Increasing evidence indicates exosomes are capable of mediating cell∶cell communication, leading to intercellular transfer of molecules [44], [45]. Having established that T. vaginalis exosomes contain several proteins potentially involved in pathogenesis, we hypothesized that these exosomes could be used by the parasite during infection. The host cells first encountered by T. vaginalis during infection that are the primary site of replication and survival are ectocervical cells (Ects) [46]. We examined whether T. vaginalis exosomes associates with Ects by labeling exosomal membranes with BODIPY-PC [47], followed by incubation with Ects. After extensive washing to remove free exosomes, Ects were examined by microscopy and the fluorescent BODIPY-PC was found to label the Ects (Fig. 5A). This is in contrast with that observed using BODIPY-PC labeled hydrogenosomes [48] (Fig. 5A) which are not capable of transferring BODIPY-PC to Ects. These data indicate that T. vaginalis exosomes deliver their contents to host cells.
To directly examine whether soluble protein in the parasite exosomes are delivered to host cells we utilized a split-GFP system [49]. Ects were transiently transfected with the large S1-10 fragment of GFP. Exosomes were purified from T.vaginalis expressing a soluble exosome protein (Tvag_180840, TcTP or Tvag_137880, peptidyl proyl isomerase A) tagged with the small S11 fragment of GFP. These exosomes were then incubated with GFPS1-10 transfected Ects. GFP fluorescence will only be observed if the S11 fragment fused in frame with an exosomal protein is delivered to the cytoplasm of an Ect containing the S1-10 fragment. As shown in Fig. 5B, fluorescence was observed specifically in exosome-treated and not in vehicle treated Ects. After normalization by transfection efficiency which was ∼30%, quantification of the data estimates that exosomes fused with ∼50% of the Ects (Fig. 5C). These results provide strong evidence that T. vaginalis exosomes fuse with and deliver their contents to host cells.
We hypothesized that T. vaginalis exosomes may modulate the Ect immune response, as these cells are involved in host innate immunity [12], [50]–[53]. Specifically, we examined the proinflammatory cytokines IL6 and IL8 secreted by Ects in response to both exosomes and parasites. To allow comparison, we used the concentration of exosomes predicted to be produced by the number of parasites used in the same experiment as calculated from exosome isolation yields. Quantification of IL6 secretion by Ects demonstrated that T. vaginalis exosomes induce this cytokine to approximately the same extent as parasites (Fig. 6A). Exosomes were also shown to elicit IL8 secretion; however, the response is only ∼50% of that observed when Ects are exposed to parasites (Fig. 6B). It is notable that using equivalent amount (9 ug) of either T. vaginalis hydrogenosomes, cytosol or exosomal supernatant did not induce a cytokine response (Fig. 6E). Because IL6 is an acute inflammation protein and IL8 is involved in a long-term inflammatory process [54] we further hypothesized that the exosomes secreted by T. vaginalis could potentially prime host cells for parasite infection. We found pretreating Ects with exosomes did not affect their subsequent IL6 cytokine production (Fig. 6C). However, as shown in Fig. 6D, preincubation of Ects with exosomes prior to the addition of T.vaginalis parasites led to a significant inhibition of IL8 secretion by the Ects. These results indicate that T. vaginalis exosomes specifically modulate IL8, but not IL6 production by Ects.
Secreted exosomes can potentially interact with other parasites in the population such that exosomes from one parasite might influence another. As attachment to host cells is critical for establishing infection [6], [7], we examined whether exosomes purified from B7RC2, a strain 20-fold more adherent than the lab strain G3 [15], affects the attachment of the less adherent strain G3 to Ects. The ability of exosomes to mediate both parasite∶parasite and parasite∶host cell adherence was assessed. The following were preincubated with B7RC2 exosomes for 1 hour prior to performing attachment assays: 1) G3 parasites only 2) Ects only or 3) both Ects and G3 parasites. Unincorporated exosomes were washed away before performing the attachment assay and as a negative control BSA was used instead of exosomes. Based on average yields of exosomes per parasite number, the amount of exosomes oredicted to be secreted by the quantity of parasites used in our assays was used for all preincubation treatments. Preincubation of B7RC2 exosomes with G3 parasites resulted in a 2-fold increase in G3 attachment to Ects showing that exosomes from one parasite can affect attachment of another parasite (Fig. 7A). Preincubation of B7RC2 exosomes with Ects resulted in a 3-fold increase in G3 attachment to Ects showing that exosomes alter the host cell and result in increased parasite attachment. Interestingly, this effect is additive as a fivefold increase in G3 attachment was observed when both the G3 parasites and Ects are preincubated with B7RC2 exosomes (Fig. 7A). As losses result during isolation of exosomes, we believe the amount of exosomes utilized in our experiments is conservative, however a dose curve (Fig. S2) shows that increasing amounts of B7RC2 exosomes does increase G3 adherence. Contrary to that observed using B7RC2 exosomes, preincubation of parasites, Ects, or both, with exosomes from the poorly-adherent G3 strain did not increase G3 attachment to Ects (Fig. 7B). Furthermore preincubation of B7RC2 exosomes on B7RC2 parasites only had a slight effect on attachment to Ects (Fig. 7C). This is likely due to a saturation effect as the B7RC2 parasites produce and secrete their own exosomes. Doing the identical experiment except replacing exosomes with equivalent amounts of T. vaginalis hydrogenosomes or cytosol does not result in increased parasite attachment (Fig. S1A and Fig. S1B, respectively). These data indicate that exosomes from a highly adherent strain can increase parasite attachment of a less adherent strain to host cells. Moreover, they indicate that exosomes can mediate both intraspecies and interspecies interactions.
To test whether the parasite∶parasite and parasite∶host interactions observed using the B7RC2 strain is generally observed with other parasite strains, we expanded our G3 attachment assay to include exosomes purified from several strains. We found that exosomes from poorly adherent strains like T1 or RU384 [55] do not significantly increase adherence of G3 parasites while those from more highly adherent strains like MSA1103 or LSU160 [55] do increase attachment of G3 to Ects (Fig. 8). As MSA1103 and LSU160 strains are more adherent (3 and 2 fold respectively) to the benign prostate epithelium cell line (BPH1) than the female Ect host cells [55], we tested whether exosomes from these two strains result in a greater increase in G3 parasite adherence to BPH1 cells as compared to Ects. We found that while exosomes from LSU160 and MSA1103 increased attachment to Ects ∼2 fold, they increased attachment to BPH1 cells by ∼6 and ∼4 fold respectively (Fig. 9). In contrast a difference was not observed when preincubating with B7RC2 exosomes consistent with B7BC2 parasites displaying no preference for attachment to Ects versus BPH1 [55]. Together, these results indicate that exosomes package pathogenic factors specific to the strain that produces them and that exosomes from highly adherent strains may contribute to the parasites' ability to colonize specific niches of the male and female urogenital tract.
This study, the first to identify and characterize exosomes from the parasite T. vaginalis, reveals a role for these secreted vesicles in host∶parasite interactions. We have isolated T. vaginalis exosomes from the adherent B7RC2 strain and shown that they are remarkably similar to mammalian exosomes in size, structure and core protein components [22], [25], [56]. The extensive (73%) overlap between the T. vaginalis and mammalian exosomal proteomes demonstrate the specific packaging of proteins within exosomes as the majority of T. vaginalis proteins have no orthologs in humans [40]. Minimal overlap between the T. vaginalis exosomal proteome and other reported T. vaginalis proteomes [7], [15], [57] and the presence of only 1–2 proteins from large, expressed protein families [6], [40] likewise argues for specificity in exosomal protein content. Exosome biogenesis and the selective packaging of exosomal proteins is poorly understood [20], [58], [59]; however, the overlap of the exosome proteomes of this highly divergent parasite and that of humans indicates the conservation of underlying mechanisms. Thus the ease of culturing T. vaginalis and its high exosomes yield makes this parasite a good model for studying the basic biological properties of exosomes.
In addition to the core proteins conserved between T. vaginalis and mammalian exosomes, many proteins unique to T. vaginalis exosomes were also identified. Thirty-two are conserved hypothetical proteins of unknown function. Others, including several surface proteins and proteases, have been implicated in pathogenesis. Future experiments characterizing the exosomal contents from strains of parasites of varying virulence will assist in identifying and characterizing specific exosomeal proteins that affect pathogenesis. Unique T. vaginalis exosomal proteins may be critical for mediating host∶parasite interactions In this regard, another parallel can be drawn between T. vaginalis and mammalian exosomes as the latter are involved in interactions that lead to pathologies such as cancer proliferation and HIV transmission between different cell types [28], [60]–[63].
Using a split-GFP assay, parasite exosomes were shown to fuse with and deliver their contents to host cells. T. vaginalis exosomes were also shown to modulate host cell cytokine production. T. vaginalis may use exosomes to manipulate host defense responses similar to the secretion of virulence factors and vesicles by bacteria [22], [64], [65]. Exosomes were found to induce an IL6 response in Ects and to down regulate the IL8 response to parasites. IL8 is involved in the recruitment of neutrophils to the site of infection and persists in its active form within the immediate environment for longer than other chemoattractants [12], [66]. Thus by dampening the IL8 response of Ects to parasites, T. vaginalis exosomes may play a critical role in establishing a successful chronic infection. IL6 is a chief stimulator of proteins in acute inflammation and suppresses the level of other proinflammatory cytokines in an acute response [67]. IL6 can also stimulate IL-1 receptor antagonist, an anti-inflammatory mediator, to control tissue inflammatory responses [68]. Thus T. vaginalis exosomes may lead to the regulation of IL6 and IL8 secretion, thus priming the urogenital tract for parasite colonization. Both proinflammatory and immunosuppressive responses to T. vaginalis infection or its double-stranded RNA virus have been described in various in vitro and mouse studies [69], [70]. Future studies aimed at investigating the effects of exosomes on various host effector cells recruited during infection will be necessary to understand the molecular mechanisms by which exosomes modulate host immune responses.
T. vaginalis exosomes were found to substantially increase the adherence of this sexually-transmitted parasite to female (Ect) and male (BPH1) epithelial cells via an effect on both parasites and host cells. As an extracellular parasite, attachment of T. vaginalis to host cells is vital for survival and pathogenesis. Similar to that described for exosomes produced by different mammalian tissues and cells [23], [56], [71], T. vaginalis exosomes have strain-specific characteristics. Exosomes from highly adherent parasites are capable of increasing the adherence of poorly adhering parasite to host cells, whereas those from poorly adhering parasites are not. Furthermore, exosomes from parasite strains that preferentially bind BPH1 cells are more effective in increasing adherence to BPH1 versus Ects. Taken together, our data indicate T. vaginalis exosomes may package strain specific, perhaps even host cell specific, virulence factors. Future studies on differential packaging of factors in exosomes between strains may lend insight into differences attributed to virulent and less virulent phenotypes.
T. vaginalis exosomes could potentially be found in vaginal secretions or urine of infected individuals and thus serve as biomarkers for infection. Testing for infection in females currently requires gynecology visits [2], [72] and there is a lack of a non-invasive, quick method for diagnosing T. vaginalis infection in men [73]. Exosomes have been successfully isolated from a variety of easily obtained bodily fluids including blood, amniotic fluid, saliva, and urine [74]. It has been shown that urine contains abundant amounts of exosomes from prostatic secretions [75] and hence could contain exosomes from T.vaginalis in infected patients. Thus the isolation of T. vaginalis exosomes or detection of exosomal contents may provide a non-invasive means to diagnose infection. Furthermore, the ability of T. vaginalis exosomes to deliver soluble substances to host cells provides the potential for parasite exosomes to be used therapeutically.
Mammalian exosomes and their RNA and protein contents have been shown to regulate a variety of cellular pathways by modulating gene expression in recipient cells [76], [77]. The characterization of T. vaginalis exosomes and their protein cargo sets the stage for determining specific parasite factors likely to modulate host cell response and affect infection outcomes. The small RNAs packaged inside T. vaginalis exosomes may also modulate parasite∶parasite or parasite∶host interactions. The possibility that these small RNAs are novel parasite miRNAs that are delivered to the host cell to modulate gene activity is an appealing idea.
The use of exosomes by a strictly extracellular parasite represents a novel method by which the parasite may deliver proteins and/or RNA to the host to manipulate host cell responses while remaining extracellular. The delivery of strain specific exosomal contents can impact both host immune response as well as parasite attachment to host cells (Fig. 10). A better understanding of how exosomes increase parasite adherence to host cells and modulate host cell responses will provide insights into pathogenesis and possibly new avenues for diagnosis and therapy.
T. vaginalis strains (B7RC2, G3, T1, RU384, MSA1103, LSU160) were cultured in TYM medium supplemented with 10% horse serum, 10 U/ml penicillin/10 ug/ml streptomycin (Invitrogen), 180 uM ferrous ammonium sulphate and 28 uM sulfosalicylic acid [78]. 100 ug/mL G418 (Invitrogen) was added to culture of the Tsp1-HA (Tvag_019180), Tvag_180840, and Tvag_137880 transfectants. Parasites were grown at 37°C and passaged daily for ≤2 weeks. The human cervical epithelial cell line Ect1 E6/E7 (ATCC CRL-2614) was grown as described [46] except without additional CaCl2. The human benign prostate epithelial line BPH1 [79] was grown as described [80].
T.vaginalis at a density of ∼1.0×106 parasites/ml, were washed 3X and resuspended in TYM media without serum for 4 hrs after which parasites were removed by centrifugation at 500×g. The cell-free media was filtered through 0.22 um filter and concentrated using a Vivaflow 200 100,000 MWCO PES (Sartorious Stedium). Exosomes were pelleted by ultracentrifugation at 100,000×g for 75 min using a TLA100 rotor followed by resuspension in 2 mL cold PBS+1X HALT protease inhibitors (Thermo Scientific). Exosomes were concentrated further by ultracentrifugation at 100,000×g for 70 min and resuspended in 100–300 uL PBS to be purified by floatation on a linear sucrose gradient as described (Raposo, 1996). Pellets were resuspended by boiling for 5 min in Laemmli sample buffer and stored at −20°C in PBS or immediately resolved by SDS-PAGE followed by western blot analysis or silver staining using standard procedures. Fractions transferred to PVDF membranes were blocked with 5% TBST-milk and probed with an anti-HA antibody (1∶5000, Covance) followed by reaction with anti-mouse-HRP (Jackson Labs). Exosome protein concentration was assayed using the Pierce BCA Kit (Thermo Scientific).
Freshly isolated exosomes were directly adsorbed onto charged carbon-coated grids, contrasted with 1% uranyl acetate, and examined using a transmission electron microscope.
Total RNA was isolated from freshly isolated exosomes or whole parasites using the MirVana Paris kit (Invitrogen) according to manufacturer's protocol. The eukaryote total RNA Nano assay (RNA 6000 Nano kit, Agilent) was used to analyze the RNA on an Agilent 2100 Bioanalyzer at the UCLA Clinical Microarray Core.
Parasites were incubated with Ects for 15 or 60 min. Cells were fixed in 4% formaldehyde for 20 min, permeabilized with 0.2% Triton X-100 in PBS, blocked with 3% BSA in PBS (PBS-BSA), incubated with a 1∶1,000 dilution of anti-HA primary antibody (Covance), washed, and then incubated with a 1∶5000 dilution of Alexa Fluor-conjugated secondary antibody (Molecular Probes). The coverslips were mounted using ProLong Gold Antifade reagent with DAPI (Invitrogen). Stained parasites were examined using an Axioskop 2 epifluorescence microscope (Zeiss), and images were recorded with an AxioCam camera and processed with the AxioVision 3.2 program (Zeiss).
Parasites at log growth were resuspended in 5% PBS-sucrose at a density of 1×106 parasites/mL at 16°C or 37°C for 1 hour. Parasites were pelleted by centrifugation and the supernatant was filtered through 0.22 um filter to remove cell debris and concentrated using an Amicon filter. The pellet was resuspened to the same volume as the filtered supernatant (700 uL), and 15 uL of each was subjected to SDS-PAGE. Western blot analysis was performed using anti-HA (1∶5000; Covance) to detect Tsp1, anti-β-tubulin (1∶1000; Sigma) and anti-neomycin phosphotransferase II (1∶2500; Jackson Labs). Secondary antibodies were anti-mouse-HRP (1∶25000) and anti-rabbit-HRP (1∶25000).
Exosomes were fractionated on 12% Tris-glycine gels (Invitrogen) followed by fixation with 10% methanol and 7% acetic acid. Gel slices were excised and treated with 100 mM ammonium bicarbonate (Fisher) and 50% acetonitrile (ACN). Disulphide bonds were reduced with 10 mM DTT and SH groups alkylated with 50 mM iodoacetamide (Sigma). After washing, gel pieces were dehydrated with ACN then rehydrated on ice with 20 ng/uL trypsin, 40 mM ammonium bicarbonate, 9% ACN, and incubated overnight at 37°C. Acidic peptides were extracted by the addition of 100 mM ammonium bicarbonate and extraction of basic peptides was performed with 2.5% trifluoroacetic acid (TFA). The supernatants were combined, dried in a speed-vac. The dried samples were resuspended in digestion buffer (100 mM Tris-HCl, pH 8, 8M urea), proteolytically digested by the sequential addition of Lys-C and trypsin proteases and subjected to MudPIT analyses as described [15].
10 uM BODIPY-PC (2-decanoyl-1-(O-(11-(4,4-difluoro-5,7-dimethyl-4-bora-3a,4a-diaza-s-indacene-3-propionyl)amino)undecyl)-sn-glycero-3-phosphocholine; Invitrogen) was used to label exosomal or hydrogenosomal membranes for 30 min at 4°C in the dark. Excess lipids were removed by two 500X volume washes with PBS and ultracentrifugation at 50,000 rpm for 1 hr. BODIPY-PC labeled exosomes or hydrogenosomes were added to Ects for 24 hr after which cells were washed with warm media 3X followed by fixation with 4% formaldehyde and mounting with ProLong Antifade reagent with DAPI (Invitrogen). Cells were visualized as described for immunolocalization.
The pCMV-mGFP-S1-10 mammalian optimized plasmid (Theranostech, Inc) was transiently transfected into Ect cells using Fugene HD (Promega) according to manufacturer's protocol. Tvag_137880 and Tvag_180840 were cloned into Master-Neo plasmid with S11 fused in frame using PCR and parasites were transfected and cultured as described [81]. Exosomes containing S11-tagged proteins were added 48 hr post transfection, allowed to interact for 4 hr and cultures were then washed 3X to remove unfused exosomes. Cells were fixed with 4% formaldehyde and mounted with ProLong Gold antifade reagent with DAPI (Invitrogen) and viewed as described for immunolocalization.
Ects were plated in 48-well plates and grown to confluence. 5×105 parasites or 9 ug exosomes were added for 6 hr and the supernatant was then removed. Cells/debris were removed by spinning at 5000 rpm for 10 min. Supernatants were stored at −20°C. IL6 and IL8 was quantified using IL6 or IL8 ELISA kits from AssayPro following manufacturer's directions. For priming experiments, exosomes, BSA, or PBS was added to host cells for 12 hr after which host cells were rinsed 3X with prewarmed KSFM (Invitrogen) and 5×105 parasites were added for 6 hr. Data were normalized as percent of cytokine secretion using 9 ug BSA without the addition of parasites.
Attachment of parasites to Ect or BPH1 cells was performed as described [15]. Briefly, CellTracker Blue CMAC (Invitrogen) labeled parasites were added to confluent monolayer of host cells (1∶3 parasite∶host cell ratio) for 30 min. Coverslips were subsequently rinsed in PBS to remove unattached parasites, fixed with 4% formaldehyde (Polysciences, Inc), and mounted on slides with Mowiol (Calbiochem).Fifteen 10X magnification fields were analyzed per coverslip with three coverslips per treatment per experiment. Fluorescent parasites adhered to host cells were quantified using ImageJ.. When examining the role of exosomes in attachment, exosomes were preincubated with Ects, parasites, or both for 1 hr prior to attachment followed by washing with warm media to remove remaining exosomes. All experiments were performed 3–5 times with 3 coverslips per treatment per experiment. Fold change in parasite number was calculated by totaling the number of parasites for 15 images/coverslip, and averaging all coverslips per treatment condition and dividing by the same number derived using negative control BSA samples.
Graphs were made and statistical analyses performed using Microsoft Excel 2010. Independent experiments were performed a minimum of 3 times with at least three technical replicates per experiment. Two sample t-tests were used to determine significance. Data are expressed as standard error of the mean (± SEM).
|
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.